+
+### [Version 1.56.5](https://github.com/lobehub/lobe-chat/compare/v1.56.4...v1.56.5)
+
+Released on **2025-02-16**
+
+#### 🐛 Bug Fixes
+
+- **misc**: Match o1 series models more robust in Azure OpenAI provider, set max_completion_tokens to null for Azure OpenAI.
+
+
+
+
+Improvements and Fixes
+
+#### What's fixed
+
+- **misc**: Match o1 series models more robust in Azure OpenAI provider, closes [#6193](https://github.com/lobehub/lobe-chat/issues/6193) ([f444e66](https://github.com/lobehub/lobe-chat/commit/f444e66))
+- **misc**: Set max_completion_tokens to null for Azure OpenAI, closes [#6198](https://github.com/lobehub/lobe-chat/issues/6198) ([e9e8da4](https://github.com/lobehub/lobe-chat/commit/e9e8da4))
+
+
+
+
+
+### [Version 1.53.8](https://github.com/lobehub/lobe-chat/compare/v1.53.7...v1.53.8)
+
+Released on **2025-02-13**
+
+#### 🐛 Bug Fixes
+
+- **misc**: Fix model fetch for spark and fix the support of model reset.
+
+
+
+
+Improvements and Fixes
+
+#### What's fixed
+
+- **misc**: Fix model fetch for spark and fix the support of model reset, closes [#6080](https://github.com/lobehub/lobe-chat/issues/6080) ([257fda1](https://github.com/lobehub/lobe-chat/commit/257fda1))
+
+
+
+
+
+### [Version 1.50.1](https://github.com/lobehub/lobe-chat/compare/v1.50.0...v1.50.1)
+
+Released on **2025-02-03**
+
+#### 🐛 Bug Fixes
+
+- **misc**: Bind the selected group name in the rename modal..
+
+
+
+
+Improvements and Fixes
+
+#### What's fixed
+
+- **misc**: Bind the selected group name in the rename modal., closes [#5159](https://github.com/lobehub/lobe-chat/issues/5159) ([7da05ce](https://github.com/lobehub/lobe-chat/commit/7da05ce))
+
+
+
+
+
+### [Version 1.49.8](https://github.com/lobehub/lobe-chat/compare/v1.49.7...v1.49.8)
+
+Released on **2025-02-01**
+
+#### 💄 Styles
+
+- **misc**: Support thinking for all non DeepSeek official api R1 models.
+
+
+
+
+Improvements and Fixes
+
+#### Styles
+
+- **misc**: Support thinking for all non DeepSeek official api R1 models, closes [#5654](https://github.com/lobehub/lobe-chat/issues/5654) ([9b32137](https://github.com/lobehub/lobe-chat/commit/9b32137))
+
+
+
+
+
+### [Version 1.47.14](https://github.com/lobehub/lobe-chat/compare/v1.47.13...v1.47.14)
+
+Released on **2025-01-22**
+
+#### 💄 Styles
+
+- **misc**: Support model list with model fetcher settings.
+
+
+
+
+Improvements and Fixes
+
+#### Styles
+
+- **misc**: Support model list with model fetcher settings, closes [#5540](https://github.com/lobehub/lobe-chat/issues/5540) ([637fbf4](https://github.com/lobehub/lobe-chat/commit/637fbf4))
+
+
+
+
+
+## [Version 1.47.0](https://github.com/lobehub/lobe-chat/compare/v1.46.7...v1.47.0)
+
+Released on **2025-01-17**
+
+#### ✨ Features
+
+- **misc**: Support new ai provider in client pglite.
+
+
+
+
+Improvements and Fixes
+
+#### What's improved
+
+- **misc**: Support new ai provider in client pglite, closes [#5488](https://github.com/lobehub/lobe-chat/issues/5488) ([08f505f](https://github.com/lobehub/lobe-chat/commit/08f505f))
+
+
+
+
+
+## [Version 1.46.0](https://github.com/lobehub/lobe-chat/compare/v1.45.17...v1.46.0)
+
+Released on **2025-01-15**
+
+#### ✨ Features
+
+- **misc**: Add lm studio provider, support to customize Embedding model with env.
+
+
+
+
+Improvements and Fixes
+
+#### What's improved
+
+- **misc**: Add lm studio provider, closes [#4483](https://github.com/lobehub/lobe-chat/issues/4483) ([0ac5802](https://github.com/lobehub/lobe-chat/commit/0ac5802))
+- **misc**: Support to customize Embedding model with env, closes [#5177](https://github.com/lobehub/lobe-chat/issues/5177) ([eed9357](https://github.com/lobehub/lobe-chat/commit/eed9357))
+
+
+
+
We provide a [Docker image][docker-release-link] for deploying the LobeChat service on your private device.
+ ### Install Docker Container Environment
-### Install Docker Container Environment
-
-(Skip this step if already installed)
-
-
-
-
-
-```fish
-$ apt install docker.io
-```
-
-
-
-
-
-```fish
-$ yum install docker
-```
-
-
+ (Skip this step if already installed)
-
+
+
+ ```fish
+ $ apt install docker.io
+ ```
+
-### Run Docker Compose Deployment Command
+
+ ```fish
+ $ yum install docker
+ ```
+
+
-When using `docker-compose`, the configuration file is as follows:
+ ### Run Docker Compose Deployment Command
-```yml
-version: '3.8'
+ When using `docker-compose`, the configuration file is as follows:
-services:
- lobe-chat:
- image: lobehub/lobe-chat
- container_name: lobe-chat
- restart: always
- ports:
- - '3210:3210'
- environment:
- OPENAI_API_KEY: sk-xxxx
- OPENAI_PROXY_URL: https://api-proxy.com/v1
- ACCESS_CODE: lobe66
-```
+ ```yml
+ version: '3.8'
-Run the following command to start the Lobe Chat service:
+ services:
+ lobe-chat:
+ image: lobehub/lobe-chat
+ container_name: lobe-chat
+ restart: always
+ ports:
+ - '3210:3210'
+ environment:
+ OPENAI_API_KEY: sk-xxxx
+ OPENAI_PROXY_URL: https://api-proxy.com/v1
+ ACCESS_CODE: lobe66
+ ```
-```bash
-$ docker-compose up -d
-```
+ Run the following command to start the Lobe Chat service:
-### Crontab Automatic Update Script (Optional)
+ ```bash
+ $ docker-compose up -d
+ ```
-Similarly, you can use the following script to automatically update Lobe Chat. When using `Docker Compose`, no additional configuration of environment variables is required.
+ ### Crontab Automatic Update Script (Optional)
-```bash
-#!/bin/bash
-# auto-update-lobe-chat.sh
+ Similarly, you can use the following script to automatically update Lobe Chat. When using `Docker Compose`, no additional configuration of environment variables is required.
-# Set proxy (optional)
-export https_proxy=http://127.0.0.1:7890 http_proxy=http://127.0.0.1:7890 all_proxy=socks5://127.0.0.1:7890
+ ```bash
+ #!/bin/bash
+ # auto-update-lobe-chat.sh
-# Pull the latest image and store the output in a variable
-output=$(docker pull lobehub/lobe-chat:latest 2>&1)
+ # Set proxy (optional)
+ export https_proxy=http://127.0.0.1:7890 http_proxy=http://127.0.0.1:7890 all_proxy=socks5://127.0.0.1:7890
-# Check if the pull command was executed successfully
-if [ $? -ne 0 ]; then
- exit 1
-fi
+ # Pull the latest image and store the output in a variable
+ output=$(docker pull lobehub/lobe-chat:latest 2>&1)
-# Check if the output contains a specific string
-echo "$output" | grep -q "Image is up to date for lobehub/lobe-chat:latest"
+ # Check if the pull command was executed successfully
+ if [ $? -ne 0 ]; then
+ exit 1
+ fi
-# If the image is already up to date, do nothing
-if [ $? -eq 0 ]; then
- exit 0
-fi
+ # Check if the output contains a specific string
+ echo "$output" | grep -q "Image is up to date for lobehub/lobe-chat:latest"
-echo "Detected Lobe-Chat update"
+ # If the image is already up to date, do nothing
+ if [ $? -eq 0 ]; then
+ exit 0
+ fi
-# Remove the old container
-echo "Removed: $(docker rm -f Lobe-Chat)"
+ echo "Detected Lobe-Chat update"
-# You may need to navigate to the directory where `docker-compose.yml` is located first
-# cd /path/to/docker-compose-folder
+ # Remove the old container
+ echo "Removed: $(docker rm -f Lobe-Chat)"
-# Run the new container
-echo "Started: $(docker-compose up)"
+ # You may need to navigate to the directory where `docker-compose.yml` is located first
+ # cd /path/to/docker-compose-folder
-# Print the update time and version
-echo "Update time: $(date)"
-echo "Version: $(docker inspect lobehub/lobe-chat:latest | grep 'org.opencontainers.image.version' | awk -F'"' '{print $4}')"
+ # Run the new container
+ echo "Started: $(docker-compose up)"
-# Clean up unused images
-docker images | grep 'lobehub/lobe-chat' | grep -v 'lobehub/lobe-chat-database' | grep -v 'latest' | awk '{print $3}' | xargs -r docker rmi > /dev/null 2>&1
-echo "Removed old images."
-```
+ # Print the update time and version
+ echo "Update time: $(date)"
+ echo "Version: $(docker inspect lobehub/lobe-chat:latest | grep 'org.opencontainers.image.version' | awk -F'"' '{print $4}')"
-This script can also be used in Crontab, but ensure that your Crontab can find the correct Docker command. It is recommended to use absolute paths.
+ # Clean up unused images
+ docker images | grep 'lobehub/lobe-chat' | grep -v 'lobehub/lobe-chat-database' | grep -v 'latest' | awk '{print $3}' | xargs -r docker rmi > /dev/null 2>&1
+ echo "Removed old images."
+ ```
-Configure Crontab to execute the script every 5 minutes:
+ This script can also be used in Crontab, but ensure that your Crontab can find the correct Docker command. It is recommended to use absolute paths.
-```bash
-*/5 * * * * /path/to/auto-update-lobe-chat.sh >> /path/to/auto-update-lobe-chat.log 2>&1
-```
+ Configure Crontab to execute the script every 5 minutes:
+ ```bash
+ */5 * * * * /path/to/auto-update-lobe-chat.sh >> /path/to/auto-update-lobe-chat.log 2>&1
+ ```
[docker-pulls-link]: https://hub.docker.com/r/lobehub/lobe-chat
diff --git a/docs/self-hosting/platform/docker-compose.zh-CN.mdx b/docs/self-hosting/platform/docker-compose.zh-CN.mdx
index ac281712c24d8..3694ddd79718d 100644
--- a/docs/self-hosting/platform/docker-compose.zh-CN.mdx
+++ b/docs/self-hosting/platform/docker-compose.zh-CN.mdx
@@ -12,122 +12,112 @@ tags:
# Docker Compose 部署指引
We provide a [Docker image][docker-release-link] for you to deploy the LobeChat service on your private device.
@@ -27,138 +26,134 @@ We provide a [Docker image][docker-release-link] for you to deploy the LobeChat
### Install Docker Container Environment
-(If already installed, skip this step)
+ (If already installed, skip this step)
```fish
$ apt install docker.io
```
-
-
+
```fish
$ yum install docker
```
+
+
-
-
-
-
-### Docker Command Deployment
-
-Use the following command to start the LobeChat service with one click:
+ ### Docker Command Deployment
-```fish
-$ docker run -d -p 3210:3210 \
- -e OPENAI_API_KEY=sk-xxxx \
- -e ACCESS_CODE=lobe66 \
- --name lobe-chat \
- lobehub/lobe-chat
-```
+ Use the following command to start the LobeChat service with one click:
-Command explanation:
+ ```fish
+ $ docker run -d -p 3210:3210 \
+ -e OPENAI_API_KEY=sk-xxxx \
+ -e ACCESS_CODE=lobe66 \
+ --name lobe-chat \
+ lobehub/lobe-chat
+ ```
-- The default port mapping is `3210`, please ensure it is not occupied or manually change the port mapping.
+ Command explanation:
-- Replace `sk-xxxx` in the above command with your OpenAI API Key.
+ - The default port mapping is `3210`, please ensure it is not occupied or manually change the port mapping.
-- For the complete list of environment variables supported by LobeChat, please refer to the [Environment Variables](/docs/self-hosting/environment-variables) section.
+ - Replace `sk-xxxx` in the above command with your OpenAI API Key.
-
- Since the official Docker image build takes about half an hour, if you see the "update available"
- prompt after deployment, you can wait for the image to finish building before deploying again.
-
+ - For the complete list of environment variables supported by LobeChat, please refer to the [Environment Variables](/docs/self-hosting/environment-variables) section.
-
- The official Docker image does not have a password set. It is strongly recommended to add a
- password to enhance security, otherwise you may encounter situations like [My API Key was
- stolen!!!](https://github.com/lobehub/lobe-chat/issues/1123).
-
+
+ Since the official Docker image build takes about half an hour, if you see the "update available"
+ prompt after deployment, you can wait for the image to finish building before deploying again.
+
-
- Note that when the **deployment architecture is inconsistent with the image**, you need to
- cross-compile **Sharp**, see [Sharp
- Cross-Compilation](https://sharp.pixelplumbing.com/install#cross-platform) for details.
-
+
+ The official Docker image does not have a password set. It is strongly recommended to add a
+ password to enhance security, otherwise you may encounter situations like [My API Key was
+ stolen!!!](https://github.com/lobehub/lobe-chat/issues/1123).
+
-#### Using a Proxy Address
+
+ Note that when the **deployment architecture is inconsistent with the image**, you need to
+ cross-compile **Sharp**, see [Sharp
+ Cross-Compilation](https://sharp.pixelplumbing.com/install#cross-platform) for details.
+
-If you need to use the OpenAI service through a proxy, you can configure the proxy address using the `OPENAI_PROXY_URL` environment variable:
+ #### Using a Proxy Address
-```fish
-$ docker run -d -p 3210:3210 \
- -e OPENAI_API_KEY=sk-xxxx \
- -e OPENAI_PROXY_URL=https://api-proxy.com/v1 \
- -e ACCESS_CODE=lobe66 \
- --name lobe-chat \
- lobehub/lobe-chat
-```
+ If you need to use the OpenAI service through a proxy, you can configure the proxy address using the `OPENAI_PROXY_URL` environment variable:
-### Crontab Automatic Update Script (Optional)
+ ```fish
+ $ docker run -d -p 3210:3210 \
+ -e OPENAI_API_KEY=sk-xxxx \
+ -e OPENAI_PROXY_URL=https://api-proxy.com/v1 \
+ -e ACCESS_CODE=lobe66 \
+ --name lobe-chat \
+ lobehub/lobe-chat
+ ```
-If you want to automatically obtain the latest image, you can follow these steps.
+ ### Crontab Automatic Update Script (Optional)
-First, create a `lobe.env` configuration file with various environment variables, for example:
+ If you want to automatically obtain the latest image, you can follow these steps.
-```env
-OPENAI_API_KEY=sk-xxxx
-OPENAI_PROXY_URL=https://api-proxy.com/v1
-ACCESS_CODE=arthals2333
-OPENAI_MODEL_LIST=-gpt-4,-gpt-4-32k,-gpt-3.5-turbo-16k,gpt-3.5-turbo-1106=gpt-3.5-turbo-16k,gpt-4-0125-preview=gpt-4-turbo,gpt-4-vision-preview=gpt-4-vision
-```
+ First, create a `lobe.env` configuration file with various environment variables, for example:
-Then, you can use the following script to automate the update:
+ ```env
+ OPENAI_API_KEY=sk-xxxx
+ OPENAI_PROXY_URL=https://api-proxy.com/v1
+ ACCESS_CODE=arthals2333
+ OPENAI_MODEL_LIST=-gpt-4,-gpt-4-32k,-gpt-3.5-turbo-16k,gpt-3.5-turbo-1106=gpt-3.5-turbo-16k,gpt-4-0125-preview=gpt-4-turbo,gpt-4-vision-preview=gpt-4-vision
+ ```
-```bash
-#!/bin/bash
-# auto-update-lobe-chat.sh
+ Then, you can use the following script to automate the update:
-# Set up proxy (optional)
-export https_proxy=http://127.0.0.1:7890 http_proxy=http://127.0.0.1:7890 all_proxy=socks5://127.0.0.1:7890
+ ```bash
+ #!/bin/bash
+ # auto-update-lobe-chat.sh
-# Pull the latest image and store the output in a variable
-output=$(docker pull lobehub/lobe-chat:latest 2>&1)
+ # Set up proxy (optional)
+ export https_proxy=http://127.0.0.1:7890 http_proxy=http://127.0.0.1:7890 all_proxy=socks5://127.0.0.1:7890
-# Check if the pull command was executed successfully
-if [ $? -ne 0 ]; then
- exit 1
-fi
+ # Pull the latest image and store the output in a variable
+ output=$(docker pull lobehub/lobe-chat:latest 2>&1)
-# Check if the output contains a specific string
-echo "$output" | grep -q "Image is up to date for lobehub/lobe-chat:latest"
+ # Check if the pull command was executed successfully
+ if [ $? -ne 0 ]; then
+ exit 1
+ fi
-# If the image is already up to date, do nothing
-if [ $? -eq 0 ]; then
- exit 0
-fi
+ # Check if the output contains a specific string
+ echo "$output" | grep -q "Image is up to date for lobehub/lobe-chat:latest"
-echo "Detected Lobe-Chat update"
+ # If the image is already up to date, do nothing
+ if [ $? -eq 0 ]; then
+ exit 0
+ fi
-# Remove the old container
-echo "Removed: $(docker rm -f Lobe-Chat)"
+ echo "Detected Lobe-Chat update"
-# Run the new container
-echo "Started: $(docker run -d --network=host --env-file /path/to/lobe.env --name=Lobe-Chat --restart=always lobehub/lobe-chat)"
+ # Remove the old container
+ echo "Removed: $(docker rm -f Lobe-Chat)"
-# Print the update time and version
-echo "Update time: $(date)"
-echo "Version: $(docker inspect lobehub/lobe-chat:latest | grep 'org.opencontainers.image.version' | awk -F'"' '{print $4}')"
+ # Run the new container
+ echo "Started: $(docker run -d --network=host --env-file /path/to/lobe.env --name=Lobe-Chat --restart=always lobehub/lobe-chat)"
-# Clean up unused images
-docker images | grep 'lobehub/lobe-chat' | grep -v 'lobehub/lobe-chat-database' | grep -v 'latest' | awk '{print $3}' | xargs -r docker rmi > /dev/null 2>&1
-echo "Removed old images."
-```
+ # Print the update time and version
+ echo "Update time: $(date)"
+ echo "Version: $(docker inspect lobehub/lobe-chat:latest | grep 'org.opencontainers.image.version' | awk -F'"' '{print $4}')"
-This script can be used in Crontab, but please ensure that your Crontab can find the correct Docker command. It is recommended to use absolute paths.
+ # Clean up unused images
+ docker images | grep 'lobehub/lobe-chat' | grep -v 'lobehub/lobe-chat-database' | grep -v 'latest' | awk '{print $3}' | xargs -r docker rmi > /dev/null 2>&1
+ echo "Removed old images."
+ ```
-Configure Crontab to execute the script every 5 minutes:
+ This script can be used in Crontab, but please ensure that your Crontab can find the correct Docker command. It is recommended to use absolute paths.
-```bash
-*/5 * * * * /path/to/auto-update-lobe-chat.sh >> /path/to/auto-update-lobe-chat.log 2>&1
-```
+ Configure Crontab to execute the script every 5 minutes:
+ ```bash
+ */5 * * * * /path/to/auto-update-lobe-chat.sh >> /path/to/auto-update-lobe-chat.log 2>&1
+ ```
[docker-pulls-link]: https://hub.docker.com/r/lobehub/lobe-chat
diff --git a/docs/self-hosting/platform/docker.zh-CN.mdx b/docs/self-hosting/platform/docker.zh-CN.mdx
index f0de6c82dc99d..66feb8c0c7c8d 100644
--- a/docs/self-hosting/platform/docker.zh-CN.mdx
+++ b/docs/self-hosting/platform/docker.zh-CN.mdx
@@ -15,10 +15,9 @@ tags:
我们提供了 [Docker 镜像][docker-release-link]，供你在自己的私有设备上部署 LobeChat 服务。
@@ -28,138 +27,134 @@ tags:
### 安装 Docker 容器环境
-（如果已安装，请跳过此步）
+ （如果已安装，请跳过此步）
```fish
$ apt install docker.io
```
-
-
+
```fish
$ yum install docker
```
+
+
-
-
-
+ ### Docker 指令部署
-### Docker 指令部署
+ 使用以下命令即可使用一键启动 LobeChat 服务：
-使用以下命令即可使用一键启动 LobeChat 服务：
+ ```fish
+ $ docker run -d -p 3210:3210 \
+ -e OPENAI_API_KEY=sk-xxxx \
+ -e ACCESS_CODE=lobe66 \
+ --name lobe-chat \
+ lobehub/lobe-chat
+ ```
-```fish
-$ docker run -d -p 3210:3210 \
- -e OPENAI_API_KEY=sk-xxxx \
- -e ACCESS_CODE=lobe66 \
- --name lobe-chat \
- lobehub/lobe-chat
-```
+ 指令说明：
-指令说明：
-
-- 默认映射端口为 `3210`, 请确保未被占用或手动更改端口映射
-- 使用你的 OpenAI API Key 替换上述命令中的 `sk-xxxx` ，获取 API Key 的方式详见最后一节。
-
-
- LobeChat 支持的完整环境变量列表请参考 [📘 环境变量](/zh/docs/self-hosting/environment-variables)
- 部分
-
+ - 默认映射端口为 `3210`, 请确保未被占用或手动更改端口映射
+ - 使用你的 OpenAI API Key 替换上述命令中的 `sk-xxxx` ，获取 API Key 的方式详见最后一节。
-
- 由于官方的 Docker
- 镜像构建大约需要半小时左右，如果在更新部署后会出现「存在更新」的提示，可以等待镜像构建完成后再次部署。
-
+
+ LobeChat 支持的完整环境变量列表请参考 [📘 环境变量](/zh/docs/self-hosting/environment-variables)
+ 部分
+
-
- 官方 Docker 镜像中未设定密码，强烈建议添加密码以提升安全性，否则你可能会遇到 [My API Key was
- stolen!!!](https://github.com/lobehub/lobe-chat/issues/1123) 这样的情况
-
+
+ 由于官方的 Docker
+ 镜像构建大约需要半小时左右，如果在更新部署后会出现「存在更新」的提示，可以等待镜像构建完成后再次部署。
+
-
- 注意，当**部署架构与镜像的不一致时**，需要对 **Sharp** 进行交叉编译，详见 [Sharp
- 交叉编译](https://sharp.pixelplumbing.com/install#cross-platform)
-
+
+ 官方 Docker 镜像中未设定密码，强烈建议添加密码以提升安全性，否则你可能会遇到 [My API Key was
+ stolen!!!](https://github.com/lobehub/lobe-chat/issues/1123) 这样的情况
+
-#### 使用代理地址
+
+ 注意，当**部署架构与镜像的不一致时**，需要对 **Sharp** 进行交叉编译，详见 [Sharp
+ 交叉编译](https://sharp.pixelplumbing.com/install#cross-platform)
+
-如果你需要通过代理使用 OpenAI 服务，你可以使用 `OPENAI_PROXY_URL` 环境变量来配置代理地址：
+ #### 使用代理地址
-```fish
-$ docker run -d -p 3210:3210 \
- -e OPENAI_API_KEY=sk-xxxx \
- -e OPENAI_PROXY_URL=https://api-proxy.com/v1 \
- -e ACCESS_CODE=lobe66 \
- --name lobe-chat \
- lobehub/lobe-chat
-```
+ 如果你需要通过代理使用 OpenAI 服务，你可以使用 `OPENAI_PROXY_URL` 环境变量来配置代理地址：
-### Crontab 自动更新脚本（可选）
+ ```fish
+ $ docker run -d -p 3210:3210 \
+ -e OPENAI_API_KEY=sk-xxxx \
+ -e OPENAI_PROXY_URL=https://api-proxy.com/v1 \
+ -e ACCESS_CODE=lobe66 \
+ --name lobe-chat \
+ lobehub/lobe-chat
+ ```
-如果你想自动获得最新的镜像，你可以如下操作。
+ ### Crontab 自动更新脚本（可选）
-首先，新建一个 `lobe.env` 配置文件，内容为各种环境变量，例如：
+ 如果你想自动获得最新的镜像，你可以如下操作。
-```env
-OPENAI_API_KEY=sk-xxxx
-OPENAI_PROXY_URL=https://api-proxy.com/v1
-ACCESS_CODE=arthals2333
-OPENAI_MODEL_LIST=-gpt-4,-gpt-4-32k,-gpt-3.5-turbo-16k,gpt-3.5-turbo-1106=gpt-3.5-turbo-16k,gpt-4-0125-preview=gpt-4-turbo,gpt-4-vision-preview=gpt-4-vision
-```
+ 首先，新建一个 `lobe.env` 配置文件，内容为各种环境变量，例如：
-然后，你可以使用以下脚本来自动更新：
+ ```env
+ OPENAI_API_KEY=sk-xxxx
+ OPENAI_PROXY_URL=https://api-proxy.com/v1
+ ACCESS_CODE=arthals2333
+ OPENAI_MODEL_LIST=-gpt-4,-gpt-4-32k,-gpt-3.5-turbo-16k,gpt-3.5-turbo-1106=gpt-3.5-turbo-16k,gpt-4-0125-preview=gpt-4-turbo,gpt-4-vision-preview=gpt-4-vision
+ ```
-```bash
-#!/bin/bash
-# auto-update-lobe-chat.sh
+ 然后，你可以使用以下脚本来自动更新：
-# 设置代理（可选）
-export https_proxy=http://127.0.0.1:7890 http_proxy=http://127.0.0.1:7890 all_proxy=socks5://127.0.0.1:7890
+ ```bash
+ #!/bin/bash
+ # auto-update-lobe-chat.sh
-# 拉取最新的镜像并将输出存储在变量中
-output=$(docker pull lobehub/lobe-chat:latest 2>&1)
+ # 设置代理（可选）
+ export https_proxy=http://127.0.0.1:7890 http_proxy=http://127.0.0.1:7890 all_proxy=socks5://127.0.0.1:7890
-# 检查拉取命令是否成功执行
-if [ $? -ne 0 ]; then
- exit 1
-fi
+ # 拉取最新的镜像并将输出存储在变量中
+ output=$(docker pull lobehub/lobe-chat:latest 2>&1)
-# 检查输出中是否包含特定的字符串
-echo "$output" | grep -q "Image is up to date for lobehub/lobe-chat:latest"
+ # 检查拉取命令是否成功执行
+ if [ $? -ne 0 ]; then
+ exit 1
+ fi
-# 如果镜像已经是最新的，则不执行任何操作
-if [ $? -eq 0 ]; then
- exit 0
-fi
+ # 检查输出中是否包含特定的字符串
+ echo "$output" | grep -q "Image is up to date for lobehub/lobe-chat:latest"
-echo "Detected Lobe-Chat update"
+ # 如果镜像已经是最新的，则不执行任何操作
+ if [ $? -eq 0 ]; then
+ exit 0
+ fi
-# 删除旧的容器
-echo "Removed: $(docker rm -f Lobe-Chat)"
+ echo "Detected Lobe-Chat update"
-# 运行新的容器
-echo "Started: $(docker run -d --network=host --env-file /path/to/lobe.env --name=Lobe-Chat --restart=always lobehub/lobe-chat)"
+ # 删除旧的容器
+ echo "Removed: $(docker rm -f Lobe-Chat)"
-# 打印更新的时间和版本
-echo "Update time: $(date)"
-echo "Version: $(docker inspect lobehub/lobe-chat:latest | grep 'org.opencontainers.image.version' | awk -F'"' '{print $4}')"
+ # 运行新的容器
+ echo "Started: $(docker run -d --network=host --env-file /path/to/lobe.env --name=Lobe-Chat --restart=always lobehub/lobe-chat)"
-# 清理不再使用的镜像
-docker images | grep 'lobehub/lobe-chat' | grep -v 'lobehub/lobe-chat-database' | grep -v 'latest' | awk '{print $3}' | xargs -r docker rmi > /dev/null 2>&1
-echo "Removed old images."
-```
+ # 打印更新的时间和版本
+ echo "Update time: $(date)"
+ echo "Version: $(docker inspect lobehub/lobe-chat:latest | grep 'org.opencontainers.image.version' | awk -F'"' '{print $4}')"
-此脚本可以在 Crontab 中使用，但请确认你的 Crontab 可以找到正确的 Docker 命令。建议使用绝对路径。
+ # 清理不再使用的镜像
+ docker images | grep 'lobehub/lobe-chat' | grep -v 'lobehub/lobe-chat-database' | grep -v 'latest' | awk '{print $3}' | xargs -r docker rmi > /dev/null 2>&1
+ echo "Removed old images."
+ ```
-配置 Crontab，每 5 分钟执行一次脚本：
+ 此脚本可以在 Crontab 中使用，但请确认你的 Crontab 可以找到正确的 Docker 命令。建议使用绝对路径。
-```bash
-*/5 * * * * /path/to/auto-update-lobe-chat.sh >> /path/to/auto-update-lobe-chat.log 2>&1
-```
+ 配置 Crontab，每 5 分钟执行一次脚本：
+ ```bash
+ */5 * * * * /path/to/auto-update-lobe-chat.sh >> /path/to/auto-update-lobe-chat.log 2>&1
+ ```
## 获取 OpenAI API Key
@@ -172,31 +167,17 @@ API Key 是使用 LobeChat 进行大语言模型会话的必要信息，本节
- 注册完毕后，前往 [API Keys](https://platform.openai.com/api-keys) 页面，点击 `Create new secret key` 创建新的 API Key:
+ #### 步骤 1：打开创建窗口
-#### 步骤 1：打开创建窗口
-
-
-
-#### 步骤 2：创建 API Key
+
-
+ #### 步骤 2：创建 API Key
-#### 步骤 3：获取 API Key
+
-
+ #### 步骤 3：获取 API Key
+
将此 API Key 填写到 LobeChat 的 API Key 配置中，即可开始使用。
@@ -211,8 +192,8 @@ API Key 是使用 LobeChat 进行大语言模型会话的必要信息，本节
如果你发现注册 OpenAI 账户或者绑定外币信用卡比较麻烦，可以考虑借助一些知名的 OpenAI 第三方代理商来获取 API Key，这可以有效降低获取 OpenAI API Key 的门槛。但与此同时，一旦使用三方服务，你可能也需要承担潜在的风险，请根据你自己的实际情况自行决策。以下是常见的第三方模型代理商列表，供你参考：
-| Logo | 服务商 | 特性说明 | Proxy 代理地址 | 链接 |
-| --- | --- | --- | --- | --- |
+| Logo | 服务商 | 特性说明 | Proxy 代理地址 | 链接 |
+| ------------------------------------------------------------------------------------------------------------------------------------------------- | ------------ | -------------------------------------------------------- | ------------------------- | ----------------------------- |
| | **AiHubMix** | 使用 OpenAI 企业接口，全站模型价格为官方 **86 折**（含 GPT-4 、Cluade 3.5 等） | `https://aihubmix.com/v1` | [获取](https://lobe.li/CnsM6fH) |
diff --git a/docs/self-hosting/platform/netlify.mdx b/docs/self-hosting/platform/netlify.mdx
index f20084b5f2b18..7b86e482e7d77 100644
--- a/docs/self-hosting/platform/netlify.mdx
+++ b/docs/self-hosting/platform/netlify.mdx
@@ -22,129 +22,79 @@ If you want to deploy LobeChat on Netlify, you can follow these steps:
### Fork the LobeChat Repository
-Click the Fork button to fork the LobeChat repository to your GitHub account.
+ Click the Fork button to fork the LobeChat repository to your GitHub account.
-### Prepare your OpenAI API Key
+ ### Prepare your OpenAI API Key
-Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to obtain your OpenAI API Key.
+ Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to obtain your OpenAI API Key.
-### Import to Netlify Workspace
+ ### Import to Netlify Workspace
-
- After testing, it is currently not supported to have a one-click deployment button similar to
- Vercel/Zeabur. The reason is unknown. Therefore, manual import is required.
-
+
+ After testing, it is currently not supported to have a one-click deployment button similar to
+ Vercel/Zeabur. The reason is unknown. Therefore, manual import is required.
+
-Click "Import from git"
+ Click "Import from git"
-
+
-Then click "Deploy with Github" and authorize Netlify to access your GitHub account.
+ Then click "Deploy with Github" and authorize Netlify to access your GitHub account.
-
+
-Next, select the LobeChat project:
+ Next, select the LobeChat project:
-
+
-### Configure Site Name and Environment Variables
+ ### Configure Site Name and Environment Variables
-In this step, you need to configure your site, including the site name, build command, and publish directory. Fill in your site name in the "Site Name" field. If there are no special requirements, you do not need to modify the remaining configurations as we have already set the default configurations.
+ In this step, you need to configure your site, including the site name, build command, and publish directory. Fill in your site name in the "Site Name" field. If there are no special requirements, you do not need to modify the remaining configurations as we have already set the default configurations.
-
+
-Click the "Add environment variables" button to add site environment variables:
+ Click the "Add environment variables" button to add site environment variables:
-
+
-Taking OpenAI as an example, the environment variables you need to add are as follows:
+ Taking OpenAI as an example, the environment variables you need to add are as follows:
-| Environment Variable | Type | Description | Example |
-| --- | --- | --- | --- |
-| `OPENAI_API_KEY` | Required | This is the API key you applied for on the OpenAI account page | `sk-xxxxxx...xxxxxx` |
-| `ACCESS_CODE` | Required | Add a password to access this service. You can set a long password to prevent brute force attacks. When this value is separated by commas, it becomes an array of passwords | `awCT74` or `e3@09!` or `code1,code2,code3` |
-| `OPENAI_PROXY_URL` | Optional | If you manually configure the OpenAI interface proxy, you can use this configuration to override the default OpenAI API request base URL | `https://aihubmix.com/v1`, default value: `https://api.openai.com/v1` |
+ | Environment Variable | Type | Description | Example |
+ | -------------------- | -------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------- |
+ | `OPENAI_API_KEY` | Required | This is the API key you applied for on the OpenAI account page | `sk-xxxxxx...xxxxxx` |
+ | `ACCESS_CODE` | Required | Add a password to access this service. You can set a long password to prevent brute force attacks. When this value is separated by commas, it becomes an array of passwords | `awCT74` or `e3@09!` or `code1,code2,code3` |
+ | `OPENAI_PROXY_URL` | Optional | If you manually configure the OpenAI interface proxy, you can use this configuration to override the default OpenAI API request base URL | `https://aihubmix.com/v1`, default value: `https://api.openai.com/v1` |
-
- For a complete list of environment variables supported by LobeChat, please refer to the [📘
- Environment Variables](/docs/self-hosting/environment-variables)
-
+
+ For a complete list of environment variables supported by LobeChat, please refer to the [📘
+ Environment Variables](/docs/self-hosting/environment-variables)
+
-Afteradding the variables, finally click "Deploy lobe-chat" to enter the deployment phase
+ Afteradding the variables, finally click "Deploy lobe-chat" to enter the deployment phase
-
+
-### Wait for Deployment to Complete
+ ### Wait for Deployment to Complete
-After clicking deploy, you will enter the site details page, where you can click the "Deploying your site" in blue or the "Building" in yellow to view the deployment progress.
+ After clicking deploy, you will enter the site details page, where you can click the "Deploying your site" in blue or the "Building" in yellow to view the deployment progress.
-
+
-Upon entering the deployment details, you will see the following interface, indicating that your LobeChat is currently being deployed. Simply wait for the deployment to complete.
+ Upon entering the deployment details, you will see the following interface, indicating that your LobeChat is currently being deployed. Simply wait for the deployment to complete.
-
+
-During the deployment and build process:
+ During the deployment and build process:
-
-
-### Deployment Successful, Start Using
+
-If your Deploy Log in the interface looks like the following, it means your LobeChat has been successfully deployed.
-
-
+ ### Deployment Successful, Start Using
-At this point, you can click on "Open production deploy" to access your LobeChat site.
+ If your Deploy Log in the interface looks like the following, it means your LobeChat has been successfully deployed.
+
+
+
+ At this point, you can click on "Open production deploy" to access your LobeChat site.
diff --git a/docs/self-hosting/platform/netlify.zh-CN.mdx b/docs/self-hosting/platform/netlify.zh-CN.mdx
index 3760bb95e1075..0078a80ebb2f4 100644
--- a/docs/self-hosting/platform/netlify.zh-CN.mdx
+++ b/docs/self-hosting/platform/netlify.zh-CN.mdx
@@ -20,126 +20,76 @@ tags:
### Fork LobeChat 仓库
-点击 Fork 按钮，将 LobeChat 仓库 Fork 到你的 GitHub 账号下。
+ 点击 Fork 按钮，将 LobeChat 仓库 Fork 到你的 GitHub 账号下。
-### 准备好你的 OpenAI API Key
+ ### 准备好你的 OpenAI API Key
-前往 [OpenAI API Key](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
+ 前往 [OpenAI API Key](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
-### 在 Netlify 工作台导入
+ ### 在 Netlify 工作台导入
-经过测试，暂不支持类似 Vercel/Zeabur 的一键部署按钮，原因未知。因此需要手动导入
+ 经过测试，暂不支持类似 Vercel/Zeabur 的一键部署按钮，原因未知。因此需要手动导入
-点击 「Import from git」
+ 点击 「Import from git」
-
+
-然后点击 「Deploy with Github」，并授权 Netlify 访问你的 GitHub 账号
+ 然后点击 「Deploy with Github」，并授权 Netlify 访问你的 GitHub 账号
-
+
-然后选择 LobeChat 项目：
+ 然后选择 LobeChat 项目：
-
+
-### 配置站点名称与环境变量
+ ### 配置站点名称与环境变量
-在这一步，你需要配置你的站点，包括站点名称、构建命令、发布目录等。在「Site Name」字段填写上你的站点名称。其余配置如果没有特殊要求，无需修改，我们已经设定好了默认配置。
+ 在这一步，你需要配置你的站点，包括站点名称、构建命令、发布目录等。在「Site Name」字段填写上你的站点名称。其余配置如果没有特殊要求，无需修改，我们已经设定好了默认配置。
-
+
-点击 「Add environment variables」按钮，添加站点环境变量：
+ 点击 「Add environment variables」按钮，添加站点环境变量：
-
+
-以配置 OpenAI 为例，你需要添加的环境变量如下：
+ 以配置 OpenAI 为例，你需要添加的环境变量如下：
-| 环境变量 | 类型 | 描述 | 示例 |
-| --- | --- | --- | --- |
-| `OPENAI_API_KEY` | 必选 | 这是你在 OpenAI 账户页面申请的 API 密钥 | `sk-xxxxxx...xxxxxx` |
-| `ACCESS_CODE` | 必选 | 添加访问此服务的密码，你可以设置一个长密码以防被爆破，该值用逗号分隔时为密码数组 | `awCT74` 或 `e3@09!` or `code1,code2,code3` |
-| `OPENAI_PROXY_URL` | 可选 | 如果你手动配置了 OpenAI 接口代理，可以使用此配置项来覆盖默认的 OpenAI API 请求基础 URL | `https://aihubmix.com/v1` ，默认值：`https://api.openai.com/v1` |
+ | 环境变量 | 类型 | 描述 | 示例 |
+ | ------------------ | -- | ------------------------------------------------------- | ---------------------------------------------------------- |
+ | `OPENAI_API_KEY` | 必选 | 这是你在 OpenAI 账户页面申请的 API 密钥 | `sk-xxxxxx...xxxxxx` |
+ | `ACCESS_CODE` | 必选 | 添加访问此服务的密码，你可以设置一个长密码以防被爆破，该值用逗号分隔时为密码数组 | `awCT74` 或 `e3@09!` or `code1,code2,code3` |
+ | `OPENAI_PROXY_URL` | 可选 | 如果你手动配置了 OpenAI 接口代理，可以使用此配置项来覆盖默认的 OpenAI API 请求基础 URL | `https://aihubmix.com/v1` ，默认值：`https://api.openai.com/v1` |
-
- LobeChat 支持的完整环境变量列表请参考 [📘 环境变量](/zh/docs/self-hosting/environment-variables)
- 部分
-
+
+ LobeChat 支持的完整环境变量列表请参考 [📘 环境变量](/zh/docs/self-hosting/environment-variables)
+ 部分
+
-添加完成后，最后点击「Deploy lobe-chat」 进入部署阶段。
+ 添加完成后，最后点击「Deploy lobe-chat」 进入部署阶段。
-
+
-### 等待部署完成
+ ### 等待部署完成
-点击部署后，会进入站点详情页面，你可以点击青色字样的「Deploying your site」或者 「Building」 黄色标签查看部署进度。
+ 点击部署后，会进入站点详情页面，你可以点击青色字样的「Deploying your site」或者 「Building」 黄色标签查看部署进度。
-
+
-进入部署详情，你会看到下述界面，这意味着你的 LobeChat 正在部署中，只需等待部署完成即可。
+ 进入部署详情，你会看到下述界面，这意味着你的 LobeChat 正在部署中，只需等待部署完成即可。
-
+
-部署构建过程中：
+ 部署构建过程中：
-
+
-### 部署成功，开始使用
+ ### 部署成功，开始使用
-如果你的界面中的 Deploy Log 如下所示，意味着你的 LobeChat 部署成功了。
-
-
+ 如果你的界面中的 Deploy Log 如下所示，意味着你的 LobeChat 部署成功了。
-此时，你可以点击「Open production deploy」，即可访问你的 LobeChat 站点
+
+
+ 此时，你可以点击「Open production deploy」，即可访问你的 LobeChat 站点
diff --git a/docs/self-hosting/platform/railway.mdx b/docs/self-hosting/platform/railway.mdx
index f312329f86515..3c6d683a5fc56 100644
--- a/docs/self-hosting/platform/railway.mdx
+++ b/docs/self-hosting/platform/railway.mdx
@@ -20,16 +20,15 @@ If you want to deploy LobeChat on Railway, you can follow the steps below:
### Prepare your OpenAI API Key
-Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
+ Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
-### Click the button below to deploy
+ ### Click the button below to deploy
-[](https://railway.app/template/FB6HrV?referralCode=9bD9mT)
+ [](https://railway.app/template/FB6HrV?referralCode=9bD9mT)
-### Once deployed, you can start using it
+ ### Once deployed, you can start using it
-### Bind a custom domain (optional)
-
-You can use the subdomain provided by Railway, or choose to bind a custom domain. Currently, the domains provided by Railway have not been contaminated, and most regions can connect directly.
+ ### Bind a custom domain (optional)
+ You can use the subdomain provided by Railway, or choose to bind a custom domain. Currently, the domains provided by Railway have not been contaminated, and most regions can connect directly.
diff --git a/docs/self-hosting/platform/railway.zh-CN.mdx b/docs/self-hosting/platform/railway.zh-CN.mdx
index 57fb29a21ace3..4f466ede01600 100644
--- a/docs/self-hosting/platform/railway.zh-CN.mdx
+++ b/docs/self-hosting/platform/railway.zh-CN.mdx
@@ -19,16 +19,15 @@ tags:
### 准备好你的 OpenAI API Key
-前往 [OpenAI API Key](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
+ 前往 [OpenAI API Key](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
-### 点击下方按钮进行部署
+ ### 点击下方按钮进行部署
-[](https://railway.app/template/FB6HrV?referralCode=9bD9mT)
+ [](https://railway.app/template/FB6HrV?referralCode=9bD9mT)
-### 部署完毕后，即可开始使用
+ ### 部署完毕后，即可开始使用
-### 绑定自定义域名（可选）
-
-你可以使用 Railway 提供的子域名，也可以选择绑定自定义域名。目前 Railway 提供的域名还未被污染，大多数地区都可以直连。
+ ### 绑定自定义域名（可选）
+ 你可以使用 Railway 提供的子域名，也可以选择绑定自定义域名。目前 Railway 提供的域名还未被污染，大多数地区都可以直连。
diff --git a/docs/self-hosting/platform/repocloud.mdx b/docs/self-hosting/platform/repocloud.mdx
index 2b07870c4942c..a0bf8bc08aa80 100644
--- a/docs/self-hosting/platform/repocloud.mdx
+++ b/docs/self-hosting/platform/repocloud.mdx
@@ -20,18 +20,17 @@ If you want to deploy LobeChat on RepoCloud, you can follow the steps below:
### Prepare your OpenAI API Key
-Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
+ Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
-### One-click to deploy
+ ### One-click to deploy
-[![][deploy-button-image]][deploy-link]
+ [![][deploy-button-image]][deploy-link]
-### Once deployed, you can start using it
+ ### Once deployed, you can start using it
-### Bind a custom domain (optional)
-
-You can use the subdomain provided by RepoCloud, or choose to bind a custom domain. Currently, the domains provided by RepoCloud have not been contaminated, and most regions can connect directly.
+ ### Bind a custom domain (optional)
+ You can use the subdomain provided by RepoCloud, or choose to bind a custom domain. Currently, the domains provided by RepoCloud have not been contaminated, and most regions can connect directly.
[deploy-button-image]: https://d16t0pc4846x52.cloudfront.net/deploy.svg
diff --git a/docs/self-hosting/platform/repocloud.zh-CN.mdx b/docs/self-hosting/platform/repocloud.zh-CN.mdx
index 0bdaabccef2a4..58d0e8097ee18 100644
--- a/docs/self-hosting/platform/repocloud.zh-CN.mdx
+++ b/docs/self-hosting/platform/repocloud.zh-CN.mdx
@@ -18,18 +18,17 @@ tags:
### 准备好你的 OpenAI API Key
-前往 [OpenAI API Key](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
+ 前往 [OpenAI API Key](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
-### 点击下方按钮进行部署
+ ### 点击下方按钮进行部署
-[![][deploy-button-image]][deploy-link]
+ [![][deploy-button-image]][deploy-link]
-### 部署完毕后，即可开始使用
+ ### 部署完毕后，即可开始使用
-### 绑定自定义域名（可选）
-
-你可以使用 RepoCloud 提供的子域名，也可以选择绑定自定义域名。目前 RepoCloud 提供的域名还未被污染，大多数地区都可以直连。
+ ### 绑定自定义域名（可选）
+ 你可以使用 RepoCloud 提供的子域名，也可以选择绑定自定义域名。目前 RepoCloud 提供的域名还未被污染，大多数地区都可以直连。
[deploy-button-image]: https://d16t0pc4846x52.cloudfront.net/deploy.svg
diff --git a/docs/self-hosting/platform/sealos.mdx b/docs/self-hosting/platform/sealos.mdx
index 440510e444aac..8ad9dae019f1d 100644
--- a/docs/self-hosting/platform/sealos.mdx
+++ b/docs/self-hosting/platform/sealos.mdx
@@ -19,18 +19,17 @@ If you want to deploy LobeChat on SealOS, you can follow the steps below:
### Prepare your OpenAI API Key
-Go to [OpenAI](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
+ Go to [OpenAI](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
-### Click the button below to deploy
+ ### Click the button below to deploy
-[![][deploy-button-image]][deploy-link]
+ [![][deploy-button-image]][deploy-link]
-### After deployment, you can start using it
+ ### After deployment, you can start using it
-### Bind a custom domain (optional)
-
-You can use the subdomain provided by SealOS, or choose to bind a custom domain. Currently, the domains provided by SealOS have not been contaminated, and can be directly accessed in most regions.
+ ### Bind a custom domain (optional)
+ You can use the subdomain provided by SealOS, or choose to bind a custom domain. Currently, the domains provided by SealOS have not been contaminated, and can be directly accessed in most regions.
[deploy-button-image]: https://raw.githubusercontent.com/labring-actions/templates/main/Deploy-on-Sealos.svg
diff --git a/docs/self-hosting/platform/sealos.zh-CN.mdx b/docs/self-hosting/platform/sealos.zh-CN.mdx
index b0bf457a1ea7e..9896d9a5adcb8 100644
--- a/docs/self-hosting/platform/sealos.zh-CN.mdx
+++ b/docs/self-hosting/platform/sealos.zh-CN.mdx
@@ -18,18 +18,17 @@ tags:
### 准备好你的 OpenAI API Key
-前往 [OpenAI](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
+ 前往 [OpenAI](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
-### 点击下方按钮进行部署
+ ### 点击下方按钮进行部署
-[![][deploy-button-image]][deploy-link]
+ [![][deploy-button-image]][deploy-link]
-### 部署完毕后，即可开始使用
+ ### 部署完毕后，即可开始使用
-### 绑定自定义域名（可选）
-
-你可以使用 SealOS 提供的子域名，也可以选择绑定自定义域名。目前 SealOS 提供的域名还未被污染，大多数地区都可以直连。
+ ### 绑定自定义域名（可选）
+ 你可以使用 SealOS 提供的子域名，也可以选择绑定自定义域名。目前 SealOS 提供的域名还未被污染，大多数地区都可以直连。
[deploy-button-image]: https://raw.githubusercontent.com/labring-actions/templates/main/Deploy-on-Sealos.svg
diff --git a/docs/self-hosting/platform/tencentcloud-lighthouse.mdx b/docs/self-hosting/platform/tencentcloud-lighthouse.mdx
new file mode 100644
index 0000000000000..85d8f108d638c
--- /dev/null
+++ b/docs/self-hosting/platform/tencentcloud-lighthouse.mdx
@@ -0,0 +1,33 @@
+---
+title: Deploy LobeChat on TencentCloud Lighthouse
+description: >-
+ Learn how to deploy the LobeChat application on TencentCloud Lighthouse,
+ including preparing the large model API Key, clicking the deploy button, and
+ other operations.
+tags:
+ - TencentCloud Lighthouse
+ - TencentCloud
+ - LobeChat
+ - API Key
+---
+
+# Deploy LobeChat with TencentCloud Lighthouse
+
+If you want to deploy LobeChat on TencentCloud Lighthouse, you can follow the steps below:
+
+## Tencent Cloud Deployment Process
+
+
+ ### Prepare your API Key
+
+ Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
+
+ ### One-click to deploy
+
+ [![][deploy-button-image]][deploy-link]
+
+ ### Once deployed, you can start using it
+
+
+[deploy-button-image]: https://cloudcache.tencent-cloud.com/qcloud/ui/static/static_source_business/d65fb782-4fb0-4348-ad85-f2943d6bee8f.svg
+[deploy-link]: https://buy.tencentcloud.com/lighthouse?blueprintType=APP_OS&blueprintOfficialId=lhbp-6u0ti132®ionId=9&zone=ap-singapore-3&bundleId=bundle_starter_nmc_lin_med2_01&loginSet=AUTO&rule=true&from=lobechat
diff --git a/docs/self-hosting/platform/tencentcloud-lighthouse.zh-CN.mdx b/docs/self-hosting/platform/tencentcloud-lighthouse.zh-CN.mdx
new file mode 100644
index 0000000000000..395ca10a976ac
--- /dev/null
+++ b/docs/self-hosting/platform/tencentcloud-lighthouse.zh-CN.mdx
@@ -0,0 +1,31 @@
+---
+title: 在 腾讯轻量云 上部署 LobeChat
+description: 学习如何快速在腾讯轻量云上部署LobeChat应用，包括准备大模型 API Key、点击部署按钮等操作。
+tags:
+ - 腾讯云
+ - 腾讯轻量云
+ - LobeChat
+ - 部署流程
+ - 大模型 API Key
+---
+
+# 使用 腾讯轻量云 部署
+
+如果想在 腾讯云 上部署 LobeChat，可以按照以下步骤进行操作：
+
+## 腾讯轻量云 部署流程
+
+
+ ### 准备好你的 API Key
+
+ 前往 [OpenAI API Key](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
+
+ ### 点击下方按钮进行部署
+
+ [![][deploy-button-image]][deploy-link]
+
+ ### 部署完毕后，即可开始使用
+
+
+[deploy-button-image]: https://cloudcache.tencent-cloud.com/qcloud/ui/static/static_source_business/d65fb782-4fb0-4348-ad85-f2943d6bee8f.svg
+[deploy-link]: https://buy.cloud.tencent.com/lighthouse?blueprintType=APP_OS&blueprintOfficialId=lhbp-6u0ti132®ionId=8&zone=ap-beijing-3&bundleId=bundle_starter_mc_med2_01&loginSet=AUTO&rule=true&from=lobechat
diff --git a/docs/self-hosting/platform/vercel.mdx b/docs/self-hosting/platform/vercel.mdx
index a521cf70d63d5..0c9efe23c6cb1 100644
--- a/docs/self-hosting/platform/vercel.mdx
+++ b/docs/self-hosting/platform/vercel.mdx
@@ -20,20 +20,19 @@ If you want to deploy LobeChat on Vercel, you can follow the steps below:
### Prepare your OpenAI API Key
-Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
+ Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
-### Click the button below to deploy
+ ### Click the button below to deploy
-[](https://vercel.com/new/clone?repository-url=https%3A%2F%2Fgithub.com%2Flobehub%2Flobe-chat&env=OPENAI_API_KEY,ACCESS_CODE&envDescription=Find%20your%20OpenAI%20API%20Key%20by%20click%20the%20right%20Learn%20More%20button.%20%7C%20Access%20Code%20can%20protect%20your%20website&envLink=https%3A%2F%2Fplatform.openai.com%2Faccount%2Fapi-keys&project-name=lobe-chat&repository-name=lobe-chat)
+ [](https://vercel.com/new/clone?repository-url=https%3A%2F%2Fgithub.com%2Flobehub%2Flobe-chat\&env=OPENAI_API_KEY,ACCESS_CODE\&envDescription=Find%20your%20OpenAI%20API%20Key%20by%20click%20the%20right%20Learn%20More%20button.%20%7C%20Access%20Code%20can%20protect%20your%20website\&envLink=https%3A%2F%2Fplatform.openai.com%2Faccount%2Fapi-keys\&project-name=lobe-chat\&repository-name=lobe-chat)
-Simply log in with your GitHub account, and remember to fill in `OPENAI_API_KEY` (required) and `ACCESS_CODE` (recommended) in the environment variables page.
+ Simply log in with your GitHub account, and remember to fill in `OPENAI_API_KEY` (required) and `ACCESS_CODE` (recommended) in the environment variables page.
-### After deployment, you can start using it
+ ### After deployment, you can start using it
-### Bind a custom domain (optional)
-
-Vercel's assigned domain DNS may be polluted in some regions, so binding a custom domain can establish a direct connection.
+ ### Bind a custom domain (optional)
+ Vercel's assigned domain DNS may be polluted in some regions, so binding a custom domain can establish a direct connection.
## Automatic Synchronization of Updates
diff --git a/docs/self-hosting/platform/vercel.zh-CN.mdx b/docs/self-hosting/platform/vercel.zh-CN.mdx
index 60fc6d012509f..acf68749bbe4d 100644
--- a/docs/self-hosting/platform/vercel.zh-CN.mdx
+++ b/docs/self-hosting/platform/vercel.zh-CN.mdx
@@ -19,20 +19,19 @@ tags:
### 准备好你的 OpenAI API Key
-前往 [OpenAI API Key](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
+ 前往 [OpenAI API Key](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
-### 点击下方按钮进行部署
+ ### 点击下方按钮进行部署
-[![][deploy-button-image]][deploy-link]
+ [![][deploy-button-image]][deploy-link]
-直接使用 GitHub 账号登录即可，记得在环境变量页填入 `OPENAI_API_KEY` （必填） and `ACCESS_CODE`（推荐）；
+ 直接使用 GitHub 账号登录即可，记得在环境变量页填入 `OPENAI_API_KEY` （必填） and `ACCESS_CODE`（推荐）；
-### 部署完毕后，即可开始使用
+ ### 部署完毕后，即可开始使用
-### 绑定自定义域名（可选）
-
-Vercel 分配的域名 DNS 在某些区域被污染了，绑定自定义域名即可直连。
+ ### 绑定自定义域名（可选）
+ Vercel 分配的域名 DNS 在某些区域被污染了，绑定自定义域名即可直连。
## 自动同步更新
diff --git a/docs/self-hosting/platform/zeabur.mdx b/docs/self-hosting/platform/zeabur.mdx
index d6a4705a3e4bf..50baf76ffc585 100644
--- a/docs/self-hosting/platform/zeabur.mdx
+++ b/docs/self-hosting/platform/zeabur.mdx
@@ -19,23 +19,19 @@ If you want to deploy LobeChat on Zeabur, you can follow the steps below:
### Prepare your OpenAI API Key
-Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
+ Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
-### Click the button below to deploy
+ ### Click the button below to deploy
-[![][deploy-button-image]][deploy-link]
+ [![][deploy-button-image]][deploy-link]
-### Once deployed, you can start using it
+ ### Once deployed, you can start using it
-### Bind a custom domain (optional)
-
-You can use the subdomain provided by Zeabur, or choose to bind a custom domain. Currently, the domains provided by Zeabur have not been contaminated, and most regions can connect directly.
+ ### Bind a custom domain (optional)
+ You can use the subdomain provided by Zeabur, or choose to bind a custom domain. Currently, the domains provided by Zeabur have not been contaminated, and most regions can connect directly.
-[deploy-button-image]: https://zeabur.com/button.svg
-[deploy-link]: https://zeabur.com/templates/VZGGTI
-
# Deploy LobeChat with Zeabur as serverless function
> Note: There are still issues with [middlewares and rewrites of next.js on Zeabur](https://github.com/lobehub/lobe-chat/pull/2775?notification_referrer_id=NT_kwDOAdi2DrQxMDkyODQ4MDc2NTozMDk3OTU5OA#issuecomment-2146713899), use at your own risk!
@@ -45,40 +41,41 @@ Since Zeabur does NOT officially support FREE users deploy containerized service
## Zeabur Deployment Process
+ ### Fork LobeChat
-### Fork LobeChat
-
-### Add Zeabur pack config file
+ ### Add Zeabur pack config file
-Add a `zbpack.json` configuration file with the following content to the root dir of your fork:
+ Add a `zbpack.json` configuration file with the following content to the root dir of your fork:
-```json
-{
- "ignore_dockerfile": true,
- "serverless": true
-}
-```
+ ```json
+ {
+ "ignore_dockerfile": true,
+ "serverless": true
+ }
+ ```
-### Prepare your OpenAI API Key
-
-Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
+ ### Prepare your OpenAI API Key
-### Login to your [Zeabur dashboard](https://dash.zeabur.com)
+ Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
-If you do not already have an account, you will need to register one.
+ ### Login to your [Zeabur dashboard](https://dash.zeabur.com)
-### Create a project and service
+ If you do not already have an account, you will need to register one.
-Create a project, then create a service under this project.
+ ### Create a project and service
-### Link your fork of LobeChat to the just created Zeabur service.
+ Create a project, then create a service under this project.
-When adding service, choose github. This may triger a oAuth depend on varies factors like how you login to Zeabur and if you have already authorized Zeabur to access all your repos
+ ### Link your fork of LobeChat to the just created Zeabur service.
-### Bind a custom domain (optional)
+ When adding service, choose github. This may triger a oAuth depend on varies factors like how you login to Zeabur and if you have already authorized Zeabur to access all your repos
-You can create a subdomain provided by Zeabur, or choose to bind a custom domain. Currently, the domains provided by Zeabur have not been contaminated, and most regions can connect directly.
+ ### Bind a custom domain (optional)
-### Zeabur shall start auto build and you should be able to access it by the domain of your choice after a while.
+ You can create a subdomain provided by Zeabur, or choose to bind a custom domain. Currently, the domains provided by Zeabur have not been contaminated, and most regions can connect directly.
+ ### Zeabur shall start auto build and you should be able to access it by the domain of your choice after a while.
+
+[deploy-button-image]: https://zeabur.com/button.svg
+[deploy-link]: https://zeabur.com/templates/VZGGTI
diff --git a/docs/self-hosting/platform/zeabur.zh-CN.mdx b/docs/self-hosting/platform/zeabur.zh-CN.mdx
index 00ee9aa463755..8b6fc7e2100b7 100644
--- a/docs/self-hosting/platform/zeabur.zh-CN.mdx
+++ b/docs/self-hosting/platform/zeabur.zh-CN.mdx
@@ -18,23 +18,19 @@ tags:
### 准备好你的 OpenAI API Key
-前往 [OpenAI API Key](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
+ 前往 [OpenAI API Key](https://platform.openai.com/account/api-keys) 获取你的 OpenAI API Key
-### 点击下方按钮进行部署
+ ### 点击下方按钮进行部署
-[![][deploy-button-image]][deploy-link]
+ [![][deploy-button-image]][deploy-link]
-### 部署完毕后，即可开始使用
+ ### 部署完毕后，即可开始使用
-### 绑定自定义域名（可选）
-
-你可以使用 Zeabur 提供的子域名，也可以选择绑定自定义域名。目前 Zeabur 提供的域名还未被污染，大多数地区都可以直连。
+ ### 绑定自定义域名（可选）
+ 你可以使用 Zeabur 提供的子域名，也可以选择绑定自定义域名。目前 Zeabur 提供的域名还未被污染，大多数地区都可以直连。
-[deploy-button-image]: https://zeabur.com/button.svg
-[deploy-link]: https://zeabur.com/templates/VZGGTI
-
# 使用 Zeabur 将 LobeChat 部署为无服务器函数
> **注意：** 仍然存在关于 [Zeabur 上 next.js 的中间件和重写问题](https://github.com/lobehub/lobe-chat/pull/2775?notification_referrer_id=NT_kwDOAdi2DrQxMDkyODQ4MDc2NTozMDk3OTU5OA#issuecomment-2146713899)，请自担风险！
@@ -44,40 +40,41 @@ tags:
## Zeabur 部署流程
+ ### Fork LobeChat
-### Fork LobeChat
-
-### 添加 Zeabur 打包配置文件
+ ### 添加 Zeabur 打包配置文件
-在您的分支的根目录下添加一个 `zbpack.json` 配置文件，内容如下：
+ 在您的分支的根目录下添加一个 `zbpack.json` 配置文件，内容如下：
-```json
-{
- "ignore_dockerfile": true,
- "serverless": true
-}
-```
+ ```json
+ {
+ "ignore_dockerfile": true,
+ "serverless": true
+ }
+ ```
-### 准备您的 OpenAI API 密钥
+ ### 准备您的 OpenAI API 密钥
-前往 [OpenAI API 密钥](https://platform.openai.com/account/api-keys) 获取您的 OpenAI API 密钥。
+ 前往 [OpenAI API 密钥](https://platform.openai.com/account/api-keys) 获取您的 OpenAI API 密钥。
-### 登录到您的 [Zeabur 仪表板](https://dash.zeabur.com)
+ ### 登录到您的 [Zeabur 仪表板](https://dash.zeabur.com)
-如果您尚未拥有一个账号，您需要注册一个。
+ 如果您尚未拥有一个账号，您需要注册一个。
-### 创建项目与服务。
+ ### 创建项目与服务。
-创建一个项目，并再这个项目下新建一个服务。
+ 创建一个项目，并再这个项目下新建一个服务。
-### 将您的 LobeChat 分支链接到刚创建的 Zeabur 服务。
+ ### 将您的 LobeChat 分支链接到刚创建的 Zeabur 服务。
-在添加服务时，选择 github。这可能会触发一个 oAuth，取决于诸如您如何登录到 Zeabur以及您是否已经授权 Zeabur 访问所有您的存储库等各种因素。
+ 在添加服务时，选择 github。这可能会触发一个 oAuth，取决于诸如您如何登录到 Zeabur 以及您是否已经授权 Zeabur 访问所有您的存储库等各种因素。
-### 绑定自定义域名（可选）
+ ### 绑定自定义域名（可选）
-您可以创建 Zeabur 提供的子域名，或选择绑定自定义域名。目前，Zeabur 提供的域名尚未受到污染，大多数地区可以直接连接。
-
-### Zeabur 将开始自动构建，您应该可以在一段时间后通过您选择的域名访问它。
+ 您可以创建 Zeabur 提供的子域名，或选择绑定自定义域名。目前，Zeabur 提供的域名尚未受到污染，大多数地区可以直接连接。
+ ### Zeabur 将开始自动构建，您应该可以在一段时间后通过您选择的域名访问它。
+
+[deploy-button-image]: https://zeabur.com/button.svg
+[deploy-link]: https://zeabur.com/templates/VZGGTI
diff --git a/docs/self-hosting/server-database.mdx b/docs/self-hosting/server-database.mdx
index d995f10efe08a..833720b8a9791 100644
--- a/docs/self-hosting/server-database.mdx
+++ b/docs/self-hosting/server-database.mdx
@@ -47,58 +47,56 @@ Before deployment, make sure you have a Postgres database instance ready. You ca
Since we support file-based conversations/knowledge base conversations, we need to install the `pgvector` plugin for Postgres. This plugin provides vector search capabilities and is a key component for LobeChat to implement RAG.
-### `NEXT_PUBLIC_SERVICE_MODE`
+ ### `NEXT_PUBLIC_SERVICE_MODE`
-LobeChat supports both client-side and server-side databases, so we provide an environment variable for switching modes, which is `NEXT_PUBLIC_SERVICE_MODE`, with a default value of `client`.
+ LobeChat supports both client-side and server-side databases, so we provide an environment variable for switching modes, which is `NEXT_PUBLIC_SERVICE_MODE`, with a default value of `client`.
-For server-side database deployment scenarios, you need to set `NEXT_PUBLIC_SERVICE_MODE` to `server`.
+ For server-side database deployment scenarios, you need to set `NEXT_PUBLIC_SERVICE_MODE` to `server`.
-
- In the official `lobe-chat-database` Docker image, this environment variable is already set to
- `server` by default. Therefore, if you deploy using the Docker image, you do not need to configure
- this environment variable again.
-
-
-
-Since environment variables starting with `NEXT_PUBLIC` take effect in the front-end code, they cannot be modified through container runtime injection. (Refer to the `next.js` documentation [Configuring: Environment Variables | Next.js (nextjs.org)](https://nextjs.org/docs/pages/building-your-application/configuring/environment-variables)). This is why we chose to create a separate DB version image.
-
-If you need to modify variables with the `NEXT_PUBLIC` prefix in a Docker deployment, you must build the image yourself and inject your own `NEXT_PUBLIC` prefixed environment variables during the build.
+
+ In the official `lobe-chat-database` Docker image, this environment variable is already set to
+ `server` by default. Therefore, if you deploy using the Docker image, you do not need to configure
+ this environment variable again.
+
-
+
+ Since environment variables starting with `NEXT_PUBLIC` take effect in the front-end code, they cannot be modified through container runtime injection. (Refer to the `next.js` documentation [Configuring: Environment Variables | Next.js (nextjs.org)](https://nextjs.org/docs/pages/building-your-application/configuring/environment-variables)). This is why we chose to create a separate DB version image.
-### `DATABASE_URL`
+ If you need to modify variables with the `NEXT_PUBLIC` prefix in a Docker deployment, you must build the image yourself and inject your own `NEXT_PUBLIC` prefixed environment variables during the build.
+
-The core of configuring the database is to add the `DATABASE_URL` environment variable and fill in the Postgres database connection URL you have prepared. The typical format of the database connection URL is `postgres://username:password@host:port/database`.
+ ### `DATABASE_URL`
-
- If you want to enable SSL when connecting to the database, please refer to the
- [documentation](https://stackoverflow.com/questions/14021998/using-psql-to-connect-to-postgresql-in-ssl-mode)
- for setup instructions.
-
+ The core of configuring the database is to add the `DATABASE_URL` environment variable and fill in the Postgres database connection URL you have prepared. The typical format of the database connection URL is `postgres://username:password@host:port/database`.
-### `DATABASE_DRIVER`
+
+ If you want to enable SSL when connecting to the database, please refer to the
+ [documentation](https://stackoverflow.com/questions/14021998/using-psql-to-connect-to-postgresql-in-ssl-mode)
+ for setup instructions.
+
-The `DATABASE_DRIVER` environment variable is used to distinguish between the two types of Postgres database instances, with values of `node` or `neon`.
+ ### `DATABASE_DRIVER`
-To streamline deployment, we have set default values based on the characteristics of different platforms:
+ The `DATABASE_DRIVER` environment variable is used to distinguish between the two types of Postgres database instances, with values of `node` or `neon`.
-- On the Vercel platform, `DATABASE_DRIVER` defaults to `neon`;
-- In our provided Docker image `lobe-chat-database`, `DATABASE_DRIVER` defaults to `node`.
+ To streamline deployment, we have set default values based on the characteristics of different platforms:
-Therefore, if you follow the standard deployment methods below, you do not need to manually configure the `DATABASE_DRIVER` environment variable:
+ - On the Vercel platform, `DATABASE_DRIVER` defaults to `neon`;
+ - In our provided Docker image `lobe-chat-database`, `DATABASE_DRIVER` defaults to `node`.
-- Vercel + Serverless Postgres
-- Docker image + Node Postgres
+ Therefore, if you follow the standard deployment methods below, you do not need to manually configure the `DATABASE_DRIVER` environment variable:
-### `KEY_VAULTS_SECRET`
+ - Vercel + Serverless Postgres
+ - Docker image + Node Postgres
-Considering that users will store sensitive information such as their API Key and baseURL in the database, we need a key to encrypt this information to prevent leakage in case of a database breach. Hence, the `KEY_VAULTS_SECRET` environment variable is used to encrypt sensitive information like user-stored apikeys.
+ ### `KEY_VAULTS_SECRET`
-
- You can generate a random 32-character string as the value of `KEY_VAULTS_SECRET` using `openssl
- rand -base64 32`.
-
+ Considering that users will store sensitive information such as their API Key and baseURL in the database, we need a key to encrypt this information to prevent leakage in case of a database breach. Hence, the `KEY_VAULTS_SECRET` environment variable is used to encrypt sensitive information like user-stored apikeys.
+
+ You can generate a random 32-character string as the value of `KEY_VAULTS_SECRET` using `openssl
+ rand -base64 32`.
+
## Configuring Authentication Services
@@ -107,7 +105,7 @@ In the server-side database mode, we need an authentication service to distingui
### Clerk
-[Clerk](https://clerk.com?utm_source=lobehub&utm_medium=docs) is an authentication SaaS service that provides out-of-the-box authentication capabilities with high productization, low integration costs, and a great user experience. For those who offer SaaS products, Clerk is a good choice. Our official [LobeChat Cloud](https://lobechat.com) uses Clerk as the authentication service.
+[Clerk](https://clerk.com?utm_source=lobehub\&utm_medium=docs) is an authentication SaaS service that provides out-of-the-box authentication capabilities with high productization, low integration costs, and a great user experience. For those who offer SaaS products, Clerk is a good choice. Our official [LobeChat Cloud](https://lobechat.com) uses Clerk as the authentication service.
The integration of Clerk is relatively simple, requiring only the configuration of these environment variables:
diff --git a/docs/self-hosting/server-database.zh-CN.mdx b/docs/self-hosting/server-database.zh-CN.mdx
index 8f0c966d0d361..fc8a0c94e224a 100644
--- a/docs/self-hosting/server-database.zh-CN.mdx
+++ b/docs/self-hosting/server-database.zh-CN.mdx
@@ -15,7 +15,7 @@ tags:
LobeChat 默认使用客户端数据库（IndexedDB），同时也支持使用服务端数据库（下简称 DB 版）。LobeChat 采用了 Postgres 作为后端存储数据库。
- PostgreSQL是一种强大的开源关系型数据库管理系统，具备高度扩展性和标准SQL支持。它提供了丰富的数据类型、并发处理、数据完整性、安全性及可编程性，适用于复杂应用和大规模数据管理。
+ PostgreSQL 是一种强大的开源关系型数据库管理系统，具备高度扩展性和标准 SQL 支持。它提供了丰富的数据类型、并发处理、数据完整性、安全性及可编程性，适用于复杂应用和大规模数据管理。
本文将从框架角度介绍在任何一个平台中部署 DB 版 LobeChat 的流程和原理，让你知其然也知其所以然，最后可以根据自己的实际情况进行部署。
@@ -41,59 +41,56 @@ LobeChat 默认使用客户端数据库（IndexedDB），同时也支持使用
两者的配置方式在环境变量的取值上会略有一点区别，其他方面是一样的。
-同时，由于我们支持了文件对话/知识库对话的能力，因此我们需要为 Postgres 安装 `pgvector` 插件，该插件提供了向量搜索的能力，是 LobeChat 实现 RAG 的重要构件之一。
+同时，由于我们支持了文件对话 / 知识库对话的能力，因此我们需要为 Postgres 安装 `pgvector` 插件，该插件提供了向量搜索的能力，是 LobeChat 实现 RAG 的重要构件之一。
+ ### `NEXT_PUBLIC_SERVICE_MODE`
-### `NEXT_PUBLIC_SERVICE_MODE`
+ LobeChat 同时支持了客户端数据库和服务端数据库，因此我们提供了一个环境变量用于切换模式，这个变量为 `NEXT_PUBLIC_SERVICE_MODE`，该值默认为 `client`。
-LobeChat 同时支持了客户端数据库和服务端数据库，因此我们提供了一个环境变量用于切换模式，这个变量为 `NEXT_PUBLIC_SERVICE_MODE`，该值默认为 `client`。
+ 针对服务端数据库部署场景，你需要将 `NEXT_PUBLIC_SERVICE_MODE` 设置为 `server`。
-针对服务端数据库部署场景，你需要将 `NEXT_PUBLIC_SERVICE_MODE` 设置为 `server`。
+
+ 在官方的 `lobe-chat-database` Docker 镜像中，已经默认将该环境变量设为 `server`，因此如果你使用
+ Docker 镜像部署，则无需再配置该环境变量。
+
-
- 在官方的 `lobe-chat-database` Docker 镜像中，已经默认将该环境变量设为 `server`，因此如果你使用
- Docker 镜像部署，则无需再配置该环境变量。
-
+
+ 由于 `NEXT_PUBLIC` 开头的环境变量是在前端代码中生效的，而因此无法通过容器运行时注入进行修改。 （`next.js`的参考文档 [Configuring: Environment Variables | Next.js (nextjs.org)](https://nextjs.org/docs/pages/building-your-application/configuring/environment-variables) ) 这也是为什么我们选择再打一个 DB 版镜像的原因。
-
- 由于 `NEXT_PUBLIC` 开头的环境变量是在前端代码中生效的，而因此无法通过容器运行时注入进行修改。 （`next.js`的参考文档 [Configuring: Environment Variables | Next.js (nextjs.org)](https://nextjs.org/docs/pages/building-your-application/configuring/environment-variables) ) 这也是为什么我们选择再打一个 DB 版镜像的原因。
+ 如果你需要在 Docker 部署中修改 `NEXT_PUBLIC` 前缀的变量，你必须自行构建镜像，在 build 时就把自己的 `NEXT_PUBLIC` 开头的环境变量打进去。
+
-如果你需要在 Docker 部署中修改 `NEXT_PUBLIC` 前缀的变量，你必须自行构建镜像，在 build 时就把自己的 `NEXT_PUBLIC` 开头的环境变量打进去。
+ ### `DATABASE_URL`
-
+ 配置数据库，核心是添加 `DATABASE_URL` 环境变量，将你准备好的 Postgres 数据库连接 URL 填入其中。数据库连接 URL 的通常格式为 `postgres://username:password@host:port/database`。
-### `DATABASE_URL`
+
+ 如果希望连接数据库时启用 SSL
+ ，请自行参考[文档](https://stackoverflow.com/questions/14021998/using-psql-to-connect-to-postgresql-in-ssl-mode)进行设置
+
-配置数据库，核心是添加 `DATABASE_URL` 环境变量，将你准备好的 Postgres 数据库连接 URL 填入其中。数据库连接 URL 的通常格式为 `postgres://username:password@host:port/database`。
+ ### `DATABASE_DRIVER`
-
- 如果希望连接数据库时启用 SSL
- ，请自行参考[文档](https://stackoverflow.com/questions/14021998/using-psql-to-connect-to-postgresql-in-ssl-mode)进行设置
-
+ `DATABASE_DRIVER` 环境变量用于区分两种 Postgres 数据库实例，`DATABASE_DRIVER` 的取值为 `node` 或 `neon`。
-### `DATABASE_DRIVER`
+ 为提升部署便捷性，我们根据不同的平台特点设置了默认值：
-`DATABASE_DRIVER` 环境变量用于区分两种 Postgres 数据库实例，`DATABASE_DRIVER` 的取值为 `node` 或 `neon`。
+ - 在 Vercel 平台下，`DATABASE_DRIVER` 默认为 `neon`；
+ - 在我们提供的 Docker 镜像 `lobe-chat-database` 中，`DATABASE_DRIVER` 默认为 `node`。
-为提升部署便捷性，我们根据不同的平台特点设置了默认值：
+ 因此如果你采用了以下标准的部署方式，你无需手动配置 `DATABASE_DRIVER` 环境变量：
-- 在 Vercel 平台下，`DATABASE_DRIVER` 默认为 `neon`；
-- 在我们提供的 Docker 镜像 `lobe-chat-database` 中，`DATABASE_DRIVER` 默认为 `node`。
+ - Vercel + Serverless Postgres
+ - Docker 镜像 + Node Postgres
-因此如果你采用了以下标准的部署方式，你无需手动配置 `DATABASE_DRIVER` 环境变量：
+ ### `KEY_VAULTS_SECRET`
-- Vercel + Serverless Postgres
-- Docker 镜像 + Node Postgres
-
-### `KEY_VAULTS_SECRET`
-
-考虑到用户会存储自己的 API Key 和 baseURL 等敏感信息到数据库中，因此我们需要一个密钥来加密这些信息，避免数据库被爆破/脱库时这些关键信息被泄露。 因此有了 `KEY_VAULTS_SECRET` 环境变量，用于加密用户存储的 apikey 等敏感信息。
-
-
- 你可以使用 `openssl rand -base64 32` 生成一个随机的 32 位字符串作为 `KEY_VAULTS_SECRET` 的值。
-
+ 考虑到用户会存储自己的 API Key 和 baseURL 等敏感信息到数据库中，因此我们需要一个密钥来加密这些信息，避免数据库被爆破 / 脱库时这些关键信息被泄露。 因此有了 `KEY_VAULTS_SECRET` 环境变量，用于加密用户存储的 apikey 等敏感信息。
+
+ 你可以使用 `openssl rand -base64 32` 生成一个随机的 32 位字符串作为 `KEY_VAULTS_SECRET` 的值。
+
## 配置身份验证服务
@@ -102,7 +99,7 @@ LobeChat 同时支持了客户端数据库和服务端数据库，因此我们
### Clerk
-[Clerk](https://clerk.com?utm_source=lobehub&utm_medium=docs) 是一个身份验证 SaaS 服务，提供了开箱即用的身份验证能力，产品化程度很高，集成成本较低，体验很好。对于提供 SaaS 化产品的诉求来说，Clerk 是一个不错的选择。我们官方提供的 [LobeChat Cloud](https://lobechat.com)，就是使用了 Clerk 作为身份验证服务。
+[Clerk](https://clerk.com?utm_source=lobehub\&utm_medium=docs) 是一个身份验证 SaaS 服务，提供了开箱即用的身份验证能力，产品化程度很高，集成成本较低，体验很好。对于提供 SaaS 化产品的诉求来说，Clerk 是一个不错的选择。我们官方提供的 [LobeChat Cloud](https://lobechat.com)，就是使用了 Clerk 作为身份验证服务。
Clerk 的集成也相对简单，只需要配置 `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY` 、 `CLERK_SECRET_KEY` 和 `CLERK_WEBHOOK_SECRET` 环境变量即可，这三个环境变量可以在 Clerk 控制台中获取。
@@ -132,7 +129,7 @@ NextAuth 是一个开源的身份验证库，支持多种身份验证提供商
LobeChat 在 [很早以前](https://x.com/lobehub/status/1724289575672291782) 就支持了多模态的 AI 会话，其中涉及到图片上传给大模型的功能。在客户端数据库方案中，图片文件直接以二进制数据存储在浏览器 IndexedDB 数据库，但在服务端数据库中这个方案并不可行。因为在 Postgres 中直接存储文件类二进制数据会大大浪费宝贵的数据库存储空间，并拖慢计算性能。
-这块最佳实践是使用文件存储服务（S3）来存储图片文件，同时 S3 也是文件上传/知识库功能所依赖的大容量静态文件存储方案。
+这块最佳实践是使用文件存储服务（S3）来存储图片文件，同时 S3 也是文件上传 / 知识库功能所依赖的大容量静态文件存储方案。
在本文档库中，S3 所指代的是指兼容 S3 存储方案，即支持 Amazon S3 API 的对象存储系统，常见例如
diff --git a/docs/self-hosting/server-database/docker-compose.mdx b/docs/self-hosting/server-database/docker-compose.mdx
index e6f1e004a3ccc..8a15f917874da 100644
--- a/docs/self-hosting/server-database/docker-compose.mdx
+++ b/docs/self-hosting/server-database/docker-compose.mdx
@@ -6,7 +6,7 @@ description: >-
tags:
- Docker Compose
- LobeChat
- - Docker Containers
+ - Docker Container
- Deployment Guide
---
@@ -15,135 +15,433 @@ tags:
+
+ **Note on Docker Deployment Limitations**
+ The Docker and Docker Compose deployment options do not support injecting the `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY` through environment variables, which prevents enabling the Clerk authentication service. Recommended alternatives include:
+
+ - Hosting deployment via Vercel
+
+ - Running a local image build process
+
+
+## Quick Start
+
- This article assumes you are familiar with the basic principles and processes of deploying the
- LobeChat server database version (hereinafter referred to as DB version), so it only includes core
- environment variable configurations. If you are not familiar with the deployment principles of
- LobeChat DB version, please refer to [Deploying with a Server
- Database](/docs/self-hosting/server-database).
+ **System Compatibility Notes**
+
+ - One-click deployment is supported in Unix environments (Linux/macOS).
+
+ - Windows users must run through [WSL 2](https://aka.ms/wsl).
+
+ - The one-click startup script is only for initial deployment; for subsequent deployments, please refer to the [Custom Deployment](#custom-deployment) section.
+
+ - Port occupation check: Ensure that ports `3210`, `8000`, `9000`, and `9001` are available.
-
- Due to the inability to expose `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY` using Docker environment variables, you cannot use Clerk as a login authentication service when deploying LobeChat with Docker / Docker Compose.
+Execute the following commands to set up the deployment environment; the directory `lobe-chat-db` will be used to store your configuration files and subsequent database files.
+
+```sh
+mkdir lobe-chat-db && cd lobe-chat-db
+```
+
+Fetch and execute the deployment script:
+
+```sh
+bash <(curl -fsSL https://lobe.li/setup.sh) -l en
+```
+
+The script supports the following deployment modes; please choose the appropriate mode based on your needs and read the rest of the documentation.
-If you need Clerk as a login authentication service, consider deploying with Vercel or building your own image.
+- [Local Mode (default)](#local-mode): Accessible only locally, not supporting LAN/public access; suitable for initial experiences.
+- [Port Mode](#port-mode): Supports LAN/public `http` access; suitable for no domain or private network use.
+- [Domain Mode](#domain-mode): Supports LAN/public `http/https` access with reverse proxy; suitable for personal or team use.
+
+ In the script's options prompt `(Option1/Option2)[Option1]`: `(Option1 / Option2)` indicates selectable options, while `[Option1]` indicates the default option; simply press enter to choose the default.
-In general, to fully run the LobeChat database version, you need at least the following four services:
+### Local Mode
-- The LobeChat database version itself
-- A PostgreSQL database with the PGVector plugin
-- An object storage service that supports S3 protocol
-- An SSO login authentication service supported by LobeChat
+
+ ### Complete Remaining Configuration in Interactive Script
-These services can be combined through self-built or online cloud services to meet different levels of deployment needs.
+ Continue pressing enter to use the default configuration.
-In this article, we provide a Docker Compose configuration entirely based on open-source self-built services. You can use this configuration file directly to start the LobeChat database version or modify it to suit your needs.
+ ### Check Configuration Generation Report
-We default to using [MinIO](https://github.com/minio/minio) as the local S3 object storage service and [Casdoor](https://github.com/casdoor/casdoor) as the local authentication service.
+ After the script finishes running, you need to check the configuration generation report, which includes the accounts and initial login passwords for the Casdoor administrator and user.
-## Quick Start
+
+ Please log in to LobeChat using the user account; the administrator account is only for managing Casdoor.
+
-To facilitate a quick start, this section uses the `docker-compose/local` directory's `docker-compose.yml` configuration file. The started LobeChat application runs at `http://localhost:3210` and can be run locally.
+ ```log
+ The results of the secure key generation are as follows:
+ LobeChat:
+ - URL: http://localhost:3210
+ - Username: user
+ - Password: c66f8c
+ Casdoor:
+ - URL: http://localhost:8000
+ - Username: admin
+ - Password: c66f8c
+ Minio:
+ - URL: http://localhost:9000
+ - Username: admin
+ - Password: 8c82ea41
+ ```
-
- For a quick start, this `docker-compose.yml` contains a large number of sensitive Secret/Password
- fields, suitable only for quick demonstrations or personal local use. Do not use it directly in a
- production environment! Otherwise, you will be responsible for any security issues encountered!
-
+ ### Start Docker
+
+ ```sh
+ docker compose up -d
+ ```
+
+ ### Check Logs
+
+ ```sh
+ docker logs -f lobe-chat
+ ```
+
+ If you see the following logs in the container, it means the startup was successful:
+
+ ```log
+ [Database] Start to migration...
+ ✅ database migration pass.
+ -------------------------------------
+ ▲ Next.js 14.x.x
+ - Local: http://localhost:3210
+ - Network: http://0.0.0.0:3210
+
+ ✓ Starting...
+ ✓ Ready in 95ms
+ ```
+
+ ### Access Application
+
+ Visit your LobeChat service at [http://localhost:3210](http://localhost:3210). The account credentials for the application can be found in the report from step `2`.
+
+
+### Port Mode
- ### One-click Startup Script
+ ### Complete Remaining Configuration in Interactive Script
+
+ In port mode, you need to complete the following based on the script prompts:
+
+ - Server IP address settings: for LAN/public access.
+ - Regenerate secure keys: We highly recommend regenerating the secure keys; if you lack the key generation library required by the script, we suggest referring to the [Custom Deployment](#custom-deployment) section for key modifications.
+
+ ### Check Configuration Generation Report
+
+ After the script finishes running, please check the configuration generation report for the Casdoor administrator account, user account, and their initial login passwords.
+
+
+ Please log in to LobeChat using the user account; the administrator account is only for managing Casdoor.
+
+
+ ```log
+ The results of the secure key generation are as follows:
+ LobeChat:
+ - URL: http://your_server_ip:3210
+ - Username: user
+ - Password: 837e26
+ Casdoor:
+ - URL: http://your_server_ip:8000
+ - Username: admin
+ - Password: 837e26
+ Minio:
+ - URL: http://your_server_ip:9000
+ - Username: admin
+ - Password: dbac8440
+ ```
+
+ ### Start Docker
+
+ ```sh
+ docker compose up -d
+ ```
+
+ ### Check Logs
+
+ ```sh
+ docker logs -f lobe-chat
+ ```
+
+ If you see the following logs in the container, it means the startup was successful:
+
+ ```log
+ [Database] Start to migration...
+ ✅ database migration pass.
+ -------------------------------------
+ ▲ Next.js 14.x.x
+ - Local: http://your_server_ip:3210
+ - Network: http://0.0.0.0:3210
+ ✓ Starting...
+ ✓ Ready in 95ms
+ ```
+
+ ### Access Application
+
+ You can access your LobeChat service at `http://your_server_ip:3210`. The account credentials for the application can be found in the report from step `2`.
+
+
+ If your service can accessed via the public network,
+ we strongly recommend disabling the registration,
+ refer to the [documentation](https://lobehub.com/docs/self-hosting/advanced/auth/next-auth/casdoor)
+ for more information.
+
+
-Create a new directory named `lobe-chat-db` to store your configuration files and subsequent database files.
+### Domain Mode
-```sh
-mkdir lobe-chat-db
-cd lobe-chat-db
-```
+
+ ### Complete Reverse Proxy Configuration
+
+ In domain mode, you need to complete the reverse proxy configuration and ensure that the LAN/public can access the following services. Please use a reverse proxy to map the following service ports to the domain names:
+
+ | Domain | Proxy Port | Required |
+ | ---------------------- | ---------- | -------- |
+ | `lobe.example.com` | `3210` | Yes |
+ | `auth.example.com` | `8000` | Yes |
+ | `minio.example.com` | `9000` | Yes |
+ | `minio-ui.example.com` | `9001` | |
+
+
+ If you are using panel software like [aaPanel](https://www.bt.cn/) for reverse proxy configuration,
+ ensure it does not intercept requests to the `.well-known` path to facilitate the proper functioning of Casdoor's OAuth2 configuration.
+ Below is a whitelist configuration for the Nginx server block concerning paths for Casdoor reverse proxy:
+
+ ```nginx
+ location /.well-known/openid-configuration {
+ proxy_pass http://localhost:8000; # Forward to localhost:8000
+ proxy_set_header Host $host; # Keep the original host header
+ proxy_set_header X-Real-IP $remote_addr; # Keep the client's real IP
+ proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; # Keep the forwarded IP
+ proxy_set_header X-Forwarded-Proto $scheme; # Keep the request protocol
+ }
+ ```
+
+ ⚠️ If you are using such panel software,
+ please do not enable any form of caching in the reverse proxy settings of such panel software to avoid affecting the normal operation of the service.
+ Read more at https://github.com/lobehub/lobe-chat/discussions/5986
+
+
+ ### Complete Remaining Configuration in Interactive Script
+
+ In domain mode, you need to complete the following configurations based on script prompts:
+
+ - Domain setup for the LobeChat service: `lobe.example.com`
+ - Domain setup for the Minio service: `minio.example.com`
+ - Domain setup for the Casdoor service: `auth.example.com`
+ - Choose the access protocol: `http` or `https`
+ - Regenerate secure keys: We highly recommend regenerating the secure keys; if you lack the key generation library required by the script, we suggest referring to the [Custom Deployment](#custom-deployment) section for key modifications.
+
+
+ The following issues may impede access to your service:
+
+ - The domain configuration here must match the reverse proxy configuration in step `1`.
+
+ - If you are using Cloudflare for domain resolution and have activated `full proxy`, please use the `https` protocol.
+
+ - If you have used the HTTPS protocol, ensure that your domain certificate is correctly configured; one-click deployment does not support self-signed certificates by default.
+
+
+ ### Check Configuration Generation Report
+
+ After the script finishes running, you need to check the configuration generation report, which includes the initial login password for the Casdoor administrator.
+
+
+ Please log in to LobeChat using the user account; the administrator account is only for managing Casdoor.
+
+
+ ```log
+ The results of the secure key generation are as follows:
+ LobeChat:
+ - URL: https://lobe.example.com
+ - Username: user
+ - Password: 837e26
+ Casdoor:
+ - URL: https://auth.example.com
+ - Username: admin
+ - Password: 837e26
+ Minio:
+ - URL: https://minio.example.com
+ - Username: admin
+ - Password: dbac8440
+ ```
+
+ ### Start Docker
+
+ ```sh
+ docker compose up -d
+ ```
+
+ ### Check Logs
+
+ ```sh
+ docker logs -f lobe-chat
+ ```
+
+ If you see the following logs in the container, it indicates a successful startup:
+
+ ```log
+ [Database] Start to migration...
+ ✅ database migration pass.
+ -------------------------------------
+ ▲ Next.js 14.x.x
+ - Local: https://localhost:3210
+ - Network: http://0.0.0.0:3210
+ ✓ Starting...
+ ✓ Ready in 95ms
+ ```
+
+ ### Access Application
+
+ You can access your LobeChat service via `https://lobe.example.com`. The account credentials for the application can be found in the report from step `3`.
+
+
+ If your service can accessed via the public network,
+ we strongly recommend disabling the registration,
+ refer to the [documentation](https://lobehub.com/docs/self-hosting/advanced/auth/next-auth/casdoor)
+ for more information.
+
+
-We provide a one-click startup script `setup.sh`, which can automatically use the default configuration and start the service:
+## Custom Deployment
+
+This section mainly introduces the configurations that need to be modified to customize the deployment of the LobeChat service in different network environments. Before starting, you can download the [Docker Compose configuration file](https://raw.githubusercontent.com/lobehub/lobe-chat/HEAD/docker-compose/local/docker-compose.yml) and the [environment variable configuration file](https://raw.githubusercontent.com/lobehub/lobe-chat/refs/heads/main/docker-compose/local/.env.example).
```sh
-bash <(curl -fsSL https://raw.githubusercontent.com/lobehub/lobe-chat/HEAD/docker-compose/local/setup.sh) -f
+curl -O https://raw.githubusercontent.com/lobehub/lobe-chat/HEAD/docker-compose/local/docker-compose.yml
+curl -O https://raw.githubusercontent.com/lobehub/lobe-chat/HEAD/docker-compose/local/.env.en_US.example
+mv .env.en_US.example .env
```
-### Start Docker
+
+ This section does not cover all complete variables; remaining variables can be referenced in [Deploying with the Server Database](/en/docs/self-hosting/server-database).
+
-```sh
-docker compose up -d
+### Prerequisites
+
+Generally, to fully run the LobeChat database version, you will need at least the following four services:
+
+- The LobeChat database version itself
+- PostgreSQL database with PGVector plugin
+- Object storage service that supports S3 protocol
+- An SSO authentication service supported by LobeChat
+
+These services can be combined through self-hosting or online cloud services to meet various deployment needs. In this article, we provide a Docker Compose configuration entirely based on open-source self-hosted services, which can be used directly to start the LobeChat database version or modified to suit your requirements.
+
+We use [MinIO](https://github.com/minio/minio) as the local S3 object storage service and [Casdoor](https://github.com/casdoor/casdoor) as the local authentication service by default.
+
+
+ If your network topology is complex, please make sure these services can communicate properly within your network environment.
+
+
+### Necessary Configuration
+
+Now, we will introduce the necessary configurations for running these services:
+
+1. Casdoor
+
+- LobeChat requires communication with Casdoor, so you need to configure Casdoor's Issuer.
+
+```env
+AUTH_CASDOOR_ISSUER=https://auth.example.com
```
-The default login account is the default account of Casdoor, with the username `admin` and password `123`.
+This configuration will affect LobeChat's login authentication service, and you need to ensure that the URL of the Casdoor service is correct. You can find common manifestations and solutions for errors in this configuration in the [FAQ](#faq).
-### Check Logs
+- Additionally, you need to allow the callback URL in Casdoor to point to the LobeChat address:
+
+Please add a line in the `Authentication -> Application` -> `` -> `Redirect URI` in Casdoor's web panel:
-```sh
-docker logs -f lobe-chat
+```
+https://auth.example.com/api/auth/callback/casdoor
```
-If you see the following logs in the container, it means the startup was successful:
+- Casdoor needs to provide the Origin information for access in the environment variables:
-```log
-[Database] Start to migration...
-✅ database migration pass.
--------------------------------------
- ▲ Next.js 14.x.x
- - Local: http://localhost:3210
- - Network: http://0.0.0.0:3210
-
- ✓ Starting...
- ✓ Ready in 95ms
+```env
+origin=https://auth.example.com
```
-
+2. MinIO
+
+- LobeChat needs to provide a public access URL for object files for the LLM service provider, hence you need to configure MinIO's Endpoint.
+
+```env
+S3_PUBLIC_DOMAIN=https://minio.example.com
+S3_ENDPOINT=https://minio.example.com
+```
+
+3. PostgreSQL
-At this point, you have successfully deployed the LobeChat database version, and you can access your LobeChat service at `http://localhost:3210`.
+This configuration is found in the `docker-compose.yml` file, and you will need to configure the database name and password:
-If you encounter problems, try checking the Docker logs and console logs for troubleshooting.
+```yaml
+services:
+ lobe:
+ environment:
+ - 'DATABASE_URL=postgresql://postgres:${POSTGRES_PASSWORD}@postgresql:5432/${LOBE_DB_NAME}'
+```
-## Publishing to Production (IP Mode)
+## FAQ
-The following assumes that the machine IP running the LobeChat service on the LAN/public network is `your_server_ip`.
+#### Unable to Log In Properly
-To access via the local network IP, you first need to run the local startup:
+Check for the following errors based on the container logs:
```sh
-bash <(curl -fsSL https://raw.githubusercontent.com/lobehub/lobe-chat/HEAD/docker-compose/local/setup.sh) -f
-docker compose up -d
+docker logs -f lobe-chat
```
-Next, you need to modify the downloaded `docker-compose.yml` file, performing a global replacement to change `localhost` to `your_server_ip`, and then restart:
+- r3: "response" is not a conform Authorization Server Metadata response (unexpected HTTP status code)
-```sh
-sed -i 's/localhost/your_server_ip/g' docker-compose.yml
-docker compose up -d
+```log
+lobe-chat | [auth][error] r3: "response" is not a conform Authorization Server Metadata response (unexpected HTTP status code)
```
-### Configuring Casdoor
+Cause: This issue is typically caused by improper reverse proxy configuration; you need to ensure your reverse proxy configuration does not intercept the Casdoor OAuth2 configuration requests.
-1. After starting with the `setup.sh` script, the default port for Casdoor WebUI is `8000`. You can access it via `http://your_server_ip:8000`, with the default username `admin` and password `123`.
+Solutions:
-2. In `Identity -> Applications`, add a new line:
+- Please refer to the reverse proxy configuration notes in the [Domain Mode](#domain-mode) section.
- ```
- http://your_server_ip:3210/api/auth/callback/casdoor
- ```
+- A direct troubleshooting method is to access `https://auth.example.com/.well-known/openid-configuration` directly; if:
+ - Non-JSON format data is returned, it indicates your reverse proxy configuration is incorrect.
+ - If the returned JSON format data contains an `"issuer": "URL"` field that does not match your configured `https://auth.example.com`, it indicates your environment variable configuration is incorrect.
-3. Most other configurations can remain default; you can modify the default configurations in `Identity -> Applications`.
+- TypeError: fetch failed
-### Configuring MinIO S3
+```log
+lobe-chat | [auth][error] TypeError: fetch failed
+```
-1. After starting with the `setup.sh` script, the default port for MinIO WebUI is `9001`. You can access it via `http://your_server_ip:9001`, with the default username `YOUR_MINIO_USER` and password `YOUR_MINIO_PASSWORD`.
+Cause: LobeChat cannot access the authentication service.
-2. Most configurations can remain default. If you wish to allow users to change their avatars, you need to create a bucket named `casdoor` in `buckets`, select custom policy, and copy and paste the following content (if you have changed the bucket name, please find and replace):
+Solutions:
+
+- Check whether your authentication service is running properly and whether LobeChat's network can reach the authentication service.
+
+- A straightforward troubleshooting method is to use the `curl` command in the LobeChat container terminal to access your authentication service at `https://auth.example.com/.well-known/openid-configuration`. If JSON format data is returned, it indicates your authentication service is functioning correctly.
+
+````markdown
+## Extended Configuration
+
+To enhance your LobeChat service, you can perform the following extended configurations according to your needs.
+
+### Use MinIO to Store Casdoor Avatars
+
+Allow users to change their avatars in Casdoor.
+
+1. First, create a bucket named `casdoor` in `buckets`, select a custom policy, and copy and paste the content below (if you modify the bucket name, please find and replace accordingly).
```json
{
@@ -180,66 +478,103 @@ docker compose up -d
],
"Version": "2012-10-17"
}
- ```
+````
-3. Create a new access key and store the generated `Access Key` and `Secret Key`.
+2. Create a new access key and store the generated `Access Key` and `Secret Key`.
-4. In Casdoor's `Identity -> Providers`, associate the MinIO S3 service. Here is a sample configuration:
+3. In Casdoor's `Authentication -> Providers`, associate the MinIO S3 service. Below is an example configuration:
- The Client ID and Client Secret should be the `Access Key` and `Secret Key` from the previous step, and `192.168.31.251` should be replaced with `your_server_ip`.
+ Here, the client ID and client secret correspond to the `Access Key` and `Secret Key` from the previous step; replace `192.168.31.251` with `your_server_ip`.
+
+4. In Casdoor's `Authentication -> Apps`, add a provider to the `app-built-in` application, select `minio`, and save and exit.
-5. In Casdoor's `Identity -> Applications`, add the provider to the `app-built-in` application, select `minio`, save and exit.
+5. You can attempt to upload a file in Casdoor's `Authentication -> Resources` to test if the configuration is correct.
-6. You can try uploading a file in Casdoor's `Identity -> Resources` to test if the configuration is correct.
+### Migrating from `logto` to `Casdoor` in Production Deployment
-## Publishing to Production (Domain Name Mode)
+This is applicable for users who have been using `logto` as their login and authentication service in a production environment.
- For deployments with a domain name, we recommend using [Logto](https://logto.io/) as the login
- authentication service. The remainder of this article will use it as an example. If you use other
- authentication services like Casdoor, the process should be similar, but note that different
- authentication services may have different port configurations.
+ Due to significant instability when using [Logto](https://logto.io/)
+ as a login and authentication service, the following tutorial is based on deploying with an IP mode, implementing a domain release solution using Casdoor as the authentication service provider.
+ The remainder of this article will illustrate using this as an example. If you are using other login authentication services like Logto, the process should be similar, but be aware that port configurations may differ among different services.
-In the following text, we assume that in addition to the above services, you are also running a layer of Nginx for reverse proxy and SSL configuration.
+In the following, it is assumed that in addition to the above services, you are also running an **Nginx** layer for reverse proxy and SSL configuration.
-The domain and associated service port descriptions are as follows:
+The domain and corresponding service port descriptions are as follows:
-- `lobe.example.com`: Your LobeChat service domain, which needs to reverse proxy to the LobeChat service port, defaulting to `3210`.
-- `lobe-auth-api.example.com`: Your Logto service domain, which needs to reverse proxy to the Logto API service port, defaulting to `3001`.
-- `lobe-auth-ui.example.com`: Your Logto UI domain, which needs to reverse proxy to the Logto WebUI service port, defaulting to `3002`.
-- `lobe-s3-api.example.com`: Your MinIO API domain, which needs to reverse proxy to the MinIO API service port, defaulting to `9000`.
-- `lobe-s3-ui.example.com`: Optional, your MinIO UI domain, which needs to reverse proxy to the MinIO WebUI service port, defaulting to `9001`.
+- `lobe.example.com`: This is your LobeChat service domain, which needs to reverse proxy to the LobeChat service port, default is `3210`.
+- `auth.example.com`: This is your Logto UI domain, which needs to reverse proxy to the Logto WebUI service port, default is `8000`.
+- `minio.example.com`: This is your MinIO API domain, which needs to reverse proxy to the MinIO API service port, default is `9000`.
+- `minio-ui.example.com`: Optional, this is your MinIO UI domain, which needs to reverse proxy to the MinIO WebUI service port, default is `9001`.
-And the service ports that do not require reverse proxy:
+#### Configuration File
-- `postgresql`: Your PostgreSQL database service port, defaulting to `5432`.
+```sh
+bash <(curl -fsSL https://raw.githubusercontent.com/lobehub/lobe-chat/HEAD/docker-compose/local/setup.sh) -f -l zh_CN
+docker compose up -d
+```
-
- Please note that CORS cross-domain is configured internally in MinIO / Logto services. Do not configure CORS additionally in your reverse proxy, as this will cause errors.
+Make sure to save the newly generated password at this time!
-For MinIO not on port 443, the Host must be `$http_host` (with port number), otherwise a 403 error will occur: `proxy_set_header Host $http_host`.
+After running, you will get three files:
-If you need to configure an SSL certificate, please configure it uniformly in the outer Nginx reverse proxy and not in MinIO.
+- init\_data.json
+- docker-compose.yml
+- .env
-
+Next, modify the configuration files to achieve domain release.
-### Configuration Files
+1. Modify the `docker-compose.yml` file.
+ 1. Change the `MINIO_API_CORS_ALLOW_ORIGIN` field of `minio`.
+ ```yaml
+ 'MINIO_API_CORS_ALLOW_ORIGIN=https://lobe.example.com'
+ ```
+ 2. Modify the `origin` field of `casdoor`.
+ ```yaml
+ origin: 'https://auth.example.com'
+ ```
+ 3. Modify the `environment` field of `lobe`.
+ ```yaml
+ # - 'APP_URL=http://localhost:3210'
+ - 'APP_URL=https://lobe.example.com'
+
+ - 'NEXT_AUTH_SSO_PROVIDERS=casdoor'
+ - 'KEY_VAULTS_SECRET=Kix2wcUONd4CX51E/ZPAd36BqM4wzJgKjPtz2sGztqQ='
+ - 'NEXT_AUTH_SECRET=NX2kaPE923dt6BL2U8e9oSre5RfoT7hg'
+ # - 'AUTH_URL=http://localhost:${LOBE_PORT}/api/auth'
+ - 'AUTH_URL=https://lobe.example.com/api/auth'
+
+ # - 'AUTH_CASDOOR_ISSUER=http://localhost:${CASDOOR_PORT}'
+ - 'AUTH_CASDOOR_ISSUER=https://auth.example.com'
+
+ - 'DATABASE_URL=postgresql://postgres:${POSTGRES_PASSWORD}@postgresql:5432/${LOBE_DB_NAME}'
+ # - 'S3_ENDPOINT=http://localhost:${MINIO_PORT}'
+ - 'S3_ENDPOINT=https://minio.example.com'
+
+ - 'S3_BUCKET=${MINIO_LOBE_BUCKET}'
+ # - 'S3_PUBLIC_DOMAIN=http://localhost:${MINIO_PORT}'
+ - 'S3_PUBLIC_DOMAIN=https://minio.example.com'
+
+ - 'S3_ENABLE_PATH_STYLE=1'
+ - 'LLM_VISION_IMAGE_USE_BASE64=1'
+ ```
+2. Modify the `.env` file.
+
+
+ For security reasons, modify the ROOT USER field in the `.env` file.
+
```sh
-curl -fsSL https://raw.githubusercontent.com/lobehub/lobe-chat/HEAD/docker-compose/production/logto/docker-compose.yml > docker-compose.yml
-curl -fsSL https://raw.githubusercontent.com/lobehub/lobe-chat/HEAD/docker-compose/production/logto/.env.example > .env
+# MinIO S3 configuration
+MINIO_ROOT_USER=XXXX
+MINIO_ROOT_PASSWORD=XXXX
```
-The configuration files include both `.env` and `docker-compose.yml`, where the `.env` file is used to configure the environment variables for LobeChat, and the `docker-compose.yml` file is used to configure the Postgres, MinIO, and Logto services.
-
-Generally, you should only modify sensitive information such as domain names and account passwords, while keeping other configuration items at their default values.
-
-Refer to the appendix for example configurations.
-
-### Postgres Database Configuration
+#### Postgres Database Configuration
You can check the logs with the following command:
@@ -248,13 +583,11 @@ docker logs -f lobe-chat
```
- In our official Docker image, database schema migration is automatically executed before starting
- the image. Our official image promises stability in the automatic creation of tables from "empty
- database -> complete tables." Therefore, we recommend that your database instance use an empty
- table instance to avoid the hassle of manual table structure maintenance or migration.
+ In our official Docker images, automatic migration of the database schema is performed before starting the images.
+ Our official images guarantee the stability of "empty database -> complete tables" for automatic table creation. Therefore, we recommend your database instance use an empty table instance to avoid the trouble of manually maintaining table structure or migrations.
-If you encounter issues when creating tables, you can try the following command to force remove the database container and restart:
+If you encounter issues during table creation, you can try the following command to forcibly remove the database container and restart:
```sh
docker compose down # Stop the service
@@ -262,109 +595,59 @@ sudo rm -rf ./data # Remove mounted database data
docker compose up -d # Restart
```
-### Login Authentication Service Configuration
-
-This article uses Logto as an example to explain the configuration process. If you are using other authentication service providers, please refer to their documentation for configuration.
+#### Login Authentication Service Configuration
-
- Please remember to configure the CORS cross-domain settings for the corresponding login authentication service provider to ensure LobeChat can access the authentication service properly.
-
-In this article, you need to allow cross-domain requests from `https://lobe.example.com`.
-
-
+You first need to access the WebUI for configuration:
-You first need to visit the WebUI for configuration:
+- If you have set up the reverse proxy as mentioned before, open `https://auth.example.com`
+- Otherwise, after port mapping, open `http://localhost:8000`
-- If you have configured the reverse proxy as mentioned earlier, open `https://lobe-auth-ui.example.com`.
-- Otherwise, open `http://localhost:3002` after performing port mapping.
+Log in to the admin account:
-1. Register a new account; the first registered account will automatically become an administrator.
+- The default username is admin.
+- The default password is the random password generated when downloading the config file. If forgotten, you can find it in the `init_data.json` file.
-2. In `Applications`, create a `Next.js (App Router)` application with any name.
+After logging in, perform the following actions:
-3. Set `Redirect URI` to `https://lobe.example.com/api/auth/callback/logto`, and `Post sign-out redirect URI` to `https://lobe.example.com/`.
+1. In `User Management -> Organizations`, add a new organization with the name and display name `Lobe Users`. Keep the rest as default.
+2. In `Authentication -> Apps`, add a new application.
-4. Set `CORS allowed origins` to `https://lobe.example.com`.
+- Name and display name should be `LobeChat`.
+- Organization should be `Lobe Users`.
+- Add a line in Redirect URLs as `https://lobe.example.com/api/auth/callback/casdoor`.
+- Disable all login methods except password.
+- Fill in the client ID and client secret in the `.env` file under `AUTH_CASDOOR_ID` and `AUTH_CASDOOR_SECRET`.
+- (Optional) Design the appearance of the login and registration pages by mimicking the `built-in` application configuration.
+- Save and exit.
-
-
-5. Obtain `App ID` and `App secrets`, and fill them into your `.env` file under `AUTH_LOGTO_ID` and `AUTH_LOGTO_SECRET`.
-
-6. Set `AUTH_LOGTO_ISSUER` in your `.env` file to `https://lobe-auth-api.example.com/oidc`.
-
-
-
-7. Optionally, in the left panel under `Sign-in experience`, you can disable `Enable user registration` in `Sign-up and sign-in - Advanced Options` to prevent users from registering on their own. If you disable user registration, you will need to manually add users in the left panel under `User Management`.
-
-
-
-8. Restart the LobeChat service:
-
- ```sh
- docker compose up -d
- ```
-
-
- Please note that the administrator account is not the same as a registered user. Do not log in to
- LobeChat with your administrator account, as this will only result in an error.
+
+ Following the steps above ensures that not all users are administrators by default, leading to an unsafe situation.
-### S3 Object Storage Service Configuration
+#### S3 Object Storage Service Configuration
-This article uses MinIO as an example to explain the configuration process. If you are using other S3 service providers, please refer to their documentation for configuration.
+This article uses MinIO as an example to explain the configuration process. If you are using another S3 service provider, please refer to their documentation for configuration.
- Please remember to configure the CORS cross-domain settings for the corresponding S3 service provider to ensure LobeChat can access the S3 service properly.
+ Please remember to configure the corresponding S3 service provider's CORS settings to ensure that LobeChat can access the S3 service correctly.
-In this article, you need to allow cross-domain requests from `https://lobe.example.com`. This can be configured in MinIO WebUI under `Configuration - API - Cors Allow Origin`, or in the Docker Compose under `minio - environment - MINIO_API_CORS_ALLOW_ORIGIN`.
-
-If you use the second method (which is also the default method) for configuration, you will not be able to configure it in MinIO WebUI anymore.
+ In this document, you need to allow cross-origin requests from `https://lobe.example.com`. This can either be configured in MinIO WebUI under `Configuration - API - Cors Allow Origin`, or in the Docker Compose configuration under `minio - environment - MINIO_API_CORS_ALLOW_ORIGIN`.
+ If you use the second method (which is also the default), you will no longer be able to configure it in the MinIO WebUI.
-You first need to visit the WebUI for configuration:
-
-- If you have configured the reverse proxy as mentioned earlier, open `https://lobe-s3-ui.example.com`.
-- Otherwise, open `http://localhost:9001` after performing port mapping.
-
-1. Enter your set `MINIO_ROOT_USER` and `MINIO_ROOT_PASSWORD` on the login page, then click login.
-
-2. In the left panel under Administer / Buckets, click `Create Bucket`, enter `lobe` (which corresponds to your `S3_BUCKET` environment variable), and then click `Create`.
-
-
-
-3. Select your bucket, click Summary - Access Policy, edit, choose `Custom`, input the content from `minio-bucket-config.json` (see appendix), and save (assuming your bucket name is `lobe`):
+You first need to access the WebUI for configuration:
-
+- If you have set up the reverse proxy as mentioned before, open `https://minio-ui.example.com`
+- Otherwise, after port mapping, open `http://localhost:9001`
-
+1. Enter the `MINIO_ROOT_USER` and `MINIO_ROOT_PASSWORD` you set in the login interface, then click login.
-4. In the left panel under User / Access Keys, click `Create New Access Key`, without any extra modifications, and fill the generated `Access Key` and `Secret Key` into your `.env` file under `S3_ACCESS_KEY_ID` and `S3_SECRET_ACCESS_KEY`.
+2. In the left panel under User / Access Keys, click `Create New Access Key`, no additional modifications needed, and fill the generated `Access Key` and `Secret Key` into your `.env` file under `S3_ACCESS_KEY_ID` and `S3_SECRET_ACCESS_KEY`.
-
+
-5. Restart the LobeChat service:
+3. Restart the LobeChat service:
```sh
docker compose up -d
@@ -372,76 +655,65 @@ You first need to visit the WebUI for configuration:
At this point, you have successfully deployed the LobeChat database version, and you can access your LobeChat service at `https://lobe.example.com`.
-## Appendix
+#### Configuration Files
-For convenience, here is a summary of the example configuration files needed for deploying the server database in domain name mode (for local quick start, local IP access, etc., please directly use the `setup.sh` script to generate and modify it yourself):
+For convenience, here is a summary of example configuration files required for the production deployment using the Casdoor authentication scheme:
-### `.env`
+- `.env`
```sh
-# Required: LobeChat domain for tRPC calls
-# Ensure this domain is whitelisted in your NextAuth providers and S3 service CORS settings
-APP_URL=https://lobe.example.com/
-
-# Postgres related environment variables
-# Required: Secret key for encrypting sensitive information. Generate with: openssl rand -base64 32
-KEY_VAULTS_SECRET=Kix2wcUONd4CX51E/ZPAd36BqM4wzJgKjPtz2sGztqQ=
-# Required: Postgres database connection string
-# Format: postgresql://username:password@host:port/dbname
-# If using Docker, you can use the container name as the host
-DATABASE_URL=postgresql://postgres:uWNZugjBqixf8dxC@postgresql:5432/lobe
-
-# NEXT_AUTH related environment variables
-# Supports auth0, Azure AD, GitHub, Authentik, Zitadel, Logto, etc.
-# For supported providers, see: https://lobehub.com/docs/self-hosting/advanced/auth#next-auth
-# If you have ACCESS_CODE, please remove it. We use NEXT_AUTH as the sole authentication source
-# Required: NextAuth secret key. Generate with: openssl rand -base64 32
-NEXT_AUTH_SECRET=NX2kaPE923dt6BL2U8e9oSre5RfoT7hg
-# Required: Specify the authentication provider (e.g., Logto)
-NEXT_AUTH_SSO_PROVIDERS=logto
-# Required: NextAuth URL for callbacks
-NEXTAUTH_URL=https://lobe.example.com/api/auth
-
-# NextAuth providers configuration (example using Logto)
-# For other providers, see: https://lobehub.com/docs/self-hosting/environment-variables/auth
-AUTH_LOGTO_ID=YOUR_LOGTO_ID
-AUTH_LOGTO_SECRET=YOUR_LOGTO_SECRET
-AUTH_LOGTO_ISSUER=https://lobe-auth-api.example.com/oidc
-
-# Proxy settings (if needed, e.g., when using GitHub as an auth provider)
+# Proxy, if you need it
# HTTP_PROXY=http://localhost:7890
# HTTPS_PROXY=http://localhost:7890
-# S3 related environment variables (example using MinIO)
-# Required: S3 Access Key ID (for MinIO, invalid until manually created in MinIO UI)
-S3_ACCESS_KEY_ID=YOUR_S3_ACCESS_KEY_ID
-# Required: S3 Secret Access Key (for MinIO, invalid until manually created in MinIO UI)
-S3_SECRET_ACCESS_KEY=YOUR_S3_SECRET_ACCESS_KEY
-# Required: S3 Endpoint for server/client connections to S3 API
-S3_ENDPOINT=https://lobe-s3-api.example.com
-# Required: S3 Bucket (invalid until manually created in MinIO UI)
-S3_BUCKET=lobe
-# Required: S3 Public Domain for client access to unstructured data
-S3_PUBLIC_DOMAIN=https://lobe-s3-api.example.com
-# Optional: S3 Enable Path Style
-# Use 0 for mainstream S3 cloud providers; use 1 for self-hosted MinIO
-# See: https://lobehub.com/docs/self-hosting/advanced/s3#s-3-enable-path-style
-S3_ENABLE_PATH_STYLE=1
-
-# Other basic environment variables (as needed)
-# See: https://lobehub.com/docs/self-hosting/environment-variables/basic
-# Note: For server versions, the API must support embedding models (OpenAI text-embedding-3-small) for file processing
-# You don't need to specify this model in OPENAI_MODEL_LIST
+# Other environment variables, as needed. You can refer to the environment variables configuration for the client version, making sure not to have ACCESS_CODE.
# OPENAI_API_KEY=sk-xxxx
# OPENAI_PROXY_URL=https://api.openai.com/v1
# OPENAI_MODEL_LIST=...
+
+# ===========================
+# ====== Preset config ======
+# ===========================
+# if no special requirements, no need to change
+LOBE_PORT=3210
+CASDOOR_PORT=8000
+MINIO_PORT=9000
+
+# Postgres related, which are the necessary environment variables for DB
+LOBE_DB_NAME=lobechat
+POSTGRES_PASSWORD=uWNZugjBqixf8dxC
+
+# Casdoor secret
+AUTH_CASDOOR_ID=943e627d79d5dd8a22a1
+AUTH_CASDOOR_SECRET=6ec24ac304e92e160ef0d0656ecd86de8cb563f1
+
+# MinIO S3 configuration
+MINIO_ROOT_USER=Joe
+MINIO_ROOT_PASSWORD=Crj1570768
+
+# Configure the bucket information of MinIO
+MINIO_LOBE_BUCKET=lobe
+S3_ACCESS_KEY_ID=dB6Uq9CYZPdWSZouPyEd
+S3_SECRET_ACCESS_KEY=aPBW8CVULkh8bw1GatlT0GjLihcXHLNwRml4pieS
```
-### `docker-compose.yml`
+- `docker-compose.yml`
```yaml
name: lobe-chat-database
services:
+ network-service:
+ image: alpine
+ container_name: lobe-network
+ ports:
+ - '${MINIO_PORT}:${MINIO_PORT}' # MinIO API
+ - '9001:9001' # MinIO Console
+ - '${CASDOOR_PORT}:${CASDOOR_PORT}' # Casdoor
+ - '${LOBE_PORT}:3210' # LobeChat
+ command: tail -f /dev/null
+ networks:
+ - lobe-network
+
postgresql:
image: pgvector/pgvector:pg16
container_name: lobe-postgres
@@ -450,57 +722,87 @@ services:
volumes:
- './data:/var/lib/postgresql/data'
environment:
- - 'POSTGRES_DB=lobe'
- - 'POSTGRES_PASSWORD=uWNZugjBqixf8dxC'
+ - 'POSTGRES_DB=${LOBE_DB_NAME}'
+ - 'POSTGRES_PASSWORD=${POSTGRES_PASSWORD}'
healthcheck:
test: ['CMD-SHELL', 'pg_isready -U postgres']
interval: 5s
timeout: 5s
retries: 5
restart: always
+ networks:
+ - lobe-network
minio:
image: minio/minio
container_name: lobe-minio
- ports:
- - '9000:9000'
- - '9001:9001'
+ network_mode: 'service:network-service'
volumes:
- './s3_data:/etc/minio/data'
environment:
- - 'MINIO_ROOT_USER=YOUR_MINIO_USER'
- - 'MINIO_ROOT_PASSWORD=YOUR_MINIO_PASSWORD'
- - 'MINIO_DOMAIN=lobe-s3-api.example.com'
- - 'MINIO_API_CORS_ALLOW_ORIGIN=https://lobe.example.com' # Your LobeChat's domain name.
+ - 'MINIO_ROOT_USER=${MINIO_ROOT_USER}'
+ - 'MINIO_ROOT_PASSWORD=${MINIO_ROOT_PASSWORD}'
+ # - 'MINIO_API_CORS_ALLOW_ORIGIN=http://localhost:${LOBE_PORT}'
+ - 'MINIO_API_CORS_ALLOW_ORIGIN=https://lobe.example.com'
restart: always
command: >
- server /etc/minio/data --address ":9000" --console-address ":9001"
+ server /etc/minio/data --address ":${MINIO_PORT}" --console-address ":9001"
- logto:
- image: svhd/logto
- container_name: lobe-logto
- ports:
- - '3001:3001'
- - '3002:3002'
+ casdoor:
+ image: casbin/casdoor
+ container_name: lobe-casdoor
+ entrypoint: /bin/sh -c './server --createDatabase=true'
+ network_mode: 'service:network-service'
depends_on:
postgresql:
condition: service_healthy
environment:
- - 'TRUST_PROXY_HEADER=1'
- - 'DB_URL=postgresql://postgres:uWNZugjBqixf8dxC@postgresql:5432/logto'
- - 'ENDPOINT=https://lobe-auth-api.example.com'
- - 'ADMIN_ENDPOINT=https://lobe-auth-ui.example.com'
- entrypoint: ['sh', '-c', 'npm run cli db seed -- --swe && npm start']
+ RUNNING_IN_DOCKER: 'true'
+ driverName: 'postgres'
+ dataSourceName: 'user=postgres password=${POSTGRES_PASSWORD} host=postgresql port=5432 sslmode=disable dbname=casdoor'
+ # origin: 'http://localhost:${CASDOOR_PORT}'
+ origin: 'https://auth.example.com'
+ runmode: 'dev'
+ volumes:
+ - ./init_data.json:/init_data.json
lobe:
image: lobehub/lobe-chat-database
- container_name: lobe-chat
- ports:
- - '3210:3210'
+ container_name: lobe-chat-database
+ network_mode: 'service:network-service'
depends_on:
- - postgresql
- - minio
- - logto
+ postgresql:
+ condition: service_healthy
+ network-service:
+ condition: service_started
+ minio:
+ condition: service_started
+ casdoor:
+ condition: service_started
+
+ environment:
+ # - 'APP_URL=http://localhost:3210'
+ - 'APP_URL=https://lobe.example.com'
+
+ - 'NEXT_AUTH_SSO_PROVIDERS=casdoor'
+ - 'KEY_VAULTS_SECRET=Kix2wcUONd4CX51E/ZPAd36BqM4wzJgKjPtz2sGztqQ='
+ - 'NEXT_AUTH_SECRET=NX2kaPE923dt6BL2U8e9oSre5RfoT7hg'
+ # - 'AUTH_URL=http://localhost:${LOBE_PORT}/api/auth'
+ - 'AUTH_URL=https://lobe.example.com/api/auth'
+
+ # - 'AUTH_CASDOOR_ISSUER=http://localhost:${CASDOOR_PORT}'
+ - 'AUTH_CASDOOR_ISSUER=https://auth.example.com'
+
+ - 'DATABASE_URL=postgresql://postgres:${POSTGRES_PASSWORD}@postgresql:5432/${LOBE_DB_NAME}'
+ # - 'S3_ENDPOINT=http://localhost:${MINIO_PORT}'
+ - 'S3_ENDPOINT=https://minio.example.com'
+
+ - 'S3_BUCKET=${MINIO_LOBE_BUCKET}'
+ # - 'S3_PUBLIC_DOMAIN=http://localhost:${MINIO_PORT}'
+ - 'S3_PUBLIC_DOMAIN=https://minio.example.com'
+
+ - 'S3_ENABLE_PATH_STYLE=1'
+ - 'LLM_VISION_IMAGE_USE_BASE64=1'
env_file:
- .env
restart: always
@@ -510,45 +812,10 @@ volumes:
driver: local
s3_data:
driver: local
-```
-### `minio-bucket-config.json`
-
-```json
-{
- "Statement": [
- {
- "Effect": "Allow",
- "Principal": {
- "AWS": ["*"]
- },
- "Action": ["s3:GetBucketLocation"],
- "Resource": ["arn:aws:s3:::lobe"]
- },
- {
- "Effect": "Allow",
- "Principal": {
- "AWS": ["*"]
- },
- "Action": ["s3:ListBucket"],
- "Resource": ["arn:aws:s3:::lobe"],
- "Condition": {
- "StringEquals": {
- "s3:prefix": ["files/*"]
- }
- }
- },
- {
- "Effect": "Allow",
- "Principal": {
- "AWS": ["*"]
- },
- "Action": ["s3:PutObject", "s3:DeleteObject", "s3:GetObject"],
- "Resource": ["arn:aws:s3:::lobe/files/**"]
- }
- ],
- "Version": "2012-10-17"
-}
+networks:
+ lobe-network:
+ driver: bridge
```
[docker-pulls-link]: https://hub.docker.com/r/lobehub/lobe-chat-database
diff --git a/docs/self-hosting/server-database/docker-compose.zh-CN.mdx b/docs/self-hosting/server-database/docker-compose.zh-CN.mdx
index 9f28b74ac0130..8cde5595d8a42 100644
--- a/docs/self-hosting/server-database/docker-compose.zh-CN.mdx
+++ b/docs/self-hosting/server-database/docker-compose.zh-CN.mdx
@@ -13,26 +13,317 @@ tags:
@@ -32,8 +31,7 @@ tags:
Due to the inability to expose `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY` using Docker environment variables, you cannot use Clerk as an authentication service when deploying LobeChat using Docker / Docker Compose.
-If you do need Clerk as an authentication service, you might consider deploying using Vercel or building your own image.
-
+ If you do need Clerk as an authentication service, you might consider deploying using Vercel or building your own image.
## Deploying on a Linux Server
@@ -41,89 +39,87 @@ If you do need Clerk as an authentication service, you might consider deploying
Here is the process for deploying the LobeChat server database version on a Linux server:
-
-### Create a Postgres Database Instance
-
-Please create a Postgres database instance with the PGVector plugin according to your needs, for example:
-
-```sh
-docker network create pg
-
-docker run --name my-postgres --network pg -e POSTGRES_PASSWORD=mysecretpassword -p 5432:5432 -d pgvector/pgvector:pg16
-```
-
-The above command will create a PG instance named `my-postgres` on the network `pg`, where `pgvector/pgvector:pg16` is a Postgres 16 image with the pgvector plugin installed by default.
-
-
- The pgvector plugin provides vector search capabilities for Postgres, which is an important
- component for LobeChat to implement RAG.
-
-
-
- The above command does not specify a persistent storage location for the pg instance, so it is
- only for testing/demonstration purposes. Please configure persistent storage for production
- environments.
-
-
-### Create a file named `lobe-chat.env` to store environment variables:
-
-```shell
-# Website domain
-APP_URL=https://your-prod-domain.com
-
-# DB required environment variables
-KEY_VAULTS_SECRET=jgwsK28dspyVQoIf8/M3IIHl1h6LYYceSYNXeLpy6uk=
-# Postgres database connection string
-# Format: postgres://username:password@host:port/dbname; if your pg instance is a Docker container, use the container name
-DATABASE_URL=postgres://postgres:mysecretpassword@my-postgres:5432/postgres
-
-# NEXT_AUTH related, can use auth0, Azure AD, GitHub, Authentik, zitadel, etc. If you have other access requirements, feel free to submit a PR
-NEXT_AUTH_SECRET=3904039cd41ea1bdf6c93db0db96e250
-NEXT_AUTH_SSO_PROVIDERS=auth0
-NEXTAUTH_URL=https://your-prod-domain.com/api/auth
-AUTH_AUTH0_ID=xxxxxx
-AUTH_AUTH0_SECRET=cSX_xxxxx
-AUTH_AUTH0_ISSUER=https://lobe-chat-demo.us.auth0.com
-
-# S3 related
-S3_ACCESS_KEY_ID=xxxxxxxxxx
-S3_SECRET_ACCESS_KEY=xxxxxxxxxx
-S3_ENDPOINT=https://xxxxxxxxxx.r2.cloudflarestorage.com
-S3_BUCKET=lobechat
-S3_PUBLIC_DOMAIN=https://s3-for-lobechat.your-domain.com
-
-# Other environment variables, as needed. You can refer to the environment variables configuration for the client version, making sure not to have ACCESS_CODE.
-# OPENAI_API_KEY=sk-xxxx
-# OPENAI_PROXY_URL=https://api.openai.com/v1
-# OPENAI_MODEL_LIST=...
-```
-
-### Start the lobe-chat-database Docker image
-
-```sh
-docker run -it -d -p 3210:3210 --network pg --env-file lobe-chat.env --name lobe-chat-database lobehub/lobe-chat-database
-```
-
-You can use the following command to check the logs:
-
-```sh
-docker logs -f lobe-chat-database
-```
-
-If you see the following logs in the container, it means it has started successfully:
-
-```log
-[Database] Start to migration...
-✅ database migration pass.
--------------------------------------
- ▲ Next.js 14.x.x
- - Local: http://localhost:3210
- - Network: http://0.0.0.0:3210
-
- ✓ Starting...
- ✓ Ready in 95ms
-```
-
+ ### Create a Postgres Database Instance
+
+ Please create a Postgres database instance with the PGVector plugin according to your needs, for example:
+
+ ```sh
+ docker network create pg
+
+ docker run --name my-postgres --network pg -e POSTGRES_PASSWORD=mysecretpassword -p 5432:5432 -d pgvector/pgvector:pg16
+ ```
+
+ The above command will create a PG instance named `my-postgres` on the network `pg`, where `pgvector/pgvector:pg16` is a Postgres 16 image with the pgvector plugin installed by default.
+
+
+ The pgvector plugin provides vector search capabilities for Postgres, which is an important
+ component for LobeChat to implement RAG.
+
+
+
+ The above command does not specify a persistent storage location for the pg instance, so it is
+ only for testing/demonstration purposes. Please configure persistent storage for production
+ environments.
+
+
+ ### Create a file named `lobe-chat.env` to store environment variables:
+
+ ```shell
+ # Website domain
+ APP_URL=https://your-prod-domain.com
+
+ # DB required environment variables
+ KEY_VAULTS_SECRET=jgwsK28dspyVQoIf8/M3IIHl1h6LYYceSYNXeLpy6uk=
+ # Postgres database connection string
+ # Format: postgres://username:password@host:port/dbname; if your pg instance is a Docker container, use the container name
+ DATABASE_URL=postgres://postgres:mysecretpassword@my-postgres:5432/postgres
+
+ # NEXT_AUTH related, can use auth0, Azure AD, GitHub, Authentik, zitadel, etc. If you have other access requirements, feel free to submit a PR
+ NEXT_AUTH_SECRET=3904039cd41ea1bdf6c93db0db96e250
+ NEXT_AUTH_SSO_PROVIDERS=auth0
+ NEXTAUTH_URL=https://your-prod-domain.com/api/auth
+ AUTH_AUTH0_ID=xxxxxx
+ AUTH_AUTH0_SECRET=cSX_xxxxx
+ AUTH_AUTH0_ISSUER=https://lobe-chat-demo.us.auth0.com
+
+ # S3 related
+ S3_ACCESS_KEY_ID=xxxxxxxxxx
+ S3_SECRET_ACCESS_KEY=xxxxxxxxxx
+ S3_ENDPOINT=https://xxxxxxxxxx.r2.cloudflarestorage.com
+ S3_BUCKET=lobechat
+ S3_PUBLIC_DOMAIN=https://s3-for-lobechat.your-domain.com
+
+ # Other environment variables, as needed. You can refer to the environment variables configuration for the client version, making sure not to have ACCESS_CODE.
+ # OPENAI_API_KEY=sk-xxxx
+ # OPENAI_PROXY_URL=https://api.openai.com/v1
+ # OPENAI_MODEL_LIST=...
+ ```
+
+ ### Start the lobe-chat-database Docker image
+
+ ```sh
+ docker run -it -d -p 3210:3210 --network pg --env-file lobe-chat.env --name lobe-chat-database lobehub/lobe-chat-database
+ ```
+
+ You can use the following command to check the logs:
+
+ ```sh
+ docker logs -f lobe-chat-database
+ ```
+
+ If you see the following logs in the container, it means it has started successfully:
+
+ ```log
+ [Database] Start to migration...
+ ✅ database migration pass.
+ -------------------------------------
+ ▲ Next.js 14.x.x
+ - Local: http://localhost:3210
+ - Network: http://0.0.0.0:3210
+
+ ✓ Starting...
+ ✓ Ready in 95ms
+ ```
diff --git a/docs/self-hosting/server-database/docker.zh-CN.mdx b/docs/self-hosting/server-database/docker.zh-CN.mdx
index d6e9d3103a671..a930d2750d6c0 100644
--- a/docs/self-hosting/server-database/docker.zh-CN.mdx
+++ b/docs/self-hosting/server-database/docker.zh-CN.mdx
@@ -13,10 +13,9 @@ tags:
@@ -31,8 +30,7 @@ tags:
由于无法使用 Docker 环境变量暴露 `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY`，使用 Docker / Docker Compose
部署 LobeChat 时，你不能使用 Clerk 作为登录鉴权服务。
-如果你确实需要 Clerk 作为登录鉴权服务，你可以考虑使用 Vercel 部署或者自行构建镜像。
-
+ 如果你确实需要 Clerk 作为登录鉴权服务，你可以考虑使用 Vercel 部署或者自行构建镜像。
## 在 Linux 服务器上部署
@@ -40,92 +38,90 @@ tags:
以下是在 Linux 服务器上部署 LobeChat DB 版的流程：
-
-### 创建 Postgres 数据库实例
-
-请按照你自己的诉求创建一个带有 PGVector 插件的 Postgres 数据库实例，例如：
-
-```sh
-docker network create pg
-
-docker run --name my-postgres --network pg -e POSTGRES_PASSWORD=mysecretpassword -p 5432:5432 -d pgvector/pgvector:pg16
-```
-
-上述指令会创建一个名为 `my-postgres`，并且网络为 `pg` 的 PG 实例，其中 `pgvector/pgvector:pg16` 是一个 Postgres 16 的镜像，且默认安装了 pgvector 插件。
-
-
- pgvector 插件为 Postgres 提供了向量搜索的能力，是 LobeChat 实现 RAG 的重要构件之一。
-
-
-
- 以上指令得到的 pg
- 实例并没有指定持久化存储位置，因此仅用于测试/演示，生产环境请自行配置持久化存储。
-
-
-### 创建名为 `lobe-chat.env` 文件用于存放环境变量：
-
-```shell
-# 网站域名
-APP_URL=https://your-prod-domain.com
-
-# DB 必须的环境变量
-# 用于加密敏感信息的密钥，可以使用 openssl rand -base64 32 生成
-KEY_VAULTS_SECRET='jgwsK28dspyVQoIf8/M3IIHl1h6LYYceSYNXeLpy6uk='
-# Postgres 数据库连接字符串
-# 格式：postgres://username:password@host:port/dbname，如果你的 pg 实例为 Docker 容器，请使用容器名
-DATABASE_URL=postgres://postgres:mysecretpassword@my-postgres:5432/postgres
-
-# NEXT_AUTH 相关，可以使用 auth0、Azure AD、GitHub、Authentik、zitadel 等，如有其他接入诉求欢迎提 PR
-NEXT_AUTH_SECRET=3904039cd41ea1bdf6c93db0db96e250
-NEXT_AUTH_SSO_PROVIDERS=auth0
-NEXTAUTH_URL=https://your-prod-domain.com/api/auth
-AUTH_AUTH0_ID=xxxxxx
-AUTH_AUTH0_SECRET=cSX_xxxxx
-AUTH_AUTH0_ISSUER=https://lobe-chat-demo.us.auth0.com
-
-# S3 相关
-S3_ACCESS_KEY_ID=xxxxxxxxxx
-S3_SECRET_ACCESS_KEY=xxxxxxxxxx
-# 用于 S3 API 访问的域名
-S3_ENDPOINT=https://xxxxxxxxxx.r2.cloudflarestorage.com
-S3_BUCKET=lobechat
-# 用于外网访问 S3 的公共域名，需配置 CORS
-S3_PUBLIC_DOMAIN=https://s3-for-lobechat.your-domain.com
-# S3_REGION=ap-chengdu # 如果需要指定地域
-
-# 其他环境变量，视需求而定
-# OPENAI_API_KEY=sk-xxxx
-# OPENAI_PROXY_URL=https://api.openai.com/v1
-# OPENAI_MODEL_LIST=...
-# ...
-```
-
-### 启动 lobe-chat-database docker 镜像
-
-```sh
-docker run -it -d -p 3210:3210 --network pg --env-file lobe-chat.env --name lobe-chat-database lobehub/lobe-chat-database
-```
-
-你可以使用下述指令检查日志：
-
-```sh
-docker logs -f lobe-chat-database
-```
-
-如果你在容器中看到了以下日志，则说明已经启动成功：
-
-```log
-[Database] Start to migration...
-✅ database migration pass.
--------------------------------------
- ▲ Next.js 14.x.x
- - Local: http://localhost:3210
- - Network: http://0.0.0.0:3210
-
- ✓ Starting...
- ✓ Ready in 95ms
-```
-
+ ### 创建 Postgres 数据库实例
+
+ 请按照你自己的诉求创建一个带有 PGVector 插件的 Postgres 数据库实例，例如：
+
+ ```sh
+ docker network create pg
+
+ docker run --name my-postgres --network pg -e POSTGRES_PASSWORD=mysecretpassword -p 5432:5432 -d pgvector/pgvector:pg16
+ ```
+
+ 上述指令会创建一个名为 `my-postgres`，并且网络为 `pg` 的 PG 实例，其中 `pgvector/pgvector:pg16` 是一个 Postgres 16 的镜像，且默认安装了 pgvector 插件。
+
+
+ pgvector 插件为 Postgres 提供了向量搜索的能力，是 LobeChat 实现 RAG 的重要构件之一。
+
+
+
+ 以上指令得到的 pg
+ 实例并没有指定持久化存储位置，因此仅用于测试 / 演示，生产环境请自行配置持久化存储。
+
+
+ ### 创建名为 `lobe-chat.env` 文件用于存放环境变量：
+
+ ```shell
+ # 网站域名
+ APP_URL=https://your-prod-domain.com
+
+ # DB 必须的环境变量
+ # 用于加密敏感信息的密钥，可以使用 openssl rand -base64 32 生成
+ KEY_VAULTS_SECRET='jgwsK28dspyVQoIf8/M3IIHl1h6LYYceSYNXeLpy6uk='
+ # Postgres 数据库连接字符串
+ # 格式：postgres://username:password@host:port/dbname，如果你的 pg 实例为 Docker 容器，请使用容器名
+ DATABASE_URL=postgres://postgres:mysecretpassword@my-postgres:5432/postgres
+
+ # NEXT_AUTH 相关，可以使用 auth0、Azure AD、GitHub、Authentik、zitadel 等，如有其他接入诉求欢迎提 PR
+ NEXT_AUTH_SECRET=3904039cd41ea1bdf6c93db0db96e250
+ NEXT_AUTH_SSO_PROVIDERS=auth0
+ NEXTAUTH_URL=https://your-prod-domain.com/api/auth
+ AUTH_AUTH0_ID=xxxxxx
+ AUTH_AUTH0_SECRET=cSX_xxxxx
+ AUTH_AUTH0_ISSUER=https://lobe-chat-demo.us.auth0.com
+
+ # S3 相关
+ S3_ACCESS_KEY_ID=xxxxxxxxxx
+ S3_SECRET_ACCESS_KEY=xxxxxxxxxx
+ # 用于 S3 API 访问的域名
+ S3_ENDPOINT=https://xxxxxxxxxx.r2.cloudflarestorage.com
+ S3_BUCKET=lobechat
+ # 用于外网访问 S3 的公共域名，需配置 CORS
+ S3_PUBLIC_DOMAIN=https://s3-for-lobechat.your-domain.com
+ # S3_REGION=ap-chengdu # 如果需要指定地域
+
+ # 其他环境变量，视需求而定
+ # OPENAI_API_KEY=sk-xxxx
+ # OPENAI_PROXY_URL=https://api.openai.com/v1
+ # OPENAI_MODEL_LIST=...
+ # ...
+ ```
+
+ ### 启动 lobe-chat-database docker 镜像
+
+ ```sh
+ docker run -it -d -p 3210:3210 --network pg --env-file lobe-chat.env --name lobe-chat-database lobehub/lobe-chat-database
+ ```
+
+ 你可以使用下述指令检查日志：
+
+ ```sh
+ docker logs -f lobe-chat-database
+ ```
+
+ 如果你在容器中看到了以下日志，则说明已经启动成功：
+
+ ```log
+ [Database] Start to migration...
+ ✅ database migration pass.
+ -------------------------------------
+ ▲ Next.js 14.x.x
+ - Local: http://localhost:3210
+ - Network: http://0.0.0.0:3210
+
+ ✓ Starting...
+ ✓ Ready in 95ms
+ ```
diff --git a/docs/self-hosting/server-database/dokploy.mdx b/docs/self-hosting/server-database/dokploy.mdx
index 67d80cb25f918..849e403ad201d 100644
--- a/docs/self-hosting/server-database/dokploy.mdx
+++ b/docs/self-hosting/server-database/dokploy.mdx
@@ -11,14 +11,19 @@ tags:
---
# Deploying Server Database Version on Dokploy.
+
This article will detail how to deploy the server database version of LobeChat.
+
## 1. Preparation Work
+
### Deploy Dokploy and configure related settings.
+
```shell
curl -sSL https://dokploy.com/install.sh | sh
```
+
1. Connect your GitHub to Dokploy in the Settings / Git section according to the prompt.
-
+
2. Enter the Projects interface to create a Project.
@@ -26,7 +31,9 @@ curl -sSL https://dokploy.com/install.sh | sh
### Configure S3 Storage Service
+
In the server-side database, we need to configure the S3 storage service to store files. For detailed configuration instructions, please refer to the section [Configure S3 Storage Service](https://lobehub.com/docs/self-hosting/server-database/vercel#3-configure-s-3-storage-service) in the Vercel deployment guide。After the configuration is complete, you will obtain the following environment variables:
+
```shell
S3_ACCESS_KEY_ID=
S3_SECRET_ACCESS_KEY=
@@ -35,28 +42,36 @@ S3_BUCKET=
S3_PUBLIC_DOMAIN=
S3_ENABLE_PATH_STYLE=
```
+
### Configure the Clerk authentication service.
+
Obtain the three environment variables: `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY`, `CLERK_SECRET_KEY`, and `CLERK_WEBHOOK_SECRET`. For detailed configuration steps for Clerk, please refer to the section [Configure Authentication Service](https://lobehub.com/docs/self-hosting/server-database/vercel#2-configure-authentication-service) in the Vercel deployment guide.
+
```shell
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_live_xxxxxxxxxxx
CLERK_SECRET_KEY=sk_live_xxxxxxxxxxxxxxxxxxxxxx
CLERK_WEBHOOK_SECRET=whsec_xxxxxxxxxxxxxxxxxxxxxx
```
+
## 2. Deploying the database on Dokploy
+
Enter the previously created Project, click on Create Service, and select Database. In the Database interface, choose PostgreSQL, then set the database name, user, and password. In the Docker image field, enter `pgvector/pgvector:pg17`, and finally click Create to create the database.
Enter the created database and set an unused port in External Credentials to allow external access; otherwise, LobeChat will not be able to connect to the database.
You can view the Postgres database connection URL in External Host, as shown below:
+
```shell
postgresql://postgres:wAbLxfXSwkxxxxxx@45.577.281.48:5432/postgres
```
+
Finally, click Deploy to deploy the database.
## Deploy LobeChat on Dokploy.
+
Click "Create Service", select "Application", and create the LobeChat application.
@@ -106,6 +121,7 @@ OPENAI_API_KEY=
OPENAI_MODEL_LIST=
OPENAI_PROXY_URL=
```
+
After adding the environment variables and saving, click Deploy to initiate the deployment. You can check the deployment progress and log information under Deployments.
@@ -115,6 +131,7 @@ After a successful deployment, bind your own domain to your LobeChat application
## Check if LobeChat is working properly.
+
Go to your LobeChat website, and if you click on the login button in the upper left corner and the login pop-up appears normally, it means you have configured it successfully. Enjoy it to the fullest!
diff --git a/docs/self-hosting/server-database/dokploy.zh-CN.mdx b/docs/self-hosting/server-database/dokploy.zh-CN.mdx
index 3c7a2b8cbfeec..61c0055e27e3e 100644
--- a/docs/self-hosting/server-database/dokploy.zh-CN.mdx
+++ b/docs/self-hosting/server-database/dokploy.zh-CN.mdx
@@ -10,67 +10,82 @@ tags:
- 身份验证服务
- 环境变量配置
---
-
-# 在 Dokploy 上部署服务端数据库版
-本文将详细介绍如何在 Dokploy 中部署服务端数据库版 LobeChat。
-## 一、准备工作
-### 部署 Dokploy 并进行相关设置
-```shell
-curl -sSL https://dokploy.com/install.sh | sh
-```
-1. 在 Dokploy 的 Settings / Git 处根据提示将 Github 绑定到 Dokploy
-
-
-
-2. 进入 Projects 界面创建一个 Project
-
-
-
-### 配置 S3 存储服务
-在服务端数据库中我们需要配置 S3 存储服务来存储文件，详细配置教程请参考 使用 Vercel 部署中 [配置S3储存服务](https://lobehub.com/zh/docs/self-hosting/server-database/vercel#%E4%B8%89%E3%80%81-%E9%85%8D%E7%BD%AE-s-3-%E5%AD%98%E5%82%A8%E6%9C%8D%E5%8A%A1)。配置完成后你将获得以下环境变量：
-```shell
+
+# 在 Dokploy 上部署服务端数据库版
+
+本文将详细介绍如何在 Dokploy 中部署服务端数据库版 LobeChat。
+
+## 一、准备工作
+
+### 部署 Dokploy 并进行相关设置
+
+```shell
+curl -sSL https://dokploy.com/install.sh | sh
+```
+
+1. 在 Dokploy 的 Settings / Git 处根据提示将 Github 绑定到 Dokploy
+
+
+
+2. 进入 Projects 界面创建一个 Project
+
+
+
+### 配置 S3 存储服务
+
+在服务端数据库中我们需要配置 S3 存储服务来存储文件，详细配置教程请参考 使用 Vercel 部署中 [配置 S3 储存服务](https://lobehub.com/zh/docs/self-hosting/server-database/vercel#%E4%B8%89%E3%80%81-%E9%85%8D%E7%BD%AE-s-3-%E5%AD%98%E5%82%A8%E6%9C%8D%E5%8A%A1)。配置完成后你将获得以下环境变量：
+
+```shell
S3_ACCESS_KEY_ID=
S3_SECRET_ACCESS_KEY=
S3_ENDPOINT=
S3_BUCKET=
S3_PUBLIC_DOMAIN=
-S3_ENABLE_PATH_STYLE=
-```
-### 配置 Clerk 身份验证服务
-获取 `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY` 、`CLERK_SECRET_KEY` 、`CLERK_WEBHOOK_SECRET` 这三个环境变量，Clerk的详细配置流程请参考 使用 Vercel 部署中 [配置身份验证服务](https://lobehub.com/zh/docs/self-hosting/server-database/vercel#二、-配置身份验证服务)
-```shell
+S3_ENABLE_PATH_STYLE=
+```
+
+### 配置 Clerk 身份验证服务
+
+获取 `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY` 、`CLERK_SECRET_KEY` 、`CLERK_WEBHOOK_SECRET` 这三个环境变量，Clerk 的详细配置流程请参考 使用 Vercel 部署中 [配置身份验证服务](https://lobehub.com/zh/docs/self-hosting/server-database/vercel#二、-配置身份验证服务)
+
+```shell
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_live_xxxxxxxxxxx
CLERK_SECRET_KEY=sk_live_xxxxxxxxxxxxxxxxxxxxxx
-CLERK_WEBHOOK_SECRET=whsec_xxxxxxxxxxxxxxxxxxxxxx
-```
-## 二、在 Dokploy 上部署数据库
-进入前面创建的Project，点击 Create Service 选择 Database，在 Database 界面选择 PostgreSQL ，然后设置数据库名、用户、密码，在 Docker image 中填入 `pgvector/pgvector:pg17` 最后点击 Create 创建数据库。
-
-
-
+CLERK_WEBHOOK_SECRET=whsec_xxxxxxxxxxxxxxxxxxxxxx
+```
+
+## 二、在 Dokploy 上部署数据库
+
+进入前面创建的 Project，点击 Create Service 选择 Database，在 Database 界面选择 PostgreSQL ，然后设置数据库名、用户、密码，在 Docker image 中填入 `pgvector/pgvector:pg17` 最后点击 Create 创建数据库。
+
+
+
进入创建的数据库，在 External Credentials 设置一个未被占用的端口，使其能能通过外部访问，否则 LobeChat 将无法连接到该数据库。
-你可以在 External Host 查看 Postgres 数据库连接 URL ，如下：
-```shell
-postgresql://postgres:wAbLxfXSwkxxxxxx@45.577.281.48:5432/postgres
-```
-最后点击 Deploy 部署数据库
-
-
-
-## 在 Dokploy 上部署 LobeChat
-点击 Create Service 选择 Application，创建 LobeChat 应用
-
-
-
-进入创建的 LobeChat 应用，选择你 fork 的 lobe-chat 项目及分支，点击 Save 保存
-
-
-
-切换到 Environment ，在其中填入环境变量，点击保存。
-
-
-
-```shell
+你可以在 External Host 查看 Postgres 数据库连接 URL ，如下：
+
+```shell
+postgresql://postgres:wAbLxfXSwkxxxxxx@45.577.281.48:5432/postgres
+```
+
+最后点击 Deploy 部署数据库
+
+
+
+## 在 Dokploy 上部署 LobeChat
+
+点击 Create Service 选择 Application，创建 LobeChat 应用
+
+
+
+进入创建的 LobeChat 应用，选择你 fork 的 lobe-chat 项目及分支，点击 Save 保存
+
+
+
+切换到 Environment ，在其中填入环境变量，点击保存。
+
+
+
+```shell
# 构建所必需的环境变量
NIXPACKS_PKGS="pnpm bun"
NIXPACKS_INSTALL_CMD="pnpm install"
@@ -105,18 +120,19 @@ S3_ENABLE_PATH_STYLE=
# OpenAI 相关配置
OPENAI_API_KEY=
OPENAI_MODEL_LIST=
-OPENAI_PROXY_URL=
-```
-添加完环境变量并保存后，点击 Deploy 进行部署，你可以在 Deployments 处查看部署进程及日志信息
-
-
-
-部署成功后在 Domains 页面，为你的 LobeChat 应用绑定自己的域名并申请证书。
-
-
-
-## 验证 LobeChat 是否正常工作
-进入你的 LobeChat 网址，如果你点击左上角登录，可以正常显示登录弹窗，那么说明你已经配置成功了，尽情享用吧～
-
-
-
+OPENAI_PROXY_URL=
+```
+
+添加完环境变量并保存后，点击 Deploy 进行部署，你可以在 Deployments 处查看部署进程及日志信息
+
+
+
+部署成功后在 Domains 页面，为你的 LobeChat 应用绑定自己的域名并申请证书。
+
+
+
+## 验证 LobeChat 是否正常工作
+
+进入你的 LobeChat 网址，如果你点击左上角登录，可以正常显示登录弹窗，那么说明你已经配置成功了，尽情享用吧～
+
+
diff --git a/docs/self-hosting/server-database/repocloud.mdx b/docs/self-hosting/server-database/repocloud.mdx
index 55e04e33ed776..ee95f9601f366 100644
--- a/docs/self-hosting/server-database/repocloud.mdx
+++ b/docs/self-hosting/server-database/repocloud.mdx
@@ -17,19 +17,17 @@ If you want to deploy LobeChat Database Edition on RepoCloud, you can follow the
## RepoCloud Deployment Process
+ ### Prepare your OpenAI API Key
-### Prepare your OpenAI API Key
+ Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
-Go to [OpenAI API Key](https://platform.openai.com/account/api-keys) to get your OpenAI API Key.
+ ### One-click to deploy
-### One-click to deploy
+ [](https://repocloud.io/details/?app_id=248)
-[](https://repocloud.io/details/?app_id=248)
+ ### Once deployed, you can start using it
-### Once deployed, you can start using it
-
-### Bind a custom domain (optional)
-
-You can use the subdomain provided by RepoCloud, or choose to bind a custom domain. Currently, the domains provided by RepoCloud have not been contaminated, and most regions can connect directly.
+ ### Bind a custom domain (optional)
+ You can use the subdomain provided by RepoCloud, or choose to bind a custom domain. Currently, the domains provided by RepoCloud have not been contaminated, and most regions can connect directly.
diff --git a/docs/self-hosting/server-database/repocloud.zh-CN.mdx b/docs/self-hosting/server-database/repocloud.zh-CN.mdx
index ae397c86216fd..dd9bfbc7304c0 100644
--- a/docs/self-hosting/server-database/repocloud.zh-CN.mdx
+++ b/docs/self-hosting/server-database/repocloud.zh-CN.mdx
@@ -11,24 +11,22 @@ tags:
# 在 RepoCloud 上部署 LobeChat 数据库版
-如果您想在RepoCloud上部署LobeChat数据库版，可以按照以下步骤进行操作：
+如果您想在 RepoCloud 上部署 LobeChat 数据库版，可以按照以下步骤进行操作：
-## RepoCloud部署流程
+## RepoCloud 部署流程
+ ### 准备您的 OpenAI API 密钥
-### 准备您的 OpenAI API 密钥
+ 请访问 [OpenAI API 密钥](https://platform.openai.com/account/api-keys) 获取您的 OpenAI API 密钥。
-请访问 [OpenAI API密钥](https://platform.openai.com/account/api-keys) 获取您的OpenAI API密钥。
+ ### 一键部署
-### 一键部署
+ [](https://repocloud.io/details/?app_id=248)
-[](https://repocloud.io/details/?app_id=248)
+ ### 部署完成后，您可以开始使用
-### 部署完成后，您可以开始使用
-
-### 绑定自定义域名（可选）
-
-您可以使用RepoCloud提供的子域名，或选择绑定自定义域名。目前，RepoCloud提供的域名尚未被污染，大多数地区可以直接连接。
+ ### 绑定自定义域名（可选）
+ 您可以使用 RepoCloud 提供的子域名，或选择绑定自定义域名。目前，RepoCloud 提供的域名尚未被污染，大多数地区可以直接连接。
diff --git a/docs/self-hosting/server-database/vercel.mdx b/docs/self-hosting/server-database/vercel.mdx
index 909b2448885f6..29a50560399bf 100644
--- a/docs/self-hosting/server-database/vercel.mdx
+++ b/docs/self-hosting/server-database/vercel.mdx
@@ -15,107 +15,90 @@ tags:
This article will detail how to deploy the server database version of LobeChat on Vercel, including: 1) database configuration; 2) identity authentication service configuration; 3) steps for setting up the S3 storage service.
+ Before proceeding, please make sure of the following:
-Before proceeding, please make sure of the following:
-
-- Export all data, as after deploying the server-side database, existing user data cannot be automatically migrated and can only be manually imported after backup!
-- The `ACCESS_CODE` in the environment variables is either unset or cleared!
-- When configuring the environment variables required for the server-side database, make sure to fill in all of them before deployment, otherwise you may encounter database migration issues!
-
+ - Export all data, as after deploying the server-side database, existing user data cannot be automatically migrated and can only be manually imported after backup!
+ - The `ACCESS_CODE` in the environment variables is either unset or cleared!
+ - When configuring the environment variables required for the server-side database, make sure to fill in all of them before deployment, otherwise you may encounter database migration issues!
## 1. Configure the Database
+ ### Prepare the Server Database Instance and Obtain the Connection URL
-### Prepare the Server Database Instance and Obtain the Connection URL
-
-Before deployment, make sure you have prepared a Postgres database instance. You can choose one of the following methods:
-
-- `A.` Use Serverless Postgres instances like Vercel / Neon;
-- `B.` Use self-deployed Postgres instances like Docker.
-
-The configuration for both methods is slightly different, and will be distinguished in the next step.
+ Before deployment, make sure you have prepared a Postgres database instance. You can choose one of the following methods:
-### Add Environment Variables in Vercel
+ - `A.` Use Serverless Postgres instances like Vercel / Neon;
+ - `B.` Use self-deployed Postgres instances like Docker.
-In Vercel's deployment environment variables, add `DATABASE_URL` and other environment variables, and fill in the Postgres database connection URL prepared in the previous step. The typical format for the database connection URL is `postgres://username:password@host:port/database`.
-
-
-
-
-
-
- Please confirm the `Postgres` type provided by your vendor. If it is `Node Postgres`, switch to
- the `Node Postgres` Tab.
-
+ The configuration for both methods is slightly different, and will be distinguished in the next step.
-Variables to be filled for Serverless Postgres are as follows:
+ ### Add Environment Variables in Vercel
-```shell
-# Serverless Postgres DB Url
-DATABASE_URL=
+ In Vercel's deployment environment variables, add `DATABASE_URL` and other environment variables, and fill in the Postgres database connection URL prepared in the previous step. The typical format for the database connection URL is `postgres://username:password@host:port/database`.
-# Specify service mode as server, otherwise it will not enter the server-side database
-NEXT_PUBLIC_SERVICE_MODE=server
-```
+
+
+
+ Please confirm the `Postgres` type provided by your vendor. If it is `Node Postgres`, switch to
+ the `Node Postgres` Tab.
+
-An example of filling in Vercel is as follows:
+ Variables to be filled for Serverless Postgres are as follows:
-
+ ```shell
+ # Serverless Postgres DB Url
+ DATABASE_URL=
-
+ # Specify service mode as server, otherwise it will not enter the server-side database
+ NEXT_PUBLIC_SERVICE_MODE=server
+ ```
-
- Variables to be filled for Node Postgres are as follows:
+ An example of filling in Vercel is as follows:
-```shell
-# Node Postgres DB Url
-DATABASE_URL=
+
+
-# Specify Postgres database driver as node
-DATABASE_DRIVER=node
+
+ Variables to be filled for Node Postgres are as follows:
-# Specify service mode as server, otherwise it will not enter the server-side database
-NEXT_PUBLIC_SERVICE_MODE=server
-```
+ ```shell
+ # Node Postgres DB Url
+ DATABASE_URL=
-An example of filling in Vercel is as follows:
+ # Specify Postgres database driver as node
+ DATABASE_DRIVER=node
-
+ # Specify service mode as server, otherwise it will not enter the server-side database
+ NEXT_PUBLIC_SERVICE_MODE=server
+ ```
-
+ An example of filling in Vercel is as follows:
-
+
+
+
-
If you wish to enable SSL when connecting to the database, please refer to the
[link](https://stackoverflow.com/questions/14021998/using-psql-to-connect-to-postgresql-in-ssl-mode)
for setup instructions.
+
-
+ ### Add the `KEY_VAULTS_SECRET` Environment Variable
-### Add the `KEY_VAULTS_SECRET` Environment Variable
+ After adding the `DATABASE_URL` environment variable for the database, you need to add a `KEY_VAULTS_SECRET` environment variable. This variable is used to encrypt sensitive information such as apikeys stored by users. You can generate a random 32-character string as the key using `openssl rand -base64 32`.
-After adding the DATABASE_URL environment variable for the database, you need to add a `KEY_VAULTS_SECRET` environment variable. This variable is used to encrypt sensitive information such as apikeys stored by users. You can generate a random 32-character string as the key using `openssl rand -base64 32`.
+ ```shell
+ KEY_VAULTS_SECRET=jgwsK28dspyVQoIf8/M3IIHl1h6LYYceSYNXeLpy6uk=
+ ```
-```shell
-KEY_VAULTS_SECRET=jgwsK28dspyVQoIf8/M3IIHl1h6LYYceSYNXeLpy6uk=
-```
+ Make sure to add this to the Vercel environment variables as well.
-Make sure to add this to the Vercel environment variables as well.
-
-### Add the `APP_URL` Environment Variable
-
-Finally, you need to add the `APP_URL` environment variable, which specifies the URL address of the LobeChat application.
+ ### Add the `APP_URL` Environment Variable
+ Finally, you need to add the `APP_URL` environment variable, which specifies the URL address of the LobeChat application.
## 2. Configure Authentication Service
@@ -123,87 +106,63 @@ Finally, you need to add the `APP_URL` environment variable, which specifies the
The server-side database needs to be paired with a user authentication service to function properly. Therefore, the corresponding authentication service needs to be configured.
+ ### Prepare Clerk Authentication Service
-### Prepare Clerk Authentication Service
-
-Go to [Clerk](https://clerk.com?utm_source=lobehub&utm_medium=docs) to register and create an application to obtain the corresponding Public Key and Secret Key.
+ Go to [Clerk](https://clerk.com?utm_source=lobehub\&utm_medium=docs) to register and create an application to obtain the corresponding Public Key and Secret Key.
-
If you are not familiar with Clerk, you can refer to [Authentication
Service-Clerk](/en/docs/self-hosting/advanced/authentication#clerk) for details on using Clerk.
+
-
-
-### Add Public and Private Key Environment Variables in Vercel
+ ### Add Public and Private Key Environment Variables in Vercel
-In Vercel's deployment environment variables, add the `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY` and `CLERK_SECRET_KEY` environment variables. You can click on "API Keys" in the menu, then copy the corresponding values and paste them into Vercel's environment variables.
+ In Vercel's deployment environment variables, add the `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY` and `CLERK_SECRET_KEY` environment variables. You can click on "API Keys" in the menu, then copy the corresponding values and paste them into Vercel's environment variables.
-
+
-The environment variables required for this step are as follows:
+ The environment variables required for this step are as follows:
-```shell
-NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_live_xxxxxxxxxxx
-CLERK_SECRET_KEY=sk_live_xxxxxxxxxxxxxxxxxxxxxx
-```
+ ```shell
+ NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_live_xxxxxxxxxxx
+ CLERK_SECRET_KEY=sk_live_xxxxxxxxxxxxxxxxxxxxxx
+ ```
-Add the above variables to Vercel:
+ Add the above variables to Vercel:
-
+
-### Create and Configure Webhook in Clerk
+ ### Create and Configure Webhook in Clerk
-Since we let Clerk fully handle user authentication and management, we need Clerk to notify our application and store data in the database when there are changes in the user's lifecycle (create, update, delete). We achieve this requirement through the Webhook provided by Clerk.
+ Since we let Clerk fully handle user authentication and management, we need Clerk to notify our application and store data in the database when there are changes in the user's lifecycle (create, update, delete). We achieve this requirement through the Webhook provided by Clerk.
-We need to add an endpoint in Clerk's Webhooks to inform Clerk to send notifications to this endpoint when a user's information changes.
+ We need to add an endpoint in Clerk's Webhooks to inform Clerk to send notifications to this endpoint when a user's information changes.
-
+
-Fill in the endpoint with the URL of your Vercel project, such as `https://your-project.vercel.app/api/webhooks/clerk`. Then, subscribe to events by checking the three user events (`user.created`, `user.deleted`, `user.updated`), and click create.
+ Fill in the endpoint with the URL of your Vercel project, such as `https://your-project.vercel.app/api/webhooks/clerk`. Then, subscribe to events by checking the three user events (`user.created`, `user.deleted`, `user.updated`), and click create.
-
The `https://` in the URL is essential to maintain the integrity of the URL.
+
-
-
-
-
-### Add Webhook Secret to Vercel Environment Variables
+
-After creation, you can find the secret of this Webhook in the bottom right corner:
+ ### Add Webhook Secret to Vercel Environment Variables
-
+ After creation, you can find the secret of this Webhook in the bottom right corner:
-The environment variable corresponding to this secret is `CLERK_WEBHOOK_SECRET`:
+
-```shell
-CLERK_WEBHOOK_SECRET=whsec_xxxxxxxxxxxxxxxxxxxxxx
-```
+ The environment variable corresponding to this secret is `CLERK_WEBHOOK_SECRET`:
-Add it to Vercel's environment variables:
+ ```shell
+ CLERK_WEBHOOK_SECRET=whsec_xxxxxxxxxxxxxxxxxxxxxx
+ ```
-
+ Add it to Vercel's environment variables:
+
By completing these steps, you have successfully configured the Clerk authentication service. Next, we will configure the S3 storage service.
@@ -213,155 +172,121 @@ By completing these steps, you have successfully configured the Clerk authentica
In the server-side database, we need to configure the S3 storage service to store files.
-
-In this article, S3 refers to a compatible S3 storage solution, which supports object storage systems that comply with the Amazon S3 API. Common examples include Cloudflare R2, Alibaba Cloud OSS, etc., all of which support S3-compatible APIs.
-
+ In this article, S3 refers to a compatible S3 storage solution, which supports object storage systems that comply with the Amazon S3 API. Common examples include Cloudflare R2, Alibaba Cloud OSS, etc., all of which support S3-compatible APIs.
+ ### Configure and Obtain S3 Bucket
-### Configure and Obtain S3 Bucket
+ You need to go to your S3 service provider (such as AWS S3, Cloudflare R2, etc.) and create a new storage bucket. The following steps will use Cloudflare R2 as an example to explain the creation process.
-You need to go to your S3 service provider (such as AWS S3, Cloudflare R2, etc.) and create a new storage bucket. The following steps will use Cloudflare R2 as an example to explain the creation process.
+ The interface of Cloudflare R2 is shown below:
-The interface of Cloudflare R2 is shown below:
+
-
+ When creating a storage bucket, specify its name and then click create.
-When creating a storage bucket, specify its name and then click create.
+
-
+ ### Obtain Environment Variables for the Bucket
-### Obtain Environment Variables for the Bucket
+ In the settings of the R2 storage bucket, you can view the bucket configuration information:
-In the settings of the R2 storage bucket, you can view the bucket configuration information:
+
-
+ The corresponding environment variables are:
-The corresponding environment variables are:
+ ```shell
+ # Storage bucket name
+ S3_BUCKET=lobechat
+ # Storage bucket request endpoint (note that the path in this link includes the bucket name, which must be removed, or use the link provided on the S3 API token application page)
+ S3_ENDPOINT=https://0b33a03b5c993fd2f453379dc36558e5.r2.cloudflarestorage.com
+ # Public access domain for the storage bucket
+ S3_PUBLIC_DOMAIN=https://s3-for-lobechat.your-domain.com
+ ```
-```shell
-# Storage bucket name
-S3_BUCKET=lobechat
-# Storage bucket request endpoint (note that the path in this link includes the bucket name, which must be removed, or use the link provided on the S3 API token application page)
-S3_ENDPOINT=https://0b33a03b5c993fd2f453379dc36558e5.r2.cloudflarestorage.com
-# Public access domain for the storage bucket
-S3_PUBLIC_DOMAIN=https://s3-for-lobechat.your-domain.com
-```
+
+ `S3_ENDPOINT` must have its path removed, otherwise uploaded files will not be accessible
+
-
- `S3_ENDPOINT` must have its path removed, otherwise uploaded files will not be accessible
-
+ ### Obtain S3 Key Environment Variables
-### Obtain S3 Key Environment Variables
+ You need to obtain the access key for S3 so that the LobeChat server has permission to access the S3 storage service. In R2, you can configure the access key in the account details:
-You need to obtain the access key for S3 so that the LobeChat server has permission to access the S3 storage service. In R2, you can configure the access key in the account details:
+
-
+ Click the button in the upper right corner to create an API token and enter the create API Token page.
-Click the button in the upper right corner to create an API token and enter the create API Token page.
+
-
+ Since our server-side database needs to read and write to the S3 storage service, the permission needs to be set to `Object Read and Write`, then click create.
-Since our server-side database needs to read and write to the S3 storage service, the permission needs to be set to `Object Read and Write`, then click create.
+
-
+ After creation, you can see the corresponding S3 API token.
-After creation, you can see the corresponding S3 API token.
+
-
+ The corresponding environment variables are:
-The corresponding environment variables are:
+ ```shell
+ S3_ACCESS_KEY_ID=9998d6757e276cf9f1edbd325b7083a6
+ S3_SECRET_ACCESS_KEY=55af75d8eb6b99f189f6a35f855336ea62cd9c4751a5cf4337c53c1d3f497ac2
+ ```
-```shell
-S3_ACCESS_KEY_ID=9998d6757e276cf9f1edbd325b7083a6
-S3_SECRET_ACCESS_KEY=55af75d8eb6b99f189f6a35f855336ea62cd9c4751a5cf4337c53c1d3f497ac2
-```
+ ### Adding Corresponding Environment Variables in Vercel
-### Adding Corresponding Environment Variables in Vercel
+ The steps to obtain the required environment variables may vary for different S3 service providers, but the obtained environment variables should be consistent:
-The steps to obtain the required environment variables may vary for different S3 service providers, but the obtained environment variables should be consistent:
+
+ The `https://` in the URL is essential and must be maintained for the completeness of the URL.
+
-
- The `https://` in the URL is essential and must be maintained for the completeness of the URL.
-
+ ```shell
+ # S3 Keys
+ S3_ACCESS_KEY_ID=9998d6757e276cf9f1edbd325b7083a6
+ S3_SECRET_ACCESS_KEY=55af75d8eb6b99f189f6a35f855336ea62cd9c4751a5cf4337c53c1d3f497ac2
-```shell
-# S3 Keys
-S3_ACCESS_KEY_ID=9998d6757e276cf9f1edbd325b7083a6
-S3_SECRET_ACCESS_KEY=55af75d8eb6b99f189f6a35f855336ea62cd9c4751a5cf4337c53c1d3f497ac2
+ # Bucket name
+ S3_BUCKET=lobechat
+ # Bucket request endpoint
+ S3_ENDPOINT=https://0b33a03b5c993fd2f453379dc36558e5.r2.cloudflarestorage.com
+ # Public domain for bucket access
+ S3_PUBLIC_DOMAIN=https://s3-dev.your-domain.com
-# Bucket name
-S3_BUCKET=lobechat
-# Bucket request endpoint
-S3_ENDPOINT=https://0b33a03b5c993fd2f453379dc36558e5.r2.cloudflarestorage.com
-# Public domain for bucket access
-S3_PUBLIC_DOMAIN=https://s3-dev.your-domain.com
-
-# Bucket region, such as us-west-1, generally not required, but some providers may need to configure
-# S3_REGION=us-west-1
-```
-
-Then, insert the above environment variables into Vercel's environment variables:
+ # Bucket region, such as us-west-1, generally not required, but some providers may need to configure
+ # S3_REGION=us-west-1
+ ```
-
+ Then, insert the above environment variables into Vercel's environment variables:
-### Configuring Cross-Origin Resource Sharing (CORS)
+
-Since S3 storage services are often on a separate domain, cross-origin access needs to be configured.
+ ### Configuring Cross-Origin Resource Sharing (CORS)
-In R2, you can find the CORS configuration in the bucket settings:
+ Since S3 storage services are often on a separate domain, cross-origin access needs to be configured.
-
+ In R2, you can find the CORS configuration in the bucket settings:
-Add a CORS rule to allow requests from your domain (in this case, `https://your-project.vercel.app`):
+
-
+ Add a CORS rule to allow requests from your domain (in this case, `https://your-project.vercel.app`):
-Example configuration:
+
-```json
-[
- {
- "AllowedOrigins": ["https://your-project.vercel.app"],
- "AllowedMethods": ["GET", "PUT", "HEAD", "POST", "DELETE"],
- "AllowedHeaders": ["*"]
- }
-]
-```
+ Example configuration:
-After configuring, click save.
+ ```json
+ [
+ {
+ "AllowedOrigins": ["https://your-project.vercel.app"],
+ "AllowedMethods": ["GET", "PUT", "HEAD", "POST", "DELETE"],
+ "AllowedHeaders": ["*"]
+ }
+ ]
+ ```
+ After configuring, click save.
## Four, Deployment and Verification
@@ -371,27 +296,17 @@ After completing the steps above, the configuration of the server-side database
### Redeploy the latest commit
-After configuring the environment variables, you need to redeploy the latest commit and wait for the deployment to complete.
-
-
+ After configuring the environment variables, you need to redeploy the latest commit and wait for the deployment to complete.
-### Check if the features are working properly
+
-If you click on the login button in the top left corner and the login popup appears normally, then you have successfully configured it. Enjoy using it\~
+ ### Check if the features are working properly
-
+ If you click on the login button in the top left corner and the login popup appears normally, then you have successfully configured it. Enjoy using it\~
-
+
+
## Appendix
diff --git a/docs/self-hosting/server-database/vercel.zh-CN.mdx b/docs/self-hosting/server-database/vercel.zh-CN.mdx
index bbcc7a86531b1..22502646ba35e 100644
--- a/docs/self-hosting/server-database/vercel.zh-CN.mdx
+++ b/docs/self-hosting/server-database/vercel.zh-CN.mdx
@@ -17,37 +17,34 @@ tags:
本文将详细介绍如何在 Vercel 中部署服务端数据库版 LobeChat，包括： 1）数据库配置；2）身份验证服务配置；3） S3 存储服务的设置步骤。
- 进行后续操作前，请务必确认以下事项： -
- 导出所有数据，部署服务端数据库后，原有用户数据无法自动迁移，只能提前备份后进行手动导入！ -
- 环境变量中的`ACCESS_CODE`未设置或已清除！ -
- 配置服务端数据库所需要的环境变量时，需全部填入后再进行部署，否则可能遭遇数据库迁移问题！
+ 进行后续操作前，请务必确认以下事项：
+
+ - 导出所有数据，部署服务端数据库后，原有用户数据无法自动迁移，只能提前备份后进行手动导入！
+ - 环境变量中的`ACCESS_CODE`未设置或已清除！
+ - 配置服务端数据库所需要的环境变量时，需全部填入后再进行部署，否则可能遭遇数据库迁移问题！
## 一、 配置数据库
+ ### 准备服务端数据库实例，获取连接 URL
-### 准备服务端数据库实例，获取连接 URL
-
-在部署之前，请确保你已经准备好 Postgres 数据库实例，你可以选择以下任一方式：
+ 在部署之前，请确保你已经准备好 Postgres 数据库实例，你可以选择以下任一方式：
-- `A.` 使用 Vercel / Neon 等 Serverless Postgres 实例；
-- `B.` 使用 Docker 等自部署 Postgres 实例。
+ - `A.` 使用 Vercel / Neon 等 Serverless Postgres 实例；
+ - `B.` 使用 Docker 等自部署 Postgres 实例。
-两者的配置方式略有不同，在下一步会有所区分。
+ 两者的配置方式略有不同，在下一步会有所区分。
-### 在 Vercel 中添加环境变量
+ ### 在 Vercel 中添加环境变量
-在 Vercel 的部署环境变量中，添加 `DATABASE_URL` 等环境变量，将上一步准备好的 Postgres 数据库连接 URL 填入其中。数据库连接 URL 的通常格式为 `postgres://username:password@host:port/database`。
+ 在 Vercel 的部署环境变量中，添加 `DATABASE_URL` 等环境变量，将上一步准备好的 Postgres 数据库连接 URL 填入其中。数据库连接 URL 的通常格式为 `postgres://username:password@host:port/database`。
-
-
请确认您的供应商所提供的 `Postgres` 类型，若为 `Node Postgres`，请切换到 `Node Postgres` Tab 。
-
-
+
Serverless Postgres 需要填写的变量如下：
@@ -61,9 +58,8 @@ tags:
在 Vercel 中填写的示例如下：
-
-
-
+
+
Node Postgres 需要填写的变量如下：
@@ -81,31 +77,28 @@ tags:
在 Vercel 中填写的示例如下：
-
+
+
+
-
+
+ 如果希望连接数据库时启用 SSL
+ ，请自行参考[链接](https://stackoverflow.com/questions/14021998/using-psql-to-connect-to-postgresql-in-ssl-mode)进行设置
+
-
+ ### 添加 `KEY_VAULTS_SECRET` 环境变量
-
- 如果希望连接数据库时启用 SSL
- ，请自行参考[链接](https://stackoverflow.com/questions/14021998/using-psql-to-connect-to-postgresql-in-ssl-mode)进行设置
-
-
-### 添加 `KEY_VAULTS_SECRET` 环境变量
-
-在完成数据库 DATABASE_URL 环境变量添加后，需要添加一个 `KEY_VAULTS_SECRET` 环境变量。该变量用于加密用户存储的 apikey 等敏感信息。你可以使用 `openssl rand -base64 32` 生成一个随机的 32 位字符串作为密钥。
-
-```shell
-KEY_VAULTS_SECRET=jgwsK28dspyVQoIf8/M3IIHl1h6LYYceSYNXeLpy6uk=
-```
+ 在完成数据库 `DATABASE_URL` 环境变量添加后，需要添加一个 `KEY_VAULTS_SECRET` 环境变量。该变量用于加密用户存储的 apikey 等敏感信息。你可以使用 `openssl rand -base64 32` 生成一个随机的 32 位字符串作为密钥。
-同样需要将其添加到 Vercel 环境变量中。
+ ```shell
+ KEY_VAULTS_SECRET=jgwsK28dspyVQoIf8/M3IIHl1h6LYYceSYNXeLpy6uk=
+ ```
-### 添加 `APP_URL` 环境变量
+ 同样需要将其添加到 Vercel 环境变量中。
-该部分最后需要添加 APP_URL 环境变量，用于指定LobeChat 应用的 URL 地址。
+ ### 添加 `APP_URL` 环境变量
+ 该部分最后需要添加 `APP_URL` 环境变量，用于指定 LobeChat 应用的 URL 地址。
## 二、 配置身份验证服务
@@ -113,81 +106,61 @@ KEY_VAULTS_SECRET=jgwsK28dspyVQoIf8/M3IIHl1h6LYYceSYNXeLpy6uk=
服务端数据库需要搭配用户身份验证服务才可以正常使用。因此需要配置对应的身份验证服务。
+ ### 准备 Clerk 身份验证服务
-### 准备 Clerk 身份验证服务
+ 前往 [Clerk](https://clerk.com?utm_source=lobehub\&utm_medium=docs) 注册并创建应用，获取相应的 Public Key 和 Secret Key。
-前往 [Clerk](https://clerk.com?utm_source=lobehub&utm_medium=docs) 注册并创建应用，获取相应的 Public Key 和 Secret Key。
+
+ 如果对 Clerk 不太了解，可以查阅
+ [身份验证服务 - Clerk](/zh/docs/self-hosting/advanced/authentication#clerk) 了解 Clerk 的使用详情。
+
-
- 如果对 Clerk 不太了解，可以查阅
- [身份验证服务-Clerk](/zh/docs/self-hosting/advanced/authentication#clerk) 了解 Clerk 的使用详情。
-
+ ### 在 Vercel 中添加公、私钥环境变量
-### 在 Vercel 中添加公、私钥环境变量
+ 在 Vercel 的部署环境变量中，添加 `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY` 和 `CLERK_SECRET_KEY` 环境变量。你可以在菜单中点击「API Keys」，然后复制对应的值填入 Vercel 的环境变量中。
-在 Vercel 的部署环境变量中，添加 `NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY` 和 `CLERK_SECRET_KEY` 环境变量。你可以在菜单中点击「API Keys」，然后复制对应的值填入 Vercel 的环境变量中。
+
-
+ 此步骤所需的环境变量如下：
-此步骤所需的环境变量如下：
+ ```shell
+ NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_live_xxxxxxxxxxx
+ CLERK_SECRET_KEY=sk_live_xxxxxxxxxxxxxxxxxxxxxx
+ ```
-```shell
-NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_live_xxxxxxxxxxx
-CLERK_SECRET_KEY=sk_live_xxxxxxxxxxxxxxxxxxxxxx
-```
-
-添加上述变量到 Vercel 中：
-
-
+ 添加上述变量到 Vercel 中：
-### 在 Clerk 中创建并配置 Webhook
+
-由于我们让 Clerk 完全接管用户鉴权与管理，因此我们需要在 Clerk 用户生命周期变更时（创建、更新、删除）中通知我们的应用并存储落库。我们通过 Clerk 提供的 Webhook 来实现这一诉求。
+ ### 在 Clerk 中创建并配置 Webhook
-我们需要在 Clerk 的 Webhooks 中添加一个端点（Endpoint），告诉 Clerk 当用户发生变更时，向这个端点发送通知。
+ 由于我们让 Clerk 完全接管用户鉴权与管理，因此我们需要在 Clerk 用户生命周期变更时（创建、更新、删除）中通知我们的应用并存储落库。我们通过 Clerk 提供的 Webhook 来实现这一诉求。
-
+ 我们需要在 Clerk 的 Webhooks 中添加一个端点（Endpoint），告诉 Clerk 当用户发生变更时，向这个端点发送通知。
-在 endppint 中填写你的 Vercel 项目的 URL，如 `https://your-project.vercel.app/api/webhooks/clerk`。然后在订阅事件（Subscribe to events）中，勾选 user 的三个事件（`user.created` 、`user.deleted`、`user.updated`），然后点击创建。
+
-URL的`https://`不可缺失，须保持URL的完整性
+ 在 endppint 中填写你的 Vercel 项目的 URL，如 `https://your-project.vercel.app/api/webhooks/clerk`。然后在订阅事件（Subscribe to events）中，勾选 user 的三个事件（`user.created` 、`user.deleted`、`user.updated`），然后点击创建。
-
+ URL 的`https://`不可缺失，须保持 URL 的完整性
-### 将 Webhook 秘钥添加到 Vercel 环境变量
+
-创建完毕后，可以在右下角找到该 Webhook 的秘钥：
+ ### 将 Webhook 秘钥添加到 Vercel 环境变量
-
+ 创建完毕后，可以在右下角找到该 Webhook 的秘钥：
-这个秘钥所对应的环境变量名为 `CLERK_WEBHOOK_SECRET`：
+
-```shell
-CLERK_WEBHOOK_SECRET=whsec_xxxxxxxxxxxxxxxxxxxxxx
-```
+ 这个秘钥所对应的环境变量名为 `CLERK_WEBHOOK_SECRET`：
-将其添加到 Vercel 的环境变量中：
+ ```shell
+ CLERK_WEBHOOK_SECRET=whsec_xxxxxxxxxxxxxxxxxxxxxx
+ ```
-
+ 将其添加到 Vercel 的环境变量中：
+
这样，你已经成功配置了 Clerk 身份验证服务。接下来我们将配置 S3 存储服务。
@@ -197,149 +170,118 @@ CLERK_WEBHOOK_SECRET=whsec_xxxxxxxxxxxxxxxxxxxxxx
在服务端数据库中我们需要配置 S3 存储服务来存储文件。
- 在本文，S3所指代的是指兼容 S3 存储方案，即支持 Amazon S3 API 的对象存储系统，常见例如 Cloudflare
+ 在本文，S3 所指代的是指兼容 S3 存储方案，即支持 Amazon S3 API 的对象存储系统，常见例如 Cloudflare
R2 、阿里云 OSS 等均支持 S3 兼容 API。
+ ### 配置并获取 S3 存储桶
- ### 配置并获取 S3 存储桶
+ 你需要前往你的 S3 服务提供商（如 AWS S3、Cloudflare R2 等）并创建一个新的存储桶（Bucket）。接下来以 Cloudflare R2 为例，介绍创建流程。
- 你需要前往你的 S3 服务提供商（如 AWS S3、Cloudflare R2 等）并创建一个新的存储桶（Bucket）。接下来以 Cloudflare R2 为例，介绍创建流程。
+ 下图是 Cloudflare R2 的界面：
- 下图是 Cloudflare R2 的界面：
+
-
+ 创建存储桶时将指定其名称，然后点击创建。
- 创建存储桶时将指定其名称，然后点击创建。
-
+
- ### 获取存储桶相关环境变量
+ ### 获取存储桶相关环境变量
- 在 R2 存储桶的设置中，可以看到桶配置的信息：
+ 在 R2 存储桶的设置中，可以看到桶配置的信息：
-
+
-其对应的环境变量为：
+ 其对应的环境变量为：
-```shell
-# 存储桶的名称
-S3_BUCKET=lobechat
-# 存储桶的请求端点(注意此处链接的路径带存储桶名称，必须删除该路径，或使用申请 S3 API token 页面所提供的链接)
-S3_ENDPOINT=https://0b33a03b5c993fd2f453379dc36558e5.r2.cloudflarestorage.com
-# 存储桶对外的访问域名
-S3_PUBLIC_DOMAIN=https://s3-for-lobechat.your-domain.com
-```
+ ```shell
+ # 存储桶的名称
+ S3_BUCKET=lobechat
+ # 存储桶的请求端点(注意此处链接的路径带存储桶名称，必须删除该路径，或使用申请 S3 API token 页面所提供的链接)
+ S3_ENDPOINT=https://0b33a03b5c993fd2f453379dc36558e5.r2.cloudflarestorage.com
+ # 存储桶对外的访问域名
+ S3_PUBLIC_DOMAIN=https://s3-for-lobechat.your-domain.com
+ ```
-`S3_ENDPOINT`必须删除其路径，否则会无法访问所上传文件
+ `S3_ENDPOINT`必须删除其路径，否则会无法访问所上传文件
- ### 获取 S3 密钥环境变量
+ ### 获取 S3 密钥环境变量
- 你需要获取 S3 的访问密钥，以便 LobeChat 的服务端有权限访问 S3 存储服务。在 R2 中，你可以在账户详情中配置访问密钥：
+ 你需要获取 S3 的访问密钥，以便 LobeChat 的服务端有权限访问 S3 存储服务。在 R2 中，你可以在账户详情中配置访问密钥：
-
+
- 点击右上角按钮创建 API token，进入创建 API Token 页面
+ 点击右上角按钮创建 API token，进入创建 API Token 页面
-
+
- 鉴于我们的服务端数据库需要读写 S3 存储服务，因此权限需要选择`对象读与写`，然后点击创建。
+ 鉴于我们的服务端数据库需要读写 S3 存储服务，因此权限需要选择`对象读与写`，然后点击创建。
-
+
- 创建完成后，就可以看到对应的 S3 API token
+ 创建完成后，就可以看到对应的 S3 API token
-
+
- 其对应的环境变量为：
+ 其对应的环境变量为：
-```shell
-S3_ACCESS_KEY_ID=9998d6757e276cf9f1edbd325b7083a6
-S3_SECRET_ACCESS_KEY=55af75d8eb6b99f189f6a35f855336ea62cd9c4751a5cf4337c53c1d3f497ac2
-```
-
-### 在 Vercel 中添加对应的环境变量
+ ```shell
+ S3_ACCESS_KEY_ID=9998d6757e276cf9f1edbd325b7083a6
+ S3_SECRET_ACCESS_KEY=55af75d8eb6b99f189f6a35f855336ea62cd9c4751a5cf4337c53c1d3f497ac2
+ ```
- 不同 S3 服务商获取所需环境变量的步骤可能有所不同，但最终获得到的环境变量应该都是一致的：
+ ### 在 Vercel 中添加对应的环境变量
-URL的`https://`不可缺失，须保持URL的完整性
+ 不同 S3 服务商获取所需环境变量的步骤可能有所不同，但最终获得到的环境变量应该都是一致的：
-```shell
-# S3 秘钥
-S3_ACCESS_KEY_ID=9998d6757e276cf9f1edbd325b7083a6
-S3_SECRET_ACCESS_KEY=55af75d8eb6b99f189f6a35f855336ea62cd9c4751a5cf4337c53c1d3f497ac2
+ URL 的`https://`不可缺失，须保持 URL 的完整性
-# 存储桶的名称
-S3_BUCKET=lobechat
-# 存储桶的请求端点
-S3_ENDPOINT=https://0b33a03b5c993fd2f453379dc36558e5.r2.cloudflarestorage.com
-# 存储桶对外的访问域名
-S3_PUBLIC_DOMAIN=https://s3-dev.your-domain.com
+ ```shell
+ # S3 秘钥
+ S3_ACCESS_KEY_ID=9998d6757e276cf9f1edbd325b7083a6
+ S3_SECRET_ACCESS_KEY=55af75d8eb6b99f189f6a35f855336ea62cd9c4751a5cf4337c53c1d3f497ac2
-# 桶的区域，如 us-west-1，一般来说不需要添加，但某些服务商则需要配置
-# S3_REGION=us-west-1
-```
+ # 存储桶的名称
+ S3_BUCKET=lobechat
+ # 存储桶的请求端点
+ S3_ENDPOINT=https://0b33a03b5c993fd2f453379dc36558e5.r2.cloudflarestorage.com
+ # 存储桶对外的访问域名
+ S3_PUBLIC_DOMAIN=https://s3-dev.your-domain.com
-然后将上述环境变量填入 Vercel 的环境变量中：
+ # 桶的区域，如 us-west-1，一般来说不需要添加，但某些服务商则需要配置
+ # S3_REGION=us-west-1
+ ```
-
+ 然后将上述环境变量填入 Vercel 的环境变量中：
- ### 配置跨域
+
- 由于 S3 存储服务往往是一个独立的网址，因此需要配置跨域访问。
+ ### 配置跨域
- 在 R2 中，你可以在存储桶的设置中找到跨域配置：
+ 由于 S3 存储服务往往是一个独立的网址，因此需要配置跨域访问。
-
+ 在 R2 中，你可以在存储桶的设置中找到跨域配置：
- 添加跨域规则，允许你的域名（在上文是 `https://your-project.vercel.app`）来源的请求：
+
-
+ 添加跨域规则，允许你的域名（在上文是 `https://your-project.vercel.app`）来源的请求：
-示例配置如下：
+
-```json
-[
- {
- "AllowedOrigins": ["https://your-project.vercel.app"],
- "AllowedMethods": ["GET", "PUT", "HEAD", "POST", "DELETE"],
- "AllowedHeaders": ["*"]
- }
-]
-```
+ 示例配置如下：
-配置后点击保存即可。
+ ```json
+ [
+ {
+ "AllowedOrigins": ["https://your-project.vercel.app"],
+ "AllowedMethods": ["GET", "PUT", "HEAD", "POST", "DELETE"],
+ "AllowedHeaders": ["*"]
+ }
+ ]
+ ```
+ 配置后点击保存即可。
## 四、部署并验证
@@ -349,27 +291,17 @@ S3_PUBLIC_DOMAIN=https://s3-dev.your-domain.com
### 重新部署最新的 commit
-配置好环境变量后，你需要重新部署最新的 commit，并等待部署完成。
-
-
+ 配置好环境变量后，你需要重新部署最新的 commit，并等待部署完成。
-### 检查功能是否正常
+
-如果你点击左上角登录，可以正常显示登录弹窗，那么说明你已经配置成功了，尽情享用吧~
+ ### 检查功能是否正常
-
+ 如果你点击左上角登录，可以正常显示登录弹窗，那么说明你已经配置成功了，尽情享用吧～
-
+
+
## 附录
diff --git a/docs/self-hosting/server-database/zeabur.mdx b/docs/self-hosting/server-database/zeabur.mdx
index e5e4fa411adb4..5f9dc5503c191 100644
--- a/docs/self-hosting/server-database/zeabur.mdx
+++ b/docs/self-hosting/server-database/zeabur.mdx
@@ -32,47 +32,45 @@ The template on Zeabur includes 4 services:
Here is the process for deploying the LobeChat server database version on Zeabur:
+ ### Go to the template page on Zeabur
-### Go to the template page on Zeabur
+ Go to the [LobeChat Database template page](https://zeabur.com/templates/RRSPSD) on Zeabur and click on the "Deploy" button.
-Go to the [LobeChat Database template page](https://zeabur.com/templates/RRSPSD) on Zeabur and click on the "Deploy" button.
+ ### Fill in the required environment variables
-### Fill in the required environment variables
+ After you click on the "Deploy" button, you will see a modal pop-up where you can fill in the required environment variables.
-After you click on the "Deploy" button, you will see a modal pop-up where you can fill in the required environment variables.
+ Here are the environment variables you need to fill in:
-Here are the environment variables you need to fill in:
+ - OpenAI API key: Your OpenAI API key to get responses from OpenAI.
-- OpenAI API key: Your OpenAI API key to get responses from OpenAI.
+ - LobeChat Domain: A free subdomain with `.zeabur.app` suffix.
-- LobeChat Domain: A free subdomain with `.zeabur.app` suffix.
+ - MinIO Public Domain: A free subdomain with `.zeabur.app` suffix for yout MinIO web port to enable public access for the uploaded files.
-- MinIO Public Domain: A free subdomain with `.zeabur.app` suffix for yout MinIO web port to enable public access for the uploaded files.
+ - Logto Console Domain: A free subdomain with `.zeabur.app` suffix for your Logto console.
-- Logto Console Domain: A free subdomain with `.zeabur.app` suffix for your Logto console.
+ - Logto API Domain: A free subdomain with `.zeabur.app` suffix for your Logto api.
-- Logto API Domain: A free subdomain with `.zeabur.app` suffix for your Logto api.
+ ### Select a region and deploy
-### Select a region and deploy
+ After you fill all the required environment variables, select a region where you want to deploy your LobeChat Database and click on the "Deploy" button.
-After you fill all the required environment variables, select a region where you want to deploy your LobeChat Database and click on the "Deploy" button.
+ You will see another modal pop-up where you can see the deployment progress.
-You will see another modal pop-up where you can see the deployment progress.
+ ### Configure Logto
-### Configure Logto
+ After the deployment is done, you need to configure your Logto service to enable authrization.
-After the deployment is done, you need to configure your Logto service to enable authrization.
+ Access your Logto console with the console domain you just binded, and then create a `Next.js 14(App router)` application to get the client ID and client secret, and fill in the cors and callback URLs. You can check [this document](../advanced/auth.mdx) for a more detailed guide.
-Access your Logto console with the console domain you just binded, and then create a `Next.js 14(App router)` application to get the client ID and client secret, and fill in the cors and callback URLs. You can check [this document](../advanced/auth.mdx) for a more detailed guide.
+ Fill in those variables into your LobeChat service on Zeabur, here is a more detailed guide for [editing environment variables on Zeabur](https://zeabur.com/docs/deploy/variables).
-Fill in those variables into your LobeChat service on Zeabur, here is a more detailed guide for [editing environment variables on Zeabur](https://zeabur.com/docs/deploy/variables).
+ For detailed configuration of Logto, refer to [this document](/docs/self-hosting/advanced/auth/next-auth/logto).
-For detailed configuration of Logto, refer to [this document](/docs/self-hosting/advanced/auth/next-auth/logto).
+ ### Access your LobeChat Instance
-### Access your LobeChat Instance
-
-Press on the `LobeChat-Database` and you can see the public domain you just created, click on it to access your LobeChat Database.
-
-You can also bind a custom domain for your services if you want, here is a guide on how to [bind a custom domain on Zeabur](https://zeabur.com/docs/deploy/domain-binding).
+ Press on the `LobeChat-Database` and you can see the public domain you just created, click on it to access your LobeChat Database.
+ You can also bind a custom domain for your services if you want, here is a guide on how to [bind a custom domain on Zeabur](https://zeabur.com/docs/deploy/domain-binding).
diff --git a/docs/self-hosting/server-database/zeabur.zh-CN.mdx b/docs/self-hosting/server-database/zeabur.zh-CN.mdx
index b1c572c4ef52c..a1d7fab2ae83a 100644
--- a/docs/self-hosting/server-database/zeabur.zh-CN.mdx
+++ b/docs/self-hosting/server-database/zeabur.zh-CN.mdx
@@ -29,41 +29,39 @@ tags:
这里是在 Zeabur 上部署 LobeChat 服务器数据库版的流程：
+ ### 前往 Zeabur 上的模板页面
-### 前往 Zeabur 上的模板页面
+ 前往 [Zeabur 上的 LobeChat 数据库模板页面](https://zeabur.com/templates/RRSPSD) 并点击 "Deploy" 按钮。
-前往 [Zeabur 上的 LobeChat 数据库模板页面](https://zeabur.com/templates/RRSPSD) 并点击 "Deploy" 按钮。
+ ### 填写必要的环境变量
-### 填写必要的环境变量
+ 在你点击 “部署 “按钮后，你会看到一个模态弹窗，你可以在这里填写必要的环境变量。
-在你点击“部署“按钮后，你会看到一个模态弹窗，你可以在这里填写必要的环境变量。
+ 以下是你需要填写的环境变量：
-以下是你需要填写的环境变量：
+ - OpenAI API key: 你的 OpenAI API key 用于获取模型的访问权限。
+ - LobeChat Domain: 一个免费的 `.zeabur.app` 后缀的域名。
+ - MinIO Public Domain: 一个免费的 `.zeabur.app` 后缀的域名为了暴露 MinIO 服务以公开访问资源。
+ - Logto Console Domain: 一个免费的 `.zeabur.app` 后缀的域名来访问 Logto 的控制台。
+ - Logto API Domain: 一个免费的 `.zeabur.app` 后缀的域名来访问 Logto 的 API。
-- OpenAI API key: 你的 OpenAI API key 用于获取模型的访问权限。
-- LobeChat Domain: 一个免费的 `.zeabur.app` 后缀的域名。
-- MinIO Public Domain: 一个免费的 `.zeabur.app` 后缀的域名为了暴露 MinIO 服务以公开访问资源。
-- Logto Console Domain: 一个免费的 `.zeabur.app` 后缀的域名来访问 Logto 的控制台。
-- Logto API Domain: 一个免费的 `.zeabur.app` 后缀的域名来访问 Logto 的 API。
+ ### 选择一个区域并部署
-### 选择一个区域并部署
+ 在你填写完所有必要的环境变量后，选择一个你想要部署 LobeChat 数据库的区域并点击 “部署” 按钮。
-在你填写完所有必要的环境变量后，选择一个你想要部署 LobeChat 数据库的区域并点击“部署”按钮。
+ 你会看到另一个模态弹窗，你可以在这里看到部署的进度。
-你会看到另一个模态弹窗，你可以在这里看到部署的进度。
+ ### 配置 Logto
-### 配置 Logto
+ 当部署完成后，你会被自动导航到你在 Zeabur 控制台上刚刚创建的项目。你需要再进一步配置你的 Logto 服务。
-当部署完成后，你会被自动导航到你在 Zeabur 控制台上刚刚创建的项目。你需要再进一步配置你的 Logto 服务。
+ 使用你刚绑定的域名来访问你的 Logto 控制台，创建一个新项目以获得对应的客户端 ID 与密钥，将它们填入你的 LobeChat 服务的变量中。关于如何填入变量，可以参照 [Zeabur 的官方文档](https://zeabur.com/docs/deploy/variables)。
-使用你刚绑定的域名来访问你的 Logto 控制台，创建一个新项目以获得对应的客户端 ID 与密钥，将它们填入你的 LobeChat 服务的变量中。关于如何填入变量，可以参照 [Zeabur 的官方文档](https://zeabur.com/docs/deploy/variables)。
+ Logto 的详细配置可以参考[这篇文档](/zh/docs/self-hosting/advanced/auth/next-auth/logto)。
-Logto 的详细配置可以参考[这篇文档](/zh/docs/self-hosting/advanced/auth/next-auth/logto)。
+ ### 访问你的 LobeChat
-### 访问你的 LobeChat
-
-按下 `LobeChat-Database` 你会看到你刚刚创建的公共域名，点击它以访问你的 LobeChat 数据库。
-
-你可以选择绑定一个自定义域名，这里有一个关于如何在 Zeabur 上[绑定自定义域名](https://zeabur.com/docs/deploy/domain-binding)的指南。
+ 按下 `LobeChat-Database` 你会看到你刚刚创建的公共域名，点击它以访问你的 LobeChat 数据库。
+ 你可以选择绑定一个自定义域名，这里有一个关于如何在 Zeabur 上[绑定自定义域名](https://zeabur.com/docs/deploy/domain-binding)的指南。
diff --git a/docs/self-hosting/start.mdx b/docs/self-hosting/start.mdx
index c1af7e400f8e9..4f5aa21e8a69f 100644
--- a/docs/self-hosting/start.mdx
+++ b/docs/self-hosting/start.mdx
@@ -35,8 +35,5 @@ You can follow the guide below for quick deployment of LobeChat:
For users who are already familiar with LobeChat or need cross-device synchronization, you can deploy a version with a server-side database to access a more complete and powerful LobeChat.
-
+
diff --git a/docs/self-hosting/start.zh-CN.mdx b/docs/self-hosting/start.zh-CN.mdx
index 914a06aca81e2..9643918dd81bf 100644
--- a/docs/self-hosting/start.zh-CN.mdx
+++ b/docs/self-hosting/start.zh-CN.mdx
@@ -11,15 +11,17 @@ tags:
- Vercel
- 阿里云计算巢
- 个性化
+ - 腾讯云
+ - 腾讯轻量云
---
# 构建属于自己的 Lobe Chat
-LobeChat 支持多种部署平台，包括 Vercel、Docker、 Docker Compose 和 阿里云计算巢 等，你可以选择适合自己的部署平台进行部署，构建属于自己的 Lobe Chat。
+LobeChat 支持多种部署平台，包括 Vercel、Docker、 Docker Compose 、阿里云计算巢 和腾讯轻量云 等，你可以选择适合自己的部署平台进行部署，构建属于自己的 Lobe Chat。
## 快速部署
-对于第一次了解 LobeChat 的用户，我们推荐使用客户端数据库的模式快速部署，该模式的优势是一行指令/一个按钮即可快捷完成部署，便于你快速上手与体验 LobeChat。
+对于第一次了解 LobeChat 的用户，我们推荐使用客户端数据库的模式快速部署，该模式的优势是一行指令 / 一个按钮即可快捷完成部署，便于你快速上手与体验 LobeChat。
你可以通过以下指南快速部署 LobeChat：
diff --git a/docs/usage/agents/agent-organization.mdx b/docs/usage/agents/agent-organization.mdx
index 477dcccec2264..3ecdeff625d75 100644
--- a/docs/usage/agents/agent-organization.mdx
+++ b/docs/usage/agents/agent-organization.mdx
@@ -14,11 +14,7 @@ tags:
# Assistant Organization Guide
-
+
LobeChat provides a rich variety of AI assistant resources. Users can easily add various assistants through the assistant market, offering a wide range of application scenarios for AI applications.
@@ -30,39 +26,27 @@ Firstly, LobeChat's AI assistants support organization through grouping. You can
### Assistant Settings
-
+
- In the menu of an individual assistant, selecting the `Move to Group` option can quickly categorize the assistant into the specified group.
- If you don't find the group you want, you can choose `Add Group` to quickly create a new group.
### Group Settings
-
+
- In the group menu, you can quickly create a new assistant under that group.
- Clicking the `Group Management` button allows you to `rename`, `delete`, `sort`, and perform other operations on all groups.
## Assistant Search
-
+
- At the top of the assistant list, you can use the assistant search function to easily locate the assistant you need using keywords.
## Assistant Pinning
-
+
- In the assistant menu, you can use the `Pin` function to pin the assistant to the top.
- After pinning an assistant, a pinned area will appear at the top of the assistant list, displaying all pinned assistants.
diff --git a/docs/usage/agents/agent-organization.zh-CN.mdx b/docs/usage/agents/agent-organization.zh-CN.mdx
index 5ce6aa52e4e60..5bff869c49e5e 100644
--- a/docs/usage/agents/agent-organization.zh-CN.mdx
+++ b/docs/usage/agents/agent-organization.zh-CN.mdx
@@ -12,11 +12,7 @@ tags:
# 助手组织指南
-
+
LobeChat 提供了丰富的 AI 助手资源，用户可以通过助手市场方便地添加各类助手，为 AI 应用提供了广泛的应用场景。
@@ -28,39 +24,27 @@ LobeChat 提供了丰富的 AI 助手资源，用户可以通过助手市场方
### 助手设置
-
+
- 在单个助手的菜单中，选择`移动到分组`选项可以快速将该助手归类到指定分组。
- 如果没有你想要的分组，可以选择`添加分组`，快速创建一个新的分组。
### 分组设置
-
+
- 在分组菜单中，可以快速在该分组下新建助手
- 点击`分组管理`按钮可以对所有分组进行`重命名`、`删除`、`排序`等操作。
## 助手搜索
-
+
- 在助手列表的顶部，您可以通过助手搜索功能，方便地使用关键词定位到您所需的助手。
## 助手固定
-
+
- 在助手菜单中，你可以使用`固定`功能将该助手固定在顶部。
- 固定助手后，助手列表的上方将出现一个固定区域，显示所有已固定的助手列表。
diff --git a/docs/usage/agents/concepts.mdx b/docs/usage/agents/concepts.mdx
index 3589579784516..61cba73950938 100644
--- a/docs/usage/agents/concepts.mdx
+++ b/docs/usage/agents/concepts.mdx
@@ -21,8 +21,8 @@ In the official ChatGPT application, there is only the concept of "topics." As s
However, in our usage, we have found that this model has many issues. For example, the information indexing of historical conversations is too scattered. Additionally, when dealing with repetitive tasks, it is difficult to have a stable entry point. For instance, if I want ChatGPT to help me translate a document, in this model, I would need to constantly create new topics and then set up the translation prompt I had previously created. When there are high-frequency tasks, this will result in a very inefficient interaction format.
@@ -34,8 +34,8 @@ Therefore, in LobeChat, we have introduced the concept of **Agents**. An agent i
At the same time, we have integrated topics into each agent. The benefit of this approach is that each agent has an independent topic list. You can choose the corresponding agent based on the current task and quickly switch between historical conversation records. This method is more in line with users' habits in common chat software, improving interaction efficiency.
diff --git a/docs/usage/agents/concepts.zh-CN.mdx b/docs/usage/agents/concepts.zh-CN.mdx
index d76dcad0f2dd5..cfee63f897629 100644
--- a/docs/usage/agents/concepts.zh-CN.mdx
+++ b/docs/usage/agents/concepts.zh-CN.mdx
@@ -18,8 +18,8 @@ tags:
但在我们的使用过程中其实会发现这种模式存在很多问题，比如历史对话的信息索引过于分散问题，同时当处理一些重复任务时很难有一个稳定的入口，比如我希望有一个稳定的入口可以让 ChatGPT 帮助我翻译文档，在这个模式下，我需要不断新建新的话题同时再设置我之前创建好的翻译 Prompt 设定，当有高频任务存在时，这将是一个效率很低的交互形式。
@@ -31,8 +31,8 @@ tags:
与此同时，我们将话题索引到每个助手内部。这样做的好处是，每个助手都有一个独立的话题列表，你可以根据当前任务选择对应的助手，并快速切换历史对话记录。这种方式更符合用户对常见聊天软件的使用习惯，提高了交互的效率。
diff --git a/docs/usage/agents/custom-agent.mdx b/docs/usage/agents/custom-agent.mdx
index c34eac5a68fe8..34a36e2e9ef7c 100644
--- a/docs/usage/agents/custom-agent.mdx
+++ b/docs/usage/agents/custom-agent.mdx
@@ -24,8 +24,8 @@ If you are a beginner in Prompt writing, you might want to browse the assistant
## `B` Create a custom assistant
diff --git a/docs/usage/agents/custom-agent.zh-CN.mdx b/docs/usage/agents/custom-agent.zh-CN.mdx
index 2011e0708e7b2..62b6dc7a2a50f 100644
--- a/docs/usage/agents/custom-agent.zh-CN.mdx
+++ b/docs/usage/agents/custom-agent.zh-CN.mdx
@@ -22,8 +22,8 @@ tags:
## `B` 通过新建自定义助手
diff --git a/docs/usage/agents/model.mdx b/docs/usage/agents/model.mdx
index b5131fb0cad61..e8056cbbae7c6 100644
--- a/docs/usage/agents/model.mdx
+++ b/docs/usage/agents/model.mdx
@@ -43,10 +43,10 @@ This parameter controls the randomness of the model's output. The higher the val
### `top_p`
-Top_p is also a sampling parameter, but it differs from temperature in its sampling method. Before outputting, the model generates a bunch of tokens, and these tokens are ranked based on their quality. In the top-p sampling mode, the candidate word list is dynamic, and tokens are selected from the tokens based on a percentage. Top_p introduces randomness in token selection, allowing other high-scoring tokens to have a chance of being selected, rather than always choosing the highest-scoring one.
+`top_p` is also a sampling parameter, but it differs from temperature in its sampling method. Before outputting, the model generates a bunch of tokens, and these tokens are ranked based on their quality. In the top-p sampling mode, the candidate word list is dynamic, and tokens are selected from the tokens based on a percentage. Top\_p introduces randomness in token selection, allowing other high-scoring tokens to have a chance of being selected, rather than always choosing the highest-scoring one.
- Top\_p is similar to randomness, and it is generally not recommended to change it together with
+ `top_p` is similar to randomness, and it is generally not recommended to change it together with
the randomness of temperature.
@@ -75,5 +75,21 @@ It is a mechanism that penalizes frequently occurring new vocabulary in the text
- `-2.0` When the morning news started broadcasting, I found that my TV now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now now **(The highest frequency word is "now", accounting for 44.79%)**
- `-1.0` He always watches the news in the early morning, in front of the TV watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch watch **(The highest frequency word is "watch", accounting for 57.69%)**
- `0.0` When the morning sun poured into the small diner, a tired postman appeared at the door, carrying a bag of letters in his hands. The owner warmly prepared a breakfast for him, and he started sorting the mail while enjoying his breakfast. **(The highest frequency word is "of", accounting for 8.45%)**
-- `1.0` A girl in deep sleep was woken up by a warm ray of sunshine, she saw the first ray of morning light, surrounded by birdsong and flowers, everything was full of vitality. \_ (The highest frequency word is "of", accounting for 5.45%)
-- `2.0` Every morning, he would sit on the balcony to have breakfast. Under the soft setting sun, everything looked very peaceful. However, one day, when he was about to pick up his breakfast, an optimistic little bird flew by, bringing him a good mood for the day. \_ (The highest frequency word is "of", accounting for 4.94%)
+- `1.0` A girl in deep sleep was woken up by a warm ray of sunshine, she saw the first ray of morning light, surrounded by birdsong and flowers, everything was full of vitality. (The highest frequency word is "of", accounting for 5.45%)
+- `2.0` Every morning, he would sit on the balcony to have breakfast. Under the soft setting sun, everything looked very peaceful. However, one day, when he was about to pick up his breakfast, an optimistic little bird flew by, bringing him a good mood for the day. (The highest frequency word is "of", accounting for 4.94%)
+
+
+
+### `reasoning_effort`
+
+The `reasoning_effort` parameter controls the strength of the reasoning process. This setting affects the depth of reasoning the model performs when generating a response. The available values are **`low`**, **`medium`**, and **`high`**, with the following meanings:
+
+- **low**: Lower reasoning effort, resulting in faster response times. Suitable for scenarios where quick responses are needed, but it may sacrifice some reasoning accuracy.
+- **medium** (default): Balances reasoning accuracy and response speed, suitable for most scenarios.
+- **high**: Higher reasoning effort, producing more detailed and complex responses, but slower response times and greater token consumption.
+
+By adjusting the `reasoning_effort` parameter, you can find an appropriate balance between response speed and reasoning depth based on your needs. For example, in conversational scenarios, if fast responses are a priority, you can choose low reasoning effort; if more complex analysis or reasoning is needed, you can opt for high reasoning effort.
+
+
+ This parameter is only applicable to reasoning models, such as OpenAI's `o1`, `o1-mini`, `o3-mini`, etc.
+
diff --git a/docs/usage/agents/model.zh-CN.mdx b/docs/usage/agents/model.zh-CN.mdx
index 95264e4f4e910..578dd4096ce5a 100644
--- a/docs/usage/agents/model.zh-CN.mdx
+++ b/docs/usage/agents/model.zh-CN.mdx
@@ -41,9 +41,9 @@ LLM 看似很神奇，但本质还是一个概率问题，神经网络根据输
### `top_p`
-核采样 top_p 也是采样参数，跟 temperature 不一样的采样方式。模型在输出之前，会生成一堆 token，这些 token 根据质量高低排名，核采样模式中候选词列表是动态的，从 tokens 里按百分比选择候选词。 top_p 为选择 token 引入了随机性，让其他高分的 token 有被选择的机会，不会总是选最高分的。
+核采样 `top_p` 也是采样参数，跟 temperature 不一样的采样方式。模型在输出之前，会生成一堆 token，这些 token 根据质量高低排名，核采样模式中候选词列表是动态的，从 tokens 里按百分比选择候选词。 top\_p 为选择 token 引入了随机性，让其他高分的 token 有被选择的机会，不会总是选最高分的。
-top\_p 与随机性类似，一般来说不建议和随机性 temperature 一起更改
+`top_p` 与随机性类似，一般来说不建议和随机性 temperature 一起更改
@@ -67,8 +67,24 @@ Presence Penalty 参数可以看作是对生成文本中重复内容的一种惩
是一种机制，通过对文本中频繁出现的新词汇施加惩罚，以减少模型重复同一词语的可能性，值越大，越有可能降低重复字词。
-- `-2.0` 当早间新闻开始播出，我发现我家电视现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在 _（频率最高的词是 “现在”，占比 44.79%）_
-- `-1.0` 他总是在清晨看新闻，在电视前看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看 _（频率最高的词是 “看”，占比 57.69%）_
+- `-2.0` 当早间新闻开始播出，我发现我家电视现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在现在 *（频率最高的词是 “现在”，占比 44.79%）*
+- `-1.0` 他总是在清晨看新闻，在电视前看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看看 *（频率最高的词是 “看”，占比 57.69%）*
- `0.0` 当清晨的阳光洒进小餐馆时，一名疲倦的邮递员出现在门口，他的手中提着一袋信件。店主热情地为他准备了一份早餐，他在享用早餐的同时开始整理邮件。**（频率最高的词是 “的”，占比 8.45%）**
-- `1.0` 一个深度睡眠的女孩被一阵温暖的阳光唤醒，她看到了早晨的第一缕阳光，周围是鸟语花香，一切都充满了生机。_（频率最高的词是 “的”，占比 5.45%）_
-- `2.0` 每天早上，他都会在阳台上坐着吃早餐。在柔和的夕阳照耀下，一切看起来都非常宁静。然而有一天，当他准备端起早餐的时候，一只乐观的小鸟飞过，给他带来了一天的好心情。 _（频率最高的词是 “的”，占比 4.94%）_
+- `1.0` 一个深度睡眠的女孩被一阵温暖的阳光唤醒，她看到了早晨的第一缕阳光，周围是鸟语花香，一切都充满了生机。*（频率最高的词是 “的”，占比 5.45%）*
+- `2.0` 每天早上，他都会在阳台上坐着吃早餐。在柔和的夕阳照耀下，一切看起来都非常宁静。然而有一天，当他准备端起早餐的时候，一只乐观的小鸟飞过，给他带来了一天的好心情。 *（频率最高的词是 “的”，占比 4.94%）*
+
+
+
+### `reasoning_effort`
+
+`reasoning_effort` 参数用于控制推理过程的强度。此参数的设置会影响模型在生成回答时的推理深度。可选值包括 **`low`**、**`medium`** 和 **`high`**，具体含义如下：
+
+- **low（低）**：推理强度较低，生成速度较快，适用于需要快速响应的场景，但可能牺牲一定的推理精度。
+- **medium（中，默认值）**：平衡推理精度与响应速度，适用于大多数场景。
+- **high（高）**：推理强度较高，生成更为详细和复杂的回答，但响应时间较长，且消耗更多的 Token。
+
+通过调整 `reasoning_effort` 参数，可以根据需求在生成速度与推理深度之间找到适合的平衡。例如，在对话场景中，如果更关注快速响应，可以选择低推理强度；如果需要更复杂的分析或推理，可以选择高推理强度。
+
+
+ 该参数仅适用于推理模型，如 OpenAI 的 `o1`、`o1-mini`、`o3-mini` 等。
+
diff --git a/docs/usage/agents/prompt.mdx b/docs/usage/agents/prompt.mdx
index 6d38cf6b787e0..9a9e09413b45a 100644
--- a/docs/usage/agents/prompt.mdx
+++ b/docs/usage/agents/prompt.mdx
@@ -30,7 +30,7 @@ Generative AI is very useful, but it requires human guidance. In most cases, gen
Let's look at a basic discussion prompt example:
-> _"What are the most urgent environmental issues facing our planet, and what actions can individuals take to help address these issues?"_
+> *"What are the most urgent environmental issues facing our planet, and what actions can individuals take to help address these issues?"*
We can convert it into a simple prompt for the assistant to answer the following questions: placed at the front.
@@ -54,17 +54,16 @@ The second prompt generates longer output and better structure. The use of the t
There are several ways to improve the quality and effectiveness of prompts:
-- **Be Clear About Your Needs:** The model's output will strive to meet your needs, so if your needs are not clear, the output may not meet expectations.
-- **Use Correct Grammar and Spelling:** The model will try to mimic your language style, so if your language style is problematic, the output may also be problematic.
-- **Provide Sufficient Contextual Information:** The model will generate output based on the contextual information you provide, so if the information is insufficient, it may not produce the desired results.
-
+ - **Be Clear About Your Needs:** The model's output will strive to meet your needs, so if your needs are not clear, the output may not meet expectations.
+ - **Use Correct Grammar and Spelling:** The model will try to mimic your language style, so if your language style is problematic, the output may also be problematic.
+ - **Provide Sufficient Contextual Information:** The model will generate output based on the contextual information you provide, so if the information is insufficient, it may not produce the desired results.
After formulating effective prompts for discussing issues, you now need to refine the generated results. This may involve adjusting the output to fit constraints such as word count or combining concepts from different generated results.
A simple method of iteration is to generate multiple outputs and review them to understand the concepts and structures being used. Once the outputs have been evaluated, you can select the most suitable ones and combine them into a coherent response. Another iterative method is to start small and **gradually expand**. This requires more than one prompt: an initial prompt for drafting the initial one or two paragraphs, followed by additional prompts to expand on the content already written. Here is a potential philosophical discussion prompt:
-> _"Is mathematics an invention or a discovery? Use careful reasoning to explain your answer."_
+> *"Is mathematics an invention or a discovery? Use careful reasoning to explain your answer."*
Add it to a simple prompt as follows:
diff --git a/docs/usage/agents/prompt.zh-CN.mdx b/docs/usage/agents/prompt.zh-CN.mdx
index 31578925bf490..5c2e7de6cf4eb 100644
--- a/docs/usage/agents/prompt.zh-CN.mdx
+++ b/docs/usage/agents/prompt.zh-CN.mdx
@@ -24,7 +24,7 @@ tags:
让我们看一个基本的讨论问题的例子:
-> _"我们星球面临的最紧迫的环境问题是什么，个人可以采取哪些措施来帮助解决这些问题？"_
+> *"我们星球面临的最紧迫的环境问题是什么，个人可以采取哪些措施来帮助解决这些问题？"*
我们可以将其转化为简单的助手提示，将回答以下问题：放在前面。
@@ -50,17 +50,16 @@ tags:
提升 prompt 质量和效果的方法主要有以下几点：
-- **尽量明确你的需求：** 模型的输出会尽可能满足你的需求，所以如果你的需求不明确，输出可能会不如预期。
-- **使用正确的语法和拼写：** 模型会尽可能模仿你的语言风格，所以如果你的语言风格有问题，输出可能也会有问题。
-- **提供足够的上下文信息：** 模型会根据你提供的上下文信息生成输出，所以如果你提供的上下文信息不足，可能无法生成你想要的结果。
-
+ - **尽量明确你的需求：** 模型的输出会尽可能满足你的需求，所以如果你的需求不明确，输出可能会不如预期。
+ - **使用正确的语法和拼写：** 模型会尽可能模仿你的语言风格，所以如果你的语言风格有问题，输出可能也会有问题。
+ - **提供足够的上下文信息：** 模型会根据你提供的上下文信息生成输出，所以如果你提供的上下文信息不足，可能无法生成你想要的结果。
在为讨论问题制定有效的提示后，您现在需要细化生成的结果。这可能涉及到调整输出以符合诸如字数等限制，或将不同生成的结果的概念组合在一起。
迭代的一个简单方法是生成多个输出并查看它们，以了解正在使用的概念和结构。一旦评估了输出，您就可以选择最合适的输出并将它们组合成一个连贯的回答。另一种迭代的方法是逐步开始，然后**逐步扩展**。这需要不止一个提示：一个起始提示，用于撰写最初的一两段，然后是其他提示，以扩展已经写过的内容。以下是一个潜在的哲学讨论问题:
-> _"数学是发明还是发现？用仔细的推理来解释你的答案。"_
+> *"数学是发明还是发现？用仔细的推理来解释你的答案。"*
将其添加到一个简单的提示中，如下所示:
diff --git a/docs/usage/agents/topics.mdx b/docs/usage/agents/topics.mdx
index 1c8f2965faa24..88048a8dfc9cd 100644
--- a/docs/usage/agents/topics.mdx
+++ b/docs/usage/agents/topics.mdx
@@ -19,8 +19,8 @@ tags:
borderless
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/279602496-fd72037a-735e-4cc2-aa56-2994bceaba81.png'
- }
+'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/279602496-fd72037a-735e-4cc2-aa56-2994bceaba81.png'
+}
/>
- **Save Topic:** During a conversation, if you want to save the current context and start a new topic, you can click the save button next to the send button.
diff --git a/docs/usage/agents/topics.zh-CN.mdx b/docs/usage/agents/topics.zh-CN.mdx
index 2a903387f9e33..bdf15271bf5fd 100644
--- a/docs/usage/agents/topics.zh-CN.mdx
+++ b/docs/usage/agents/topics.zh-CN.mdx
@@ -17,8 +17,8 @@ tags:
borderless
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/279602496-fd72037a-735e-4cc2-aa56-2994bceaba81.png'
- }
+'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/279602496-fd72037a-735e-4cc2-aa56-2994bceaba81.png'
+}
/>
- **保存话题：** 在聊天过程中，如果想要保存当前上下文并开启新的话题，可以点击发送按钮旁边的保存按钮。
diff --git a/docs/usage/features/agent-market.mdx b/docs/usage/features/agent-market.mdx
index fca028fb494e6..803769058dc84 100644
--- a/docs/usage/features/agent-market.mdx
+++ b/docs/usage/features/agent-market.mdx
@@ -19,8 +19,8 @@ tags:
borderless
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/268670869-f1ffbf66-42b6-42cf-a937-9ce1f8328514.png'
- }
+'https://github.com/user-attachments/assets/b3ab6e35-4fbc-468d-af10-e3e0c687350f'
+}
/>
In LobeChat's Assistant Market, creators can discover a vibrant and innovative community that brings together numerous carefully designed assistants. These assistants not only play a crucial role in work scenarios but also provide great convenience in the learning process. Our market is not just a showcase platform, but also a collaborative space. Here, everyone can contribute their wisdom and share their personally developed assistants.
diff --git a/docs/usage/features/agent-market.zh-CN.mdx b/docs/usage/features/agent-market.zh-CN.mdx
index 91d6927109bb3..38f113643abb5 100644
--- a/docs/usage/features/agent-market.zh-CN.mdx
+++ b/docs/usage/features/agent-market.zh-CN.mdx
@@ -18,8 +18,8 @@ tags:
alt={'助手市场'}
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/268670869-f1ffbf66-42b6-42cf-a937-9ce1f8328514.png'
- }
+'https://github.com/user-attachments/assets/b3ab6e35-4fbc-468d-af10-e3e0c687350f'
+}
/>
在 LobeChat 的助手市场中，创作者们可以发现一个充满活力和创新的社区，它汇聚了众多精心设计的助手，这些助手不仅在工作场景中发挥着重要作用，也在学习过程中提供了极大的便利。我们的市场不仅是一个展示平台，更是一个协作的空间。在这里，每个人都可以贡献自己的智慧，分享个人开发的助手。
diff --git a/docs/usage/features/artifacts.mdx b/docs/usage/features/artifacts.mdx
new file mode 100644
index 0000000000000..dcbe9236a4cb5
--- /dev/null
+++ b/docs/usage/features/artifacts.mdx
@@ -0,0 +1,23 @@
+---
+title: Artifacts Support
+description: >-
+ Discover the power of Claude Artifacts for dynamic content creation and
+ visualization.
+tags:
+ - Claude Artifacts
+ - LobeChat
+ - AI Interaction
+ - Dynamic Content
+---
+
+# Artifacts Support
+
+
+
+Experience the power of Claude Artifacts, now integrated into LobeChat. This revolutionary feature expands the boundaries of AI-human interaction, enabling real-time creation and visualization of diverse content formats.
+
+Create and visualize with unprecedented flexibility:
+
+- Generate and display dynamic SVG graphics
+- Build and render interactive HTML pages in real-time
+- Produce professional documents in multiple formats
diff --git a/docs/usage/features/artifacts.zh-CN.mdx b/docs/usage/features/artifacts.zh-CN.mdx
new file mode 100644
index 0000000000000..07b2b3e7f49f2
--- /dev/null
+++ b/docs/usage/features/artifacts.zh-CN.mdx
@@ -0,0 +1,22 @@
+---
+title: 支持白板 (Artifacts)
+description: 体验 LobeChat 的 Claude Artifacts，实时创建和可视化内容。
+tags:
+ - Claude Artifacts
+ - LobeChat
+ - 实时创作
+ - 动态 SVG
+ - 交互式 HTML
+---
+
+# 支持白板 (Artifacts)
+
+
+
+体验集成于 LobeChat 的 Claude Artifacts 能力。这项革命性功能突破了 AI 人机交互的边界，让您能够实时创建和可视化各种格式的内容。
+
+以前所未有的灵活度进行创作与可视化：
+
+- 生成并展示动态 SVG 图形
+- 实时构建与渲染交互式 HTML 页面
+- 输出多种格式的专业文档
diff --git a/docs/usage/features/auth.mdx b/docs/usage/features/auth.mdx
index ba9afb9719525..4bc91b4ee6570 100644
--- a/docs/usage/features/auth.mdx
+++ b/docs/usage/features/auth.mdx
@@ -17,11 +17,7 @@ tags:
# Support Multi-User Management
-
+
In modern applications, user management and identity verification are essential functions. To meet the diverse needs of different users, LobeChat provides two main user authentication and management solutions: `next-auth` and `Clerk`. Whether you are looking for simple user registration and login or need advanced multi-factor authentication and user management, LobeChat can flexibly accommodate your requirements.
diff --git a/docs/usage/features/auth.zh-CN.mdx b/docs/usage/features/auth.zh-CN.mdx
index 983c0c7b4dc75..32dda43ee6648 100644
--- a/docs/usage/features/auth.zh-CN.mdx
+++ b/docs/usage/features/auth.zh-CN.mdx
@@ -12,11 +12,7 @@ tags:
# 身份验证系统 / 多用户管理支持
-
+
在现代应用中，用户管理和身份验证是至关重要的功能。为满足不同用户的多样化需求，LobeChat 提供了两种主要的用户认证和管理方案：`next-auth` 和 `Clerk`。无论您是追求简便的用户注册登录，还是需要更高级的多因素认证和用户管理，LobeChat 都可以灵活实现。
diff --git a/docs/usage/features/branching-conversations.mdx b/docs/usage/features/branching-conversations.mdx
new file mode 100644
index 0000000000000..1cc5d4f135a2a
--- /dev/null
+++ b/docs/usage/features/branching-conversations.mdx
@@ -0,0 +1,21 @@
+---
+title: Branching Conversations
+description: Explore dynamic AI chats with Branching Conversations for deeper interactions.
+tags:
+ - Branching Conversations
+ - AI Chat
+ - Dynamic Conversations
+---
+
+# Branching Conversations
+
+
+
+Introducing a more natural and flexible way to chat with AI. With Branch Conversations, your discussions can flow in multiple directions, just like human conversations do. Create new conversation branches from any message, giving you the freedom to explore different paths while preserving the original context.
+
+Choose between two powerful modes:
+
+- **Continuation Mode:** Seamlessly extend your current discussion while maintaining valuable context
+- **Standalone Mode:** Start fresh with a new topic based on any previous message
+
+This groundbreaking feature transforms linear conversations into dynamic, tree-like structures, enabling deeper exploration of ideas and more productive interactions.
diff --git a/docs/usage/features/branching-conversations.zh-CN.mdx b/docs/usage/features/branching-conversations.zh-CN.mdx
new file mode 100644
index 0000000000000..16caf102f55a0
--- /dev/null
+++ b/docs/usage/features/branching-conversations.zh-CN.mdx
@@ -0,0 +1,21 @@
+---
+title: 分支对话
+description: 探索分支对话功能，提升 AI 交流的自然性与灵活性。
+tags:
+ - 分支对话
+ - AI 交流
+ - 对话模式
+---
+
+# 分支对话
+
+
+
+为您带来更自然、更灵活的 AI 对话方式。通过分支对话功能，您的讨论可以像人类对话一样自然延伸。在任意消息处创建新的对话分支，让您在保留原有上下文的同时，自由探索不同的对话方向。
+
+两种强大模式任您选择：
+
+- **延续模式**：无缝延展当前讨论，保持宝贵的对话上下文
+- **独立模式**：基于任意历史消息，开启全新话题探讨
+
+这项突破性功能将线性对话转变为动态的树状结构，让您能够更深入地探索想法，实现更高效的互动体验。
diff --git a/docs/usage/features/cot.mdx b/docs/usage/features/cot.mdx
new file mode 100644
index 0000000000000..f901182774e11
--- /dev/null
+++ b/docs/usage/features/cot.mdx
@@ -0,0 +1,18 @@
+---
+title: Chain of Thought
+description: >-
+ Explore AI's decision-making with Chain of Thought visualization for clear
+ insights.
+tags:
+ - AI Reasoning
+ - Chain of Thought
+ - CoT Visualization
+---
+
+# Chain of Thought
+
+
+
+Experience AI reasoning like never before. Watch as complex problems unfold step by step through our innovative Chain of Thought (CoT) visualization. This breakthrough feature provides unprecedented transparency into AI's decision-making process, allowing you to observe how conclusions are reached in real-time.
+
+By breaking down complex reasoning into clear, logical steps, you can better understand and validate the AI's problem-solving approach. Whether you're debugging, learning, or simply curious about AI reasoning, CoT visualization transforms abstract thinking into an engaging, interactive experience.
diff --git a/docs/usage/features/cot.zh-CN.mdx b/docs/usage/features/cot.zh-CN.mdx
new file mode 100644
index 0000000000000..c2890b4614a71
--- /dev/null
+++ b/docs/usage/features/cot.zh-CN.mdx
@@ -0,0 +1,18 @@
+---
+title: 思维链 (CoT)
+description: 体验思维链 (CoT) 的 AI 推理过程，了解复杂问题的解析步骤。
+tags:
+ - 思维链
+ - AI 推理
+ - 可视化
+ - 逻辑步骤
+ - 决策过程
+---
+
+# 思维链 (CoT)
+
+
+
+体验前所未有的 AI 推理过程。通过创新的思维链（CoT）可视化功能，您可以实时观察复杂问题是如何一步步被解析的。这项突破性的功能为 AI 的决策过程提供了前所未有的透明度，让您能够清晰地了解结论是如何得出的。
+
+通过将复杂的推理过程分解为清晰的逻辑步骤，您可以更好地理解和验证 AI 的解题思路。无论您是在调试问题、学习知识，还是单纯对 AI 推理感兴趣，思维链可视化都能将抽象思维转化为一种引人入胜的互动体验。
diff --git a/docs/usage/features/database.mdx b/docs/usage/features/database.mdx
index 83ef17bd15a5e..db0d345699038 100644
--- a/docs/usage/features/database.mdx
+++ b/docs/usage/features/database.mdx
@@ -16,11 +16,7 @@ tags:
# Local / Cloud Database
-
+
In modern application development, the choice of data storage solution is crucial. To meet the needs of different users, LobeChat offers flexible configurations that support both local and server-side databases. Whether you prioritize data privacy and control or seek a convenient user experience, LobeChat can provide excellent solutions for you.
diff --git a/docs/usage/features/database.zh-CN.mdx b/docs/usage/features/database.zh-CN.mdx
index 172afcadc6d1c..68b7dc4734acb 100644
--- a/docs/usage/features/database.zh-CN.mdx
+++ b/docs/usage/features/database.zh-CN.mdx
@@ -12,11 +12,7 @@ tags:
# 本地 / 云端数据存储
-
+
在现代应用开发中，数据存储方案的选择至关重要。为了满足不同用户的需求，LobeChat 提供了同时支持本地数据库和服务端数据库的灵活配置。无论您是注重数据隐私与掌控，还是追求便捷的使用体验，LobeChat 都能为您提供卓越的解决方案。
diff --git a/docs/usage/features/knowledge-base.mdx b/docs/usage/features/knowledge-base.mdx
new file mode 100644
index 0000000000000..151fadfd6d0e3
--- /dev/null
+++ b/docs/usage/features/knowledge-base.mdx
@@ -0,0 +1,24 @@
+---
+title: File Upload / Knowledge Base
+description: >-
+ Discover LobeChat's file upload and knowledge base features for enhanced user
+ experience.
+tags:
+ - File Upload
+ - Knowledge Base
+ - LobeChat
+ - User Management
+ - File Management
+---
+
+# File Upload / Knowledge Base
+
+
+
+LobeChat supports file upload and knowledge base functionality. You can upload various types of files including documents, images, audio, and video, as well as create knowledge bases, making it convenient for users to manage and search for files. Additionally, you can utilize files and knowledge base features during conversations, enabling a richer dialogue experience.
+
+
+
+
+ Learn more on [📘 LobeChat Knowledge Base Launch — From Now On, Every Step Counts](https://lobehub.com/blog/knowledge-base)
+
diff --git a/docs/usage/features/knowledge-base.zh-CN.mdx b/docs/usage/features/knowledge-base.zh-CN.mdx
new file mode 100644
index 0000000000000..0506a24bc6826
--- /dev/null
+++ b/docs/usage/features/knowledge-base.zh-CN.mdx
@@ -0,0 +1,21 @@
+---
+title: 文件上传 / 知识库
+description: 了解LobeChat的文件上传与知识库功能，提升对话体验。
+tags:
+ - 文件上传
+ - 知识库
+ - LobeChat
+ - 对话体验
+---
+
+# 文件上传 / 知识库
+
+
+
+LobeChat 支持文件上传与知识库功能，你可以上传文件、图片、音频、视频等多种类型的文件，以及创建知识库，方便用户管理和查找文件。同时在对话中使用文件和知识库功能，实现更加丰富的对话体验。
+
+
+
+
+ 查阅 [📘 LobeChat 知识库上线 —— 此刻起，跬步千里](https://lobehub.com/zh/blog/knowledge-base) 了解详情。
+
diff --git a/docs/usage/features/local-llm.mdx b/docs/usage/features/local-llm.mdx
index a84df3c9374fe..3a07271c98fca 100644
--- a/docs/usage/features/local-llm.mdx
+++ b/docs/usage/features/local-llm.mdx
@@ -15,12 +15,7 @@ tags:
# Local Large Language Model (LLM) Support
-
+Available in >=0.127.0, currently only supports Docker deployment
@@ -52,5 +47,6 @@ Now, let's embark on this exciting journey together! Through the collaboration o
+
diff --git a/docs/usage/features/local-llm.zh-CN.mdx b/docs/usage/features/local-llm.zh-CN.mdx
index c48643a0519eb..b4c044e0b8574 100644
--- a/docs/usage/features/local-llm.zh-CN.mdx
+++ b/docs/usage/features/local-llm.zh-CN.mdx
@@ -7,12 +7,7 @@ tags:
# 支持本地大语言模型（LLM）
-
+在 >=v0.127.0 版本中可用，目前仅支持 Docker 部署
@@ -44,5 +39,6 @@ docker run -d -p 3210:3210 -e OLLAMA_PROXY_URL=http://host.docker.internal:11434
+
diff --git a/docs/usage/features/mobile.mdx b/docs/usage/features/mobile.mdx
index d74d728daedfc..1cd4175e2ae5c 100644
--- a/docs/usage/features/mobile.mdx
+++ b/docs/usage/features/mobile.mdx
@@ -13,11 +13,7 @@ tags:
# Mobile Device Adaptation
-
+
LobeChat has undergone a series of optimized designs for mobile devices to enhance the user's mobile experience.
diff --git a/docs/usage/features/mobile.zh-CN.mdx b/docs/usage/features/mobile.zh-CN.mdx
index 5b881eb7416cc..f90421cc62ac3 100644
--- a/docs/usage/features/mobile.zh-CN.mdx
+++ b/docs/usage/features/mobile.zh-CN.mdx
@@ -12,11 +12,7 @@ tags:
# 移动设备适配
-
+
LobeChat 针对移动设备进行了一系列的优化设计，以提升用户的移动体验。
diff --git a/docs/usage/features/multi-ai-providers.mdx b/docs/usage/features/multi-ai-providers.mdx
index c212badb26013..0c02536913a0c 100644
--- a/docs/usage/features/multi-ai-providers.mdx
+++ b/docs/usage/features/multi-ai-providers.mdx
@@ -18,12 +18,7 @@ tags:
# Multi-Model Service Provider Support
-
+Available in version 0.123.0 and later
@@ -48,19 +43,16 @@ We have implemented support for the following model service providers:
- **DeepSeek**: Integrated with the DeepSeek series models, an innovative AI startup from China, The product has been designed to provide a model that balances performance with price. [Learn more](https://www.deepseek.com/)
- **Qwen**: Integrated with the Qwen series models, including the latest **qwen-turbo**, **qwen-plus** and **qwen-max**. [Learn more](https://help.aliyun.com/zh/dashscope/developer-reference/model-introduction)
-At the same time, we are also planning to support more model service providers, such as Replicate and Perplexity, to further enrich our service provider library. If you would like LobeChat to support your favorite service provider, feel free to join our [community discussion](https://github.com/lobehub/lobe-chat/discussions/1284).
+At the same time, we are also planning to support more model service providers, such as Replicate and Perplexity, to further enrich our service provider library. If you would like LobeChat to support your favorite service provider, feel free to join our [community discussion](https://github.com/lobehub/lobe-chat/discussions/6157).
## Local Model Support
-
+
To meet the specific needs of users, LobeChat also supports the use of local models based on [Ollama](https://ollama.ai), allowing users to flexibly use their own or third-party models. For more details, see [Local Model Support](/docs/usage/features/local-llm).
+
diff --git a/docs/usage/features/multi-ai-providers.zh-CN.mdx b/docs/usage/features/multi-ai-providers.zh-CN.mdx
index 761f6e7f8c536..8d490d402ed9c 100644
--- a/docs/usage/features/multi-ai-providers.zh-CN.mdx
+++ b/docs/usage/features/multi-ai-providers.zh-CN.mdx
@@ -18,12 +18,7 @@ tags:
# 多模型服务商支持
-
+在 0.123.0 及以后版本中可用
@@ -48,19 +43,16 @@ tags:
- **DeepSeek**: 接入了 DeepSeek 的 AI 模型，包括最新的 **DeepSeek-V2**，提供兼顾性能与价格的模型。[了解更多](https://www.deepseek.com/)
- **Qwen (通义千问)**: 接入了 Qwen 的 AI 模型，包括最新的 **qwen-turbo**，**qwen-plus** 和 **qwen-max** 等模型。[了解更多](https://help.aliyun.com/zh/dashscope/developer-reference/model-introduction)
-同时，我们也在计划支持更多的模型服务商，如 Replicate 和 Perplexity 等，以进一步丰富我们的服务商库。如果你希望让 LobeChat 支持你喜爱的服务商，欢迎加入我们的[社区讨论](https://github.com/lobehub/lobe-chat/discussions/1284)。
+同时，我们也在计划支持更多的模型服务商，如 Replicate 和 Perplexity 等，以进一步丰富我们的服务商库。如果你希望让 LobeChat 支持你喜爱的服务商，欢迎加入我们的[社区讨论](https://github.com/lobehub/lobe-chat/discussions/6157)。
## 本地模型支持
-
+
为了满足特定用户的需求，LobeChat 还基于 [Ollama](https://ollama.ai) 支持了本地模型的使用，让用户能够更灵活地使用自己的或第三方的模型，详见 [本地模型支持](/zh/docs/usage/features/local-llm)。
+
diff --git a/docs/usage/features/plugin-system.mdx b/docs/usage/features/plugin-system.mdx
index 80165cf0bb695..28db9243719b3 100644
--- a/docs/usage/features/plugin-system.mdx
+++ b/docs/usage/features/plugin-system.mdx
@@ -20,8 +20,8 @@ tags:
borderless
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/268670883-33c43a5c-a512-467e-855c-fa299548cce5.png'
- }
+'https://github.com/user-attachments/assets/66a891ac-01b6-4e3f-b978-2eb07b489b1b'
+}
/>
The plugin ecosystem of LobeChat is an important extension of its core functionality, greatly enhancing the practicality and flexibility of the LobeChat assistant.
@@ -32,7 +32,7 @@ By utilizing plugins, LobeChat assistants can obtain and process real-time infor
In addition, these plugins are not limited to news aggregation, but can also extend to other practical functions, such as quickly searching documents, generating images, obtaining data from various platforms like Bilibili, Steam, and interacting with various third-party services.
-Learn more about [plugin usage](/docs/usage/plugins/basic) by checking it out.
+Learn more about [plugin usage](/docs/usage/plugins/basic-usage) by checking it out.
To help developers better participate in this ecosystem, we provide comprehensive development
@@ -67,20 +67,19 @@ The plugin system of LobeChat has now entered a stable stage, and we have basica
### ✅ Phase One of Plugins
-Implementing the separation of plugins from the main body, splitting the plugins into independent repositories for maintenance, and implementing dynamic loading of plugins. [**#73**](https://github.com/lobehub/lobe-chat/issues/73)
+ Implementing the separation of plugins from the main body, splitting the plugins into independent repositories for maintenance, and implementing dynamic loading of plugins. [**#73**](https://github.com/lobehub/lobe-chat/issues/73)
-### ✅ Phase Two of Plugins
+ ### ✅ Phase Two of Plugins
-The security and stability of plugin usage, more accurate presentation of abnormal states, maintainability and developer-friendliness of the plugin architecture. [**#97**](https://github.com/lobehub/lobe-chat/issues/97)
+ The security and stability of plugin usage, more accurate presentation of abnormal states, maintainability and developer-friendliness of the plugin architecture. [**#97**](https://github.com/lobehub/lobe-chat/issues/97)
-### ✅ Phase Three of Plugins
+ ### ✅ Phase Three of Plugins
-Higher-level and improved customization capabilities, support for OpenAPI schema invocation, compatibility with ChatGPT plugins, and the addition of Midjourney plugins. [**#411**](https://github.com/lobehub/lobe-chat/discussions/#411)
+ Higher-level and improved customization capabilities, support for OpenAPI schema invocation, compatibility with ChatGPT plugins, and the addition of Midjourney plugins. [**#411**](https://github.com/lobehub/lobe-chat/discussions/#411)
-### 💭 Phase Four of Plugins
-
-Comprehensive authentication, visual configuration of plugin definitions, Plugin SDK CLI, Python language development template, any other ideas? Join the discussion: [**#1310**](https://github.com/lobehub/lobe-chat/discussions/#1310)
+ ### 💭 Phase Four of Plugins
+ Comprehensive authentication, visual configuration of plugin definitions, Plugin SDK CLI, Python language development template, any other ideas? Join the discussion: [**#1310**](https://github.com/lobehub/lobe-chat/discussions/#1310)
[chat-plugin-sdk]: https://github.com/lobehub/chat-plugin-sdk
diff --git a/docs/usage/features/plugin-system.zh-CN.mdx b/docs/usage/features/plugin-system.zh-CN.mdx
index 1313ab1ec65ce..bf0074a362c0c 100644
--- a/docs/usage/features/plugin-system.zh-CN.mdx
+++ b/docs/usage/features/plugin-system.zh-CN.mdx
@@ -15,8 +15,8 @@ tags:
borderless
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/268670883-33c43a5c-a512-467e-855c-fa299548cce5.png'
- }
+'https://github.com/user-attachments/assets/66a891ac-01b6-4e3f-b978-2eb07b489b1b'
+}
/>
LobeChat 的插件生态系统是其核心功能的重要扩展，它极大地增强了 LobeChat 助手的实用性和灵活性。
@@ -27,7 +27,7 @@ LobeChat 的插件生态系统是其核心功能的重要扩展，它极大地
此外，这些插件不仅局限于新闻聚合，还可以扩展到其他实用的功能，如快速检索文档、生成图片、获取 Bilibili 、Steam 等各种平台数据，以及与其他各式各样的第三方服务交互。
-通过查看 [插件使用](/zh/docs/usage/plugins/basic) 了解更多。
+通过查看 [插件使用](/zh/docs/usage/plugins/basic-usage) 了解更多。
为了帮助开发者更好地参与到这个生态中来，我们在提供了全面的开发资源。这包括详尽的组件开发文档、功能齐全的软件开发工具包（SDK），以及样板示例，这些都是为了简化开发过程，降低开发者的入门门槛。
@@ -56,20 +56,19 @@ LobeChat 的插件系统目前已初步进入一个稳定阶段，我们已基
### ✅ 插件一期
-实现插件与主体分离，将插件拆分为独立仓库维护，并实现插件的动态加载。 [**#73**](https://github.com/lobehub/lobe-chat/issues/73)
+ 实现插件与主体分离，将插件拆分为独立仓库维护，并实现插件的动态加载。 [**#73**](https://github.com/lobehub/lobe-chat/issues/73)
-### ✅ 插件二期
+ ### ✅ 插件二期
-插件的安全性与使用的稳定性，更加精准地呈现异常状态，插件架构的可维护性与开发者友好。[**#97**](https://github.com/lobehub/lobe-chat/issues/97)
+ 插件的安全性与使用的稳定性，更加精准地呈现异常状态，插件架构的可维护性与开发者友好。[**#97**](https://github.com/lobehub/lobe-chat/issues/97)
-### ✅ 插件三期
+ ### ✅ 插件三期
-更高阶与完善的自定义能力，支持 OpenAPI schema 调用、兼容 ChatGPT 插件、新增 Midjourney 插件。 [**#411**](https://github.com/lobehub/lobe-chat/discussions/#411)
+ 更高阶与完善的自定义能力，支持 OpenAPI schema 调用、兼容 ChatGPT 插件、新增 Midjourney 插件。 [**#411**](https://github.com/lobehub/lobe-chat/discussions/#411)
-### 💭 插件四期
-
-完善的鉴权、可视化配置插件定义、 Plugin SDK CLI 、 Python 语言研发模板、还有什么想法？欢迎参与讨论： [**#1310**](https://github.com/lobehub/lobe-chat/discussions/#1310)
+ ### 💭 插件四期
+ 完善的鉴权、可视化配置插件定义、 Plugin SDK CLI 、 Python 语言研发模板、还有什么想法？欢迎参与讨论： [**#1310**](https://github.com/lobehub/lobe-chat/discussions/#1310)
[chat-plugin-sdk]: https://github.com/lobehub/chat-plugin-sdk
diff --git a/docs/usage/features/pwa.mdx b/docs/usage/features/pwa.mdx
index 606093d61c6c1..c4ee493323a5a 100644
--- a/docs/usage/features/pwa.mdx
+++ b/docs/usage/features/pwa.mdx
@@ -14,12 +14,7 @@ tags:
# Progressive Web App (PWA)
-
+
We understand the importance of providing a seamless experience for users in today's multi-device environment. To achieve this, we have adopted Progressive Web App [PWA](https://support.google.com/chrome/answer/9658361) technology, which is a modern web technology that elevates web applications to a near-native app experience. Through PWA, LobeChat is able to provide a highly optimized user experience on both desktop and mobile devices, while maintaining lightweight and high performance characteristics. Visually and perceptually, we have also carefully designed it to ensure that its interface is indistinguishable from a native app, providing smooth animations, responsive layouts, and adaptation to different screen resolutions of various devices.
@@ -33,15 +28,13 @@ If you are unfamiliar with the installation process of PWA, you can follow the s
+ ### Run Chrome or Edge browser on your computer
-### Run Chrome or Edge browser on your computer
+ ### Visit the LobeChat webpage
-### Visit the LobeChat webpage
-
-### In the top right corner of the address bar, click the Install icon
-
-### Follow the on-screen instructions to complete the PWA installation
+ ### In the top right corner of the address bar, click the Install icon
+ ### Follow the on-screen instructions to complete the PWA installation
## Running on Safari
@@ -49,26 +42,19 @@ If you are unfamiliar with the installation process of PWA, you can follow the s
Safari PWA requires macOS Ventura or later. The PWA installed by Safari does not require Safari to be open; you can directly open the PWA app.
+ ### Run Safari browser on your computer
-### Run Safari browser on your computer
-
-### Visit the LobeChat webpage
-
-### In the top right corner of the address bar, click the Share icon
+ ### Visit the LobeChat webpage
-### Click Add to Dock
+ ### In the top right corner of the address bar, click the Share icon
-### Follow the on-screen instructions to complete the PWA installation
+ ### Click Add to Dock
+ ### Follow the on-screen instructions to complete the PWA installation
The default installed LobeChat PWA icon has a black background, you can use cmd + i to paste the following image to replace it with a white background.
-
+
diff --git a/docs/usage/features/pwa.zh-CN.mdx b/docs/usage/features/pwa.zh-CN.mdx
index b7b5564ce0062..39fe52cbd0948 100644
--- a/docs/usage/features/pwa.zh-CN.mdx
+++ b/docs/usage/features/pwa.zh-CN.mdx
@@ -14,12 +14,7 @@ tags:
# 渐进式 Web 应用（PWA）
-
+
我们深知在当今多设备环境下为用户提供无缝体验的重要性。为此，我们采用了渐进式 Web 应用 [PWA](https://support.google.com/chrome/answer/9658361) 技术，这是一种能够将网页应用提升至接近原生应用体验的现代 Web 技术。通过 PWA，LobeChat 能够在桌面和移动设备上提供高度优化的用户体验，同时保持轻量级和高性能的特点。在视觉和感觉上，我们也经过精心设计，以确保它的界面与原生应用无差别，提供流畅的动画、响应式布局和适配不同设备的屏幕分辨率。
@@ -33,15 +28,13 @@ tags:
+ ### 在电脑上运行 Chrome 或 Edge 浏览器
-### 在电脑上运行 Chrome 或 Edge 浏览器
+ ### 访问 LobeChat 网页
-### 访问 LobeChat 网页
-
-### 在地址栏的右上角，单击 安装 图标
-
-### 根据屏幕上的指示完成 PWA 的安装
+ ### 在地址栏的右上角，单击 安装 图标
+ ### 根据屏幕上的指示完成 PWA 的安装
## Safari 浏览器上运行
@@ -49,26 +42,19 @@ tags:
Safari PWA 需要 macOS Ventura 或更高版本。Safari 安装的 PWA 并不要求 Safari 是打开状态，可以直接打开 PWA 应用。
+ ### 在电脑上运行 Safari 浏览器
-### 在电脑上运行 Safari 浏览器
-
-### 访问 LobeChat 网页
-
-### 在地址栏的右上角，单击 分享 图标
+ ### 访问 LobeChat 网页
-### 点选 添加到程序坞
+ ### 在地址栏的右上角，单击 分享 图标
-### 根据屏幕上的指示完成 PWA 的安装
+ ### 点选 添加到程序坞
+ ### 根据屏幕上的指示完成 PWA 的安装
默认安装的 LobeChat PWA 图标是黑色背景的，您可以在自行使用 cmd + i 粘贴如下图片替换为白色背景的。
-
+
diff --git a/docs/usage/features/text-to-image.mdx b/docs/usage/features/text-to-image.mdx
index 01a7a2613c5c1..83aa958d0bc3c 100644
--- a/docs/usage/features/text-to-image.mdx
+++ b/docs/usage/features/text-to-image.mdx
@@ -19,8 +19,8 @@ tags:
borderless
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/297746445-0ff762b9-aa08-4337-afb7-12f932b6efbb.png'
- }
+'https://github.com/user-attachments/assets/708274a7-2458-494b-a6ec-b73dfa1fa7c2'
+}
/>
Supporting the latest text-to-image generation technology, LobeChat now enables users to directly utilize the Text to Image tool during conversations with the assistant. By harnessing the capabilities of AI tools such as [DALL-E 3](https://openai.com/dall-e-3), [MidJourney](https://www.midjourney.com/), and [Pollinations](https://pollinations.ai/), assistants can now transform your ideas into images. This allows for a more private and immersive creative process.
diff --git a/docs/usage/features/text-to-image.zh-CN.mdx b/docs/usage/features/text-to-image.zh-CN.mdx
index c512847c05114..56af4518e9e2c 100644
--- a/docs/usage/features/text-to-image.zh-CN.mdx
+++ b/docs/usage/features/text-to-image.zh-CN.mdx
@@ -19,8 +19,8 @@ tags:
borderless
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/297746445-0ff762b9-aa08-4337-afb7-12f932b6efbb.png'
- }
+'https://github.com/user-attachments/assets/708274a7-2458-494b-a6ec-b73dfa1fa7c2'
+}
/>
支持最新的文本到图片生成技术，LobeChat 现在能够让用户在与助手对话中直接调用文成图工具进行创作。通过利用 [`DALL-E 3`](https://openai.com/dall-e-3)、[`MidJourney`](https://www.midjourney.com/) 和 [`Pollinations`](https://pollinations.ai/) 等 AI 工具的能力， 助手们现在可以将你的想法转化为图像。同时可以更私密和沉浸式的完成你的创造过程。
diff --git a/docs/usage/features/theme.mdx b/docs/usage/features/theme.mdx
index 6172ff5659e8b..76be42c75eb2e 100644
--- a/docs/usage/features/theme.mdx
+++ b/docs/usage/features/theme.mdx
@@ -15,12 +15,7 @@ tags:
# Custom Themes
-
+
LobeChat places a strong emphasis on personalized user experiences in its interface design, and thus introduces flexible and diverse theme modes, including a light mode for daytime and a dark mode for nighttime.
diff --git a/docs/usage/features/theme.zh-CN.mdx b/docs/usage/features/theme.zh-CN.mdx
index 778cf2c2e191b..1a392352ed4f2 100644
--- a/docs/usage/features/theme.zh-CN.mdx
+++ b/docs/usage/features/theme.zh-CN.mdx
@@ -12,12 +12,7 @@ tags:
# 自定义主题
-
+
LobeChat 在界面设计上十分考虑用户的个性化体验，因此引入了灵活多变的主题模式，其中包括日间的亮色模式和夜间的深色模式。
diff --git a/docs/usage/features/tts.mdx b/docs/usage/features/tts.mdx
index 755c784e41e3c..fcff7d0422179 100644
--- a/docs/usage/features/tts.mdx
+++ b/docs/usage/features/tts.mdx
@@ -22,8 +22,8 @@ tags:
borderless
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/284072124-c9853d8d-f1b5-44a8-a305-45ebc0f6d19a.png'
- }
+'https://github.com/user-attachments/assets/50189597-2cc3-4002-b4c8-756a52ad5c0a'
+}
/>
LobeChat supports Text-to-Speech (TTS) and Speech-to-Text (STT) technologies. Our application can convert text information into clear voice output, allowing users to interact with our conversational agents as if they were talking to a real person. Users can choose from a variety of voices and pair the appropriate audio with the assistant. Additionally, for users who prefer auditory learning or need to obtain information while busy, TTS provides an excellent solution.
@@ -32,11 +32,7 @@ In LobeChat, we have carefully selected a series of high-quality voice options (
## Lobe TTS
-
+
[`@lobehub/tts`](https://tts.lobehub.com) is a high-quality TTS toolkit developed using the TS language, supporting usage in both server and browser environments.
diff --git a/docs/usage/features/tts.zh-CN.mdx b/docs/usage/features/tts.zh-CN.mdx
index 6b5a72d5179ec..10cceecf03b76 100644
--- a/docs/usage/features/tts.zh-CN.mdx
+++ b/docs/usage/features/tts.zh-CN.mdx
@@ -18,8 +18,8 @@ tags:
borderless
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/284072124-c9853d8d-f1b5-44a8-a305-45ebc0f6d19a.png'
- }
+'https://github.com/user-attachments/assets/50189597-2cc3-4002-b4c8-756a52ad5c0a'
+}
/>
LobeChat 支持文字转语音（Text-to-Speech，TTS）和语音转文字（Speech-to-Text，STT）技术，我们的应用能够将文本信息转化为清晰的语音输出，用户可以像与真人交谈一样与我们的对话代理进行交流。用户可以从多种声音中选择，给助手搭配合适的音源。 同时，对于那些倾向于听觉学习或者想要在忙碌中获取信息的用户来说，TTS 提供了一个极佳的解决方案。
@@ -28,11 +28,7 @@ LobeChat 支持文字转语音（Text-to-Speech，TTS）和语音转文字（Spe
## Lobe TTS
-
+
[`@lobehub/tts`](https://tts.lobehub.com) 是一个使用 TS 语言开发的，高质量 TTS 工具包，支持在服务端和浏览器中使用。
diff --git a/docs/usage/features/vision.mdx b/docs/usage/features/vision.mdx
index bf14216087e9a..6bc3c95354567 100644
--- a/docs/usage/features/vision.mdx
+++ b/docs/usage/features/vision.mdx
@@ -19,8 +19,8 @@ tags:
borderless
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/284072129-382bdf30-e3d6-4411-b5a0-249710b8ba08.png'
- }
+'https://github.com/user-attachments/assets/18574a1f-46c2-4cbc-af2c-35a86e128a07'
+}
/>
LobeChat now supports large language models with visual recognition capabilities such as OpenAI's [`gpt-4-vision`](https://platform.openai.com/docs/guides/vision), Google Gemini Pro vision, and Zhipu GLM-4 Vision, enabling LobeChat to have multimodal interaction capabilities. Users can easily upload or drag and drop images into the chat box, and the assistant will be able to recognize the content of the images and engage in intelligent conversations based on them, creating more intelligent and diverse chat scenarios.
diff --git a/docs/usage/features/vision.zh-CN.mdx b/docs/usage/features/vision.zh-CN.mdx
index 1e2bf3b2ee4b4..e2f8c9abc8084 100644
--- a/docs/usage/features/vision.zh-CN.mdx
+++ b/docs/usage/features/vision.zh-CN.mdx
@@ -16,8 +16,8 @@ tags:
borderless
cover
src={
- 'https://github-production-user-asset-6210df.s3.amazonaws.com/17870709/284072129-382bdf30-e3d6-4411-b5a0-249710b8ba08.png'
- }
+'https://github.com/user-attachments/assets/18574a1f-46c2-4cbc-af2c-35a86e128a07'
+}
/>
LobeChat 已经支持 OpenAI 的 [`gpt-4-vision`](https://platform.openai.com/docs/guides/vision) 、Google Gemini Pro vision、智谱 GLM-4 Vision 等具有视觉识别能力的大语言模型，这使得 LobeChat 具备了多模态交互的能力。用户可以轻松上传图片或者拖拽图片到对话框中，助手将能够识别图片内容，并在此基础上进行智能对话，构建更智能、更多元化的聊天场景。
diff --git a/docs/usage/foundation/basic.mdx b/docs/usage/foundation/basic.mdx
index 015032b81e3cd..5b35869c53ee1 100644
--- a/docs/usage/foundation/basic.mdx
+++ b/docs/usage/foundation/basic.mdx
@@ -15,20 +15,13 @@ tags:
# Basic Usage Guide for Conversations
-
+
In general, the basic interaction with Large Language Models (LLMs) can be done through the fundamental functions provided in this area (as shown above).
## Basic Function Description
-
+
1. **Model Selection**: Choose the Large Language Model (LLM) to be used in the current conversation. For model settings, refer to [Model Providers](/docs/usage/providers).
2. **File/Image Upload**: When the selected model supports file or image recognition, users can upload files or images during the conversation with the model.
@@ -40,16 +33,12 @@ In general, the basic interaction with Large Language Models (LLMs) can be done
8. **Start New Topic**: End the current conversation and start a new topic. For more information, refer to [Topic Usage](/docs/usage/agents/topics).
9. **Send Button**: Send the current input content to the model. The dropdown menu provides additional send operation options.
-
+
- **Send Shortcut**: Set a shortcut to send messages and line breaks using the Enter key or ⌘ +
- Enter key. - **Add an AI Message**: Manually add and edit a message input by an AI character in
- the conversation context, which will not trigger a model response. - **Add a User Message**: Add
- the current input content as a message input by the user character to the conversation context,
- which will not trigger a model response.
+ Enter key. - **Add an AI Message**: Manually add and edit a message input by an AI character in
+ the conversation context, which will not trigger a model response. - **Add a User Message**: Add
+ the current input content as a message input by the user character to the conversation context,
+ which will not trigger a model response.
diff --git a/docs/usage/foundation/basic.zh-CN.mdx b/docs/usage/foundation/basic.zh-CN.mdx
index 1d6ecf37b55c3..5c858e7a9a8ad 100644
--- a/docs/usage/foundation/basic.zh-CN.mdx
+++ b/docs/usage/foundation/basic.zh-CN.mdx
@@ -17,23 +17,16 @@ tags:
# 会话基本使用指南
-
+
通常情况下，与大型语言模型 (LLMs) 的基本交互可以通过此区域（如上图）提供的基础功能进行。
## 基本功能说明
-
+
1. **模型选择**：选择当前对话所使用的大型语言模型 (LLM)。模型的设置详见[模型服务商](/zh/docs/usage/providers)。
-2. **文件/图片上传**：当所选模型支持文件或图片识别功能时，用户可以在与模型的对话中上传文件或图片。
+2. **文件 / 图片上传**：当所选模型支持文件或图片识别功能时，用户可以在与模型的对话中上传文件或图片。
3. **温度设置**：调节模型输出的随机性程度。数值越高，输出结果越随机。详细说明请参考[大语言模型指南](/zh/docs/usage/agents/model)。
4. **历史记录设置**：设定本次对话中模型需要记忆的聊天记录数量。历史记录越长，模型能够记忆的对话内容越多，但同时也会消耗更多的上下文 token。
5. **语音输入**：点击该按钮后，可以将语音转换为文字输入。有关详细信息，请参考[语音文字转换](/zh/docs/usage/foundation/tts-stt)。
@@ -42,15 +35,12 @@ tags:
8. **新建话题**：结束当前对话并开启一个新的对话主题。有关详细信息，请参考[话题使用](/zh/docs/usage/agents/topics)。
9. **发送按钮**：将当前输入内容发送至模型。下拉菜单提供额外的发送操作选项。
-
+
- **发送快捷键**：设置使用 Enter 键或 ⌘ + Enter 键发送消息和换行的快捷方式。 -
- **添加一条AI消息**：在对话上下文中手动添加并编辑一条由 AI 角色输入的消息，该操作不会触发模型响应。
+ **添加一条 AI 消息**：在对话上下文中手动添加并编辑一条由 AI 角色输入的消息，该操作不会触发模型响应。
-
+
**添加一条用户消息**：将当前输入内容作为用户角色输入的消息添加到对话上下文中，该操作不会触发模型响应。
diff --git a/docs/usage/foundation/share.mdx b/docs/usage/foundation/share.mdx
index 25b438fee5898..c981a980c2679 100644
--- a/docs/usage/foundation/share.mdx
+++ b/docs/usage/foundation/share.mdx
@@ -14,20 +14,13 @@ tags:
# Share Conversation Records
-
+
By clicking the `Share` button in the top right corner of the chat window, you can share the current conversation records with others. LobeChat supports two sharing methods: `Screenshot Sharing` and `ShareGPT Sharing`.
## Screenshot Sharing
-
+
The screenshot sharing feature will generate and save an image of the current conversation records, with the following options:
@@ -38,9 +31,6 @@ The screenshot sharing feature will generate and save an image of the current co
## ShareGPT
-
+
[ShareGPT](https://sharegpt.com/) is an AI conversation sharing platform that allows users to easily share their conversations with Large Language Models (LLMs). Users can generate a permanent link with just one click, making it convenient to share these conversations with friends or others. By integrating ShareGPT functionality, LobeChat can generate links for conversation records with just one click, making sharing easy.
diff --git a/docs/usage/foundation/share.zh-CN.mdx b/docs/usage/foundation/share.zh-CN.mdx
index c8be6fa65f88b..397f4b41b3e3e 100644
--- a/docs/usage/foundation/share.zh-CN.mdx
+++ b/docs/usage/foundation/share.zh-CN.mdx
@@ -11,20 +11,13 @@ tags:
# 分享会话记录
-
+
通过会话窗口右上角的`分享`按钮，您可以将当前会话记录分享给其他人。LobeChat 支持两种分享方式：`截图分享`和 `ShareGPT 分享`。
## 截图分享
-
+
截图分享功能将生成当前会话记录的图片并保存，其选项说明如下：
@@ -35,9 +28,6 @@ tags:
## ShareGPT
-
+
[ShareGPT](https://sharegpt.com/) 是一个 AI 对话分享平台，允许用户便捷地分享他们与大型语言模型 (LLM) 的对话。用户只需点击即可生成永久链接，方便与朋友或其他人分享这些对话。LobeChat 通过集成 ShareGPT 功能，可以一键将对话记录生成链接，方便分享。
diff --git a/docs/usage/foundation/text2image.mdx b/docs/usage/foundation/text2image.mdx
index 767ba327a0d85..4e08631faa61d 100644
--- a/docs/usage/foundation/text2image.mdx
+++ b/docs/usage/foundation/text2image.mdx
@@ -20,10 +20,7 @@ LobeChat supports text-to-image generation through a plugin mechanism. Currently
If you have configured the OpenAI API, you can enable the DALL-E plugin directly in the assistant interface and input prompts in the conversation for AI to generate images for you.
-
+
If the DALL-E plugin is not available, please check if the OpenAI API key has been correctly configured.
@@ -31,10 +28,7 @@ If the DALL-E plugin is not available, please check if the OpenAI API key has be
LobeChat also offers the Midjourney plugin, which generates images by calling the Midjourney API. Please install the Midjourney plugin in the plugin store beforehand.
-
+
info For plugin installation, please refer to [Plugin Usage](/docs/usage/plugins/basic-usage)
@@ -42,7 +36,4 @@ LobeChat also offers the Midjourney plugin, which generates images by calling th
When using the Midjourney plugin for the first time, you will need to fill in your Midjourney API key in the plugin settings.
-
+
diff --git a/docs/usage/foundation/text2image.zh-CN.mdx b/docs/usage/foundation/text2image.zh-CN.mdx
index 0b586442a8062..18434dc6dc8fe 100644
--- a/docs/usage/foundation/text2image.zh-CN.mdx
+++ b/docs/usage/foundation/text2image.zh-CN.mdx
@@ -18,10 +18,7 @@ LobeChat 通过插件机制支持文本生成图片功能。目前，LobeChat
如果您已配置 OpenAI API，可以直接在助手界面启用 DALL-E 插件，并在对话中输入提示词，让 AI 为您生成图片。
-
+
如果 DALL-E 插件不可用，请检查 OpenAI API 密钥是否已正确配置。
@@ -29,16 +26,10 @@ LobeChat 通过插件机制支持文本生成图片功能。目前，LobeChat
LobeChat 还提供 Midjourney 插件，通过 API 调用 Midjourney 生成图片。请提前在插件商店中安装 Midjourney 插件。
-
+插件安装请参考[插件使用](/zh/docs/usage/plugins/basic-usage)
首次使用 Midjourney 插件时，您需要在插件设置中填写您的 Midjourney API 密钥。
-
+
diff --git a/docs/usage/foundation/translate.mdx b/docs/usage/foundation/translate.mdx
index 43f8b76d1fb07..90f078db75b3d 100644
--- a/docs/usage/foundation/translate.mdx
+++ b/docs/usage/foundation/translate.mdx
@@ -11,29 +11,19 @@ tags:
# Translation of Conversation Records
-
+
## Translating Conversation Content
LobeChat supports users to translate conversation content into a specified language with just one click. After selecting the target language, LobeChat will use a pre-set AI model for translation and display the translated results in real-time in the chat window.
-
+
## Translation Model Settings
You can specify the model you wish to use as a translation assistant in the settings.
-
+
- Open the `Settings` panel
- Find the `Translation Settings` option under `System Assistants`
diff --git a/docs/usage/foundation/translate.zh-CN.mdx b/docs/usage/foundation/translate.zh-CN.mdx
index 5ba9784df0ca4..3449ac5d07f3e 100644
--- a/docs/usage/foundation/translate.zh-CN.mdx
+++ b/docs/usage/foundation/translate.zh-CN.mdx
@@ -10,29 +10,19 @@ tags:
# 翻译会话记录
-
+
## 翻译对话中的内容
LobeChat 支持用户一键将对话内容翻译成指定语言。选择目标语言后，LobeChat 将调用预先设置的 AI 模型进行翻译，并将翻译结果实时显示在聊天窗口中。
-
+
## 翻译模型设置
你可以在设置中指定您希望使用的模型作为翻译助手。
-
+
- 打开`设置`面板
- 在`系统助手`中找到`翻译设置`选项
diff --git a/docs/usage/foundation/tts-stt.mdx b/docs/usage/foundation/tts-stt.mdx
index a99925d86b41b..59675a2dbfffc 100644
--- a/docs/usage/foundation/tts-stt.mdx
+++ b/docs/usage/foundation/tts-stt.mdx
@@ -19,28 +19,19 @@ LobeChat supports text-to-speech conversion, allowing users to input content thr
Select any content in the chat window, choose `Text-to-Speech`, and the AI will use the TTS model to read the text content aloud.
-
+
## Speech-to-Text (STT)
Select the voice input feature in the input window, and LobeChat will convert your speech to text and input it into the text box. After completing the input, you can send it directly to the AI.
-
+
## Text-to-Speech Conversion Settings
You can specify the model you want to use for text-to-speech conversion in the settings.
-
+
- Open the `Settings` panel
- Find the `Text-to-Speech` settings
diff --git a/docs/usage/foundation/tts-stt.zh-CN.mdx b/docs/usage/foundation/tts-stt.zh-CN.mdx
index b0f885d156b5b..f5199c9cb27e2 100644
--- a/docs/usage/foundation/tts-stt.zh-CN.mdx
+++ b/docs/usage/foundation/tts-stt.zh-CN.mdx
@@ -17,28 +17,19 @@ LobeChat 支持文字语音转换功能，允许用户通过语音输入内容
在对话窗口中选中任意内容，选择`文字转语音`，AI 将通过 TTS 模型对文本内容进行语音播报。
-
+
## 语音转文字（STT）
在输入窗口中选择语音输入功能，LobeChat 将您的语音转换为文字并输入到文本框中，完成输入后可以直接发送给 AI。
-
+
## 文字语音转换设置
你可以在设置中为文字语音转换功能指定您希望使用的模型。
-
+
- 打开`设置`面板
- 找到`文字转语音`设置
diff --git a/docs/usage/foundation/vision.mdx b/docs/usage/foundation/vision.mdx
index 319bf9989b59a..449e6af47b71a 100644
--- a/docs/usage/foundation/vision.mdx
+++ b/docs/usage/foundation/vision.mdx
@@ -16,34 +16,22 @@ tags:
The ecosystem of large language models that support visual recognition is becoming increasingly rich. Starting from `gpt-4-vision`, LobeChat now supports various large language models with visual recognition capabilities, enabling LobeChat to have multimodal interaction capabilities.
-
+
## Image Input
If the model you are currently using supports visual recognition, you can input image content by uploading a file or dragging the image directly into the input box. The model will automatically recognize the image content and provide feedback based on your prompts.
-
+
## Visual Models
In the model list, models with a `👁️` icon next to their names indicate that the model supports visual recognition. Selecting such a model allows you to send image content.
-
+
## Custom Model Configuration
If you need to add a custom model that is not currently in the list and explicitly supports visual recognition, you can enable the `Visual Recognition` feature in the `Custom Model Configuration` to allow the model to interact with images.
-
+
diff --git a/docs/usage/foundation/vision.zh-CN.mdx b/docs/usage/foundation/vision.zh-CN.mdx
index 07d1dc70e741d..8df26785379ba 100644
--- a/docs/usage/foundation/vision.zh-CN.mdx
+++ b/docs/usage/foundation/vision.zh-CN.mdx
@@ -12,34 +12,22 @@ tags:
当前支持视觉识别的大语言模型生态日益丰富。从 `gpt-4-vision` 开始，LobeChat 开始支持各类具有视觉识别能力的大语言模型，这使得 LobeChat 具备了多模态交互的能力。
-
+
## 图片输入
如果你当前使用的模型支持视觉识别功能，您可以通过上传文件或直接将图片拖入输入框的方式输入图片内容。模型会自动识别图片内容，并根据您的提示词给出反馈。
-
+
## 视觉模型
在模型列表中，模型名称后面带有`👁️`图标表示该模型支持视觉识别功能。选择该模型后即可发送图片内容。
-
+
## 自定义模型配置
如果您需要添加当前列表中没有的自定义模型，并且该模型明确支持视觉识别功能，您可以在`自定义模型配置`中开启`视觉识别`功能，使该模型能够与图片进行交互。
-
+
diff --git a/docs/usage/plugins/basic-usage.mdx b/docs/usage/plugins/basic-usage.mdx
index 51374bc97d40a..3b0146bf57603 100644
--- a/docs/usage/plugins/basic-usage.mdx
+++ b/docs/usage/plugins/basic-usage.mdx
@@ -17,38 +17,23 @@ The plugin system is a key element in expanding the capabilities of assistants i
Watch the following video to quickly get started with using LobeChat plugins:
-
+
## Plugin Store
You can access the Plugin Store by navigating to "Extension Tools" -> "Plugin Store" in the session toolbar.
-
+
The Plugin Store allows you to directly install and use plugins within LobeChat.
-
+
## Using Plugins
After installing a plugin, simply enable it under the current assistant to use it.
-
+
## Plugin Configuration
@@ -56,14 +41,6 @@ Some plugins may require specific configurations, such as API keys.
After installing a plugin, you can click on "Settings" to enter the plugin's settings and fill in the required configurations:
-
-
-
+
+
+
diff --git a/docs/usage/plugins/basic-usage.zh-CN.mdx b/docs/usage/plugins/basic-usage.zh-CN.mdx
index a6d53356b3d1e..6ec4ee51d787b 100644
--- a/docs/usage/plugins/basic-usage.zh-CN.mdx
+++ b/docs/usage/plugins/basic-usage.zh-CN.mdx
@@ -15,38 +15,23 @@ tags:
查看以下视频，快速上手使用 LobeChat 插件：
-
+
## 插件商店
你可以在会话工具条中的 「扩展工具」 -> 「插件商店」，进入插件商店。
-
+
插件商店中会在 LobeChat 中可以直接安装并使用的插件。
-
+
## 使用插件
安装完毕插件后，只需在当前助手下开启插件即可使用。
-
+
## 插件配置
@@ -54,14 +39,6 @@ tags:
你可以在安装插件后，点击设置进入插件的设置填写配置：
-
-
-
+
+
+
diff --git a/docs/usage/plugins/development.mdx b/docs/usage/plugins/development.mdx
index 286bd8ba38f1e..45b9818604b39 100644
--- a/docs/usage/plugins/development.mdx
+++ b/docs/usage/plugins/development.mdx
@@ -36,88 +36,52 @@ This section will introduce how to add and use a custom plugin in LobeChat.
### Create and Launch Plugin Project
-You need to first create a plugin project locally, you can use the template we have prepared [lobe-chat-plugin-template][lobe-chat-plugin-template-url]
+ You need to first create a plugin project locally, you can use the template we have prepared [lobe-chat-plugin-template][lobe-chat-plugin-template-url]
-```bash
-$ git clone https://github.com/lobehub/chat-plugin-template.git
-$ cd chat-plugin-template
-$ npm i
-$ npm run dev
-```
-
-When you see `ready started server on 0.0.0.0:3400, url: http://localhost:3400`, it means the plugin service has been successfully launched locally.
+ ```bash
+ $ git clone https://github.com/lobehub/chat-plugin-template.git
+ $ cd chat-plugin-template
+ $ npm i
+ $ npm run dev
+ ```
-
+ When you see `ready started server on 0.0.0.0:3400, url: http://localhost:3400`, it means the plugin service has been successfully launched locally.
-### Add Local Plugin in LobeChat Role Settings
+
-Next, go to LobeChat, create a new assistant, and go to its session settings page:
+ ### Add Local Plugin in LobeChat Role Settings
-
+ Next, go to LobeChat, create a new assistant, and go to its session settings page:
-Click the Add button on the right of the plugin list to open the custom plugin adding popup:
+
-
+ Click the Add button on the right of the plugin list to open the custom plugin adding popup:
-Fill in the **Plugin Description File Url** with `http://localhost:3400/manifest-dev.json`, which is the manifest address of the plugin we started locally.
+
-At this point, you should see that the identifier of the plugin has been automatically recognized as `chat-plugin-template`. Next, you need to fill in the remaining form fields (only the title is required), and then click the Save button to complete the custom plugin addition.
+ Fill in the **Plugin Description File Url** with `http://localhost:3400/manifest-dev.json`, which is the manifest address of the plugin we started locally.
-
+ At this point, you should see that the identifier of the plugin has been automatically recognized as `chat-plugin-template`. Next, you need to fill in the remaining form fields (only the title is required), and then click the Save button to complete the custom plugin addition.
-After adding, you can see the newly added plugin in the plugin list. If you need to modify the plugin configuration, you can click the Settings button on the far right to make changes.
+
-
+ After adding, you can see the newly added plugin in the plugin list. If you need to modify the plugin configuration, you can click the Settings button on the far right to make changes.
-### Test Plugin Function in Session
+
-Next, we need to test whether the plugin's function is working properly.
+ ### Test Plugin Function in Session
-Click the Back button to return to the session area, and then send a message to the assistant: "What should I wear?" At this point, the assistant will try to ask you about your gender and current mood.
+ Next, we need to test whether the plugin's function is working properly.
-
+ Click the Back button to return to the session area, and then send a message to the assistant: "What should I wear?" At this point, the assistant will try to ask you about your gender and current mood.
-After answering, the assistant will initiate the plugin call, retrieve recommended clothing data from the server based on your gender and mood, and push it to you. Finally, it will provide a text summary based on this information.
+
-
+ After answering, the assistant will initiate the plugin call, retrieve recommended clothing data from the server based on your gender and mood, and push it to you. Finally, it will provide a text summary based on this information.
-After completing these operations, you have understood the basic process of adding custom plugins and using them in LobeChat.
+
+ After completing these operations, you have understood the basic process of adding custom plugins and using them in LobeChat.
## Local Plugin Development
@@ -247,29 +211,17 @@ export default createLobeChatPluginGateway();
The custom UI interface for plugins is optional. For example, the official plugin [Web Content Extraction](https://github.com/lobehub/chat-plugin-web-crawler) does not have a corresponding user interface.
-
+
If you want to display richer information in plugin messages or include some interactive operations, you can customize a user interface for the plugin. For example, the following image shows the user interface for the [Search Engine](https://github.com/lobehub/chat-plugin-search-engine) plugin.
-
+
#### Implementation of Plugin UI Interface
LobeChat implements the loading of plugin UI through `iframe` and uses `postMessage` to communicate with the plugin. Therefore, the implementation of the plugin UI is consistent with regular web development. You can use any frontend framework and development language you are familiar with.
-
+
In the template we provide, we use React + Next.js + [antd](https://ant.design/) as the frontend interface framework. You can find the implementation of the user interface in [`src/pages/index.tsx`](https://github.com/lobehub/chat-plugin-template/blob/main/src/pages/index.tsx).
@@ -307,7 +259,7 @@ If you want more people to use your plugin, feel free to [submit it for listing]
### Plugin Shield
-[](https://github.com/lobehub/lobe-chat-plugins)
+[](https://github.com/lobehub/lobe-chat-plugins)
```md
[](https://github.com/lobehub/lobe-chat-plugins)
diff --git a/docs/usage/plugins/development.zh-CN.mdx b/docs/usage/plugins/development.zh-CN.mdx
index 052f516eb3651..7ac82ce7bdcca 100644
--- a/docs/usage/plugins/development.zh-CN.mdx
+++ b/docs/usage/plugins/development.zh-CN.mdx
@@ -30,88 +30,52 @@ tags:
### 创建并启动插件项目
-你需要先在本地创建一个插件项目，可以使用我们准备好的模板 [lobe-chat-plugin-template][lobe-chat-plugin-template-url]
+ 你需要先在本地创建一个插件项目，可以使用我们准备好的模板 [lobe-chat-plugin-template][lobe-chat-plugin-template-url]
-```bash
-$ git clone https://github.com/lobehub/chat-plugin-template.git
-$ cd chat-plugin-template
-$ npm i
-$ npm run dev
-```
-
-当出现`ready started server on 0.0.0.0:3400, url: http://localhost:3400` 时，说明插件服务已经在本地启动成功。
+ ```bash
+ $ git clone https://github.com/lobehub/chat-plugin-template.git
+ $ cd chat-plugin-template
+ $ npm i
+ $ npm run dev
+ ```
-
+ 当出现`ready started server on 0.0.0.0:3400, url: http://localhost:3400` 时，说明插件服务已经在本地启动成功。
-### 在 LobeChat 角色设置中添加本地插件
+
-接下来进入到 LobeChat 中，创建一个新的助手，并进入它的会话设置页：
+ ### 在 LobeChat 角色设置中添加本地插件
-
+ 接下来进入到 LobeChat 中，创建一个新的助手，并进入它的会话设置页：
-点击插件列表右侧的 添加 按钮，打开自定义插件添加弹窗：
+
-
+ 点击插件列表右侧的 添加 按钮，打开自定义插件添加弹窗：
-在 **插件描述文件 Url** 地址 中填入 `http://localhost:3400/manifest-dev.json` ，这是我们本地启动的插件描述清单地址。
+
-此时，你应该可以看到看到插件的标识符一栏已经被自动识别为 `chat-plugin-template`。接下来你需要填写剩下的表单字段（只有标题必填），然后点击 保存 按钮，即可完成自定义插件添加。
+ 在 **插件描述文件 Url** 地址 中填入 `http://localhost:3400/manifest-dev.json` ，这是我们本地启动的插件描述清单地址。
-
+ 此时，你应该可以看到看到插件的标识符一栏已经被自动识别为 `chat-plugin-template`。接下来你需要填写剩下的表单字段（只有标题必填），然后点击 保存 按钮，即可完成自定义插件添加。
-完成添加后，在插件列表中就能看到刚刚添加的插件，如果需要修改插件的配置，可以点击最右侧的 设置 按钮进行修改。
+
-
+ 完成添加后，在插件列表中就能看到刚刚添加的插件，如果需要修改插件的配置，可以点击最右侧的 设置 按钮进行修改。
-### 会话测试插件功能
+
-接来下我们需要测试这个插件的功能是否正常。
+ ### 会话测试插件功能
-点击 返回 按钮回到会话区，然后向助手发送消息：「我应该穿什么？ 」此时助手将会尝试向你询问，了解你的性别与当前的心情。
+ 接来下我们需要测试这个插件的功能是否正常。
-
+ 点击 返回 按钮回到会话区，然后向助手发送消息：「我应该穿什么？ 」此时助手将会尝试向你询问，了解你的性别与当前的心情。
-当回答完毕后，助手将会发起插件的调用，根据你的性别、心情，从服务端获取推荐的衣服数据，并推送给你。最后基于这些信息做一轮文本总结。
+
-
+ 当回答完毕后，助手将会发起插件的调用，根据你的性别、心情，从服务端获取推荐的衣服数据，并推送给你。最后基于这些信息做一轮文本总结。
-当完成这些操作后，你已经了解了添加自定义插件，并在 LobeChat 中使用的基础流程。
+
+ 当完成这些操作后，你已经了解了添加自定义插件，并在 LobeChat 中使用的基础流程。
## 本地插件开发
@@ -223,7 +187,7 @@ export default async (req: Request) => {
由于 LobeChat 默认的插件网关是云端服务 `/api/plugins`，云端服务通过 manifest 上的 `api.url` 地址发送请求，以解决跨域问题。
-针对自定义插件，插件请求需要发送给本地服务， 因此通过在 manifest 中指定网关 ([http://localhost:3400/api/gateway)，LobeChat]() 将会直接请求该地址，然后只需要在该地址下创建对应的网关即可。
+针对自定义插件，插件请求需要发送给本地服务， 因此通过在 manifest 中指定网关 ([http://localhost:3400/api/gateway)，LobeChat](http://localhost:3400/api/gateway\)，LobeChat) 将会直接请求该地址，然后只需要在该地址下创建对应的网关即可。
```ts
import { createLobeChatPluginGateway } from '@lobehub/chat-plugins-gateway';
@@ -241,29 +205,17 @@ export default createLobeChatPluginGateway();
自定义插件的 UI 界面是一个可选项。例如 官方插件 [「🧩 / 🕸 网页内容提取」](https://github.com/lobehub/chat-plugin-web-crawler)，没有实现相应的用户界面。
-
+
如果你希望在插件消息中展示更加丰富的信息，或者包含一些富交互操作，你可以为插件定制一个用户界面。例如下图则为[「搜索引擎」](https://github.com/lobehub/chat-plugin-search-engine)插件的用户界面。
-
+
#### 插件 UI 界面实现
LobeChat 通过 `iframe` 实现插件 ui 的加载，使用 `postMessage` 实现主体与插件的通信。因此， 插件 UI 的实现方式与普通的网页开发一致，你可以使用任何你熟悉的前端框架与开发语言。
-
+
在我们提供的模板中使用了 React + Next.js + [antd](https://ant.design/) 作为前端界面框架，你可以在 [`src/pages/index.tsx`](https://github.com/lobehub/chat-plugin-template/blob/main/src/pages/index.tsx) 中找到用户界面的实现。
@@ -301,7 +253,7 @@ export default Render;
### 插件 Shield
-[](https://github.com/lobehub/lobe-chat-plugins)
+[](https://github.com/lobehub/lobe-chat-plugins)
```markdown
[](https://github.com/lobehub/lobe-chat-plugins)
diff --git a/docs/usage/plugins/store.mdx b/docs/usage/plugins/store.mdx
index 049688cabaa5e..614d1cc5b1cac 100644
--- a/docs/usage/plugins/store.mdx
+++ b/docs/usage/plugins/store.mdx
@@ -15,16 +15,8 @@ tags:
You can access the plugin store by going to `Extension Tools` -> `Plugin Store` in the session toolbar.
-
+
In the plugin store, you can directly install and use plugins in LobeChat.
-
+
diff --git a/docs/usage/plugins/store.zh-CN.mdx b/docs/usage/plugins/store.zh-CN.mdx
index 75fa365cc10e3..84842312e2023 100644
--- a/docs/usage/plugins/store.zh-CN.mdx
+++ b/docs/usage/plugins/store.zh-CN.mdx
@@ -12,16 +12,8 @@ tags:
你可以在会话工具条中的 `扩展工具` -> `插件商店`，进入插件商店。
-
+
插件商店中会在 LobeChat 中可以直接安装并使用的插件。
-
+
diff --git a/docs/usage/providers.mdx b/docs/usage/providers.mdx
index 75be2d036c9dd..88743886dfb4f 100644
--- a/docs/usage/providers.mdx
+++ b/docs/usage/providers.mdx
@@ -20,12 +20,7 @@ tags:
# Using Multiple Model Providers in LobeChat
-
+
In the continuous development of LobeChat, we deeply understand the importance of diversity in model providers for providing AI conversation services to meet the needs of the community. Therefore, we have expanded our support to multiple model providers instead of being limited to a single one, in order to offer users a more diverse and rich selection of conversation options.
diff --git a/docs/usage/providers.zh-CN.mdx b/docs/usage/providers.zh-CN.mdx
index 83abd4c929317..0760cbdf85676 100644
--- a/docs/usage/providers.zh-CN.mdx
+++ b/docs/usage/providers.zh-CN.mdx
@@ -18,12 +18,7 @@ tags:
# 在 LobeChat 中使用多模型服务商
-
+
在 LobeChat 的不断发展过程中，我们深刻理解到在提供 AI 会话服务时模型服务商的多样性对于满足社区需求的重要性。因此，我们不再局限于单一的模型服务商，而是拓展了对多种模型服务商的支持，以便为用户提供更为丰富和多样化的会话选择。
diff --git a/docs/usage/providers/ai21.mdx b/docs/usage/providers/ai21.mdx
index 770835820e546..9fab65620af87 100644
--- a/docs/usage/providers/ai21.mdx
+++ b/docs/usage/providers/ai21.mdx
@@ -12,53 +12,37 @@ tags:
# Using AI21 Labs in LobeChat
-
+
[AI21 Labs](https://www.ai21.com/) is a company focused on artificial intelligence, offering advanced language models and API services designed to help developers and businesses leverage natural language processing technology. Their flagship product, the "Jamba" series of models, can perform complex language understanding and generation tasks, widely utilized in fields such as content creation and conversational systems.
This article will guide you on how to use AI21 Labs within LobeChat.
-### Step 1: Obtain the AI21 Labs API Key
+ ### Step 1: Obtain the AI21 Labs API Key
-- Register and log in to [AI21 Studio](https://studio.ai21.com)
-- Click on the `User Avatar` menu, then select `API Key`
-- Copy and save the generated API key
+ - Register and log in to [AI21 Studio](https://studio.ai21.com)
+ - Click on the `User Avatar` menu, then select `API Key`
+ - Copy and save the generated API key
-
+
-### Step 2: Configure AI21 Labs in LobeChat
+ ### Step 2: Configure AI21 Labs in LobeChat
-- Go to the `Settings` page in LobeChat
-- Under `Language Model`, find the setting for `AI21 Labs`
+ - Go to the `Settings` page in LobeChat
+ - Under `Language Model`, find the setting for `AI21 Labs`
-
+
-- Enter the API key you obtained
-- Choose an AI21 Labs model for your AI assistant to begin the conversation
+ - Enter the API key you obtained
+ - Choose an AI21 Labs model for your AI assistant to begin the conversation
-
-
-
- During use, you may need to pay the API service provider; please refer to the relevant fee policy
- of AI21 Labs.
-
+
+
+ During use, you may need to pay the API service provider; please refer to the relevant fee policy
+ of AI21 Labs.
+
Now you are ready to engage in conversations using the models provided by AI21 Labs in LobeChat.
diff --git a/docs/usage/providers/ai21.zh-CN.mdx b/docs/usage/providers/ai21.zh-CN.mdx
index d6f1927c37a25..57c0c87d82833 100644
--- a/docs/usage/providers/ai21.zh-CN.mdx
+++ b/docs/usage/providers/ai21.zh-CN.mdx
@@ -10,52 +10,36 @@ tags:
# 在 LobeChat 中使用 AI21 Labs
-
+
-[AI21 Labs](https://www.ai21.com/) 是一家专注于人工智能的公司，提供先进的语言模型和API服务，旨在帮助开发者和企业利用自然语言处理技术。其旗舰产品 "Jamba" 系列模型能够进行复杂的语言理解和生成任务，广泛应用于内容创作、对话系统等领域。
+[AI21 Labs](https://www.ai21.com/) 是一家专注于人工智能的公司，提供先进的语言模型和 API 服务，旨在帮助开发者和企业利用自然语言处理技术。其旗舰产品 "Jamba" 系列模型能够进行复杂的语言理解和生成任务，广泛应用于内容创作、对话系统等领域。
本文将指导你如何在 LobeChat 中使用 AI21 Labs。
-### 步骤一：获得 AI21 Labs 的 API Key
+ ### 步骤一：获得 AI21 Labs 的 API Key
-- 注册并登录 [AI21 Studio](https://studio.ai21.com)
-- 点击 `用户头像` 菜单，点击 `API Key`
-- 复制并保存生成的 API 密钥
+ - 注册并登录 [AI21 Studio](https://studio.ai21.com)
+ - 点击 `用户头像` 菜单，点击 `API Key`
+ - 复制并保存生成的 API 密钥
-
+
-### 步骤二：在 LobeChat 中配置 AI21 Labs
+ ### 步骤二：在 LobeChat 中配置 AI21 Labs
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `AI21labs` 的设置项
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `AI21labs` 的设置项
-
+
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 AI21 Labs 的模型即可开始对话
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 AI21 Labs 的模型即可开始对话
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 AI21 Labs 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 AI21 Labs 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 AI21 Labs 提供的模型进行对话了。
diff --git a/docs/usage/providers/ai360.mdx b/docs/usage/providers/ai360.mdx
index 9bb1cd558088c..ddae022494ff9 100644
--- a/docs/usage/providers/ai360.mdx
+++ b/docs/usage/providers/ai360.mdx
@@ -10,53 +10,37 @@ tags:
# Using the 360AI in LobeChat
-
+
The [360AI](https://ai.360.com/) is a cognitive general model independently developed by 360 Company, aimed at providing powerful natural language processing capabilities for enterprises and developers. This model has been upgraded to version 4.0 and supports various application scenarios, including conversational services, image generation, vector database services, and more.
This article will guide you on how to use the 360AI in LobeChat.
-### Step 1: Obtain the 360AI API Key
+ ### Step 1: Obtain the 360AI API Key
-- Register and log in to the [360AI API Open Platform](https://ai.360.com/platform/keys)
-- Click on the `API Keys` menu on the left
-- Create an API key and copy it
+ - Register and log in to the [360AI API Open Platform](https://ai.360.com/platform/keys)
+ - Click on the `API Keys` menu on the left
+ - Create an API key and copy it
-
+
-### Step 2: Configure 360AI in LobeChat
+ ### Step 2: Configure 360AI in LobeChat
-- Access the `Settings` interface in LobeChat
-- Under `Language Models`, find the option for `360`
+ - Access the `Settings` interface in LobeChat
+ - Under `Language Models`, find the option for `360`
-
+
-- Enter the API key you obtained
-- Choose a 360AI model for your AI assistant to start chatting
+ - Enter the API key you obtained
+ - Choose a 360AI model for your AI assistant to start chatting
-
-
-
- Please note that you may need to pay the API service provider during use, refer to the relevant
- pricing policy of the 360AI.
-
+
+
+ Please note that you may need to pay the API service provider during use, refer to the relevant
+ pricing policy of the 360AI.
+
You can now use the models provided by the 360AI for conversations in LobeChat.
diff --git a/docs/usage/providers/ai360.zh-CN.mdx b/docs/usage/providers/ai360.zh-CN.mdx
index 65e27c93b13c1..b1c7b8f490363 100644
--- a/docs/usage/providers/ai360.zh-CN.mdx
+++ b/docs/usage/providers/ai360.zh-CN.mdx
@@ -8,54 +8,38 @@ tags:
- Web UI
---
-# 在 LobeChat 中使用360智脑
+# 在 LobeChat 中使用 360 智脑
-
+
-[360智脑](https://ai.360.com/)是360公司自主研发的认知型通用大模型，旨在为企业和开发者提供强大的自然语言处理能力。该模型已升级至4.0版本，能够支持多种应用场景，包括对话服务、图片生成、向量数据库服务等。
+[360 智脑](https://ai.360.com/)是 360 公司自主研发的认知型通用大模型，旨在为企业和开发者提供强大的自然语言处理能力。该模型已升级至 4.0 版本，能够支持多种应用场景，包括对话服务、图片生成、向量数据库服务等。
-本文将指导你如何在 LobeChat 中使用360智脑。
+本文将指导你如何在 LobeChat 中使用 360 智脑。
-### 步骤一：获得360智脑的 API Key
+ ### 步骤一：获得 360 智脑的 API Key
-- 注册并登录 [360智脑 API 开放平台](https://ai.360.com/platform/keys)
-- 点击左侧 `API Keys` 菜单
-- 创建一个 API 密钥并复制
+ - 注册并登录 [360 智脑 API 开放平台](https://ai.360.com/platform/keys)
+ - 点击左侧 `API Keys` 菜单
+ - 创建一个 API 密钥并复制
-
+
-### 步骤二：在 LobeChat 中配置360智脑
+ ### 步骤二：在 LobeChat 中配置 360 智脑
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `360` 的设置项
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `360` 的设置项
-
+
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个360智脑的模型即可开始对话
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 360 智脑的模型即可开始对话
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考360智脑的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 360 智脑的相关费用政策。
+
-至此你已经可以在 LobeChat 中使用360智脑提供的模型进行对话了。
+至此你已经可以在 LobeChat 中使用 360 智脑提供的模型进行对话了。
diff --git a/docs/usage/providers/anthropic.mdx b/docs/usage/providers/anthropic.mdx
index c65215909d3a3..bd24ed159875a 100644
--- a/docs/usage/providers/anthropic.mdx
+++ b/docs/usage/providers/anthropic.mdx
@@ -13,66 +13,44 @@ tags:
# Using Anthropic Claude in LobeChat
-
+
The Anthropic Claude API is now available for everyone to use. This document will guide you on how to use [Anthropic Claude](https://www.anthropic.com/api) in LobeChat:
+ ### Step 1: Obtain Anthropic Claude API Key
-### Step 1: Obtain Anthropic Claude API Key
+ - Create an [Anthropic Claude API](https://www.anthropic.com/api) account.
+ - Get your [API key](https://console.anthropic.com/settings/keys).
-- Create an [Anthropic Claude API](https://www.anthropic.com/api) account.
-- Get your [API key](https://console.anthropic.com/settings/keys).
+
-
+
+ The Claude API currently offers $5 of free credits, but it is only available in certain specific
+ countries/regions. You can go to Dashboard > Claim to see if it is applicable to your
+ country/region.
+
-
- The Claude API currently offers $5 of free credits, but it is only available in certain specific
- countries/regions. You can go to Dashboard > Claim to see if it is applicable to your
- country/region.
-
+ - Set up your billing for the API key to work on [https://console.anthropic.com/settings/plans](https://console.anthropic.com/settings/plans) (choose the "Build" plan so you can add credits and only pay for usage).
-- Set up your billing for the API key to work on [https://console.anthropic.com/settings/plans](https://console.anthropic.com/settings/plans) (choose the "Build" plan so you can add credits and only pay for usage).
+
-
+ ### Step 2: Configure Anthropic Claude in LobeChat
-### Step 2: Configure Anthropic Claude in LobeChat
+ - Access the `Settings` interface in LobeChat.
+ - Find the setting for `Anthropic Claude` under `Language Models`.
-- Access the `Settings` interface in LobeChat.
-- Find the setting for `Anthropic Claude` under `Language Models`.
+
-
+ - Enter the obtained API key.
+ - Choose an Anthropic Claude model for your AI assistant to start the conversation.
-- Enter the obtained API key.
-- Choose an Anthropic Claude model for your AI assistant to start the conversation.
-
-
-
-
- During usage, you may need to pay the API service provider. Please refer to Anthropic Claude's
- relevant pricing policies.
-
+
+
+ During usage, you may need to pay the API service provider. Please refer to Anthropic Claude's
+ relevant pricing policies.
+
You can now engage in conversations using the models provided by Anthropic Claude in LobeChat.
diff --git a/docs/usage/providers/anthropic.zh-CN.mdx b/docs/usage/providers/anthropic.zh-CN.mdx
index 866d9e8e4e9f4..0d6ba1c251e7f 100644
--- a/docs/usage/providers/anthropic.zh-CN.mdx
+++ b/docs/usage/providers/anthropic.zh-CN.mdx
@@ -12,64 +12,42 @@ tags:
# 在 LobeChat 中使用 Anthropic Claude
-
+
-Anthropic Claude API 现在可供所有人使用, 本文档将指导你如何在 LobeChat 中使用 [Anthropic Claude](https://www.anthropic.com/api):
+Anthropic Claude API 现在可供所有人使用，本文档将指导你如何在 LobeChat 中使用 [Anthropic Claude](https://www.anthropic.com/api):
+ ### 步骤一：获取 Anthropic Claude API 密钥
-### 步骤一：获取 Anthropic Claude API 密钥
+ - 创建一个 [Anthropic Claude API](https://www.anthropic.com/api) 帐户
+ - 获取您的 [API 密钥](https://console.anthropic.com/settings/keys)
-- 创建一个 [Anthropic Claude API](https://www.anthropic.com/api) 帐户
-- 获取您的 [API 密钥](https://console.anthropic.com/settings/keys)
+
-
+
+ Claude API 现在提供 5 美元的免费积分，但是，它仅适用于某些特定国家 / 地区，您可以转到 Dashboard >
+ Claim 查看它是否适用于您所在的国家 / 地区。
+
-
- Claude API 现在提供 5 美元的免费积分，但是，它仅适用于某些特定国家/地区，您可以转到 Dashboard >
- Claim 查看它是否适用于您所在的国家/地区。
-
+ - 设置您的账单，让 API 密钥在 [https://console.anthropic.com/settings/plans](https://console.anthropic.com/settings/plans) 上工作（选择 “生成” 计划，以便您可以添加积分并仅为使用量付费）
-- 设置您的账单，让 API 密钥在 https://console.anthropic.com/settings/plans 上工作（选择“生成”计划，以便您可以添加积分并仅为使用量付费）
+
-
+ ### 步骤二：在 LobeChat 中配置 Anthropic Claude
-### 步骤二：在 LobeChat 中配置 Anthropic Claude
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`Anthropic Claude`的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到`Anthropic Claude`的设置项
+
-
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 Anthropic Claude 的模型即可开始对话
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 Anthropic Claude 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Anthropic Claude 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Anthropic Claude 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Anthropic Claude 提供的模型进行对话了。
diff --git a/docs/usage/providers/azure.mdx b/docs/usage/providers/azure.mdx
index 102039b70031f..520c7715cfe7f 100644
--- a/docs/usage/providers/azure.mdx
+++ b/docs/usage/providers/azure.mdx
@@ -14,75 +14,45 @@ tags:
# Using Azure OpenAI in LobeChat
-
+
This document will guide you on how to use [Azure OpenAI](https://oai.azure.com/) in LobeChat:
+ ### Step 1: Obtain Azure OpenAI API Key
-### Step 1: Obtain Azure OpenAI API Key
+ - If you haven't registered yet, you need to create an [Azure OpenAI account](https://oai.azure.com/).
-- If you haven't registered yet, you need to create an [Azure OpenAI account](https://oai.azure.com/).
+
-
+ - After registration, go to the `Deployments` page and create a new deployment with your selected model.
-- After registration, go to the `Deployments` page and create a new deployment with your selected model.
+ 
-
+
-
+ - Navigate to the `Chat` page and click on `View Code` to obtain your endpoint and key.
-- Navigate to the `Chat` page and click on `View Code` to obtain your endpoint and key.
+
-
+
-
+ ### Step 2: Configure Azure OpenAI in LobeChat
-### Step 2: Configure Azure OpenAI in LobeChat
+ - Access the `Settings` interface in LobeChat.
+ - Find the setting for `Azure OpenAI` under `Language Model`.
-- Access the `Settings` interface in LobeChat.
-- Find the setting for `Azure OpenAI` under `Language Model`.
+
-
+ - Enter the API key you obtained.
+ - Choose an Azure OpenAI model for your AI assistant to start the conversation.
-- Enter the API key you obtained.
-- Choose an Azure OpenAI model for your AI assistant to start the conversation.
-
-
-
-
- During usage, you may need to pay the API service provider. Please refer to Azure OpenAI's
- relevant pricing policies.
-
+
+
+ During usage, you may need to pay the API service provider. Please refer to Azure OpenAI's
+ relevant pricing policies.
+
Now you can engage in conversations using the models provided by Azure OpenAI in LobeChat.
diff --git a/docs/usage/providers/azure.zh-CN.mdx b/docs/usage/providers/azure.zh-CN.mdx
index 9346a279ba6f9..ac8d399c1f83e 100644
--- a/docs/usage/providers/azure.zh-CN.mdx
+++ b/docs/usage/providers/azure.zh-CN.mdx
@@ -9,71 +9,42 @@ tags:
# 在 LobeChat 中使用 Azure OpenAI
-
+
本文档将指导你如何在 LobeChat 中使用 [Azure OpenAI](https://oai.azure.com/):
+ ### 步骤一：获取 Azure OpenAI API 密钥
-### 步骤一：获取 Azure OpenAI API 密钥
+ - 如果尚未注册，则必须注册 [Azure OpenAI 帐户](https://oai.azure.com/)。
-- 如果尚未注册，则必须注册 [Azure OpenAI 帐户](https://oai.azure.com/)。
+
-
+ - 注册完毕后，转到 `Deployments` 页面，然后使用您选择的模型创建新部署。
-- 注册完毕后，转到 `Deployments` 页面，然后使用您选择的模型创建新部署。
+
-
+ - 转到 `Chat` 页面，然后单击 `View Code` 以获取您的终结点和密钥。
-- 转到 `Chat` 页面，然后单击 `View Code` 以获取您的终结点和密钥。
+
-
-
+
-### 步骤二：在 LobeChat 中配置 Azure OpenAI
+ ### 步骤二：在 LobeChat 中配置 Azure OpenAI
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到`Azure OpenAI`的设置项
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`Azure OpenAI`的设置项
-
+
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 Azure OpenAI 的模型即可开始对话
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 Azure OpenAI 的模型即可开始对话
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Azure OpenAI 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Azure OpenAI 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Azure OpenAI 提供的模型进行对话了。
diff --git a/docs/usage/providers/baichuan.mdx b/docs/usage/providers/baichuan.mdx
index f1ee9967c04a8..4d1579e0ae73a 100644
--- a/docs/usage/providers/baichuan.mdx
+++ b/docs/usage/providers/baichuan.mdx
@@ -13,52 +13,34 @@ tags:
# Using Baichuan in LobeChat
-
+
This article will guide you on how to use Baichuan in LobeChat:
+ ### Step 1: Obtain Baichuan Intelligent API Key
-### Step 1: Obtain Baichuan Intelligent API Key
+ - Create a [Baichuan Intelligent](https://platform.baichuan-ai.com/homePage) account
+ - Create and obtain an [API key](https://platform.baichuan-ai.com/console/apikey)
-- Create a [Baichuan Intelligent](https://platform.baichuan-ai.com/homePage) account
-- Create and obtain an [API key](https://platform.baichuan-ai.com/console/apikey)
+
-
+ ### Step 2: Configure Baichuan in LobeChat
-### Step 2: Configure Baichuan in LobeChat
+ - Visit the `Settings` interface in LobeChat
+ - Find the setting for `Baichuan` under `Language Model`
-- Visit the `Settings` interface in LobeChat
-- Find the setting for `Baichuan` under `Language Model`
+
-
+ - Enter the obtained API key
+ - Choose a Baichuan model for your AI assistant to start the conversation
-- Enter the obtained API key
-- Choose a Baichuan model for your AI assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider, please refer to Baichuan's relevant
- pricing policies.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to Baichuan's relevant
+ pricing policies.
+
You can now use the models provided by Baichuan for conversation in LobeChat.
diff --git a/docs/usage/providers/baichuan.zh-CN.mdx b/docs/usage/providers/baichuan.zh-CN.mdx
index f4101bef76650..67561265009a9 100644
--- a/docs/usage/providers/baichuan.zh-CN.mdx
+++ b/docs/usage/providers/baichuan.zh-CN.mdx
@@ -11,51 +11,33 @@ tags:
# 在 LobeChat 中使用百川
-
+
本文将指导你如何在 LobeChat 中使用百川：
+ ### 步骤一：获取百川智能 API 密钥
-### 步骤一：获取百川智能 API 密钥
+ - 创建一个[百川智能](https://platform.baichuan-ai.com/homePage)账户
+ - 创建并获取 [API 密钥](https://platform.baichuan-ai.com/console/apikey)
-- 创建一个[百川智能](https://platform.baichuan-ai.com/homePage)账户
-- 创建并获取 [API 密钥](https://platform.baichuan-ai.com/console/apikey)
+
-
+ ### 步骤二：在 LobeChat 中配置百川
-### 步骤二：在 LobeChat 中配置百川
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`百川`的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到`百川`的设置项
+
-
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个百川的模型即可开始对话
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个百川的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考百川的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考百川的相关费用政策。
+
至此你已经可以在 LobeChat 中使用百川提供的模型进行对话了。
diff --git a/docs/usage/providers/bedrock.mdx b/docs/usage/providers/bedrock.mdx
index 4770bd2857cf4..e02f2e683db21 100644
--- a/docs/usage/providers/bedrock.mdx
+++ b/docs/usage/providers/bedrock.mdx
@@ -14,126 +14,77 @@ tags:
# Using Amazon Bedrock in LobeChat
-
+
Amazon Bedrock is a fully managed foundational model API service that allows users to access models from leading AI companies (such as AI21 Labs, Anthropic, Cohere, Meta, Stability AI) and Amazon's own foundational models.
This document will guide you on how to use Amazon Bedrock in LobeChat:
-### Step 1: Grant Access to Amazon Bedrock Models in AWS
+ ### Step 1: Grant Access to Amazon Bedrock Models in AWS
-- Access and log in to the [AWS Console](https://console.aws.amazon.com/)
-- Search for `bedrock` and enter the `Amazon Bedrock` service
+ - Access and log in to the [AWS Console](https://console.aws.amazon.com/)
+ - Search for `bedrock` and enter the `Amazon Bedrock` service
-
+
-- Select `Models access` from the left menu
+ - Select `Models access` from the left menu
-
+
-- Open model access permissions based on your needs
+ - Open model access permissions based on your needs
-
+
-Some models may require additional information from you
+ Some models may require additional information from you
-### Step 2: Obtain API Access Keys
+ ### Step 2: Obtain API Access Keys
-- Continue searching for IAM in the AWS console and enter the IAM service
+ - Continue searching for IAM in the AWS console and enter the IAM service
-
+
-- In the `Users` menu, create a new IAM user
+ - In the `Users` menu, create a new IAM user
-
+
-- Enter the user name in the pop-up dialog box
+ - Enter the user name in the pop-up dialog box
-
+
-- Add permissions for this user or join an existing user group to ensure access to Amazon Bedrock
+ - Add permissions for this user or join an existing user group to ensure access to Amazon Bedrock
-
+
-- Create an access key for the added user
+ - Create an access key for the added user
-
+
-- Copy and securely store the access key and secret access key, as they will be needed later
+ - Copy and securely store the access key and secret access key, as they will be needed later
-
+
-
- Please securely store the keys as they will only be shown once. If you lose them accidentally, you
- will need to create a new access key.
-
+
+ Please securely store the keys as they will only be shown once. If you lose them accidentally, you
+ will need to create a new access key.
+
-### Step 3: Configure Amazon Bedrock in LobeChat
+ ### Step 3: Configure Amazon Bedrock in LobeChat
-- Access the `Settings` interface in LobeChat
-- Find the setting for `Amazon Bedrock` under `Language Models` and open it
+ - Access the `Settings` interface in LobeChat
+ - Find the setting for `Amazon Bedrock` under `Language Models` and open it
-
+
-- Open Amazon Bedrock and enter the obtained access key and secret access key
-- Choose an Amazon Bedrock model for your assistant to start the conversation
+ - Open Amazon Bedrock and enter the obtained access key and secret access key
+ - Choose an Amazon Bedrock model for your assistant to start the conversation
-
-
-
- You may incur charges while using the API service, please refer to Amazon Bedrock's pricing
- policy.
-
+
+
+ You may incur charges while using the API service, please refer to Amazon Bedrock's pricing
+ policy.
+
You can now engage in conversations using the models provided by Amazon Bedrock in LobeChat.
diff --git a/docs/usage/providers/bedrock.zh-CN.mdx b/docs/usage/providers/bedrock.zh-CN.mdx
index ef70dc3b60e3c..e19563b4a2b63 100644
--- a/docs/usage/providers/bedrock.zh-CN.mdx
+++ b/docs/usage/providers/bedrock.zh-CN.mdx
@@ -11,124 +11,75 @@ tags:
# 在 LobeChat 中使用 Amazon Bedrock
-
+
-Amazon Bedrock 是一个完全托管的基础模型API服务,允许用户通过API访问来自领先AI公司(如AI21 Labs、Anthropic、Cohere、Meta、Stability AI)和Amazon自家的基础模型。
+Amazon Bedrock 是一个完全托管的基础模型 API 服务，允许用户通过 API 访问来自领先 AI 公司 (如 AI21 Labs、Anthropic、Cohere、Meta、Stability AI) 和 Amazon 自家的基础模型。
本文档将指导你如何在 LobeChat 中使用 Amazon Bedrock:
-### 步骤一：在 AWS 中打开 Amazon Bedrock 模型的访问权限
+ ### 步骤一：在 AWS 中打开 Amazon Bedrock 模型的访问权限
-- 访问并登录 [AWS Console](https://console.aws.amazon.com/)
-- 搜索 beckrock 并进入 `Amazon Bedrock` 服务
+ - 访问并登录 [AWS Console](https://console.aws.amazon.com/)
+ - 搜索 beckrock 并进入 `Amazon Bedrock` 服务
-
+
-- 在左侧菜单中选择 `Models acess`
+ - 在左侧菜单中选择 `Models acess`
-
+
-- 根据你所需要的模型，打开模型访问权限
+ - 根据你所需要的模型，打开模型访问权限
-
+
-某些模型可能需要你提供额外的信息
+ 某些模型可能需要你提供额外的信息
-### 步骤二：获取 API 访问密钥
+ ### 步骤二：获取 API 访问密钥
-- 继续在 AWS console 中搜索 IAM，进入 IAM 服务
+ - 继续在 AWS console 中搜索 IAM，进入 IAM 服务
-
+
-- 在 `用户` 菜单中，创建一个新的 IAM 用户
+ - 在 `用户` 菜单中，创建一个新的 IAM 用户
-
+
-- 在弹出的对话框中，输入用户名称
+ - 在弹出的对话框中，输入用户名称
-
+
-- 为这个用户添加权限，或者加入一个已有的用户组，确保用户拥有 Amazon Bedrock 的访问权限
+ - 为这个用户添加权限，或者加入一个已有的用户组，确保用户拥有 Amazon Bedrock 的访问权限
-
+
-- 为已添加的用户创建访问密钥
+ - 为已添加的用户创建访问密钥
-
+
-- 复制并妥善保存访问密钥以及秘密访问密钥，后续将会用到
+ - 复制并妥善保存访问密钥以及秘密访问密钥，后续将会用到
-
+
-
- 请安全地存储密钥，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新访问密钥。
-
+
+ 请安全地存储密钥，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新访问密钥。
+
-### 步骤三：在 LobeChat 中配置 Amazon Bedrock
+ ### 步骤三：在 LobeChat 中配置 Amazon Bedrock
-- 访问LobeChat的`设置`界面
-- 在`语言模型`下找到`Amazon Bedrock`的设置项并打开
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`Amazon Bedrock`的设置项并打开
-
+
-- 打开 Amazon Bedrock 并填入获得的访问密钥与秘密访问密钥
-- 为你的助手选择一个 Amazone Bedrock 的模型即可开始对话
+ - 打开 Amazon Bedrock 并填入获得的访问密钥与秘密访问密钥
+ - 为你的助手选择一个 Amazone Bedrock 的模型即可开始对话
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Amazon Bedrock 的费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Amazon Bedrock 的费用政策。
+
至此你已经可以在 LobeChat 中使用 Amazone Bedrock 提供的模型进行对话了。
diff --git a/docs/usage/providers/cloudflare.mdx b/docs/usage/providers/cloudflare.mdx
index a718ccc58c4e2..98dbedd167073 100644
--- a/docs/usage/providers/cloudflare.mdx
+++ b/docs/usage/providers/cloudflare.mdx
@@ -12,71 +12,48 @@ tags:
# Using Cloudflare Workers AI in LobeChat
-
+
[Cloudflare Workers AI](https://www.cloudflare.com/developer-platform/products/workers-ai/) is a service that integrates AI capabilities into the Cloudflare Workers serverless computing platform. Its core functionality lies in delivering fast, scalable computing power through Cloudflare's global network, thereby reducing operational overhead.
This document will guide you on how to use Cloudflare Workers AI in LobeChat:
+ ### Step 1: Obtain Your Cloudflare Workers AI API Key
-### Step 1: Obtain Your Cloudflare Workers AI API Key
+ - Visit the [Cloudflare website](https://www.cloudflare.com/) and sign up for an account.
+ - Log in to the [Cloudflare dashboard](https://dash.cloudflare.com/).
+ - In the left-hand menu, locate the `AI` > `Workers AI` option.
-- Visit the [Cloudflare website](https://www.cloudflare.com/) and sign up for an account.
-- Log in to the [Cloudflare dashboard](https://dash.cloudflare.com/).
-- In the left-hand menu, locate the `AI` > `Workers AI` option.
+
-
+ - In the `Using REST API` section, click the `Create Workers AI API Token` button.
+ - In the drawer dialog, copy and save your `API token`.
+ - Also, copy and save your `Account ID`.
-- In the `Using REST API` section, click the `Create Workers AI API Token` button.
-- In the drawer dialog, copy and save your `API token`.
-- Also, copy and save your `Account ID`.
+
-
+
+ - Please store your API token securely, as it will only be displayed once. If you accidentally lose it, you will need to create a new token.
+
-
+ ### Step 2: Configure Cloudflare Workers AI in LobeChat
-- Please store your API token securely, as it will only be displayed once. If you accidentally lose it, you will need to create a new token.
+ - Go to the `Settings` interface in LobeChat.
+ - Under `Language Model`, find the `Cloudflare` settings.
-
+
-### Step 2: Configure Cloudflare Workers AI in LobeChat
+ - Enter the `API Token` you obtained.
+ - Input your `Account ID`.
+ - Choose a Cloudflare Workers AI model for your AI assistant to start the conversation.
-- Go to the `Settings` interface in LobeChat.
-- Under `Language Model`, find the `Cloudflare` settings.
-
-
-
-- Enter the `API Token` you obtained.
-- Input your `Account ID`.
-- Choose a Cloudflare Workers AI model for your AI assistant to start the conversation.
-
-
-
-
- You may incur charges while using the API service, please refer to Cloudflare's pricing policy for
- details.
-
+
+
+ You may incur charges while using the API service, please refer to Cloudflare's pricing policy for
+ details.
+
At this point, you can start conversing with the model provided by Cloudflare Workers AI in LobeChat.
diff --git a/docs/usage/providers/cloudflare.zh-CN.mdx b/docs/usage/providers/cloudflare.zh-CN.mdx
index 8d8c6467e3d89..95769aea844de 100644
--- a/docs/usage/providers/cloudflare.zh-CN.mdx
+++ b/docs/usage/providers/cloudflare.zh-CN.mdx
@@ -12,68 +12,47 @@ tags:
# 在 LobeChat 中使用 Cloudflare Workers AI
-
+
[Cloudflare Workers AI](https://www.cloudflare.com/developer-platform/products/workers-ai/) 是一种将人工智能能力集成到 Cloudflare Workers 无服务器计算平台的服务。其核心功能在于通过 Cloudflare 的全球网络提供快速、可扩展的计算能力，降低运维开销。
本文档将指导你如何在 LobeChat 中使用 Cloudflare Workers AI:
+ ### 步骤一：获取 Cloudflare Workers AI 的 API Key
-### 步骤一: 获取 Cloudflare Workers AI 的 API Key
+ - 访问 [Cloudflare 官网](https://www.cloudflare.com/) 并注册一个账号。
+ - 登录 [Cloudflare 控制台](https://dash.cloudflare.com/).
+ - 在左侧的菜单中找到 `AI` > `Workers AI` 选项。
-- 访问 [Cloudflare 官网](https://www.cloudflare.com/) 并注册一个账号。
-- 登录 [Cloudflare 控制台](https://dash.cloudflare.com/).
-- 在左侧的菜单中找到 `AI` > `Workers AI` 选项。
+
-
+ - 在 `使用 REST API` 中点击 `创建 Workers AI API 令牌` 按钮
+ - 在弹出的侧边栏中复制并保存你的 `API 令牌`
+ - 同时也复制并保存你的 `账户ID`
-- 在 `使用 REST API` 中点击 `创建 Workers AI API 令牌` 按钮
-- 在弹出的侧边栏中复制并保存你的 `API 令牌`
-- 同时也复制并保存你的 `账户ID`
+
-
+
+ - 请安全地存储 API 令牌，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新令牌。
+
-
- - 请安全地存储 API 令牌，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新令牌。
-
+ ### 步骤二：在 LobeChat 中配置 Cloudflare Workers AI
-### 步骤二: 在 LobeChat 中配置 Cloudflare Workers AI
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `Cloudflare` 的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `Cloudflare` 的设置项
+
-
+ - 填入获得的 `API 令牌`
+ - 填入你的`账户ID`
+ - 为你的 AI 助手选择一个 Cloudflare Workers AI 的模型即可开始对话
-- 填入获得的 `API 令牌`
-- 填入你的`账户ID`
-- 为你的 AI 助手选择一个 Cloudflare Workers AI 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Cloudflare 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Cloudflare 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Cloudflare Workers AI 提供的模型进行对话了。
diff --git a/docs/usage/providers/deepseek.mdx b/docs/usage/providers/deepseek.mdx
index fb5c77c42e45d..ccda945490544 100644
--- a/docs/usage/providers/deepseek.mdx
+++ b/docs/usage/providers/deepseek.mdx
@@ -13,78 +13,52 @@ tags:
# Using DeepSeek in LobeChat
-
+
[DeepSeek](https://www.deepseek.com/) is an advanced open-source Large Language Model (LLM). The latest version, DeepSeek-V2, has made significant optimizations in architecture and performance, reducing training costs by 42.5% and inference costs by 93.3%.
This document will guide you on how to use DeepSeek in LobeChat:
+ ### Step 1: Obtain DeepSeek API Key
-### Step 1: Obtain DeepSeek API Key
+ - First, you need to register and log in to the [DeepSeek](https://platform.deepseek.com/) open platform.
-- First, you need to register and log in to the [DeepSeek](https://platform.deepseek.com/) open platform.
+ New users will receive a free quota of 500M Tokens
-New users will receive a free quota of 500M Tokens
+ - Go to the `API keys` menu and click on `Create API Key`.
-- Go to the `API keys` menu and click on `Create API Key`.
+
-
+ - Enter the API key name in the pop-up dialog box.
-- Enter the API key name in the pop-up dialog box.
+
-
+ - Copy the generated API key and save it securely.
-- Copy the generated API key and save it securely.
+
-
+
+ Please store the key securely as it will only appear once. If you accidentally lose it, you will
+ need to create a new key.
+
-
- Please store the key securely as it will only appear once. If you accidentally lose it, you will
- need to create a new key.
-
+ ### Step 2: Configure DeepSeek in LobeChat
-### Step 2: Configure DeepSeek in LobeChat
+ - Access the `App Settings` interface in LobeChat.
+ - Find the setting for `DeepSeek` under `Language Models`.
-- Access the `App Settings` interface in LobeChat.
-- Find the setting for `DeepSeek` under `Language Models`.
+
-
+ - Open DeepSeek and enter the obtained API key.
+ - Choose a DeepSeek model for your assistant to start the conversation.
-- Open DeepSeek and enter the obtained API key.
-- Choose a DeepSeek model for your assistant to start the conversation.
-
-
-
-
- You may need to pay the API service provider during usage, please refer to DeepSeek's relevant
- pricing policies.
-
+
+
+ You may need to pay the API service provider during usage, please refer to DeepSeek's relevant
+ pricing policies.
+
You can now engage in conversations using the models provided by Deepseek in LobeChat.
diff --git a/docs/usage/providers/deepseek.zh-CN.mdx b/docs/usage/providers/deepseek.zh-CN.mdx
index 7cbb4a016809e..8b45ae510197b 100644
--- a/docs/usage/providers/deepseek.zh-CN.mdx
+++ b/docs/usage/providers/deepseek.zh-CN.mdx
@@ -10,76 +10,50 @@ tags:
# 在 LobeChat 中使用 DeepSeek
-
+
-[DeepSeek](https://www.deepseek.com/) 是一款先进的开源大型语言模型（LLM）。最新版本 DeepSeek-V2 在架构和性能上进行了显著优化，同时训练成本降低了42.5%,推理成本降低了93.3%。
+[DeepSeek](https://www.deepseek.com/) 是一款先进的开源大型语言模型（LLM）。最新版本 DeepSeek-V2 在架构和性能上进行了显著优化，同时训练成本降低了 42.5%, 推理成本降低了 93.3%。
本文档将指导你如何在 LobeChat 中使用 DeepSeek:
+ ### 步骤一：获取 DeepSeek API 密钥
-### 步骤一：获取 DeepSeek API 密钥
+ - 首先，你需要注册并登录 [DeepSeek](https://platform.deepseek.com/) 开放平台
-- 首先，你需要注册并登录 [DeepSeek](https://platform.deepseek.com/) 开放平台
+ 当前新用户将会获赠 500M Tokens 的免费额度
-当前新用户将会获赠 500M Tokens 的免费额度
+ - 进入 `API keys` 菜单，并点击 `创建 API Key`
-- 进入 `API keys` 菜单，并点击 `创建 API Key`
+
-
+ - 在弹出的对话框中输入 API 密钥名称
-- 在弹出的对话框中输入 API 密钥名称
+
-
+ - 复制得到的 API 密钥并妥善保存
-- 复制得到的 API 密钥并妥善保存
+
-
+
+ 请安全地存储密钥，因为它只会出现一次。如果你意外丢失它，您将需要创建一个新密钥。
+
-
- 请安全地存储密钥，因为它只会出现一次。如果你意外丢失它，您将需要创建一个新密钥。
-
+ ### 步骤二：在 LobeChat 中配置 DeepSeek
-### 步骤二：在 LobeChat 中配置 DeepSeek
+ - 访问 LobeChat 的 `应用设置`界面
+ - 在 `语言模型` 下找到 `DeepSeek` 的设置项
-- 访问 LobeChat 的 `应用设置`界面
-- 在 `语言模型` 下找到 `DeepSeek` 的设置项
+
-
+ - 打开 DeepSeek 并填入获取的 API 密钥
+ - 为你的助手选择一个 DeepSeek 模型即可开始对话
-- 打开 DeepSeek 并填入获取的 API 密钥
-- 为你的助手选择一个 DeepSeek 模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 DeepSeek 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 DeepSeek 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Deepseek 提供的模型进行对话了。
diff --git a/docs/usage/providers/fireworksai.mdx b/docs/usage/providers/fireworksai.mdx
index 64e93632897a8..297b776748878 100644
--- a/docs/usage/providers/fireworksai.mdx
+++ b/docs/usage/providers/fireworksai.mdx
@@ -12,66 +12,46 @@ tags:
# Using Fireworks AI in LobeChat
-
+
[Fireworks.ai](https://fireworks.ai/) is a high-performance generative AI model inference platform that allows users to access and utilize various models through its API. The platform supports multiple modalities, including text and visual language models, and offers features like function calls and JSON schemas to enhance the flexibility of application development.
This article will guide you on how to use Fireworks AI in LobeChat.
-### Step 1: Obtain an API Key for Fireworks AI
+ ### Step 1: Obtain an API Key for Fireworks AI
-- Log in to the [Fireworks.ai Console](https://fireworks.ai/account/api-keys)
-- Navigate to the `User` page and click on `API Keys`
-- Create a new API key
+ - Log in to the [Fireworks.ai Console](https://fireworks.ai/account/api-keys)
+ - Navigate to the `User` page and click on `API Keys`
+ - Create a new API key
-
+
-- Copy and securely save the generated API key
+ - Copy and securely save the generated API key
-
+
-
- Please store the key securely, as it will appear only once. If you accidentally lose it, you will
- need to create a new key.
-
+
+ Please store the key securely, as it will appear only once. If you accidentally lose it, you will
+ need to create a new key.
+
-### Step 2: Configure Fireworks AI in LobeChat
+ ### Step 2: Configure Fireworks AI in LobeChat
-- Access the `Settings` interface in LobeChat
-- Under `Language Model`, locate the settings for `Fireworks AI`
+ - Access the `Settings` interface in LobeChat
+ - Under `Language Model`, locate the settings for `Fireworks AI`
-
+
-- Enter the obtained API key
-- Select a Fireworks AI model for your AI assistant to start a conversation
+ - Enter the obtained API key
+ - Select a Fireworks AI model for your AI assistant to start a conversation
-
-
-
- Please note that you may need to pay fees to the API service provider during use; refer to
- Fireworks AI's pricing policy for details.
-
+
+
+ Please note that you may need to pay fees to the API service provider during use; refer to
+ Fireworks AI's pricing policy for details.
+
You are now ready to use the models provided by Fireworks AI for conversations in LobeChat.
diff --git a/docs/usage/providers/fireworksai.zh-CN.mdx b/docs/usage/providers/fireworksai.zh-CN.mdx
index 8fab3414de871..5d861917d9977 100644
--- a/docs/usage/providers/fireworksai.zh-CN.mdx
+++ b/docs/usage/providers/fireworksai.zh-CN.mdx
@@ -10,64 +10,44 @@ tags:
# 在 LobeChat 中使用 Fireworks AI
-
+
[Fireworks.ai](https://fireworks.ai/) 是一个高性能的生成式 AI 模型推理平台，允许用户通过其 API 访问和使用各种模型。该平台支持多种模态，包括文本和视觉语言模型，并提供函数调用和 JSON 模式等功能，以增强应用开发的灵活性。
本文将指导你如何在 LobeChat 中使用 Fireworks AI。
-### 步骤一：获得 Fireworks AI 的 API Key
+ ### 步骤一：获得 Fireworks AI 的 API Key
-- 登录 [Fireworks.ai 控制台](https://fireworks.ai/account/api-keys)
-- 进入 `User` 页面，点击 `API Keys`
-- 创建一个新的 API 密钥
+ - 登录 [Fireworks.ai 控制台](https://fireworks.ai/account/api-keys)
+ - 进入 `User` 页面，点击 `API Keys`
+ - 创建一个新的 API 密钥
-
+
-- 复制并保存生成的 API 密钥
+ - 复制并保存生成的 API 密钥
-
+
-
- 请安全地存储密钥，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新密钥。
-
+
+ 请安全地存储密钥，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新密钥。
+
-### 步骤二：在 LobeChat 中配置 Fireworks AI
+ ### 步骤二：在 LobeChat 中配置 Fireworks AI
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `Fireworks AI` 的设置项
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `Fireworks AI` 的设置项
-
+
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 Fireworks AI 的模型即可开始对话
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 Fireworks AI 的模型即可开始对话
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Fireworks AI 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Fireworks AI 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Fireworks AI 提供的模型进行对话了。
diff --git a/docs/usage/providers/gemini.mdx b/docs/usage/providers/gemini.mdx
index 74929374d9522..d71ffc340cd5d 100644
--- a/docs/usage/providers/gemini.mdx
+++ b/docs/usage/providers/gemini.mdx
@@ -13,70 +13,44 @@ tags:
# Using Google Gemini in LobeChat
-
+
Gemini AI is a set of large language models (LLMs) created by Google AI, known for its cutting-edge advancements in multimodal understanding and processing. It is essentially a powerful artificial intelligence tool capable of handling various tasks involving different types of data, not just text.
This document will guide you on how to use Google Gemini in LobeChat:
+ ### Step 1: Obtain Google API Key
-### Step 1: Obtain Google API Key
+ - Visit and log in to [Google AI Studio](https://aistudio.google.com/)
+ - Navigate to `Get API Key` in the menu and click on `Create API Key`
-- Visit and log in to [Google AI Studio](https://aistudio.google.com/)
-- Navigate to `Get API Key` in the menu and click on `Create API Key`
+
-
+ - Select a project and create an API key, or create one in a new project
-- Select a project and create an API key, or create one in a new project
+
-
+ - Copy the API key from the pop-up dialog
-- Copy the API key from the pop-up dialog
+
-
+ ### Step 2: Configure OpenAI in LobeChat
-### Step 2: Configure OpenAI in LobeChat
+ - Go to the `Settings` interface in LobeChat
+ - Find the setting for `Google Gemini` under `Language Models`
-- Go to the `Settings` interface in LobeChat
-- Find the setting for `Google Gemini` under `Language Models`
+
-
+ - Enable Google Gemini and enter the obtained API key
+ - Choose a Gemini model for your assistant to start the conversation
-- Enable Google Gemini and enter the obtained API key
-- Choose a Gemini model for your assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider, please refer to Google Gemini's
- pricing policy.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to Google Gemini's
+ pricing policy.
+
Congratulations! You can now use Google Gemini in LobeChat.
diff --git a/docs/usage/providers/gemini.zh-CN.mdx b/docs/usage/providers/gemini.zh-CN.mdx
index de796fc72209a..8e55e54b7ed5d 100644
--- a/docs/usage/providers/gemini.zh-CN.mdx
+++ b/docs/usage/providers/gemini.zh-CN.mdx
@@ -10,69 +10,43 @@ tags:
# 在 LobeChat 中使用 Google Gemini
-
+
-Gemini AI是由 Google AI 创建的一组大型语言模型（LLM），以其在多模式理解和处理方面的尖端进步而闻名。它本质上是一个强大的人工智能工具，可以处理涉及不同类型数据的各种任务，而不仅仅是文本。
+Gemini AI 是由 Google AI 创建的一组大型语言模型（LLM），以其在多模式理解和处理方面的尖端进步而闻名。它本质上是一个强大的人工智能工具，可以处理涉及不同类型数据的各种任务，而不仅仅是文本。
本文档将指导你如何在 LobeChat 中使用 Google Gemini:
+ ### 步骤一：获取 Google 的 API 密钥
-### 步骤一：获取 Google 的 API 密钥
+ - 访问并登录 [Google AI Studio](https://aistudio.google.com/)
+ - 在 `获取 API 密钥` 菜单中 `创建 API 密钥`
-- 访问并登录 [Google AI Studio](https://aistudio.google.com/)
-- 在 `获取 API 密钥` 菜单中 `创建 API 密钥`
+
-
+ - 选择一个项目并创建 API 密钥，或者在新项目中创建 API 密钥
-- 选择一个项目并创建 API 密钥，或者在新项目中创建 API 密钥
+
-
+ - 在弹出的对话框中复制 API 密钥
-- 在弹出的对话框中复制 API 密钥
+
-
+ ### 步骤二：在 LobeChat 中配置 OpenAI
-### 步骤二：在 LobeChat 中配置OpenAI
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`Google Gemini`的设置项
-- 访问LobeChat的`设置`界面
-- 在`语言模型`下找到`Google Gemini`的设置项
+
-
+ - 打开 Google Gemini 并填入获得的 API 密钥
+ - 为你的助手选择一个 Gemini 的模型即可开始对话
-- 打开 Google Gemini 并填入获得的 API 密钥
-- 为你的助手选择一个 Gemini 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Google Gemini 的费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Google Gemini 的费用政策。
+
至此，你已经可以在 LobeChat 中使用 Google Gemini 啦。
diff --git a/docs/usage/providers/giteeai.mdx b/docs/usage/providers/giteeai.mdx
index 2379b4e97443c..189b2f8076fd5 100644
--- a/docs/usage/providers/giteeai.mdx
+++ b/docs/usage/providers/giteeai.mdx
@@ -12,68 +12,47 @@ tags:
# Using Gitee AI in LobeChat
-
+
[Gitee AI](https://ai.gitee.com/) is an open-source platform based on Git code hosting technology, specifically designed for AI application scenarios. It aims to provide developers and businesses with a one-stop solution for AI application development services, including model experience, inference, fine-tuning, and deployment.
This article will guide you on how to use Gitee AI in LobeChat.
+ ### Step 1: Obtain the Gitee AI API Key
-### Step 1: Obtain the Gitee AI API Key
+ - Register and log in to the [Gitee AI official website](https://ai.gitee.com/)
+ - Purchase and recharge `Serverless API` from your dashboard
-- Register and log in to the [Gitee AI official website](https://ai.gitee.com/)
-- Purchase and recharge `Serverless API` from your dashboard
+
-
+ - In `Settings`, click on the `Access Tokens` section
+ - Create a new access token
+ - Save the access token in the pop-up window
-- In `Settings`, click on the `Access Tokens` section
-- Create a new access token
-- Save the access token in the pop-up window
+
-
+
+ Please keep the access token safe as it will only appear once. If you accidentally lose it, you
+ will need to create a new one.
+
-
- Please keep the access token safe as it will only appear once. If you accidentally lose it, you
- will need to create a new one.
-
+ ### Step 2: Configure Gitee AI in LobeChat
-### Step 2: Configure Gitee AI in LobeChat
+ - Access the `Settings` page in LobeChat
+ - Under `Language Models`, find the settings for `Gitee AI`
-- Access the `Settings` page in LobeChat
-- Under `Language Models`, find the settings for `Gitee AI`
+
-
+ - Enter the obtained API key
+ - Select a Gitee AI model for your AI assistant to begin chatting
-- Enter the obtained API key
-- Select a Gitee AI model for your AI assistant to begin chatting
-
-
-
-
- During usage, you may need to make payments to the API service provider; please refer to Gitee
- AI's relevant pricing policy.
-
+
+
+ During usage, you may need to make payments to the API service provider; please refer to Gitee
+ AI's relevant pricing policy.
+
Now you can start having conversations using the models provided by Gitee AI in LobeChat!
diff --git a/docs/usage/providers/giteeai.zh-CN.mdx b/docs/usage/providers/giteeai.zh-CN.mdx
index 6008bb2bde7db..9d5ae4adcffe5 100644
--- a/docs/usage/providers/giteeai.zh-CN.mdx
+++ b/docs/usage/providers/giteeai.zh-CN.mdx
@@ -10,66 +10,45 @@ tags:
# 在 LobeChat 中使用 Gitee AI
-
+
[Gitee AI](https://ai.gitee.com/) 是一个基于 Git 代码托管技术的开源平台，专为人工智能（AI）应用场景设计。它旨在为开发者和企业提供一站式的 AI 应用开发服务，包括模型体验、推理、微调和部署等功能。
本文将指导你如何在 LobeChat 中使用 Gitee AI。
+ ### 步骤一：获取 Gitee AI 的 API 密钥
-### 步骤一: 获取 Gitee AI 的 API 密钥
+ - 注册并登录 [Gitee AI 官网](https://ai.gitee.com/)
+ - 在工作台中购买并充值 `Serverless API`
-- 注册并登录 [Gitee AI 官网](https://ai.gitee.com/)
-- 在工作台中购买并充值 `Serverless API`
+
-
+ - 在 `设置` 中点击 `访问令牌` 界面
+ - 创建一个新的访问令牌
+ - 在弹出窗口中保存访问令牌
-- 在 `设置` 中点击 `访问令牌` 界面
-- 创建一个新的访问令牌
-- 在弹出窗口中保存访问令牌
+
-
+
+ 妥善保存弹窗中的访问令牌，它只会出现一次，如果不小心丢失了，你需要重新创建一个访问令牌。
+
-
- 妥善保存弹窗中的访问令牌，它只会出现一次，如果不小心丢失了，你需要重新创建一个访问令牌。
-
+ ### 步骤二：在 LobeChat 中配置 Gitee AI
-### 步骤二：在 LobeChat 中配置 Gitee AI
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `Gitee AI` 的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `Gitee AI` 的设置项
+
-
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 Gitee AI 的模型即可开始对话
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 Gitee AI 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Gitee AI 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Gitee AI 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Gitee AI 提供的模型进行对话了。
diff --git a/docs/usage/providers/github.mdx b/docs/usage/providers/github.mdx
index 644a01044ceb0..30fc3a712f946 100644
--- a/docs/usage/providers/github.mdx
+++ b/docs/usage/providers/github.mdx
@@ -11,10 +11,7 @@ tags:
# Using GitHub Models in LobeChat
-
+
[GitHub Models](https://github.com/marketplace/models) is a new feature recently launched by GitHub, designed to provide developers with a free platform to access and experiment with various AI models. GitHub Models offers an interactive sandbox environment where users can test different model parameters and prompts, and observe the responses of the models. The platform supports advanced language models, including OpenAI's GPT-4o, Meta's Llama 3.1, and Mistral's Large 2, covering a wide range of applications from large-scale language models to task-specific models.
@@ -24,11 +21,7 @@ This article will guide you on how to use GitHub Models in LobeChat.
Currently, the usage of the Playground and free API is subject to limits on the number of requests per minute, the number of requests per day, the number of tokens per request, and the number of concurrent requests. If you hit the rate limit, you will need to wait for the limit to reset before making further requests. The rate limits vary for different models (low, high, and embedding models). For model type information, please refer to the GitHub Marketplace.
-
+
These limits are subject to change at any time. For specific information, please refer to the
@@ -41,53 +34,34 @@ Currently, the usage of the Playground and free API is subject to limits on the
## Configuration Guide for GitHub Models
-### Step 1: Obtain a GitHub Access Token
+ ### Step 1: Obtain a GitHub Access Token
-- Log in to GitHub and open the [Access Tokens](https://github.com/settings/tokens) page.
-- Create and configure a new access token.
+ - Log in to GitHub and open the [Access Tokens](https://github.com/settings/tokens) page.
+ - Create and configure a new access token.
-
+
-- Copy and save the generated token from the results returned.
+ - Copy and save the generated token from the results returned.
-
+
-
-
-- During the testing phase of GitHub Models, users must apply to join the [waitlist](https://github.com/marketplace/models/waitlist/join) in order to gain access.
+
+ - During the testing phase of GitHub Models, users must apply to join the [waitlist](https://github.com/marketplace/models/waitlist/join) in order to gain access.
-- Please store the access token securely, as it will only be displayed once. If you accidentally lose it, you will need to create a new token.
+ - Please store the access token securely, as it will only be displayed once. If you accidentally lose it, you will need to create a new token.
+
-
-
-### Step 2: Configure GitHub Models in LobeChat
-
-- Navigate to the `Settings` interface in LobeChat.
-- Under `Language Models`, find the GitHub settings.
+ ### Step 2: Configure GitHub Models in LobeChat
-
+ - Navigate to the `Settings` interface in LobeChat.
+ - Under `Language Models`, find the GitHub settings.
-- Enter the access token you obtained.
-- Select a GitHub model for your AI assistant to start the conversation.
+
-
+ - Enter the access token you obtained.
+ - Select a GitHub model for your AI assistant to start the conversation.
+
You are now ready to use the models provided by GitHub for conversations within LobeChat.
diff --git a/docs/usage/providers/github.zh-CN.mdx b/docs/usage/providers/github.zh-CN.mdx
index 9258d8a2c22bb..7b4bfa42da119 100644
--- a/docs/usage/providers/github.zh-CN.mdx
+++ b/docs/usage/providers/github.zh-CN.mdx
@@ -11,10 +11,7 @@ tags:
# 在 LobeChat 中使用 GitHub Models
-
+
[GitHub Models](https://github.com/marketplace/models) 是 GitHub 最近推出的一项新功能，旨在为开发者提供一个免费的平台来访问和实验多种 AI 模型。GitHub Models 提供了一个互动沙盒环境，用户可以在此测试不同的模型参数和提示语，观察模型的响应。该平台支持多种先进的语言模型，包括 OpenAI 的 GPT-4o、Meta 的 Llama 3.1 和 Mistral 的 Large 2 等，覆盖了从大规模语言模型到特定任务模型的广泛应用。
@@ -24,11 +21,7 @@ tags:
当前 Playground 和免费 API 的使用受到每分钟请求数、每日请求数、每个请求的令牌数以及并发请求数的限制。若达到速率限制，则需等待限制重置后方可继续发出请求。不同模型（低、高及嵌入模型）的速率限制有所不同。 模型类型信息请参阅 GitHub Marketplace。
-
+
这些限制可能随时更改，具体信息请参考 [GitHub
@@ -40,52 +33,34 @@ tags:
## GitHub Models 配置指南
-### 步骤一：获得 GitHub 的访问令牌
+ ### 步骤一：获得 GitHub 的访问令牌
-- 登录 GitHub 并打开 [访问令牌](https://github.com/settings/tokens) 页面
-- 创建并设置一个新的访问令牌
+ - 登录 GitHub 并打开 [访问令牌](https://github.com/settings/tokens) 页面
+ - 创建并设置一个新的访问令牌
-
+
-- 在返回的结果中复制并保存生成的令牌
+ - 在返回的结果中复制并保存生成的令牌
-
+
-
- - GitHub Models 测试期间，要使用 GitHub Models，用户需要申请加入[等待名单（waitlist）](https://github.com/marketplace/models/waitlist/join) 通过后才能获得访问权限。
+
+ - GitHub Models 测试期间，要使用 GitHub Models，用户需要申请加入[等待名单（waitlist）](https://github.com/marketplace/models/waitlist/join) 通过后才能获得访问权限。
-- 请安全地存储访问令牌，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新令牌。
+ - 请安全地存储访问令牌，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新令牌。
+
-
-
-### 步骤二：在 LobeChat 中配置 GitHub Models
-
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `GitHub` 的设置项
+ ### 步骤二：在 LobeChat 中配置 GitHub Models
-
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `GitHub` 的设置项
-- 填入获得的访问令牌
-- 为你的 AI 助手选择一个 GitHub 的模型即可开始对话
+
-
+ - 填入获得的访问令牌
+ - 为你的 AI 助手选择一个 GitHub 的模型即可开始对话
+
至此你已经可以在 LobeChat 中使用 GitHub 提供的模型进行对话了。
diff --git a/docs/usage/providers/groq.mdx b/docs/usage/providers/groq.mdx
index 7bf7e75d103cd..c03a45d45be61 100644
--- a/docs/usage/providers/groq.mdx
+++ b/docs/usage/providers/groq.mdx
@@ -14,11 +14,7 @@ tags:
# Using Groq in LobeChat
-
+
Groq's [LPU Inference Engine](https://wow.groq.com/news_press/groq-lpu-inference-engine-leads-in-first-independent-llm-benchmark/) has excelled in the latest independent Large Language Model (LLM) benchmark, redefining the standard for AI solutions with its remarkable speed and efficiency. By integrating LobeChat with Groq Cloud, you can now easily leverage Groq's technology to accelerate the operation of large language models in LobeChat.
@@ -34,39 +30,26 @@ This document will guide you on how to use Groq in LobeChat:
### Obtaining GroqCloud API Keys
-First, you need to obtain an API Key from the [GroqCloud Console](https://console.groq.com/).
+ First, you need to obtain an API Key from the [GroqCloud Console](https://console.groq.com/).
-
+
-Create an API Key in the `API Keys` menu of the console.
+ Create an API Key in the `API Keys` menu of the console.
-
+
-
- Safely store the key from the pop-up as it will only appear once. If you accidentally lose it, you
- will need to create a new key.
-
+
+ Safely store the key from the pop-up as it will only appear once. If you accidentally lose it, you
+ will need to create a new key.
+
-### Configure Groq in LobeChat
+ ### Configure Groq in LobeChat
-You can find the Groq configuration option in `Settings` -> `Language Model`, where you can input the API Key you just obtained.
+ You can find the Groq configuration option in `Settings` -> `Language Model`, where you can input the API Key you just obtained.
Next, select a Groq-supported model in the assistant's model options, and you can experience the powerful performance of Groq in LobeChat.
-
+
diff --git a/docs/usage/providers/groq.zh-CN.mdx b/docs/usage/providers/groq.zh-CN.mdx
index 20cf478f237d7..750b301b32d41 100644
--- a/docs/usage/providers/groq.zh-CN.mdx
+++ b/docs/usage/providers/groq.zh-CN.mdx
@@ -11,11 +11,7 @@ tags:
# 在 LobeChat 中使用 Groq
-
+
Groq 的 [LPU 推理引擎](https://wow.groq.com/news_press/groq-lpu-inference-engine-leads-in-first-independent-llm-benchmark/) 在最新的独立大语言模型（LLM）基准测试中表现卓越，以其惊人的速度和效率重新定义了 AI 解决方案的标准。通过 LobeChat 与 Groq Cloud 的集成，你现在可以轻松地利用 Groq 的技术，在 LobeChat 中加速大语言模型的运行。
@@ -30,38 +26,25 @@ Groq 的 [LPU 推理引擎](https://wow.groq.com/news_press/groq-lpu-inference-e
### 获取 GroqCloud API Key
-首先，你需要到 [GroqCloud Console](https://console.groq.com/) 中获取一个 API Key。
+ 首先，你需要到 [GroqCloud Console](https://console.groq.com/) 中获取一个 API Key。
-
+
-在控制台的 `API Keys` 菜单中创建一个 API Key。
+ 在控制台的 `API Keys` 菜单中创建一个 API Key。
-
+
-
- 妥善保存弹窗中的 key，它只会出现一次，如果不小心丢失了，你需要重新创建一个 key。
-
+
+ 妥善保存弹窗中的 key，它只会出现一次，如果不小心丢失了，你需要重新创建一个 key。
+
-### 在 LobeChat 中配置 Groq
+ ### 在 LobeChat 中配置 Groq
-你可以在 `设置` -> `语言模型` 中找到 Groq 的配置选项，将刚才获取的 API Key 填入。
+ 你可以在 `设置` -> `语言模型` 中找到 Groq 的配置选项，将刚才获取的 API Key 填入。
接下来，在助手的模型选项中，选中一个 Groq 支持的模型，就可以在 LobeChat 中体验 Groq 强大的性能了。
-
+
diff --git a/docs/usage/providers/hunyuan.mdx b/docs/usage/providers/hunyuan.mdx
index 39064fde5eced..0b84e4bdc5900 100644
--- a/docs/usage/providers/hunyuan.mdx
+++ b/docs/usage/providers/hunyuan.mdx
@@ -12,61 +12,41 @@ tags:
# Using Tencent Hunyuan in LobeChat
-
+
[Tencent Hunyuan](https://hunyuan.tencent.com/) is a large model launched by Tencent, designed to provide users with intelligent assistant services. It utilizes natural language processing technology to help users solve problems, offer suggestions, and generate content. By conversing with the model, users can quickly access the information they need, thereby enhancing work efficiency.
This article will guide you on how to use Tencent Hunyuan in LobeChat.
-### Step 1: Obtain the Tencent Hunyuan API Key
+ ### Step 1: Obtain the Tencent Hunyuan API Key
-- Register and log in to the [Tencent Cloud Console](https://console.cloud.tencent.com/hunyuan/api-key)
-- Navigate to `Hunyuan Large Model` and click on `API KEY Management`
-- Create an API key
+ - Register and log in to the [Tencent Cloud Console](https://console.cloud.tencent.com/hunyuan/api-key)
+ - Navigate to `Hunyuan Large Model` and click on `API KEY Management`
+ - Create an API key
-
+
-- Click `View`, and copy the API key from the pop-up panel, ensuring you save it securely
+ - Click `View`, and copy the API key from the pop-up panel, ensuring you save it securely
-
+
-### Step 2: Configure Tencent Hunyuan in LobeChat
+ ### Step 2: Configure Tencent Hunyuan in LobeChat
-- Go to the `Settings` page in LobeChat
-- Find the `Tencent Hunyuan` settings under `Language Models`
+ - Go to the `Settings` page in LobeChat
+ - Find the `Tencent Hunyuan` settings under `Language Models`
-
+
-- Enter the API key you obtained
-- Select a Tencent Hunyuan model for your AI assistant to start the conversation
+ - Enter the API key you obtained
+ - Select a Tencent Hunyuan model for your AI assistant to start the conversation
-
-
-
- During usage, you may need to pay the API service provider, please refer to Tencent Hunyuan's
- relevant pricing policy.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to Tencent Hunyuan's
+ relevant pricing policy.
+
You can now engage in conversations using the models provided by Tencent Hunyuan in LobeChat.
diff --git a/docs/usage/providers/hunyuan.zh-CN.mdx b/docs/usage/providers/hunyuan.zh-CN.mdx
index 1e0f730311df7..2b6aea1ba9ac2 100644
--- a/docs/usage/providers/hunyuan.zh-CN.mdx
+++ b/docs/usage/providers/hunyuan.zh-CN.mdx
@@ -10,60 +10,40 @@ tags:
# 在 LobeChat 中使用腾讯混元
-
+
[腾讯混元](https://hunyuan.tencent.com/)是由腾讯推出的一款大模型，旨在为用户提供智能助手服务。它能够通过自然语言处理技术，帮助用户解决问题、提供建议以及进行内容生成等任务。用户可以通过与模型的对话，快速获取所需信息，从而提高工作效率。
本文将指导你如何在 LobeChat 中使用腾讯混元。
-### 步骤一：获得腾讯混元的 API Key
+ ### 步骤一：获得腾讯混元的 API Key
-- 注册并登录 [腾讯云控制台](https://console.cloud.tencent.com/hunyuan/api-key)
-- 进入 `混元大模型` 并点击 `API KEY 管理`
-- 创建一个 API 密钥
+ - 注册并登录 [腾讯云控制台](https://console.cloud.tencent.com/hunyuan/api-key)
+ - 进入 `混元大模型` 并点击 `API KEY 管理`
+ - 创建一个 API 密钥
-
+
-- 点击`查看`，在弹出面板中复制 API 密钥，并妥善保存
+ - 点击`查看`，在弹出面板中复制 API 密钥，并妥善保存
-
+
-### 步骤二：在 LobeChat 中配置腾讯混元
+ ### 步骤二：在 LobeChat 中配置腾讯混元
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `腾讯混元` 的设置项
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `腾讯混元` 的设置项
-
+
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个腾讯混元的模型即可开始对话
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个腾讯混元的模型即可开始对话
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考腾讯混元的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考腾讯混元的相关费用政策。
+
至此你已经可以在 LobeChat 中使用腾讯混元提供的模型进行对话了。
diff --git a/docs/usage/providers/internlm.mdx b/docs/usage/providers/internlm.mdx
index a23a813ea4085..8474b5e266f67 100644
--- a/docs/usage/providers/internlm.mdx
+++ b/docs/usage/providers/internlm.mdx
@@ -12,59 +12,42 @@ tags:
# Using InternLM in LobeChat
-
+
[InternLM](https://platform.sensenova.cn/home) is a large pre-trained language model jointly launched by the Shanghai Artificial Intelligence Laboratory and Shusheng Group. This model focuses on natural language processing, aimed at understanding and generating human language, boasting powerful semantic comprehension and text generation capabilities.
This article will guide you on how to use InternLM in LobeChat.
+ ### Step 1: Obtain the InternLM API Key
-### Step 1: Obtain the InternLM API Key
+ - Register and log in to [InternLM API](https://InternLM.intern-ai.org.cn/api/tokens)
+ - Create an API token
+ - Save the API token in the pop-up window
-- Register and log in to [InternLM API](https://InternLM.intern-ai.org.cn/api/tokens)
-- Create an API token
-- Save the API token in the pop-up window
+
-
+
+ Please store the API token shown in the pop-up securely; it will only appear once. If you lose it,
+ you will need to create a new API token.
+
-
- Please store the API token shown in the pop-up securely; it will only appear once. If you lose it,
- you will need to create a new API token.
-
+ ### Step 2: Configure InternLM in LobeChat
-### Step 2: Configure InternLM in LobeChat
+ - Go to the `Settings` interface in LobeChat
+ - Find the settings option for `InternLM` under `Language Models`
-- Go to the `Settings` interface in LobeChat
-- Find the settings option for `InternLM` under `Language Models`
+
-
+ - Enter the obtained `AccessKey ID` and `AccessKey Secret`
+ - Choose a InternLM model for your AI assistant to start a conversation
-- Enter the obtained `AccessKey ID` and `AccessKey Secret`
-- Choose a InternLM model for your AI assistant to start a conversation
-
-
-
-
- During usage, you may need to pay the API service provider; please refer to the pricing policy
- regarding InternLM.
-
+
+
+ During usage, you may need to pay the API service provider; please refer to the pricing policy
+ regarding InternLM.
+
You are now ready to engage in conversations using the models provided by InternLM in LobeChat.
diff --git a/docs/usage/providers/internlm.zh-CN.mdx b/docs/usage/providers/internlm.zh-CN.mdx
index cb73168196ced..d4a3b7cfb73ec 100644
--- a/docs/usage/providers/internlm.zh-CN.mdx
+++ b/docs/usage/providers/internlm.zh-CN.mdx
@@ -10,57 +10,40 @@ tags:
# 在 LobeChat 中使用书生浦语
-
+
[书生浦语（InternLM）](https://platform.sensenova.cn/home) 是由上海人工智能实验室与书生集团联合推出的一款大型预训练语言模型。该模型专注于自然语言处理，旨在理解和生成自然语言，具备强大的语义理解和文本生成能力。
本文将指导你如何在 LobeChat 中使用书生浦语。
+ ### 步骤一：获取书生浦语的 API 密钥
-### 步骤一: 获取书生浦语的 API 密钥
+ - 注册并登录 [浦语 API](https://internlm.intern-ai.org.cn/api/tokens)
+ - 创建一个 API 令牌
+ - 在弹出窗口中保存 API 令牌
-- 注册并登录 [浦语 API](https://internlm.intern-ai.org.cn/api/tokens)
-- 创建一个 API 令牌
-- 在弹出窗口中保存 API 令牌
+
-
+
+ 妥善保存弹窗中的 API 令牌，它只会出现一次，如果不小心丢失了，你需要重新创建一个 API 令牌。
+
-
- 妥善保存弹窗中的 API 令牌，它只会出现一次，如果不小心丢失了，你需要重新创建一个 API 令牌。
-
+ ### 步骤二：在 LobeChat 中配置书生浦语
-### 步骤二：在 LobeChat 中配置书生浦语
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `书生浦语` 的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `书生浦语` 的设置项
+
-
+ - 填入获得的 `AccessKey ID` 和 `AccessKey Secret`
+ - 为你的 AI 助手选择一个书生浦语的模型即可开始对话
-- 填入获得的 `AccessKey ID` 和 `AccessKey Secret`
-- 为你的 AI 助手选择一个书生浦语的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考书生浦语的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考书生浦语的相关费用政策。
+
至此你已经可以在 LobeChat 中使用书生浦语提供的模型进行对话了。
diff --git a/docs/usage/providers/minimax.mdx b/docs/usage/providers/minimax.mdx
index 5841e5d0690b4..1d165290c2fd9 100644
--- a/docs/usage/providers/minimax.mdx
+++ b/docs/usage/providers/minimax.mdx
@@ -13,75 +13,49 @@ tags:
# Using Minimax in LobeChat
-
+
[MiniMax](https://www.minimaxi.com/) is a general artificial intelligence technology company founded in 2021, dedicated to co-creating intelligence with users. MiniMax has independently developed universal large models of different modalities, including trillion-parameter MoE text large models, speech large models, and image large models. They have also launched applications like Hai Luo AI.
This document will guide you on how to use Minimax in LobeChat:
+ ### Step 1: Obtain MiniMax API Key
-### Step 1: Obtain MiniMax API Key
+ - Register and log in to the [MiniMax Open Platform](https://www.minimaxi.com/platform)
+ - In `Account Management`, locate the `API Key` menu and create a new key
-- Register and log in to the [MiniMax Open Platform](https://www.minimaxi.com/platform)
-- In `Account Management`, locate the `API Key` menu and create a new key
+
-
+ - Enter a name for the API key and create it
-- Enter a name for the API key and create it
+
-
+ - Copy the API key from the pop-up dialog box and save it securely
-- Copy the API key from the pop-up dialog box and save it securely
+
-
+
+ Please store the key securely as it will only appear once. If you accidentally lose it, you will
+ need to create a new key.
+
-
- Please store the key securely as it will only appear once. If you accidentally lose it, you will
- need to create a new key.
-
+ ### Step 2: Configure MiniMax in LobeChat
-### Step 2: Configure MiniMax in LobeChat
+ - Go to the `Settings` interface of LobeChat
+ - Find the setting for `MiniMax` under `Language Model`
-- Go to the `Settings` interface of LobeChat
-- Find the setting for `MiniMax` under `Language Model`
+
-
+ - Open Minimax and enter the obtained API key
+ - Choose a MiniMax model for your AI assistant to start the conversation
-- Open Minimax and enter the obtained API key
-- Choose a MiniMax model for your AI assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider, please refer to MiniMax's relevant
- pricing policies.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to MiniMax's relevant
+ pricing policies.
+
You can now use the models provided by MiniMax to have conversations in LobeChat.
diff --git a/docs/usage/providers/minimax.zh-CN.mdx b/docs/usage/providers/minimax.zh-CN.mdx
index c612b6e0312e6..a45abd6d30804 100644
--- a/docs/usage/providers/minimax.zh-CN.mdx
+++ b/docs/usage/providers/minimax.zh-CN.mdx
@@ -12,73 +12,47 @@ tags:
# 在 LobeChat 中使用 Minimax
-
+
[MiniMax](https://www.minimaxi.com/) 是 2021 年成立的通用人工智能科技公司，致力于与用户共创智能。MiniMax 自主研发了不同模态的通用大模型，其中包括万亿参数的 MoE 文本大模型、语音大模型以及图像大模型。并推出了海螺 AI 等应用。
本文档将指导你如何在 LobeChat 中使用 Minimax:
+ ### 步骤一：获取 MiniMax API 密钥
-### 步骤一：获取 MiniMax API 密钥
+ - 注册并登录 [MiniMax 开放平台](https://www.minimaxi.com/platform)
+ - 在 `账户管理` 中找到 `接口密钥` 菜单，并创建新的密钥
-- 注册并登录 [MiniMax 开放平台](https://www.minimaxi.com/platform)
-- 在 `账户管理` 中找到 `接口密钥` 菜单，并创建新的密钥
+
-
+ - 填写一个 API 密钥的名称并创建
-- 填写一个 API 密钥的名称并创建
+
-
+ - 在弹出的对话框中复制 API 密钥，并妥善保存
-- 在弹出的对话框中复制 API 密钥，并妥善保存
+
-
+
+ 请安全地存储密钥，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新密钥。
+
-
- 请安全地存储密钥，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新密钥。
-
+ ### 步骤二：在 LobeChat 中配置 MiniMax
-### 步骤二：在LobeChat 中配置 MiniMax
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`MiniMax`的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到`MiniMax`的设置项
+
-
+ - 打开 Minimax 并填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 MiniMax 的模型即可开始对话
-- 打开 Minimax 并填入获得的 API 密钥
-- 为你的 AI 助手选择一个 MiniMax 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 MiniMax 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 MiniMax 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 MiniMax 提供的模型进行对话了。
diff --git a/docs/usage/providers/mistral.mdx b/docs/usage/providers/mistral.mdx
index 11dc8d51815d0..7ad7d965be285 100644
--- a/docs/usage/providers/mistral.mdx
+++ b/docs/usage/providers/mistral.mdx
@@ -12,58 +12,40 @@ tags:
# Using Mistral AI in LobeChat
-
+
The Mistral AI API is now available for everyone to use. This document will guide you on how to use [Mistral AI](https://mistral.ai/) in LobeChat:
+ ### Step 1: Obtain Mistral AI API Key
-### Step 1: Obtain Mistral AI API Key
+ - Create a [Mistral AI](https://mistral.ai/) account
+ - Obtain your [API key](https://console.mistral.ai/user/api-keys/)
-- Create a [Mistral AI](https://mistral.ai/) account
-- Obtain your [API key](https://console.mistral.ai/user/api-keys/)
+
-
+ ### Step 2: Configure Mistral AI in LobeChat
-### Step 2: Configure Mistral AI in LobeChat
+ - Go to the `Settings` interface in LobeChat
+ - Find the setting for `Mistral AI` under `Language Model`
-- Go to the `Settings` interface in LobeChat
-- Find the setting for `Mistral AI` under `Language Model`
+
-
+
+ If you are using mistral.ai, your account must have a valid subscription for the API key to work
+ properly. Newly created API keys may take 2-3 minutes to become active. If the "Test" button
+ fails, please retry after 2-3 minutes.
+
-
- If you are using mistral.ai, your account must have a valid subscription for the API key to work
- properly. Newly created API keys may take 2-3 minutes to become active. If the "Test" button
- fails, please retry after 2-3 minutes.
-
+ - Enter the obtained API key
+ - Choose a Mistral AI model for your AI assistant to start the conversation
-- Enter the obtained API key
-- Choose a Mistral AI model for your AI assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider, please refer to Mistral AI's relevant
- pricing policies.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to Mistral AI's relevant
+ pricing policies.
+
You can now engage in conversations using the models provided by Mistral AI in LobeChat.
diff --git a/docs/usage/providers/mistral.zh-CN.mdx b/docs/usage/providers/mistral.zh-CN.mdx
index 7b691676e972f..7670ae8f1ee5a 100644
--- a/docs/usage/providers/mistral.zh-CN.mdx
+++ b/docs/usage/providers/mistral.zh-CN.mdx
@@ -9,56 +9,38 @@ tags:
# 在 LobeChat 中使用 Mistral AI
-
+
-Mistral AI API 现在可供所有人使用, 本文档将指导你如何在 LobeChat 中使用 [Mistral AI](https://mistral.ai/):
+Mistral AI API 现在可供所有人使用，本文档将指导你如何在 LobeChat 中使用 [Mistral AI](https://mistral.ai/):
+ ### 步骤一：获取 Mistral AI API 密钥
-### 步骤一：获取 Mistral AI API 密钥
+ - 创建一个 [Mistral AI](https://mistral.ai/) 帐户
+ - 获取您的 [API 密钥](https://console.mistral.ai/user/api-keys/)
-- 创建一个 [Mistral AI](https://mistral.ai/) 帐户
-- 获取您的 [API 密钥](https://console.mistral.ai/user/api-keys/)
+
-
+ ### 步骤二：在 LobeChat 中配置 Mistral AI
-### 步骤二：在 LobeChat 中配置 Mistral AI
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`Mistral AI`的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到`Mistral AI`的设置项
+
-
+
+ 如果您使用的是 mistral.ai，则您的帐户必须具有有效的订阅才能使 API 密钥正常工作。新创建的 API
+ 密钥需要 2-3 分钟才能开始工作。如果单击 “测试” 按钮但失败，请在 2-3 分钟后重试。
+
-
- 如果您使用的是 mistral.ai，则您的帐户必须具有有效的订阅才能使 API 密钥正常工作。新创建的 API
- 密钥需要 2-3 分钟才能开始工作。如果单击“测试”按钮但失败，请在 2-3 分钟后重试。
-
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 Mistral AI 的模型即可开始对话
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 Mistral AI 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Mistral AI 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Mistral AI 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Mistral AI 提供的模型进行对话了。
diff --git a/docs/usage/providers/moonshot.mdx b/docs/usage/providers/moonshot.mdx
index 1e57d7a774b96..047f24a04f610 100644
--- a/docs/usage/providers/moonshot.mdx
+++ b/docs/usage/providers/moonshot.mdx
@@ -12,57 +12,39 @@ tags:
# Using Moonshot AI in LobeChat
-
+
The Moonshot AI API is now available for everyone to use. This document will guide you on how to use [Moonshot AI](https://www.moonshot.cn/) in LobeChat:
+ ### Step 1: Get Moonshot AI API Key
-### Step 1: Get Moonshot AI API Key
+ - Apply for your [API key](https://platform.moonshot.cn/console/api-keys)
-- Apply for your [API key](https://platform.moonshot.cn/console/api-keys)
+
-
+ ### Step 2: Configure Moonshot AI in LobeChat
-### Step 2: Configure Moonshot AI in LobeChat
+ - Visit the `Settings` interface in LobeChat
+ - Find the setting for `Moonshot AI` under `Language Models`
-- Visit the `Settings` interface in LobeChat
-- Find the setting for `Moonshot AI` under `Language Models`
+
-
+
+ If you are using mistral.ai, your account must have a valid subscription for the API key to work
+ properly. Newly created API keys may take 2-3 minutes to become active. If the "Test" button
+ fails, please retry after 2-3 minutes.
+
-
- If you are using mistral.ai, your account must have a valid subscription for the API key to work
- properly. Newly created API keys may take 2-3 minutes to become active. If the "Test" button
- fails, please retry after 2-3 minutes.
-
+ - Enter the API key you obtained
+ - Choose a Moonshot AI model for your AI assistant to start the conversation
-- Enter the API key you obtained
-- Choose a Moonshot AI model for your AI assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider according to Moonshot AI's related
- pricing policies.
-
+
+
+ During usage, you may need to pay the API service provider according to Moonshot AI's related
+ pricing policies.
+
You can now engage in conversations using the models provided by Moonshot AI in LobeChat.
diff --git a/docs/usage/providers/moonshot.zh-CN.mdx b/docs/usage/providers/moonshot.zh-CN.mdx
index c2a416e069fa8..2751ad415efb2 100644
--- a/docs/usage/providers/moonshot.zh-CN.mdx
+++ b/docs/usage/providers/moonshot.zh-CN.mdx
@@ -9,55 +9,37 @@ tags:
# 在 LobeChat 中使用 Moonshot AI
-
+
-Moonshot AI API 现在可供所有人使用, 本文档将指导你如何在 LobeChat 中使用 [Moonshot AI](https://www.moonshot.cn/):
+Moonshot AI API 现在可供所有人使用，本文档将指导你如何在 LobeChat 中使用 [Moonshot AI](https://www.moonshot.cn/):
+ ### 步骤一：获取 Moonshot AI API 密钥
-### 步骤一：获取 Moonshot AI API 密钥
+ - 申请您的 [API 密钥](https://platform.moonshot.cn/console/api-keys)
-- 申请您的 [API 密钥](https://platform.moonshot.cn/console/api-keys)
+
-
+ ### 步骤二：在 LobeChat 中配置 Moonshot AI
-### 步骤二：在 LobeChat 中配置 Moonshot AI
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`Moonshot AI`的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到`Moonshot AI`的设置项
+
-
+
+ 如果您使用的是 mistral.ai，则您的帐户必须具有有效的订阅才能使 API 密钥正常工作。新创建的 API
+ 密钥需要 2-3 分钟才能开始工作。如果单击 “测试” 按钮但失败，请在 2-3 分钟后重试。
+
-
- 如果您使用的是 mistral.ai，则您的帐户必须具有有效的订阅才能使 API 密钥正常工作。新创建的 API
- 密钥需要 2-3 分钟才能开始工作。如果单击“测试”按钮但失败，请在 2-3 分钟后重试。
-
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 Moonshot AI 的模型即可开始对话
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 Moonshot AI 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Moonshot AI 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Moonshot AI 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Moonshot AI 提供的模型进行对话了。
diff --git a/docs/usage/providers/novita.mdx b/docs/usage/providers/novita.mdx
index 4776b4b447476..31f4c254231e3 100644
--- a/docs/usage/providers/novita.mdx
+++ b/docs/usage/providers/novita.mdx
@@ -15,66 +15,43 @@ tags:
# Using Novita AI in LobeChat
-
+
[Novita AI](https://novita.ai/) is an AI API platform that provides a variety of LLM and image generation APIs, supporting Llama3 (8B, 70B), Mistral, and many other cutting-edge models. We offer a variety of censored and uncensored models to meet your different needs.
This document will guide you on how to integrate Novita AI in LobeChat:
+ ### Step 1: Register and Log in to Novita AI
-### Step 1: Register and Log in to Novita AI
+ - Visit [Novita.ai](https://novita.ai/) and create an account
+ - You can log in with your Google or Github account
+ - Upon registration, Novita AI will provide a $0.5 credit.
-- Visit [Novita.ai](https://novita.ai/) and create an account
-- You can log in with your Google or Github account
-- Upon registration, Novita AI will provide a $0.5 credit.
+
-
+ ### Step 2: Obtain the API Key
-### Step 2: Obtain the API Key
+ - Visit Novita AI's [key management page](https://novita.ai/dashboard/key), create and copy an API Key.
-- Visit Novita AI's [key management page](https://novita.ai/dashboard/key), create and copy an API Key.
+
-
+ ### Step 3: Configure Novita AI in LobeChat
-### Step 3: Configure Novita AI in LobeChat
+ - Visit the `Settings` interface in LobeChat
+ - Find the setting for `novita.ai` under `Language Model`
-- Visit the `Settings` interface in LobeChat
-- Find the setting for `novita.ai` under `Language Model`
+
-
+ - Open novita.ai and enter the obtained API key
+ - Choose a Novita AI model for your assistant to start the conversation
-- Open novita.ai and enter the obtained API key
-- Choose a Novita AI model for your assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider, please refer to Novita AI's pricing
- policy.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to Novita AI's pricing
+ policy.
+
You can now engage in conversations using the models provided by Novita AI in LobeChat.
diff --git a/docs/usage/providers/novita.zh-CN.mdx b/docs/usage/providers/novita.zh-CN.mdx
index 501496b9a85ca..fc45766262da3 100644
--- a/docs/usage/providers/novita.zh-CN.mdx
+++ b/docs/usage/providers/novita.zh-CN.mdx
@@ -14,65 +14,42 @@ tags:
# 在 LobeChat 中使用 Novita AI
-
+
[Novita AI](https://novita.ai/) 是一个 AI API 平台，它提供多种大语言模型与 AI 图像生成的 API 服务。支持 Llama3 (8B, 70B)，Mistral 和其他最新的模型。
本文档将指导你如何在 LobeChat 中使用 Novita AI:
+ ### 步骤一：注册 Novita AI 账号并登录
-### 步骤一：注册 Novita AI 账号并登录
+ - 访问 [Novita.ai](https://novita.ai/) 并创建账号
+ - 你可以使用 Google 或者 Github 账号登录
+ - 注册后，Novita AI 会赠送 0.5 美元的使用额度
-- 访问 [Novita.ai](https://novita.ai/) 并创建账号
-- 你可以使用 Google 或者 Github 账号登录
-- 注册后，Novita AI 会赠送 0.5 美元的使用额度
+
-
+ ### 步骤二：创建 API 密钥
-### 步骤二：创建 API 密钥
+ - 访问 Novita AI 的 [密钥管理页面](https://novita.ai/dashboard/key) ，创建并且复制一个 API 密钥.
-- 访问 Novita AI 的 [密钥管理页面](https://novita.ai/dashboard/key) ，创建并且复制一个 API 密钥.
+
-
+ ### 步骤三：在 LobeChat 中配置 Novita AI
-### 步骤三：在 LobeChat 中配置 Novita AI
+ - 访问 LobeChat 的 `设置` 界面
+ - 在 `语言模型` 下找到 `novita.ai` 的设置项
+ - 打开 novita.ai 并填入获得的 API 密钥
-- 访问 LobeChat 的 `设置` 界面
-- 在 `语言模型` 下找到 `novita.ai` 的设置项
-- 打开 novita.ai 并填入获得的 API 密钥
+
-
+ - 为你的助手选择一个 Novita AI 模型即可开始对话
-- 为你的助手选择一个 Novita AI 模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Novita AI 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Novita AI 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Novita AI 提供的模型进行对话了。
diff --git a/docs/usage/providers/ollama.mdx b/docs/usage/providers/ollama.mdx
index 2d7f03de24350..f066fa7cf00da 100644
--- a/docs/usage/providers/ollama.mdx
+++ b/docs/usage/providers/ollama.mdx
@@ -13,151 +13,130 @@ tags:
# Using Ollama in LobeChat
-
+
Ollama is a powerful framework for running large language models (LLMs) locally, supporting various language models including Llama 2, Mistral, and more. Now, LobeChat supports integration with Ollama, meaning you can easily enhance your application by using the language models provided by Ollama in LobeChat.
This document will guide you on how to use Ollama in LobeChat:
-
+
## Using Ollama on macOS
+ ### Local Installation of Ollama
-### Local Installation of Ollama
+ [Download Ollama for macOS](https://ollama.com/download?utm_source=lobehub\&utm_medium=docs\&utm_campaign=download-macos) and unzip/install it.
-[Download Ollama for macOS](https://ollama.com/download?utm_source=lobehub&utm_medium=docs&utm_campaign=download-macos) and unzip/install it.
+ ### Configure Ollama for Cross-Origin Access
-### Configure Ollama for Cross-Origin Access
+ Due to Ollama's default configuration, which restricts access to local only, additional environment variable setting `OLLAMA_ORIGINS` is required for cross-origin access and port listening. Use `launchctl` to set the environment variable:
-Due to Ollama's default configuration, which restricts access to local only, additional environment variable setting `OLLAMA_ORIGINS` is required for cross-origin access and port listening. Use `launchctl` to set the environment variable:
+ ```bash
+ launchctl setenv OLLAMA_ORIGINS "*"
+ ```
-```bash
-launchctl setenv OLLAMA_ORIGINS "*"
-```
-
-After setting up, restart the Ollama application.
+ After setting up, restart the Ollama application.
-### Conversing with Local Large Models in LobeChat
+ ### Conversing with Local Large Models in LobeChat
-Now, you can start conversing with the local LLM in LobeChat.
-
-
+ Now, you can start conversing with the local LLM in LobeChat.
+
## Using Ollama on Windows
+ ### Local Installation of Ollama
-### Local Installation of Ollama
-
-[Download Ollama for Windows](https://ollama.com/download?utm_source=lobehub&utm_medium=docs&utm_campaign=download-windows) and install it.
-
-### Configure Ollama for Cross-Origin Access
+ [Download Ollama for Windows](https://ollama.com/download?utm_source=lobehub\&utm_medium=docs\&utm_campaign=download-windows) and install it.
-Since Ollama's default configuration allows local access only, additional environment variable setting `OLLAMA_ORIGINS` is needed for cross-origin access and port listening.
+ ### Configure Ollama for Cross-Origin Access
-On Windows, Ollama inherits your user and system environment variables.
+ Since Ollama's default configuration allows local access only, additional environment variable setting `OLLAMA_ORIGINS` is needed for cross-origin access and port listening.
-1. First, exit the Ollama program by clicking on it in the Windows taskbar.
-2. Edit system environment variables from the Control Panel.
-3. Edit or create the Ollama environment variable `OLLAMA_ORIGINS` for your user account, setting the value to `*`.
-4. Click `OK/Apply` to save and restart the system.
-5. Run `Ollama` again.
+ On Windows, Ollama inherits your user and system environment variables.
-### Conversing with Local Large Models in LobeChat
+ 1. First, exit the Ollama program by clicking on it in the Windows taskbar.
+ 2. Edit system environment variables from the Control Panel.
+ 3. Edit or create the Ollama environment variable `OLLAMA_ORIGINS` for your user account, setting the value to `*`.
+ 4. Click `OK/Apply` to save and restart the system.
+ 5. Run `Ollama` again.
-Now, you can start conversing with the local LLM in LobeChat.
+ ### Conversing with Local Large Models in LobeChat
+ Now, you can start conversing with the local LLM in LobeChat.
## Using Ollama on Linux
+ ### Local Installation of Ollama
-### Local Installation of Ollama
+ Install using the following command:
-Install using the following command:
+ ```bash
+ curl -fsSL https://ollama.com/install.sh | sh
+ ```
-```bash
-curl -fsSL https://ollama.com/install.sh | sh
-```
-
-Alternatively, you can refer to the [Linux manual installation guide](https://github.com/ollama/ollama/blob/main/docs/linux.md).
+ Alternatively, you can refer to the [Linux manual installation guide](https://github.com/ollama/ollama/blob/main/docs/linux.md).
-### Configure Ollama for Cross-Origin Access
+ ### Configure Ollama for Cross-Origin Access
-Due to Ollama's default configuration, which allows local access only, additional environment variable setting `OLLAMA_ORIGINS` is required for cross-origin access and port listening. If Ollama runs as a systemd service, use `systemctl` to set the environment variable:
+ Due to Ollama's default configuration, which allows local access only, additional environment variable setting `OLLAMA_ORIGINS` is required for cross-origin access and port listening. If Ollama runs as a systemd service, use `systemctl` to set the environment variable:
-1. Edit the systemd service by calling `sudo systemctl edit ollama.service`:
-
-```bash
-sudo systemctl edit ollama.service
-```
+ 1. Edit the systemd service by calling `sudo systemctl edit ollama.service`:
-2. Add `Environment` under `[Service]` for each environment variable:
+ ```bash
+ sudo systemctl edit ollama.service
+ ```
-```bash
-[Service]
-Environment="OLLAMA_HOST=0.0.0.0"
-Environment="OLLAMA_ORIGINS=*"
-```
+ 2. Add `Environment` under `[Service]` for each environment variable:
-3. Save and exit.
-4. Reload `systemd` and restart Ollama:
+ ```bash
+ [Service]
+ Environment="OLLAMA_HOST=0.0.0.0"
+ Environment="OLLAMA_ORIGINS=*"
+ ```
-```bash
-sudo systemctl daemon-reload
-sudo systemctl restart ollama
-```
+ 3. Save and exit.
+ 4. Reload `systemd` and restart Ollama:
-### Conversing with Local Large Models in LobeChat
+ ```bash
+ sudo systemctl daemon-reload
+ sudo systemctl restart ollama
+ ```
-Now, you can start conversing with the local LLM in LobeChat.
+ ### Conversing with Local Large Models in LobeChat
+ Now, you can start conversing with the local LLM in LobeChat.
## Deploying Ollama using Docker
+ ### Pulling Ollama Image
-### Pulling Ollama Image
+ If you prefer using Docker, Ollama provides an official Docker image that you can pull using the following command:
-If you prefer using Docker, Ollama provides an official Docker image that you can pull using the following command:
+ ```bash
+ docker pull ollama/ollama
+ ```
-```bash
-docker pull ollama/ollama
-```
-
-### Configure Ollama for Cross-Origin Access
-
-Since Ollama's default configuration allows local access only, additional environment variable setting `OLLAMA_ORIGINS` is needed for cross-origin access and port listening.
+ ### Configure Ollama for Cross-Origin Access
-If Ollama runs as a Docker container, you can add the environment variable to the `docker run` command.
+ Since Ollama's default configuration allows local access only, additional environment variable setting `OLLAMA_ORIGINS` is needed for cross-origin access and port listening.
-```bash
-docker run -d --gpus=all -v ollama:/root/.ollama -e OLLAMA_ORIGINS="*" -p 11434:11434 --name ollama ollama/ollama
-```
+ If Ollama runs as a Docker container, you can add the environment variable to the `docker run` command.
-### Conversing with Local Large Models in LobeChat
+ ```bash
+ docker run -d --gpus=all -v ollama:/root/.ollama -e OLLAMA_ORIGINS="*" -p 11434:11434 --name ollama ollama/ollama
+ ```
-Now, you can start conversing with the local LLM in LobeChat.
+ ### Conversing with Local Large Models in LobeChat
+ Now, you can start conversing with the local LLM in LobeChat.
## Installing Ollama Models
@@ -168,11 +147,7 @@ Ollama supports various models, which you can view in the [Ollama Library](https
In LobeChat, we have enabled some common large language models by default, such as llama3, Gemma, Mistral, etc. When you select a model for conversation, we will prompt you to download that model.
-
+
Once downloaded, you can start conversing.
@@ -184,20 +159,13 @@ Alternatively, you can install models by executing the following command in the
ollama pull llama3
```
-
+
## Custom Configuration
You can find Ollama's configuration options in `Settings` -> `Language Models`, where you can configure Ollama's proxy, model names, etc.
-
+
Visit [Integrating with Ollama](/docs/self-hosting/examples/ollama) to learn how to deploy
diff --git a/docs/usage/providers/ollama.zh-CN.mdx b/docs/usage/providers/ollama.zh-CN.mdx
index 2b899b47dce2d..98ac56ae56e33 100644
--- a/docs/usage/providers/ollama.zh-CN.mdx
+++ b/docs/usage/providers/ollama.zh-CN.mdx
@@ -11,151 +11,130 @@ tags:
# 在 LobeChat 中使用 Ollama
-
+
Ollama 是一款强大的本地运行大型语言模型（LLM）的框架，支持多种语言模型，包括 Llama 2, Mistral 等。现在，LobeChat 已经支持与 Ollama 的集成，这意味着你可以在 LobeChat 中轻松使用 Ollama 提供的语言模型来增强你的应用。
本文档将指导你如何在 LobeChat 中使用 Ollama：
-
+
## 在 macOS 下使用 Ollama
+ ### 本地安装 Ollama
-### 本地安装 Ollama
+ [下载 Ollama for macOS](https://ollama.com/download?utm_source=lobehub\&utm_medium=docs\&utm_campaign=download-macos) 并解压、安装。
-[下载 Ollama for macOS](https://ollama.com/download?utm_source=lobehub&utm_medium=docs&utm_campaign=download-macos) 并解压、安装。
+ ### 配置 Ollama 允许跨域访问
-### 配置 Ollama 允许跨域访问
+ 由于 Ollama 的默认参数配置，启动时设置了仅本地访问，所以跨域访问以及端口监听需要进行额外的环境变量设置 `OLLAMA_ORIGINS`。使用 `launchctl` 设置环境变量：
-由于 Ollama 的默认参数配置，启动时设置了仅本地访问，所以跨域访问以及端口监听需要进行额外的环境变量设置 `OLLAMA_ORIGINS`。使用 `launchctl` 设置环境变量：
+ ```bash
+ launchctl setenv OLLAMA_ORIGINS "*"
+ ```
-```bash
-launchctl setenv OLLAMA_ORIGINS "*"
-```
-
-完成设置后，需要重启 Ollama 应用程序。
+ 完成设置后，需要重启 Ollama 应用程序。
-### 在 LobeChat 中与本地大模型对话
+ ### 在 LobeChat 中与本地大模型对话
-接下来，你就可以使用 LobeChat 与本地 LLM 对话了。
-
-
+ 接下来，你就可以使用 LobeChat 与本地 LLM 对话了。
+
## 在 windows 下使用 Ollama
+ ### 本地安装 Ollama
-### 本地安装 Ollama
-
-[下载 Ollama for Windows](https://ollama.com/download?utm_source=lobehub&utm_medium=docs&utm_campaign=download-windows) 并安装。
-
-### 配置 Ollama 允许跨域访问
+ [下载 Ollama for Windows](https://ollama.com/download?utm_source=lobehub\&utm_medium=docs\&utm_campaign=download-windows) 并安装。
-由于 Ollama 的默认参数配置，启动时设置了仅本地访问，所以跨域访问以及端口监听需要进行额外的环境变量设置 `OLLAMA_ORIGINS`。
+ ### 配置 Ollama 允许跨域访问
-在 Windows 上，Ollama 继承了您的用户和系统环境变量。
+ 由于 Ollama 的默认参数配置，启动时设置了仅本地访问，所以跨域访问以及端口监听需要进行额外的环境变量设置 `OLLAMA_ORIGINS`。
-1. 首先通过 Windows 任务栏点击 Ollama 退出程序。
-2. 从控制面板编辑系统环境变量。
-3. 为您的用户账户编辑或新建 Ollama 的环境变量 `OLLAMA_ORIGINS`，值设为 `*` 。
-4. 点击`OK/应用`保存后重启系统。
-5. 重新运行`Ollama`。
+ 在 Windows 上，Ollama 继承了您的用户和系统环境变量。
-### 在 LobeChat 中与本地大模型对话
+ 1. 首先通过 Windows 任务栏点击 Ollama 退出程序。
+ 2. 从控制面板编辑系统环境变量。
+ 3. 为您的用户账户编辑或新建 Ollama 的环境变量 `OLLAMA_ORIGINS`，值设为 `*` 。
+ 4. 点击`OK/应用`保存后重启系统。
+ 5. 重新运行`Ollama`。
-接下来，你就可以使用 LobeChat 与本地 LLM 对话了。
+ ### 在 LobeChat 中与本地大模型对话
+ 接下来，你就可以使用 LobeChat 与本地 LLM 对话了。
## 在 linux 下使用 Ollama
+ ### 本地安装 Ollama
-### 本地安装 Ollama
+ 通过以下命令安装：
-通过以下命令安装：
+ ```bash
+ curl -fsSL https://ollama.com/install.sh | sh
+ ```
-```bash
-curl -fsSL https://ollama.com/install.sh | sh
-```
-
-或者，你也可以参考 [Linux 手动安装指南](https://github.com/ollama/ollama/blob/main/docs/linux.md)。
+ 或者，你也可以参考 [Linux 手动安装指南](https://github.com/ollama/ollama/blob/main/docs/linux.md)。
-### 配置 Ollama 允许跨域访问
+ ### 配置 Ollama 允许跨域访问
-由于 Ollama 的默认参数配置，启动时设置了仅本地访问，所以跨域访问以及端口监听需要进行额外的环境变量设置 `OLLAMA_ORIGINS`。如果 Ollama 作为 systemd 服务运行，应该使用`systemctl`设置环境变量：
+ 由于 Ollama 的默认参数配置，启动时设置了仅本地访问，所以跨域访问以及端口监听需要进行额外的环境变量设置 `OLLAMA_ORIGINS`。如果 Ollama 作为 systemd 服务运行，应该使用`systemctl`设置环境变量：
-1. 通过调用`sudo systemctl edit ollama.service`编辑 systemd 服务。
-
-```bash
-sudo systemctl edit ollama.service
-```
+ 1. 通过调用`sudo systemctl edit ollama.service`编辑 systemd 服务。
-2. 对于每个环境变量，在`[Service]`部分下添加`Environment`：
+ ```bash
+ sudo systemctl edit ollama.service
+ ```
-```bash
-[Service]
-Environment="OLLAMA_HOST=0.0.0.0"
-Environment="OLLAMA_ORIGINS=*"
-```
+ 2. 对于每个环境变量，在`[Service]`部分下添加`Environment`：
-3. 保存并退出。
-4. 重载 `systemd` 并重启 Ollama：
+ ```bash
+ [Service]
+ Environment="OLLAMA_HOST=0.0.0.0"
+ Environment="OLLAMA_ORIGINS=*"
+ ```
-```bash
-sudo systemctl daemon-reload
-sudo systemctl restart ollama
-```
+ 3. 保存并退出。
+ 4. 重载 `systemd` 并重启 Ollama：
-### 在 LobeChat 中与本地大模型对话
+ ```bash
+ sudo systemctl daemon-reload
+ sudo systemctl restart ollama
+ ```
-接下来，你就可以使用 LobeChat 与本地 LLM 对话了。
+ ### 在 LobeChat 中与本地大模型对话
+ 接下来，你就可以使用 LobeChat 与本地 LLM 对话了。
## 使用 docker 部署使用 Ollama
+ ### 拉取 Ollama 镜像
-### 拉取 Ollama 镜像
+ 如果你更倾向于使用 Docker，Ollama 也提供了官方 Docker 镜像，你可以通过以下命令拉取：
-如果你更倾向于使用 Docker，Ollama 也提供了官方 Docker 镜像，你可以通过以下命令拉取：
+ ```bash
+ docker pull ollama/ollama
+ ```
-```bash
-docker pull ollama/ollama
-```
-
-### 配置 Ollama 允许跨域访问
-
-由于 Ollama 的默认参数配置，启动时设置了仅本地访问，所以跨域访问以及端口监听需要进行额外的环境变量设置 `OLLAMA_ORIGINS`。
+ ### 配置 Ollama 允许跨域访问
-如果 Ollama 作为 Docker 容器运行，你可以将环境变量添加到 `docker run` 命令中。
+ 由于 Ollama 的默认参数配置，启动时设置了仅本地访问，所以跨域访问以及端口监听需要进行额外的环境变量设置 `OLLAMA_ORIGINS`。
-```bash
-docker run -d --gpus=all -v ollama:/root/.ollama -e OLLAMA_ORIGINS="*" -p 11434:11434 --name ollama ollama/ollama
-```
+ 如果 Ollama 作为 Docker 容器运行，你可以将环境变量添加到 `docker run` 命令中。
-### 在 LobeChat 中与本地大模型对话
+ ```bash
+ docker run -d --gpus=all -v ollama:/root/.ollama -e OLLAMA_ORIGINS="*" -p 11434:11434 --name ollama ollama/ollama
+ ```
-接下来，你就可以使用 LobeChat 与本地 LLM 对话了。
+ ### 在 LobeChat 中与本地大模型对话
+ 接下来，你就可以使用 LobeChat 与本地 LLM 对话了。
## 安装 Ollama 模型
@@ -166,11 +145,7 @@ Ollama 支持多种模型，你可以在 [Ollama Library](https://ollama.com/lib
在 LobeChat 中，我们默认开启了一些常用的大语言模型，例如 llama3、 Gemma 、 Mistral 等。当你选中模型进行对话时，我们会提示你需要下载该模型。
-
+
下载完成后即可开始对话。
@@ -182,20 +157,13 @@ Ollama 支持多种模型，你可以在 [Ollama Library](https://ollama.com/lib
ollama pull llama3
```
-
+
## 自定义配置
你可以在 `设置` -> `语言模型` 中找到 Ollama 的配置选项，你可以在这里配置 Ollama 的代理、模型名称等。
-
+
你可以前往 [与 Ollama 集成](/zh/docs/self-hosting/examples/ollama) 了解如何部署 LobeChat
diff --git a/docs/usage/providers/ollama/gemma.mdx b/docs/usage/providers/ollama/gemma.mdx
index 80997f15eb602..f7f968080ad3f 100644
--- a/docs/usage/providers/ollama/gemma.mdx
+++ b/docs/usage/providers/ollama/gemma.mdx
@@ -14,11 +14,7 @@ tags:
# Using Google Gemma Model
-
+
[Gemma](https://blog.google/technology/developers/gemma-open-models/) is an open-source large language model (LLM) from Google, designed to provide a more general and flexible model for various natural language processing tasks. Now, with the integration of LobeChat and [Ollama](https://ollama.com/), you can easily use Google Gemma in LobeChat.
@@ -27,42 +23,29 @@ This document will guide you on how to use Google Gemma in LobeChat:
### Install Ollama locally
-First, you need to install Ollama. For the installation process, please refer to the [Ollama usage documentation](/docs/usage/providers/ollama).
+ First, you need to install Ollama. For the installation process, please refer to the [Ollama usage documentation](/docs/usage/providers/ollama).
-### Pull Google Gemma model to local using Ollama
+ ### Pull Google Gemma model to local using Ollama
-After installing Ollama, you can install the Google Gemma model using the following command, using the 7b model as an example:
+ After installing Ollama, you can install the Google Gemma model using the following command, using the 7b model as an example:
-```bash
-ollama pull gemma
-```
+ ```bash
+ ollama pull gemma
+ ```
-
+
-### Select Gemma model
+ ### Select Gemma model
-In the session page, open the model panel and then select the Gemma model.
+ In the session page, open the model panel and then select the Gemma model.
-
+
If you do not see the Ollama provider in the model selection panel, please refer to [Integrating
with Ollama](/docs/self-hosting/examples/ollama) to learn how to enable the Ollama provider in
LobeChat.
-
-
+
Now, you can start conversing with the local Gemma model using LobeChat.
diff --git a/docs/usage/providers/ollama/gemma.zh-CN.mdx b/docs/usage/providers/ollama/gemma.zh-CN.mdx
index 9ecc4ea329e24..1d30432056ee2 100644
--- a/docs/usage/providers/ollama/gemma.zh-CN.mdx
+++ b/docs/usage/providers/ollama/gemma.zh-CN.mdx
@@ -13,12 +13,7 @@ tags:
# 使用 Google Gemma 模型
-
+
[Gemma](https://blog.google/technology/developers/gemma-open-models/) 是 Google 开源的一款大语言模型（LLM），旨在提供一个更加通用、灵活的模型用于各种自然语言处理任务。现在，通过 LobeChat 与 [Ollama](https://ollama.com/) 的集成，你可以轻松地在 LobeChat 中使用 Google Gemma。
@@ -27,41 +22,28 @@ tags:
### 本地安装 Ollama
-首先，你需要安装 Ollama，安装过程请查阅 [Ollama 使用文件](/zh/docs/usage/providers/ollama)。
+ 首先，你需要安装 Ollama，安装过程请查阅 [Ollama 使用文件](/zh/docs/usage/providers/ollama)。
-### 用 Ollama 拉取 Google Gemma 模型到本地
+ ### 用 Ollama 拉取 Google Gemma 模型到本地
-在安装完成 Ollama 后，你可以通过以下命令安装 Google Gemma 模型，以 7b 模型为例：
+ 在安装完成 Ollama 后，你可以通过以下命令安装 Google Gemma 模型，以 7b 模型为例：
-```bash
-ollama pull gemma
-```
+ ```bash
+ ollama pull gemma
+ ```
-
+
-### 选择 Gemma 模型
+ ### 选择 Gemma 模型
-在会话页面中，选择模型面板打开，然后选择 Gemma 模型。
+ 在会话页面中，选择模型面板打开，然后选择 Gemma 模型。
-
+
如果你没有在模型选择面板中看到 Ollama 服务商，请查阅 [与 Ollama
集成](/zh/docs/self-hosting/examples/ollama) 了解如何在 LobeChat 中开启 Ollama 服务商。
-
-
+
接下来，你就可以使用 LobeChat 与本地 Gemma 模型对话了。
diff --git a/docs/usage/providers/ollama/qwen.mdx b/docs/usage/providers/ollama/qwen.mdx
index 243c02f53846c..12df7107670a0 100644
--- a/docs/usage/providers/ollama/qwen.mdx
+++ b/docs/usage/providers/ollama/qwen.mdx
@@ -11,11 +11,7 @@ tags:
# Using the Local Qwen Model
-
+
[Qwen](https://github.com/QwenLM/Qwen1.5) is a large language model (LLM) open-sourced by Alibaba Cloud. It is officially defined as a constantly evolving AI large model, and it achieves more accurate Chinese recognition capabilities through more training set content.
@@ -26,44 +22,33 @@ Now, through the integration of LobeChat and [Ollama](https://ollama.com/), you
## Local Installation of Ollama
-First, you need to install Ollama. For the installation process, please refer to the [Ollama Usage Document](/docs/usage/providers/ollama).
+ First, you need to install Ollama. For the installation process, please refer to the [Ollama Usage Document](/docs/usage/providers/ollama).
-## Pull the Qwen Model to Local with Ollama
+ ## Pull the Qwen Model to Local with Ollama
-After installing Ollama, you can install the Qwen model with the following command, taking the 14b model as an example:
+ After installing Ollama, you can install the Qwen model with the following command, taking the 14b model as an example:
-```bash
-ollama pull qwen:14b
-```
+ ```bash
+ ollama pull qwen:14b
+ ```
-
- The local version of Qwen provides different model sizes to choose from. Please refer to the
- [Qwen's Ollama integration page](https://ollama.com/library/qwen) to understand how to choose the
- model size.
-
+
+ The local version of Qwen provides different model sizes to choose from. Please refer to the
+ [Qwen's Ollama integration page](https://ollama.com/library/qwen) to understand how to choose the
+ model size.
+
-
+
-### Select the Qwen Model
+ ### Select the Qwen Model
-In the LobeChat conversation page, open the model selection panel, and then select the Qwen model.
+ In the LobeChat conversation page, open the model selection panel, and then select the Qwen model.
-
+
If you do not see the Ollama provider in the model selection panel, please refer to [Integration with Ollama](/docs/self-hosting/examples/ollama) to learn how to enable the Ollama provider in LobeChat.
-
-
+
Next, you can have a conversation with the local Qwen model in LobeChat.
diff --git a/docs/usage/providers/ollama/qwen.zh-CN.mdx b/docs/usage/providers/ollama/qwen.zh-CN.mdx
index db797ccb48d1d..da13ad753c83c 100644
--- a/docs/usage/providers/ollama/qwen.zh-CN.mdx
+++ b/docs/usage/providers/ollama/qwen.zh-CN.mdx
@@ -11,11 +11,7 @@ tags:
# 使用本地通义千问 Qwen 模型
-
+
[通义千问](https://github.com/QwenLM/Qwen1.5) 是阿里云开源的一款大语言模型（LLM），官方定义是一个不断进化的 AI 大模型，并通过更多的训练集内容达到更精准的中文识别能力。
@@ -28,39 +24,28 @@ tags:
### 本地安装 Ollama
-首先，你需要安装 Ollama，安装过程请查阅 [Ollama 使用文件](/zh/docs/usage/providers/ollama)。
+ 首先，你需要安装 Ollama，安装过程请查阅 [Ollama 使用文件](/zh/docs/usage/providers/ollama)。
-### 用 Ollama 拉取 Qwen 模型到本地
+ ### 用 Ollama 拉取 Qwen 模型到本地
-在安装完成 Ollama 后，你可以通过以下命令安装 Qwen 模型，以 14b 模型为例：
+ 在安装完成 Ollama 后，你可以通过以下命令安装 Qwen 模型，以 14b 模型为例：
-```bash
-ollama pull qwen:14b
-```
+ ```bash
+ ollama pull qwen:14b
+ ```
-
+
-### 选择 Qwen 模型
+ ### 选择 Qwen 模型
-在会话页面中，选择模型面板打开，然后选择 Qwen 模型。
+ 在会话页面中，选择模型面板打开，然后选择 Qwen 模型。
-
+
如果你没有在模型选择面板中看到 Ollama 服务商，请查阅 [与 Ollama
集成](/zh/docs/self-hosting/examples/ollama) 了解如何在 LobeChat 中开启 Ollama 服务商。
-
-
+
接下来，你就可以使用 LobeChat 与本地 Qwen 模型对话了。
diff --git a/docs/usage/providers/openai.mdx b/docs/usage/providers/openai.mdx
index ae624bf76ce70..4d1775bfc274c 100644
--- a/docs/usage/providers/openai.mdx
+++ b/docs/usage/providers/openai.mdx
@@ -14,81 +14,67 @@ tags:
# Using OpenAI in LobeChat
-
+
This document will guide you on how to use [OpenAI](https://openai.com/) in LobeChat:
+ ### Step 1: Obtain OpenAI API Key
-### Step 1: Obtain OpenAI API Key
+ - Register for an [OpenAI account](https://platform.openai.com/signup). You will need to register using an international phone number and a non-mainland email address.
-- Register for an [OpenAI account](https://platform.openai.com/signup). You will need to register using an international phone number and a non-mainland email address.
+ - After registration, go to the [API Keys](https://platform.openai.com/api-keys) page and click on `Create new secret key` to generate a new API Key.
-- After registration, go to the [API Keys](https://platform.openai.com/api-keys) page and click on `Create new secret key` to generate a new API Key.
+ - Open the creation window
-- Open the creation window
+
-
+ - Create API Key
-- Create API Key
+
-
+ - Retrieve API Key
-- Retrieve API Key
+
-
+
+ After registering, you generally have a free credit of $5, but it is only valid for three months.
+
-
- After registering, you generally have a free credit of $5, but it is only valid for three months.
-
+ ### Step 2: Configure OpenAI in LobeChat
-### Step 2: Configure OpenAI in LobeChat
+ - Visit the `Settings` page in LobeChat
+ - Find the setting for `OpenAI` under `Language Model`
-- Visit the `Settings` page in LobeChat
-- Find the setting for `OpenAI` under `Language Model`
+
-
+ - Enter the obtained API Key
+ - Choose an OpenAI model for your AI assistant to start the conversation
-- Enter the obtained API Key
-- Choose an OpenAI model for your AI assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider. Please refer to OpenAI's relevant
- pricing policies.
-
+
+
+ During usage, you may need to pay the API service provider. Please refer to OpenAI's relevant
+ pricing policies.
+
You can now engage in conversations using the models provided by OpenAI in LobeChat.
diff --git a/docs/usage/providers/openai.zh-CN.mdx b/docs/usage/providers/openai.zh-CN.mdx
index 390ec9e5fd96f..44d7d72067283 100644
--- a/docs/usage/providers/openai.zh-CN.mdx
+++ b/docs/usage/providers/openai.zh-CN.mdx
@@ -11,77 +11,64 @@ tags:
# 在 LobeChat 中使用 OpenAI
-
+
本文档将指导你如何在 LobeChat 中使用 [OpenAI](https://openai.com/):
+ ### 步骤一：获取 OpenAI API 密钥
-### 步骤一：获取 OpenAI API 密钥
+ - 注册一个 [OpenAI 账户](https://platform.openai.com/signup)，你需要使用国际手机号、非大陆邮箱进行注册；
-- 注册一个 [OpenAI 账户](https://platform.openai.com/signup)，你需要使用国际手机号、非大陆邮箱进行注册；
-- 注册完毕后，前往 [API Keys](https://platform.openai.com/api-keys) 页面，点击 `Create new secret key` 创建新的 API Key:
+ - 注册完毕后，前往 [API Keys](https://platform.openai.com/api-keys) 页面，点击 `Create new secret key` 创建新的 API Key:
-- 打开创建窗口
+ - 打开创建窗口
-
+
-- 创建 API Key
+ - 创建 API Key
-
+
-- 获取 API Key
+ - 获取 API Key
-
+
-账户注册后，一般有 5 美元的免费额度，但有效期只有三个月。
+ 账户注册后，一般有 5 美元的免费额度，但有效期只有三个月。
-### 步骤二：在 LobeChat 中配置 OpenAI
+ ### 步骤二：在 LobeChat 中配置 OpenAI
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到`OpenAI`的设置项
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`OpenAI`的设置项
-
+
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 OpenAI 的模型即可开始对话
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 OpenAI 的模型即可开始对话
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 OpenAI 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 OpenAI 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 OpenAI 提供的模型进行对话了。
diff --git a/docs/usage/providers/openrouter.mdx b/docs/usage/providers/openrouter.mdx
index 3fc4c683552f7..550d66bd2ca51 100644
--- a/docs/usage/providers/openrouter.mdx
+++ b/docs/usage/providers/openrouter.mdx
@@ -13,98 +13,62 @@ tags:
# Using OpenRouter in LobeChat
-
+
[OpenRouter](https://openrouter.ai/) is a service that provides a variety of excellent large language model APIs, supporting models such as OpenAI (including GPT-3.5/4), Anthropic (Claude2, Instant), LLaMA 2, and PaLM Bison.
This document will guide you on how to use OpenRouter in LobeChat:
+ ### Step 1: Register and Log in to OpenRouter
-### Step 1: Register and Log in to OpenRouter
-
-- Visit [OpenRouter.ai](https://openrouter.ai/) and create an account
-- You can log in using your Google account or MetaMask wallet
-
-
-
-### Step 2: Create an API Key
-
-- Go to the `Keys` menu or visit [OpenRouter Keys](https://openrouter.ai/keys) directly
-- Click on `Create Key` to start the creation process
-- Name your API key in the pop-up dialog, for example, "LobeChat Key"
-- Leave the `Credit limit` blank to indicate no amount limit
-
-
-
-- Copy the API key from the pop-up dialog and save it securely
-
-
-
-
- Please store the key securely as it will only appear once. If you lose it accidentally, you will
- need to create a new key.
-
-
-### Step 3: Recharge Credit
-
-- Go to the `Credit` menu or visit [OpenRouter Credit](https://openrouter.ai/credits) directly
-- Click on `Manage Credits` to recharge your credit, you can check model prices at [https://openrouter.ai/models](https://openrouter.ai/models)
-- OpenRouter provides some free models that can be used without recharging
-
-
-
-### Step 4: Configure OpenRouter in LobeChat
-
-- Visit the `Settings` interface in LobeChat
-- Find the setting for `OpenRouter` under `Language Models`
-- Enable OpenRouter and enter the API key you obtained
-
-
-
-- Choose an OpenRouter model for your assistant to start the conversation
-
-
-
-
- You may need to pay the API service provider during usage, please refer to OpenRouter's relevant
- fee policies.
-
+ - Visit [OpenRouter.ai](https://openrouter.ai/) and create an account
+ - You can log in using your Google account or MetaMask wallet
+
+
+ ### Step 2: Create an API Key
+
+ - Go to the `Keys` menu or visit [OpenRouter Keys](https://openrouter.ai/keys) directly
+ - Click on `Create Key` to start the creation process
+ - Name your API key in the pop-up dialog, for example, "LobeChat Key"
+ - Leave the `Credit limit` blank to indicate no amount limit
+
+
+
+ - Copy the API key from the pop-up dialog and save it securely
+
+
+
+
+ Please store the key securely as it will only appear once. If you lose it accidentally, you will
+ need to create a new key.
+
+
+ ### Step 3: Recharge Credit
+
+ - Go to the `Credit` menu or visit [OpenRouter Credit](https://openrouter.ai/credits) directly
+ - Click on `Manage Credits` to recharge your credit, you can check model prices at [https://openrouter.ai/models](https://openrouter.ai/models)
+ - OpenRouter provides some free models that can be used without recharging
+
+
+
+ ### Step 4: Configure OpenRouter in LobeChat
+
+ - Visit the `Settings` interface in LobeChat
+ - Find the setting for `OpenRouter` under `Language Models`
+ - Enable OpenRouter and enter the API key you obtained
+
+
+
+ - Choose an OpenRouter model for your assistant to start the conversation
+
+
+
+
+ You may need to pay the API service provider during usage, please refer to OpenRouter's relevant
+ fee policies.
+
You can now engage in conversations using the models provided by OpenRouter in LobeChat.
diff --git a/docs/usage/providers/openrouter.zh-CN.mdx b/docs/usage/providers/openrouter.zh-CN.mdx
index 7db17f3c1ea0c..40670ebe814c4 100644
--- a/docs/usage/providers/openrouter.zh-CN.mdx
+++ b/docs/usage/providers/openrouter.zh-CN.mdx
@@ -9,96 +9,60 @@ tags:
# 在 LobeChat 中使用 OpenRouter
-
+
[OpenRouter](https://openrouter.ai/) 是一个提供多种优秀大语言模型 API 的服务，它支持 OpenAI (包括 GPT-3.5/4)、Anthropic (Claude2、Instant)、LLaMA 2 和 PaLM Bison 等众多模型。
本文档将指导你如何在 LobeChat 中使用 OpenRouter:
+ ### 步骤一：注册 OpenRouter 账号并登录
-### 步骤一：注册 OpenRouter 账号并登录
+ - 访问 [OpenRouter.ai](https://openrouter.ai/) 并创建一个账号
+ - 你可以用 Google 账号或 MetaMask 钱包登录
-- 访问 [OpenRouter.ai](https://openrouter.ai/) 并创建一个账号
-- 你可以用 Google 账号或 MetaMask 钱包登录
+
-
+ ### 步骤二：创建 API 密钥
-### 步骤二：创建 API 密钥
+ - 进入 `Keys` 菜单或直接访问 [OpenRouter Keys](https://openrouter.ai/keys)
+ - 点击 `Create Key` 开始创建
+ - 在弹出对话框中为 API 密钥取一个名字，例如 "LobeChat Key"
+ - 留空 `Credit limit` 表示不设置金额限制
-- 进入 `Keys` 菜单或直接访问 [OpenRouter Keys](https://openrouter.ai/keys)
-- 点击 `Create Key` 开始创建
-- 在弹出对话框中为 API 密钥取一个名字，例如 "LobeChat Key"
-- 留空 `Credit limit` 表示不设置金额限制
+
-
+ - 在弹出的对话框中复制 API 密钥，并妥善保存
-- 在弹出的对话框中复制 API 密钥，并妥善保存
+
-
+
+ 请安全地存储密钥，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新密钥。
+
-
- 请安全地存储密钥，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新密钥。
-
+ ### 步骤三：充值信用额度
-### 步骤三：充值信用额度
+ - 进入 `Credit` 菜单，或直接访问 [OpenRouter Credit](https://openrouter.ai/credits)
+ - 点击 `Manage Credits` 充值信用额度，在 [https://openrouter.ai/models](https://openrouter.ai/models) 中可以查看模型价格
+ - OpenRouter 提供了一些免费模型，未充值的情况下可以使用
-- 进入 `Credit` 菜单，或直接访问 [OpenRouter Credit](https://openrouter.ai/credits)
-- 点击 `Manage Credits` 充值信用额度，在 [https://openrouter.ai/models](https://openrouter.ai/models) 中可以查看模型价格
-- OpenRouter 提供了一些免费模型，未充值的情况下可以使用
+
-
+ ### 步骤四：在 LobeChat 中配置 OpenRouter
-### 步骤四：在 LobeChat 中配置 OpenRouter
+ - 访问 LobeChat 的 `设置` 界面
+ - 在 `语言模型` 下找到 `OpenRouter` 的设置项
+ - 打开 OpenRouter 并填入获得的 API 密钥
-- 访问 LobeChat 的 `设置` 界面
-- 在 `语言模型` 下找到 `OpenRouter` 的设置项
-- 打开 OpenRouter 并填入获得的 API 密钥
+
-
+ - 为你的助手选择一个 OpenRouter 模型即可开始对话
-- 为你的助手选择一个 OpenRouter 模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 OpenRouter 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 OpenRouter 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 OpenRouter 提供的模型进行对话了。
diff --git a/docs/usage/providers/perplexity.mdx b/docs/usage/providers/perplexity.mdx
index d35b9e03124b5..f76fd5c7fc194 100644
--- a/docs/usage/providers/perplexity.mdx
+++ b/docs/usage/providers/perplexity.mdx
@@ -11,52 +11,34 @@ tags:
# Using Perplexity AI in LobeChat
-
+
The Perplexity AI API is now available for everyone to use. This document will guide you on how to use [Perplexity AI](https://www.perplexity.ai/) in LobeChat:
+ ### Step 1: Obtain Perplexity AI API Key
-### Step 1: Obtain Perplexity AI API Key
+ - Create a [Perplexity AI](https://www.perplexity.ai/) account
+ - Obtain your [API key](https://www.perplexity.ai/settings/api)
-- Create a [Perplexity AI](https://www.perplexity.ai/) account
-- Obtain your [API key](https://www.perplexity.ai/settings/api)
+
-
+ ### Step 2: Configure Perplexity AI in LobeChat
-### Step 2: Configure Perplexity AI in LobeChat
+ - Go to the `Settings` interface in LobeChat
+ - Find the setting for `Perplexity AI` under `Language Model`
-- Go to the `Settings` interface in LobeChat
-- Find the setting for `Perplexity AI` under `Language Model`
+
-
+ - Enter the API key you obtained
+ - Choose a Perplexity AI model for your AI assistant to start the conversation
-- Enter the API key you obtained
-- Choose a Perplexity AI model for your AI assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider. Please refer to Perplexity AI's
- relevant pricing policies.
-
+
+
+ During usage, you may need to pay the API service provider. Please refer to Perplexity AI's
+ relevant pricing policies.
+
You can now engage in conversations using the models provided by Perplexity AI in LobeChat.
diff --git a/docs/usage/providers/perplexity.zh-CN.mdx b/docs/usage/providers/perplexity.zh-CN.mdx
index 7b1b7c36cf0cb..b5de71464de43 100644
--- a/docs/usage/providers/perplexity.zh-CN.mdx
+++ b/docs/usage/providers/perplexity.zh-CN.mdx
@@ -9,51 +9,33 @@ tags:
# 在 LobeChat 中使用 Perplexity AI
-
+
-Perplexity AI API 现在可供所有人使用, 本文档将指导你如何在 LobeChat 中使用 [Perplexity AI](https://www.perplexity.ai/):
+Perplexity AI API 现在可供所有人使用，本文档将指导你如何在 LobeChat 中使用 [Perplexity AI](https://www.perplexity.ai/):
+ ### 步骤一：获取 Perplexity AI API 密钥
-### 步骤一：获取 Perplexity AI API 密钥
+ - 创建一个 [Perplexity AI](https://www.perplexity.ai/) 帐户
+ - 获取您的 [API 密钥](https://www.perplexity.ai/settings/api)
-- 创建一个 [Perplexity AI](https://www.perplexity.ai/) 帐户
-- 获取您的 [API 密钥](https://www.perplexity.ai/settings/api)
+
-
+ ### 步骤二：在 LobeChat 中配置 Perplexity AI
-### 步骤二：在 LobeChat 中配置 Perplexity AI
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`Perplexity AI`的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到`Perplexity AI`的设置项
+
-
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 Perplexity AI 的模型即可开始对话
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 Perplexity AI 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Perplexity AI 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Perplexity AI 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Perplexity AI 提供的模型进行对话了。
diff --git a/docs/usage/providers/qwen.mdx b/docs/usage/providers/qwen.mdx
index 6291a6f101fb4..9563ede888a88 100644
--- a/docs/usage/providers/qwen.mdx
+++ b/docs/usage/providers/qwen.mdx
@@ -15,78 +15,52 @@ tags:
# Using Tongyi Qianwen in LobeChat
-
+
[Tongyi Qianwen](https://tongyi.aliyun.com/) is a large-scale language model independently developed by Alibaba Cloud, with powerful natural language understanding and generation capabilities. It can answer various questions, create text content, express opinions, write code, and play a role in multiple fields.
This document will guide you on how to use Tongyi Qianwen in LobeChat:
+ ### Step 1: Activate DashScope Model Service
-### Step 1: Activate DashScope Model Service
+ - Visit and log in to Alibaba Cloud's [DashScope](https://dashscope.console.aliyun.com/) platform.
+ - If it is your first time, you need to activate the DashScope service.
+ - If you have already activated it, you can skip this step.
-- Visit and log in to Alibaba Cloud's [DashScope](https://dashscope.console.aliyun.com/) platform.
-- If it is your first time, you need to activate the DashScope service.
-- If you have already activated it, you can skip this step.
+
-
+ ### Step 2: Obtain DashScope API Key
-### Step 2: Obtain DashScope API Key
+ - Go to the `API-KEY` interface and create an API key.
-- Go to the `API-KEY` interface and create an API key.
+
-
+ - Copy the API key from the pop-up dialog box and save it securely.
-- Copy the API key from the pop-up dialog box and save it securely.
+
-
+
+ Please store the key securely as it will only appear once. If you accidentally lose it, you will
+ need to create a new key.
+
-
- Please store the key securely as it will only appear once. If you accidentally lose it, you will
- need to create a new key.
-
+ ### Step 3: Configure Tongyi Qianwen in LobeChat
-### Step 3: Configure Tongyi Qianwen in LobeChat
+ - Visit the `Settings` interface in LobeChat.
+ - Find the setting for `Tongyi Qianwen` under `Language Model`.
-- Visit the `Settings` interface in LobeChat.
-- Find the setting for `Tongyi Qianwen` under `Language Model`.
+
-
+ - Open Tongyi Qianwen and enter the obtained API key.
+ - Choose a Qwen model for your AI assistant to start the conversation.
-- Open Tongyi Qianwen and enter the obtained API key.
-- Choose a Qwen model for your AI assistant to start the conversation.
-
-
-
-
- During usage, you may need to pay the API service provider. Please refer to Tongyi Qianwen's
- relevant pricing policies.
-
+
+
+ During usage, you may need to pay the API service provider. Please refer to Tongyi Qianwen's
+ relevant pricing policies.
+
You can now engage in conversations using the models provided by Tongyi Qianwen in LobeChat.
diff --git a/docs/usage/providers/qwen.zh-CN.mdx b/docs/usage/providers/qwen.zh-CN.mdx
index 09bbb441d03c7..451d6e15ee8b6 100644
--- a/docs/usage/providers/qwen.zh-CN.mdx
+++ b/docs/usage/providers/qwen.zh-CN.mdx
@@ -12,76 +12,50 @@ tags:
# 在 LobeChat 中使用通义千问
-
+
-[通义千问](https://tongyi.aliyun.com/)是阿里云自主研发的超大规模语言模型,具有强大的自然语言理解和生成能力。它可以回答各种问题、创作文字内容、表达观点看法、撰写代码等,在多个领域发挥作用。
+[通义千问](https://tongyi.aliyun.com/)是阿里云自主研发的超大规模语言模型，具有强大的自然语言理解和生成能力。它可以回答各种问题、创作文字内容、表达观点看法、撰写代码等，在多个领域发挥作用。
本文档将指导你如何在 LobeChat 中使用通义千问:
+ ### 步骤一：开通 DashScope 模型服务
-### 步骤一：开通 DashScope 模型服务
+ - 访问并登录阿里云 [DashScope](https://dashscope.console.aliyun.com/) 平台
+ - 初次进入时需要开通 DashScope 服务
+ - 若你已开通，可跳过该步骤
-- 访问并登录阿里云 [DashScope](https://dashscope.console.aliyun.com/) 平台
-- 初次进入时需要开通 DashScope 服务
-- 若你已开通，可跳过该步骤
+
-
+ ### 步骤二：获取 DashScope API 密钥
-### 步骤二：获取 DashScope API 密钥
+ - 进入`API-KEY` 界面，并创建一个 API 密钥
-- 进入`API-KEY` 界面，并创建一个 API 密钥
+
-
+ - 在弹出的对话框中复制 API 密钥，并妥善保存
-- 在弹出的对话框中复制 API 密钥，并妥善保存
+
-
+
+ 请安全地存储密钥，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新密钥。
+
-
- 请安全地存储密钥，因为它只会出现一次。如果您意外丢失它，您将需要创建一个新密钥。
-
+ ### 步骤三：在 LobeChat 中配置通义千问
-### 步骤三：在LobeChat 中配置通义千问
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`通义千问`的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到`通义千问`的设置项
+
-
+ - 打开通义千问并填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 Qwen 的模型即可开始对话
-- 打开通义千问并填入获得的 API 密钥
-- 为你的 AI 助手选择一个 Qwen 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考通义千问的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考通义千问的相关费用政策。
+
至此你已经可以在 LobeChat 中使用通义千问提供的模型进行对话了。
diff --git a/docs/usage/providers/sensenova.mdx b/docs/usage/providers/sensenova.mdx
index e66d6d63fdb42..a8a77dc037c63 100644
--- a/docs/usage/providers/sensenova.mdx
+++ b/docs/usage/providers/sensenova.mdx
@@ -12,68 +12,47 @@ tags:
# Using SenseNova in LobeChat
-
+
[SenseNova](https://platform.sensenova.cn/home) is a large model system introduced by SenseTime, aimed at promoting the rapid iteration and practical application of artificial intelligence (AI) technology.
This article will guide you on how to use SenseNova in LobeChat.
+ ### Step 1: Obtain the API Key for SenseNova
-### Step 1: Obtain the API Key for SenseNova
+ - Register and log in to the [SenseCore Development Platform](https://www.sensecore.cn/product/aistudio).
+ - Locate the `SenseNova Large Model` in the product menu and activate the service.
-- Register and log in to the [SenseCore Development Platform](https://www.sensecore.cn/product/aistudio).
-- Locate the `SenseNova Large Model` in the product menu and activate the service.
+
-
+ - Go to the [AccessKey Management](https://console.sensecore.cn/iam/Security/access-key) page.
+ - Create an access key.
+ - Save the Access Key ID and secret in the pop-up window.
-- Go to the [AccessKey Management](https://console.sensecore.cn/iam/Security/access-key) page.
-- Create an access key.
-- Save the Access Key ID and secret in the pop-up window.
+
-
+
+ Please keep the access key from the pop-up window secure, as it will only appear once. If you lose
+ it, you will need to create a new access key.
+
-
- Please keep the access key from the pop-up window secure, as it will only appear once. If you lose
- it, you will need to create a new access key.
-
+ ### Step 2: Configure SenseNova in LobeChat
-### Step 2: Configure SenseNova in LobeChat
+ - Access the `Settings` interface on LobeChat.
+ - Find the setting for `SenseNova` under `Language Models`.
-- Access the `Settings` interface on LobeChat.
-- Find the setting for `SenseNova` under `Language Models`.
+
-
+ - Input the obtained `Access Key ID` and `Access Key Secret`.
+ - Choose a SenseNova model for your AI assistant and start the conversation.
-- Input the obtained `Access Key ID` and `Access Key Secret`.
-- Choose a SenseNova model for your AI assistant and start the conversation.
-
-
-
-
- During usage, you may need to pay the API service provider, please refer to the relevant fee
- policy for SenseNova.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to the relevant fee
+ policy for SenseNova.
+
You can now have conversations using the models provided by SenseNova in LobeChat.
diff --git a/docs/usage/providers/sensenova.zh-CN.mdx b/docs/usage/providers/sensenova.zh-CN.mdx
index 454fdf40aaffa..449f4ce2117b0 100644
--- a/docs/usage/providers/sensenova.zh-CN.mdx
+++ b/docs/usage/providers/sensenova.zh-CN.mdx
@@ -10,66 +10,45 @@ tags:
# 在 LobeChat 中使用商汤日日新
-
+
[商汤日日新](https://platform.sensenova.cn/home) 是商汤科技（SenseTime）推出的一个大模型体系，旨在推动人工智能（AI）技术的快速迭代和应用落地。
本文将指导你如何在 LobeChat 中使用商汤日日新。
+ ### 步骤一：获取商汤日日新的 API 密钥
-### 步骤一: 获取商汤日日新的 API 密钥
+ - 注册并登录 [万象模型开发平台](https://www.sensecore.cn/product/aistudio)
+ - 在产品菜单中找到 `日日新大模型` 并开通服务
-- 注册并登录 [万象模型开发平台](https://www.sensecore.cn/product/aistudio)
-- 在产品菜单中找到 `日日新大模型` 并开通服务
+
-
+ - 进入 [AccessKey 访问密钥](https://console.sensecore.cn/iam/Security/access-key) 页面
+ - 创建一个访问密钥
+ - 在弹出窗口中保存访问密钥 ID 和令牌
-- 进入 [AccessKey 访问密钥](https://console.sensecore.cn/iam/Security/access-key) 页面
-- 创建一个访问密钥
-- 在弹出窗口中保存访问密钥 ID 和令牌
+
-
+
+ 妥善保存弹窗中的访问密钥，它只会出现一次，如果不小心丢失了，你需要重新创建一个访问密钥。
+
-
- 妥善保存弹窗中的访问密钥，它只会出现一次，如果不小心丢失了，你需要重新创建一个访问密钥。
-
+ ### 步骤二：在 LobeChat 中配置商汤日日新
-### 步骤二：在 LobeChat 中配置商汤日日新
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `商汤日日新` 的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `商汤日日新` 的设置项
+
-
+ - 填入获得的 `AccessKey ID` 和 `AccessKey Secret`
+ - 为你的 AI 助手选择一个商汤日日新的模型即可开始对话
-- 填入获得的 `AccessKey ID` 和 `AccessKey Secret`
-- 为你的 AI 助手选择一个商汤日日新的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考商汤日日新的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考商汤日日新的相关费用政策。
+
至此你已经可以在 LobeChat 中使用商汤日日新提供的模型进行对话了。
diff --git a/docs/usage/providers/siliconcloud.mdx b/docs/usage/providers/siliconcloud.mdx
index 36f23a62e9a92..f578610456982 100644
--- a/docs/usage/providers/siliconcloud.mdx
+++ b/docs/usage/providers/siliconcloud.mdx
@@ -12,53 +12,37 @@ tags:
# Using SiliconCloud in LobeChat
-
+
[SiliconCloud](https://siliconflow.cn/) is an AI service platform based on open-source foundational models, offering a variety of generative AI (GenAI) services.
This article will guide you on how to use SiliconCloud in LobeChat.
-### Step 1: Obtain the API Key from SiliconCloud
+ ### Step 1: Obtain the API Key from SiliconCloud
-- Sign up and log in to [SiliconCloud](https://cloud.siliconflow.cn/account/ak)
-- Click on the `API Keys` menu on the left side
-- Create an API Key and copy it
+ - Sign up and log in to [SiliconCloud](https://cloud.siliconflow.cn/account/ak)
+ - Click on the `API Keys` menu on the left side
+ - Create an API Key and copy it
-
+
-### Step 2: Configure SiliconCloud in LobeChat
+ ### Step 2: Configure SiliconCloud in LobeChat
-- Go to the `Settings` page in LobeChat
-- Under `Language Model`, find the setting for `SiliconFlow`
+ - Go to the `Settings` page in LobeChat
+ - Under `Language Model`, find the setting for `SiliconFlow`
-
+
-- Enter the API Key you obtained
-- Choose a SiliconCloud model for your AI assistant to start the conversation
+ - Enter the API Key you obtained
+ - Choose a SiliconCloud model for your AI assistant to start the conversation
-
-
-
- During usage, you may need to pay the API service provider, so please refer to SiliconCloud's
- relevant pricing policy.
-
+
+
+ During usage, you may need to pay the API service provider, so please refer to SiliconCloud's
+ relevant pricing policy.
+
At this point, you can start chatting using the models provided by SiliconCloud in LobeChat.
diff --git a/docs/usage/providers/siliconcloud.zh-CN.mdx b/docs/usage/providers/siliconcloud.zh-CN.mdx
index 8c4e882ad03c7..b9862957e21d3 100644
--- a/docs/usage/providers/siliconcloud.zh-CN.mdx
+++ b/docs/usage/providers/siliconcloud.zh-CN.mdx
@@ -10,52 +10,36 @@ tags:
# 在 LobeChat 中使用 SiliconCloud
-
+
-[SiliconCloud](https://siliconflow.cn/) 是一个基于开源基础模型的人工智能服务平台，提供多种生成式AI（GenAI）服务。
+[SiliconCloud](https://siliconflow.cn/) 是一个基于开源基础模型的人工智能服务平台，提供多种生成式 AI（GenAI）服务。
本文将指导你如何在 LobeChat 中使用 SiliconCloud。
-### 步骤一：获得 SiliconCloud 的 API Key
+ ### 步骤一：获得 SiliconCloud 的 API Key
-- 注册并登录 [SiliconCloud](https://cloud.siliconflow.cn/account/ak)
-- 点击左侧 `API 密钥` 菜单
-- 创建一个 API 密钥并复制
+ - 注册并登录 [SiliconCloud](https://cloud.siliconflow.cn/account/ak)
+ - 点击左侧 `API 密钥` 菜单
+ - 创建一个 API 密钥并复制
-
+
-### 步骤二：在 LobeChat 中配置 SiliconCloud
+ ### 步骤二：在 LobeChat 中配置 SiliconCloud
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `SiliconFlow` 的设置项
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `SiliconFlow` 的设置项
-
+
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 SiliconCloud 的模型即可开始对话
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 SiliconCloud 的模型即可开始对话
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 SiliconCloud 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 SiliconCloud 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 SiliconCloud 提供的模型进行对话了。
diff --git a/docs/usage/providers/spark.mdx b/docs/usage/providers/spark.mdx
index d2c7385d68ea2..52a03098ed033 100644
--- a/docs/usage/providers/spark.mdx
+++ b/docs/usage/providers/spark.mdx
@@ -11,61 +11,41 @@ tags:
# Using iFLYTEK Spark in LobeChat
-
+
-[iFLYTEK Spark](https://xinghuo.xfyun.cn/) is a powerful AI model launched by iFLYTEK, equipped with cross-domain knowledge and language understanding capabilities, able to perform various tasks such as Q&A, conversations, and literary creation.
+[iFLYTEK Spark](https://xinghuo.xfyun.cn/) is a powerful AI model launched by iFLYTEK, equipped with cross-domain knowledge and language understanding capabilities, able to perform various tasks such as Q\&A, conversations, and literary creation.
This guide will instruct you on how to use iFLYTEK Spark in LobeChat.
-### Step 1: Obtain the iFLYTEK Spark API Key
+ ### Step 1: Obtain the iFLYTEK Spark API Key
-- Register and log in to the [iFLYTEK Open Platform](https://console.xfyun.cn/)
-- Create an application
+ - Register and log in to the [iFLYTEK Open Platform](https://console.xfyun.cn/)
+ - Create an application
-
+
-- Select a large model to view details
-- Copy the `API Password` from the top right corner under the HTTP service interface authentication information
+ - Select a large model to view details
+ - Copy the `API Password` from the top right corner under the HTTP service interface authentication information
-
+
-### Step 2: Configure iFLYTEK Spark in LobeChat
+ ### Step 2: Configure iFLYTEK Spark in LobeChat
-- Access the `Settings` menu in LobeChat
-- Find the iFLYTEK Spark settings under `Language Model`
+ - Access the `Settings` menu in LobeChat
+ - Find the iFLYTEK Spark settings under `Language Model`
-
+
-- Input the obtained API Key
-- Choose an iFLYTEK Spark model for your AI assistant to start the conversation
+ - Input the obtained API Key
+ - Choose an iFLYTEK Spark model for your AI assistant to start the conversation
-
-
-
- During usage, you may need to pay the API service provider, please refer to the relevant pricing
- policy of iFLYTEK Spark.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to the relevant pricing
+ policy of iFLYTEK Spark.
+
Now you can use the models provided by iFLYTEK Spark for conversations in LobeChat.
diff --git a/docs/usage/providers/spark.zh-CN.mdx b/docs/usage/providers/spark.zh-CN.mdx
index e6d1fc1b0cc79..0b43463d42f3d 100644
--- a/docs/usage/providers/spark.zh-CN.mdx
+++ b/docs/usage/providers/spark.zh-CN.mdx
@@ -10,60 +10,40 @@ tags:
# 在 LobeChat 中使用讯飞星火
-
+
-[讯飞星火](https://xinghuo.xfyun.cn/)是科大讯飞推出的一款强大的AI大模型，具备跨领域的知识和语言理解能力，能够执行问答、对话和文学创作等多种任务。
+[讯飞星火](https://xinghuo.xfyun.cn/)是科大讯飞推出的一款强大的 AI 大模型，具备跨领域的知识和语言理解能力，能够执行问答、对话和文学创作等多种任务。
本文将指导你如何在 LobeChat 中使用讯飞星火。
-### 步骤一：获得讯飞星火的 API Key
+ ### 步骤一：获得讯飞星火的 API Key
-- 注册并登录 [讯飞开放平台](https://console.xfyun.cn/)
-- 创建一个应用
+ - 注册并登录 [讯飞开放平台](https://console.xfyun.cn/)
+ - 创建一个应用
-
+
-- 选择一个大模型查看详情
-- 复制右上角 http 服务接口认证信息中的 `API Password`
+ - 选择一个大模型查看详情
+ - 复制右上角 http 服务接口认证信息中的 `API Password`
-
+
-### 步骤二：在 LobeChat 中配置讯飞星火
+ ### 步骤二：在 LobeChat 中配置讯飞星火
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `讯飞星火` 的设置项
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `讯飞星火` 的设置项
-
+
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个讯飞星火的模型即可开始对话
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个讯飞星火的模型即可开始对话
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考讯飞星火的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考讯飞星火的相关费用政策。
+
至此你已经可以在 LobeChat 中使用讯飞星火提供的模型进行对话了。
diff --git a/docs/usage/providers/stepfun.mdx b/docs/usage/providers/stepfun.mdx
index 430daa6d14d92..34ddb6d84b78b 100644
--- a/docs/usage/providers/stepfun.mdx
+++ b/docs/usage/providers/stepfun.mdx
@@ -12,55 +12,37 @@ tags:
# Using Stepfun in LobeChat
-
+
[Stepfun](https://www.stepfun.com/) is a startup focusing on the research and development of Artificial General Intelligence (AGI). They have released the Step-1 billion-parameter language model, Step-1V billion-parameter multimodal model, and the Step-2 trillion-parameter MoE language model preview.
This document will guide you on how to use Stepfun in LobeChat:
+ ### Step 1: Obtain Stepfun API Key
-### Step 1: Obtain Stepfun API Key
+ - Visit and log in to the [Stepfun Open Platform](https://platform.stepfun.com/)
+ - Go to the `API Key` menu, where the system has already created an API key for you
+ - Copy the created API key
-- Visit and log in to the [Stepfun Open Platform](https://platform.stepfun.com/)
-- Go to the `API Key` menu, where the system has already created an API key for you
-- Copy the created API key
+
-
+ ### Step 2: Configure Stepfun in LobeChat
-### Step 2: Configure Stepfun in LobeChat
+ - Visit the `Settings` interface in LobeChat
+ - Find the setting for Stepfun under `Language Models`
-- Visit the `Settings` interface in LobeChat
-- Find the setting for Stepfun under `Language Models`
+
-
+ - Open Stepfun and enter the obtained API key
+ - Choose a Stepfun model for your AI assistant to start the conversation
-- Open Stepfun and enter the obtained API key
-- Choose a Stepfun model for your AI assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider, please refer to Stepfun's relevant
- pricing policies.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to Stepfun's relevant
+ pricing policies.
+
You can now use the models provided by Stepfun to have conversations in LobeChat.
diff --git a/docs/usage/providers/stepfun.zh-CN.mdx b/docs/usage/providers/stepfun.zh-CN.mdx
index df6a855725f6d..db01006d7225d 100644
--- a/docs/usage/providers/stepfun.zh-CN.mdx
+++ b/docs/usage/providers/stepfun.zh-CN.mdx
@@ -9,54 +9,36 @@ tags:
# 在 LobeChat 中使用 Stepfun 阶跃星辰
-
+
-[Stepfun 阶跃星辰](https://www.stepfun.com/)是一家专注于通用人工智能(AGI)研发的创业公司，目前已推出Step-1千亿参数语言大模型、Step-1V千亿参数多模态大模型,以及Step-2万亿参数MoE语言大模型预览版。
+[Stepfun 阶跃星辰](https://www.stepfun.com/)是一家专注于通用人工智能 (AGI) 研发的创业公司，目前已推出 Step-1 千亿参数语言大模型、Step-1V 千亿参数多模态大模型，以及 Step-2 万亿参数 MoE 语言大模型预览版。
本文档将指导你如何在 LobeChat 中使用 Stepfun 阶跃星辰:
+ ### 步骤一：获取 Stepfun 阶跃星辰 API 密钥
-### 步骤一：获取 Stepfun 阶跃星辰 API 密钥
+ - 访问并登录 [Stepfun Stepfun 阶跃星辰开放平台](https://platform.stepfun.com/)
+ - 进入`接口密钥`菜单，系统已为你创建好 API 密钥
+ - 复制已创建的 API 密钥
-- 访问并登录 [Stepfun Stepfun 阶跃星辰开放平台](https://platform.stepfun.com/)
-- 进入`接口密钥`菜单，系统已为你创建好 API 密钥
-- 复制已创建的 API 密钥
+
-
+ ### 步骤二：在 LobeChat 中配置 Stepfun Stepfun 阶跃星辰
-### 步骤二：在LobeChat 中配置 Stepfun Stepfun 阶跃星辰
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到` Stepfun 阶跃星辰`的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到` Stepfun 阶跃星辰`的设置项
+
-
+ - 打开 Stepfun 阶跃星辰并填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 Stepfun 阶跃星辰的模型即可开始对话
-- 打开 Stepfun 阶跃星辰并填入获得的 API 密钥
-- 为你的 AI 助手选择一个 Stepfun 阶跃星辰的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Stepfun 阶跃星辰的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Stepfun 阶跃星辰的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Stepfun 阶跃星辰提供的模型进行对话了。
diff --git a/docs/usage/providers/taichu.mdx b/docs/usage/providers/taichu.mdx
index 5ccee72fbc24e..8a59caf67d492 100644
--- a/docs/usage/providers/taichu.mdx
+++ b/docs/usage/providers/taichu.mdx
@@ -13,52 +13,34 @@ tags:
# Using Taichu in LobeChat
-
+
This article will guide you on how to use Taichu in LobeChat:
+ ### Step 1: Obtain Taichu API Key
-### Step 1: Obtain Taichu API Key
+ - Create an account on [Taichu](https://ai-maas.wair.ac.cn/)
+ - Create and obtain an [API key](https://ai-maas.wair.ac.cn/#/settlement/api/key)
-- Create an account on [Taichu](https://ai-maas.wair.ac.cn/)
-- Create and obtain an [API key](https://ai-maas.wair.ac.cn/#/settlement/api/key)
+
-
+ ### Step 2: Configure Taichu in LobeChat
-### Step 2: Configure Taichu in LobeChat
+ - Go to the `Settings` interface in LobeChat
+ - Find the setting for `Taichu` under `Language Model`
-- Go to the `Settings` interface in LobeChat
-- Find the setting for `Taichu` under `Language Model`
+
-
+ - Enter the obtained API key
+ - Choose a Purple Taichu model for your AI assistant to start the conversation
-- Enter the obtained API key
-- Choose a Purple Taichu model for your AI assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider, please refer to Taichu's relevant
- pricing policies.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to Taichu's relevant
+ pricing policies.
+
Now you can start conversing with the models provided by Taichu in LobeChat.
diff --git a/docs/usage/providers/taichu.zh-CN.mdx b/docs/usage/providers/taichu.zh-CN.mdx
index 6e87ce04eb59c..82e05690f5f5c 100644
--- a/docs/usage/providers/taichu.zh-CN.mdx
+++ b/docs/usage/providers/taichu.zh-CN.mdx
@@ -11,51 +11,33 @@ tags:
# 在 LobeChat 中使用紫东太初
-
+
本文将指导你如何在 LobeChat 中使用紫东太初：
+ ### 步骤一：获取紫东太初 API 密钥
-### 步骤一：获取紫东太初 API 密钥
+ - 创建一个[紫东太初](https://ai-maas.wair.ac.cn/)账户
+ - 创建并获取 [API 密钥](https://ai-maas.wair.ac.cn/#/settlement/api/key)
-- 创建一个[紫东太初](https://ai-maas.wair.ac.cn/)账户
-- 创建并获取 [API 密钥](https://ai-maas.wair.ac.cn/#/settlement/api/key)
+
-
+ ### 步骤二：在 LobeChat 中配置紫东太初
-### 步骤二：在 LobeChat 中配置紫东太初
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`紫东太初`的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到`紫东太初`的设置项
+
-
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个紫东太初的模型即可开始对话
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个紫东太初的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考紫东太初的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考紫东太初的相关费用政策。
+
至此你已经可以在 LobeChat 中使用紫东太初提供的模型进行对话了。
diff --git a/docs/usage/providers/togetherai.mdx b/docs/usage/providers/togetherai.mdx
index fd158c168339a..5b9e2ebde593b 100644
--- a/docs/usage/providers/togetherai.mdx
+++ b/docs/usage/providers/togetherai.mdx
@@ -11,62 +11,40 @@ tags:
# Using Together AI in LobeChat
-
+
[together.ai](https://www.together.ai/) is a platform focused on the field of Artificial Intelligence Generated Content (AIGC), founded in June 2022. It is dedicated to building a cloud platform for running, training, and fine-tuning open-source models, providing scalable computing power at prices lower than mainstream vendors.
This document will guide you on how to use Together AI in LobeChat:
+ ### Step 1: Obtain the API Key for Together AI
-### Step 1: Obtain the API Key for Together AI
+ - Visit and log in to [Together AI API](https://api.together.ai/)
+ - Upon initial login, the system will automatically create an API key for you and provide a $5.0 credit
-- Visit and log in to [Together AI API](https://api.together.ai/)
-- Upon initial login, the system will automatically create an API key for you and provide a $5.0 credit
+
-
+ - If you haven't saved it, you can also view the API key at any time in the `API Key` interface under `Settings`
-- If you haven't saved it, you can also view the API key at any time in the `API Key` interface under `Settings`
+
-
+ ### Step 2: Configure Together AI in LobeChat
-### Step 2: Configure Together AI in LobeChat
+ - Visit the `Settings` interface in LobeChat
+ - Find the setting for `together.ai` under `Language Model`
-- Visit the `Settings` interface in LobeChat
-- Find the setting for `together.ai` under `Language Model`
+
-
+ - Open together.ai and enter the obtained API key
+ - Choose a Together AI model for your assistant to start the conversation
-- Open together.ai and enter the obtained API key
-- Choose a Together AI model for your assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider, please refer to Together AI's pricing
- policy.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to Together AI's pricing
+ policy.
+
You can now engage in conversations using the models provided by Together AI in LobeChat.
diff --git a/docs/usage/providers/togetherai.zh-CN.mdx b/docs/usage/providers/togetherai.zh-CN.mdx
index 7e164959df7d6..149aed87546a2 100644
--- a/docs/usage/providers/togetherai.zh-CN.mdx
+++ b/docs/usage/providers/togetherai.zh-CN.mdx
@@ -10,61 +10,39 @@ tags:
# 在 LobeChat 中使用 Together AI
-
+
-[together.ai](https://www.together.ai/) 是一家专注于生成式人工智能(AIGC)领域的平台,成立于2022年6月。 它致力于构建用于运行、训练和微调开源模型的云平台,以低于主流供应商的价格提供可扩展的计算能力。
+[together.ai](https://www.together.ai/) 是一家专注于生成式人工智能 (AIGC) 领域的平台，成立于 2022 年 6 月。 它致力于构建用于运行、训练和微调开源模型的云平台，以低于主流供应商的价格提供可扩展的计算能力。
本文档将指导你如何在 LobeChat 中使用 Together AI:
+ ### 步骤一：获取 Together AI 的 API 密钥
-### 步骤一：获取 Together AI 的 API 密钥
+ - 访问并登录 [Together AI API](https://api.together.ai/)
+ - 初次登录时系统会自动为你创建好 API 密钥并赠送 $5.0 的额度
-- 访问并登录 [Together AI API](https://api.together.ai/)
-- 初次登录时系统会自动为你创建好 API 密钥并赠送 $5.0 的额度
+
-
+ - 如果你没有保存，也可以在后续任意时间，通过 `设置` 中的 `API 密钥` 界面查看
-- 如果你没有保存，也可以在后续任意时间，通过 `设置` 中的 `API 密钥` 界面查看
+
-
+ ### 步骤二：在 LobeChat 中配置 Together AI
-### 步骤二：在 LobeChat 中配置 Together AI
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`together.ai`的设置项
-- 访问LobeChat的`设置`界面
-- 在`语言模型`下找到`together.ai`的设置项
+
-
+ - 打开 together.ai 并填入获得的 API 密钥
+ - 为你的助手选择一个 Together AI 的模型即可开始对话
-- 打开 together.ai 并填入获得的 API 密钥
-- 为你的助手选择一个 Together AI 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Together AI 的费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Together AI 的费用政策。
+
至此你已经可以在 LobeChat 中使用 Together AI 提供的模型进行对话了。
diff --git a/docs/usage/providers/upstage.mdx b/docs/usage/providers/upstage.mdx
index 7d36f33ff3699..cc22cc689d93c 100644
--- a/docs/usage/providers/upstage.mdx
+++ b/docs/usage/providers/upstage.mdx
@@ -10,54 +10,38 @@ tags:
# Using Upstage in LobeChat
-
+
[Upstage](https://www.upstage.ai/) is a platform that offers AI models and services, focusing on applications in natural language processing and machine learning. It allows developers to access its powerful AI capabilities through APIs, supporting various tasks such as text generation and conversational systems.
This article will guide you on how to use Upstage in LobeChat.
-### Step 1: Obtain an Upstage API Key
+ ### Step 1: Obtain an Upstage API Key
-- Register and log in to the [Upstage Console](https://console.upstage.ai/home)
-- Navigate to the `API Keys` page
-- Create a new API key
-- Copy and save the generated API key
+ - Register and log in to the [Upstage Console](https://console.upstage.ai/home)
+ - Navigate to the `API Keys` page
+ - Create a new API key
+ - Copy and save the generated API key
-
+
-### Step 2: Configure Upstage in LobeChat
+ ### Step 2: Configure Upstage in LobeChat
-- Access the `Settings` interface in LobeChat
-- Locate the `Upstage` settings under `Language Models`
+ - Access the `Settings` interface in LobeChat
+ - Locate the `Upstage` settings under `Language Models`
-
+
-- Enter the obtained API key
-- Select an Upstage model for your AI assistant to start the conversation
+ - Enter the obtained API key
+ - Select an Upstage model for your AI assistant to start the conversation
-
-
-
- Please note that you may need to pay the API service provider for usage. Refer to Upstage's
- pricing policy for more information.
-
+
+
+ Please note that you may need to pay the API service provider for usage. Refer to Upstage's
+ pricing policy for more information.
+
You can now use the models provided by Upstage for conversations in LobeChat.
diff --git a/docs/usage/providers/upstage.zh-CN.mdx b/docs/usage/providers/upstage.zh-CN.mdx
index aa9386c0050b1..7bc2f7cf2ac0f 100644
--- a/docs/usage/providers/upstage.zh-CN.mdx
+++ b/docs/usage/providers/upstage.zh-CN.mdx
@@ -10,53 +10,37 @@ tags:
# 在 LobeChat 中使用 Upstage
-
+
[Upstage](https://www.upstage.ai/) 是一个提供 AI 模型和服务的平台，专注于自然语言处理和机器学习应用。它允许开发者通过 API 接入其强大的 AI 功能，支持多种任务，如文本生成、对话系统等。
本文将指导你如何在 LobeChat 中使用 Upstage。
-### 步骤一：获得 Upstage 的 API Key
+ ### 步骤一：获得 Upstage 的 API Key
-- 注册并登录 [Upstage 控制台](https://console.upstage.ai/home)
-- 进入 `API Keys` 页面
-- 创建一个新的 API 密钥
-- 复制并保存生成的 API 密钥
+ - 注册并登录 [Upstage 控制台](https://console.upstage.ai/home)
+ - 进入 `API Keys` 页面
+ - 创建一个新的 API 密钥
+ - 复制并保存生成的 API 密钥
-
+
-### 步骤二：在 LobeChat 中配置 Upstage
+ ### 步骤二：在 LobeChat 中配置 Upstage
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `Upstage` 的设置项
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `Upstage` 的设置项
-
+
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 Upstage 的模型即可开始对话
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 Upstage 的模型即可开始对话
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 Upstage 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 Upstage 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 Upstage 提供的模型进行对话了。
diff --git a/docs/usage/providers/wenxin.mdx b/docs/usage/providers/wenxin.mdx
index 1682525a06068..42ca5df79e036 100644
--- a/docs/usage/providers/wenxin.mdx
+++ b/docs/usage/providers/wenxin.mdx
@@ -13,63 +13,47 @@ tags:
# Using Wenxin Qianfan in LobeChat
-
+
-[Wenxin Qianfan](https://qianfan.cloud.baidu.com/) is an artificial intelligence large language model platform launched by Baidu, supporting a variety of application scenarios, including literary creation, commercial copywriting, and mathematical logic reasoning. The platform features deep semantic understanding and generation capabilities across modalities and languages, and it is widely utilized in fields such as search Q&A, content creation, and smart office applications.
+[Wenxin Qianfan](https://qianfan.cloud.baidu.com/) is an artificial intelligence large language model platform launched by Baidu, supporting a variety of application scenarios, including literary creation, commercial copywriting, and mathematical logic reasoning. The platform features deep semantic understanding and generation capabilities across modalities and languages, and it is widely utilized in fields such as search Q\&A, content creation, and smart office applications.
This article will guide you on how to use Wenxin Qianfan in LobeChat.
-### Step 1: Obtain the Wenxin Qianfan API Key
+ ### Step 1: Obtain the Wenxin Qianfan API Key
-- Register and log in to the [Baidu Intelligent Cloud Console](https://console.bce.baidu.com/)
-- Navigate to `Baidu Intelligent Cloud Qianfan ModelBuilder`
-- Choose `Application Access` from the left-side menu
-- Create an application
+ - Register and log in to the [Baidu AI Cloud Console](https://console.bce.baidu.com/)
+ - Navigate to `Baidu AI Cloud Qianfan ModelBuilder`
+ - Select `API Key` from the left menu
-
+
-- Enter the `Security Authentication` -> `Access Key` management page from the user account menu
-- Copy the `Access Key` and `Secret Key`, and store them safely
+ - Click `Create API Key`
+ - In `Service`, select `Qianfan ModelBuilder`
+ - In `Resource`, choose `All Resources`
+ - Click the `Confirm` button
+ - Copy the `API Key` and keep it safe
-
+
-### Step 2: Configure Wenxin Qianfan in LobeChat
+
-- Go to the `Settings` interface in LobeChat
-- Locate the settings for `Wenxin Qianfan` under `Language Model`
+ ### Step 2: Configure Wenxin Qianfan in LobeChat
-
+ - Go to the `Settings` page of LobeChat
+ - Under `Language Models`, find the `Wenxin Qianfan` settings
-- Enter the obtained `Access Key` and `Secret Key`
-- Select a Wenxin Qianfan model for your AI assistant to start interacting
+
-
+ - Enter the obtained `API Key`
+ - Select a Wenxin Qianfan model for your AI assistant, and you're ready to start chatting!
-
- During usage, you may need to pay the API service provider. Please refer to Wenxin Qianfan's
- relevant fee policy.
-
+
+
+ During usage, you may need to pay the API service provider. Please refer to Wenxin Qianfan's
+ relevant fee policy.
+
You can now use the models provided by Wenxin Qianfan for conversations in LobeChat.
diff --git a/docs/usage/providers/wenxin.zh-CN.mdx b/docs/usage/providers/wenxin.zh-CN.mdx
index 386c3068ae086..318e61d0fba34 100644
--- a/docs/usage/providers/wenxin.zh-CN.mdx
+++ b/docs/usage/providers/wenxin.zh-CN.mdx
@@ -11,62 +11,46 @@ tags:
# 在 LobeChat 中使用文心千帆
-
+
[文心千帆](https://qianfan.cloud.baidu.com/)是百度推出的一个人工智能大语言模型平台，支持多种应用场景，包括文学创作、商业文案生成、数理逻辑推算等。该平台具备跨模态、跨语言的深度语义理解与生成能力，广泛应用于搜索问答、内容创作和智能办公等领域。
本文将指导你如何在 LobeChat 中使用文心千帆。
-### 步骤一：获得文心千帆的 API Key
+ ### 步骤一：获得文心千帆的 API Key
-- 注册并登录 [百度智能云控制台](https://console.bce.baidu.com/)
-- 进入 `百度智能云千帆 ModelBuilder`
-- 在左侧菜单中选择`应用接入`
-- 创建一个应用
+ - 注册并登录 [百度智能云控制台](https://console.bce.baidu.com/)
+ - 进入 `百度智能云千帆 ModelBuilder`
+ - 在左侧菜单中选择 `API Key`
-
+
-- 在用户账号菜单进入 `安全认证` -> `Access Key` 管理页面
-- 复制 `Access Key` 和 `Secret Key`，并妥善保存
+ - 点击创建 API Key
+ - 在 `服务` 中选择 `千帆ModelBuilder`
+ - 在 `资源` 中选择 `所有资源`
+ - 点击 `确定` 按钮
+ - 复制 `API Key` 并妥善保存
-
+
-### 步骤二：在 LobeChat 中配置文心千帆
+
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `文心千帆` 的设置项
+ ### 步骤二：在 LobeChat 中配置文心千帆
-
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `文心千帆` 的设置项
-- 填入获得的 `Access Key` 和 `Secret Key`
-- 为你的 AI 助手选择一个文心千帆的模型即可开始对话
+
-
+ - 填入获得的 `API Key`
+ - 为你的 AI 助手选择一个文心千帆的模型即可开始对话
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考文心千帆的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考文心千帆的相关费用政策。
+
至此你已经可以在 LobeChat 中使用文心千帆提供的模型进行对话了。
diff --git a/docs/usage/providers/xai.mdx b/docs/usage/providers/xai.mdx
index 9630e33d4415d..f04d3710499d9 100644
--- a/docs/usage/providers/xai.mdx
+++ b/docs/usage/providers/xai.mdx
@@ -12,59 +12,42 @@ tags:
# Using xAI in LobeChat
-
+
[xAI](https://x.ai/) is an artificial intelligence company founded by Elon Musk in 2023, aimed at exploring and understanding the true nature of the universe. The company's mission is to solve complex scientific and mathematical problems using AI technology and to advance the field of artificial intelligence.
This article will guide you on how to use xAI in LobeChat.
+ ### Step 1: Obtain an API Key from xAI
-### Step 1: Obtain an API Key from xAI
+ - Register and login to the [xAI console](https://console.x.ai/)
+ - Create an API token
+ - Copy and save the API token
-- Register and login to the [xAI console](https://console.x.ai/)
-- Create an API token
-- Copy and save the API token
+
-
+
+ Make sure to securely save the API token displayed in the popup; it only appears once. If you
+ accidentally lose it, you will need to create a new API token.
+
-
- Make sure to securely save the API token displayed in the popup; it only appears once. If you
- accidentally lose it, you will need to create a new API token.
-
+ ### Step 2: Configure xAI in LobeChat
-### Step 2: Configure xAI in LobeChat
+ - Go to the `Settings` menu in LobeChat
+ - Locate the `xAI` settings under `Language Model`
-- Go to the `Settings` menu in LobeChat
-- Locate the `xAI` settings under `Language Model`
+
-
+ - Enter the API key you obtained
+ - Select an xAI model for your AI assistant to start a conversation
-- Enter the API key you obtained
-- Select an xAI model for your AI assistant to start a conversation
-
-
-
-
- During use, you may need to pay the API service provider, so please refer to xAI's relevant
- pricing policies.
-
+
+
+ During use, you may need to pay the API service provider, so please refer to xAI's relevant
+ pricing policies.
+
You are now ready to engage in conversations using the models provided by xAI in LobeChat.
diff --git a/docs/usage/providers/xai.zh-CN.mdx b/docs/usage/providers/xai.zh-CN.mdx
index 3470984a9bad1..ed65cfc639aa0 100644
--- a/docs/usage/providers/xai.zh-CN.mdx
+++ b/docs/usage/providers/xai.zh-CN.mdx
@@ -10,57 +10,40 @@ tags:
# 在 LobeChat 中使用 xAI
-
+
-[xAI](https://x.ai/) 是由埃隆·马斯克于2023年成立的一家人工智能公司，旨在探索和理解宇宙的真实本质。该公司的目标是通过人工智能技术解决复杂的科学和数学问题，并推动人工智能的发展。
+[xAI](https://x.ai/) 是由埃隆・马斯克于 2023 年成立的一家人工智能公司，旨在探索和理解宇宙的真实本质。该公司的目标是通过人工智能技术解决复杂的科学和数学问题，并推动人工智能的发展。
本文将指导你如何在 LobeChat 中使用 xAI。
+ ### 步骤一：获取 xAI 的 API 密钥
-### 步骤一: 获取 xAI 的 API 密钥
+ - 注册并登录 [xAI 控制台](https://console.x.ai/)
+ - 创建一个 API Token
+ - 复制并保存 API Token
-- 注册并登录 [xAI 控制台](https://console.x.ai/)
-- 创建一个 API Token
-- 复制并保存 API Token
+
-
+
+ 妥善保存弹窗中的 API 令牌，它只会出现一次，如果不小心丢失了，你需要重新创建一个 API 令牌。
+
-
- 妥善保存弹窗中的 API 令牌，它只会出现一次，如果不小心丢失了，你需要重新创建一个 API 令牌。
-
+ ### 步骤二：在 LobeChat 中配置 xAI
-### 步骤二：在 LobeChat 中配置 xAI
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到 `xAI` 的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到 `xAI` 的设置项
+
-
+ - 填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 xAI 的模型即可开始对话
-- 填入获得的 API 密钥
-- 为你的 AI 助手选择一个 xAI 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考 xAI 的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考 xAI 的相关费用政策。
+
至此你已经可以在 LobeChat 中使用 xAI 提供的模型进行对话了。
diff --git a/docs/usage/providers/zeroone.mdx b/docs/usage/providers/zeroone.mdx
index e7e4711c780e4..87baaf816396e 100644
--- a/docs/usage/providers/zeroone.mdx
+++ b/docs/usage/providers/zeroone.mdx
@@ -13,72 +13,46 @@ tags:
# Using 01 AI in LobeChat
-
+
[01 AI](https://www.01.ai/) is a global company dedicated to AI 2.0 large model technology and applications. Its billion-parameter Yi-Large closed-source model, when evaluated on Stanford University's English ranking AlpacaEval 2.0, is on par with GPT-4.
This document will guide you on how to use 01 AI in LobeChat:
+ ### Step 1: Obtain 01 AI API Key
-### Step 1: Obtain 01 AI API Key
+ - Register and log in to the [01 AI Large Model Open Platform](https://platform.lingyiwanwu.com/)
+ - Go to the `Dashboard` and access the `API Key Management` menu
+ - A system-generated API key has been created for you automatically, or you can create a new one on this interface
-- Register and log in to the [01 AI Large Model Open Platform](https://platform.lingyiwanwu.com/)
-- Go to the `Dashboard` and access the `API Key Management` menu
-- A system-generated API key has been created for you automatically, or you can create a new one on this interface
+
-
+ - Account verification is required for first-time use
-- Account verification is required for first-time use
+
-
+ - Click on the created API key
+ - Copy and save the API key in the pop-up dialog box
-- Click on the created API key
-- Copy and save the API key in the pop-up dialog box
+
-
+ ### Step 2: Configure 01 AI in LobeChat
-### Step 2: Configure 01 AI in LobeChat
+ - Access the `Settings` interface in LobeChat
+ - Find the setting for `01 AI` under `Language Model`
-- Access the `Settings` interface in LobeChat
-- Find the setting for `01 AI` under `Language Model`
+
-
+ - Open 01 AI and enter the obtained API key
+ - Choose a 01.AI model for your AI assistant to start the conversation
-- Open 01 AI and enter the obtained API key
-- Choose a 01.AI model for your AI assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider. Please refer to 01 AI's relevant fee
- policies.
-
+
+
+ During usage, you may need to pay the API service provider. Please refer to 01 AI's relevant fee
+ policies.
+
You can now use the models provided by 01 AI for conversations in LobeChat.
diff --git a/docs/usage/providers/zeroone.zh-CN.mdx b/docs/usage/providers/zeroone.zh-CN.mdx
index 0a31699a404bd..346153974d477 100644
--- a/docs/usage/providers/zeroone.zh-CN.mdx
+++ b/docs/usage/providers/zeroone.zh-CN.mdx
@@ -15,71 +15,45 @@ tags:
# 在 LobeChat 中使用零一万物
-
+
-[零一万物](https://www.01.ai/)是一家致力于AI 2.0大模型技术和应用的全球公司，其发布的千亿参数的Yi-Large闭源模型，在斯坦福大学的英语排行AlpacaEval 2.0上，与GPT-4互有第一。
+[零一万物](https://www.01.ai/)是一家致力于 AI 2.0 大模型技术和应用的全球公司，其发布的千亿参数的 Yi-Large 闭源模型，在斯坦福大学的英语排行 AlpacaEval 2.0 上，与 GPT-4 互有第一。
本文档将指导你如何在 LobeChat 中使用零一万物:
+ ### 步骤一：获取零一万物 API 密钥
-### 步骤一：获取零一万物 API 密钥
+ - 注册并登录 [零一万物大模型开放平台](https://platform.lingyiwanwu.com/)
+ - 进入`工作台`并访问`API Key管理`菜单
+ - 系统已为你自动创建了一个 API 密钥，你也可以在此界面创建新的 API 密钥
-- 注册并登录 [零一万物大模型开放平台](https://platform.lingyiwanwu.com/)
-- 进入`工作台`并访问`API Key管理`菜单
-- 系统已为你自动创建了一个 API 密钥，你也可以在此界面创建新的 API 密钥
+
-
+ - 初次使用时需要完成账号认证
-- 初次使用时需要完成账号认证
+
-
+ - 点击创建好的 API 密钥
+ - 在弹出的对话框中复制并保存 API 密钥
-- 点击创建好的 API 密钥
-- 在弹出的对话框中复制并保存 API 密钥
+
-
+ ### 步骤二：在 LobeChat 中配置零一万物
-### 步骤二：在LobeChat 中配置零一万物
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`零一万物`的设置项
-- 访问 LobeChat 的`设置`界面
-- 在`语言模型`下找到`零一万物`的设置项
+
-
+ - 打开零一万物并填入获得的 API 密钥
+ - 为你的 AI 助手选择一个 01.AI 的模型即可开始对话
-- 打开零一万物并填入获得的 API 密钥
-- 为你的 AI 助手选择一个 01.AI 的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考零一万物的相关费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考零一万物的相关费用政策。
+
至此你已经可以在 LobeChat 中使用零一万物提供的模型进行对话了。
diff --git a/docs/usage/providers/zhipu.mdx b/docs/usage/providers/zhipu.mdx
index 048cc6cf9dc83..c2b2de317c13c 100644
--- a/docs/usage/providers/zhipu.mdx
+++ b/docs/usage/providers/zhipu.mdx
@@ -13,55 +13,37 @@ tags:
# Using Zhipu ChatGLM in LobeChat
-
+
[Zhipu AI](https://www.zhipuai.cn/) is a high-tech company originating from the Department of Computer Science at Tsinghua University. Established in 2019, the company focuses on natural language processing, machine learning, and big data analysis, dedicated to expanding the boundaries of artificial intelligence technology in the field of cognitive intelligence.
This document will guide you on how to use Zhipu AI in LobeChat:
+ ### Step 1: Obtain the API Key for Zhipu AI
-### Step 1: Obtain the API Key for Zhipu AI
+ - Visit and log in to the [Zhipu AI Open Platform](https://open.bigmodel.cn/)
+ - Upon initial login, the system will automatically create an API key for you and gift you a resource package of 25M Tokens
+ - Navigate to the `API Key` section at the top to view your API key
-- Visit and log in to the [Zhipu AI Open Platform](https://open.bigmodel.cn/)
-- Upon initial login, the system will automatically create an API key for you and gift you a resource package of 25M Tokens
-- Navigate to the `API Key` section at the top to view your API key
+
-
+ ### Step 2: Configure Zhipu AI in LobeChat
-### Step 2: Configure Zhipu AI in LobeChat
+ - Visit the `Settings` interface in LobeChat
+ - Under `Language Model`, locate the settings for Zhipu AI
-- Visit the `Settings` interface in LobeChat
-- Under `Language Model`, locate the settings for Zhipu AI
+
-
+ - Open Zhipu AI and enter the obtained API key
+ - Choose a Zhipu AI model for your assistant to start the conversation
-- Open Zhipu AI and enter the obtained API key
-- Choose a Zhipu AI model for your assistant to start the conversation
-
-
-
-
- During usage, you may need to pay the API service provider, please refer to Zhipu AI's pricing
- policy.
-
+
+
+ During usage, you may need to pay the API service provider, please refer to Zhipu AI's pricing
+ policy.
+
You can now engage in conversations using the models provided by Zhipu AI in LobeChat.
diff --git a/docs/usage/providers/zhipu.zh-CN.mdx b/docs/usage/providers/zhipu.zh-CN.mdx
index 811b4263aa414..12fb80199b198 100644
--- a/docs/usage/providers/zhipu.zh-CN.mdx
+++ b/docs/usage/providers/zhipu.zh-CN.mdx
@@ -10,52 +10,36 @@ tags:
# 在 LobeChat 中使用智谱 ChatGLM
-
+
-[智谱AI](https://www.zhipuai.cn/) 是一家源自清华大学计算机系技术成果的高科技公司,成立于2019年,专注于自然语言处理、机器学习和大数据分析，致力于在认知智能领域拓展人工智能技术的边界。
+[智谱 AI](https://www.zhipuai.cn/) 是一家源自清华大学计算机系技术成果的高科技公司，成立于 2019 年，专注于自然语言处理、机器学习和大数据分析，致力于在认知智能领域拓展人工智能技术的边界。
本文档将指导你如何在 LobeChat 中使用智谱 AI:
+ ### 步骤一：获取智谱 AI 的 API 密钥
-### 步骤一：获取智谱 AI 的 API 密钥
+ - 访问并登录 [智谱 AI 开放平台](https://open.bigmodel.cn/)
+ - 初次登录时系统会自动为你创建好 API 密钥并赠送 25M Tokens 的资源包
+ - 进入顶部的 `API密钥` 可以查看你的 API
-- 访问并登录 [智谱AI开放平台](https://open.bigmodel.cn/)
-- 初次登录时系统会自动为你创建好 API 密钥并赠送 25M Tokens 的资源包
-- 进入顶部的 `API密钥` 可以查看你的 API
+
-
+ ### 步骤二：在 LobeChat 中配置智谱 AI
-### 步骤二：在 LobeChat 中配置智谱AI
+ - 访问 LobeChat 的`设置`界面
+ - 在`语言模型`下找到`智谱AI`的设置项
-- 访问LobeChat的`设置`界面
-- 在`语言模型`下找到`智谱AI`的设置项
+
-
+ - 打开智谱 AI 并填入获得的 API 密钥
+ - 为你的助手选择一个智谱 AI 的模型即可开始对话
-- 打开智谱 AI 并填入获得的 API 密钥
-- 为你的助手选择一个智谱AI的模型即可开始对话
-
-
-
-
- 在使用过程中你可能需要向 API 服务提供商付费，请参考智谱AI的费用政策。
-
+
+
+ 在使用过程中你可能需要向 API 服务提供商付费，请参考智谱 AI 的费用政策。
+
-至此你已经可以在 LobeChat 中使用智谱AI提供的模型进行对话了。
+至此你已经可以在 LobeChat 中使用智谱 AI 提供的模型进行对话了。
diff --git a/docs/usage/start.mdx b/docs/usage/start.mdx
index 856a6a312e9b1..d4c87cbb1f67c 100644
--- a/docs/usage/start.mdx
+++ b/docs/usage/start.mdx
@@ -18,8 +18,8 @@ tags:
-
+
## Experience Features
-
+
diff --git a/docs/usage/start.zh-CN.mdx b/docs/usage/start.zh-CN.mdx
index 421df1a0f9706..2f9335112bd8f 100644
--- a/docs/usage/start.zh-CN.mdx
+++ b/docs/usage/start.zh-CN.mdx
@@ -14,26 +14,9 @@ tags:
# ✨ LobeChat 功能特性一览
-
+
-
+
## 体验特性
diff --git a/docs/usage/tools-calling.mdx b/docs/usage/tools-calling.mdx
index 490bb8b360bdc..acc025dc99067 100644
--- a/docs/usage/tools-calling.mdx
+++ b/docs/usage/tools-calling.mdx
@@ -1,3 +1,12 @@
+---
+title: Tools Calling
+description: Discover the best tools to enhance your calling experience and productivity.
+tags:
+ - Calling Tools
+ - Productivity
+ - Communication
+---
+
# Tools Calling
TODO
diff --git a/docs/usage/tools-calling.zh-CN.mdx b/docs/usage/tools-calling.zh-CN.mdx
index 8b37f6c1c2847..92ef954d45efc 100644
--- a/docs/usage/tools-calling.zh-CN.mdx
+++ b/docs/usage/tools-calling.zh-CN.mdx
@@ -27,76 +27,71 @@ LobeChat 的插件实现基于模型的 Tools Calling 能力，模型本身的 T
我们基于实际真实的用户场景出发构建了两大组测试任务，第一组为简单的调用指令（天气查询），第二组为复杂调用指令（文生图）。这两组指令的系统描述如下：
-
-
-```md
-## Tools
-
-You can use these tools below:
-
-### Realtime Weather
-
-Get realtime weather information
-
-The APIs you can use:
+
+ ```md
+ ## Tools
-#### `realtime-weather____fetchCurrentWeather`
+ You can use these tools below:
-获取当前天气情况
-```
+ ### Realtime Weather
-
+ Get realtime weather information
-
+ The APIs you can use:
-```md
-## Tools
+ #### `realtime-weather____fetchCurrentWeather`
-You can use these tools below:
+ 获取当前天气情况
+ ```
+
-### DALL·E 3
+
+ ```md
+ ## Tools
-Whenever a description of an image is given, use lobe-image-designer to create the images and then summarize the prompts used to generate the images in plain text. If the user does not ask for a specific number of images, default to creating four captions to send to lobe-image-designer that are written to be as diverse as possible.
+ You can use these tools below:
-All captions sent to lobe-image-designer must abide by the following policies:
+ ### DALL·E 3
-1. If the description is not in English, then translate it.
-2. Do not create more than 4 images, even if the user requests more.
-3. Don't create images of politicians or other public figures. Recommend other ideas instead.
-4. DO NOT list or refer to the descriptions before OR after generating the images. They should ONLY ever be written out ONCE, in the `prompts` field of the request. You do not need to ask for permission to generate, just do it!
-5. Always mention the image type (photo, oil painting, watercolor painting, illustration, cartoon, drawing, vector, render, etc.) at the beginning of the caption. Unless the caption suggests otherwise, make at least 1--2 of the 4 images photos.
-6. Diversify depictions of ALL images with people to include DESCENT and GENDER for EACH person using direct terms. Adjust only human descriptions.
+ Whenever a description of an image is given, use lobe-image-designer to create the images and then summarize the prompts used to generate the images in plain text. If the user does not ask for a specific number of images, default to creating four captions to send to lobe-image-designer that are written to be as diverse as possible.
-- EXPLICITLY specify these attributes, not abstractly reference them. The attributes should be specified in a minimal way and should directly describe their physical form.
-- Your choices should be grounded in reality. For example, all of a given OCCUPATION should not be the same gender or race. Additionally, focus on creating diverse, inclusive, and exploratory scenes via the properties you choose during rewrites. Make choices that may be insightful or unique sometimes.
-- Use "various" or "diverse" ONLY IF the description refers to groups of more than 3 people. Do not change the number of people requested in the original description.
-- Don't alter memes, fictional character origins, or unseen people. Maintain the original prompt's intent and prioritize quality.
-- Do not create any imagery that would be offensive.
+ All captions sent to lobe-image-designer must abide by the following policies:
-8. Silently modify descriptions that include names or hints or references of specific people or celebrities by carefully selecting a few minimal modifications to substitute references to the people with generic descriptions that don't divulge any information about their identities, except for their genders and physiques. Do this EVEN WHEN the instructions ask for the prompt to not be changed. Some special cases:
+ 1. If the description is not in English, then translate it.
+ 2. Do not create more than 4 images, even if the user requests more.
+ 3. Don't create images of politicians or other public figures. Recommend other ideas instead.
+ 4. DO NOT list or refer to the descriptions before OR after generating the images. They should ONLY ever be written out ONCE, in the `prompts` field of the request. You do not need to ask for permission to generate, just do it!
+ 5. Always mention the image type (photo, oil painting, watercolor painting, illustration, cartoon, drawing, vector, render, etc.) at the beginning of the caption. Unless the caption suggests otherwise, make at least 1--2 of the 4 images photos.
+ 6. Diversify depictions of ALL images with people to include DESCENT and GENDER for EACH person using direct terms. Adjust only human descriptions.
-- Modify such prompts even if you don't know who the person is, or if their name is misspelled (e.g. "Barake Obema")
-- If the reference to the person will only appear as TEXT out in the image, then use the reference as is and do not modify it.
-- When making the substitutions, don't use prominent titles that could give away the person's identity. E.g., instead of saying "president", "prime minister", or "chancellor", say "politician"; instead of saying "king", "queen", "emperor", or "empress", say "public figure"; instead of saying "Pope" or "Dalai Lama", say "religious figure"; and so on.
-- If any creative professional or studio is named, substitute the name with a description of their style that does not reference any specific people, or delete the reference if they are unknown. DO NOT refer to the artist or studio's style.
+ - EXPLICITLY specify these attributes, not abstractly reference them. The attributes should be specified in a minimal way and should directly describe their physical form.
+ - Your choices should be grounded in reality. For example, all of a given OCCUPATION should not be the same gender or race. Additionally, focus on creating diverse, inclusive, and exploratory scenes via the properties you choose during rewrites. Make choices that may be insightful or unique sometimes.
+ - Use "various" or "diverse" ONLY IF the description refers to groups of more than 3 people. Do not change the number of people requested in the original description.
+ - Don't alter memes, fictional character origins, or unseen people. Maintain the original prompt's intent and prioritize quality.
+ - Do not create any imagery that would be offensive.
-The prompt must intricately describe every part of the image in concrete, objective detail. THINK about what the end goal of the description is, and extrapolate that to what would make satisfying images. All descriptions sent to lobe-image-designer should be a paragraph of text that is extremely descriptive and detailed. Each should be more than 3 sentences long.
+ 8. Silently modify descriptions that include names or hints or references of specific people or celebrities by carefully selecting a few minimal modifications to substitute references to the people with generic descriptions that don't divulge any information about their identities, except for their genders and physiques. Do this EVEN WHEN the instructions ask for the prompt to not be changed. Some special cases:
-The APIs you can use:
+ - Modify such prompts even if you don't know who the person is, or if their name is misspelled (e.g. "Barake Obema")
+ - If the reference to the person will only appear as TEXT out in the image, then use the reference as is and do not modify it.
+ - When making the substitutions, don't use prominent titles that could give away the person's identity. E.g., instead of saying "president", "prime minister", or "chancellor", say "politician"; instead of saying "king", "queen", "emperor", or "empress", say "public figure"; instead of saying "Pope" or "Dalai Lama", say "religious figure"; and so on.
+ - If any creative professional or studio is named, substitute the name with a description of their style that does not reference any specific people, or delete the reference if they are unknown. DO NOT refer to the artist or studio's style.
-#### `lobe-image-designer____text2image____builtin`
+ The prompt must intricately describe every part of the image in concrete, objective detail. THINK about what the end goal of the description is, and extrapolate that to what would make satisfying images. All descriptions sent to lobe-image-designer should be a paragraph of text that is extremely descriptive and detailed. Each should be more than 3 sentences long.
-Create images from a text-only prompt.
-```
+ The APIs you can use:
-
+ #### `lobe-image-designer____text2image____builtin`
+ Create images from a text-only prompt.
+ ```
+
-如上所示，简单调用指令在插件调用时它的系统描述(system role)相对简单，复杂调用指令的系统描述会复杂很多。这两组不同复杂度的指令可以比较好地区分出模型对于系统指令的遵循能力：
+如上所示，简单调用指令在插件调用时它的系统描述 (system role) 相对简单，复杂调用指令的系统描述会复杂很多。这两组不同复杂度的指令可以比较好地区分出模型对于系统指令的遵循能力：
- **天气查询可以测试模型的基础 Tools Calling 能力，确认模型是否存在「虚假宣传」的情况。** 就我们实际的测试来看，的确存在一些模型号称具有 Tools Calling 能力，但是处于完全不可用的状态；
-- **文生图可以测试模型指令跟随能力的上限。** 例如基础模型（例如 GPT-3.5）可能只能生成 1 张图片的 prompt，而高级模型（例如 GPT-4o）则能够生成 1~4 张图片的 prompt。
+- **文生图可以测试模型指令跟随能力的上限。** 例如基础模型（例如 GPT-3.5）可能只能生成 1 张图片的 prompt，而高级模型（例如 GPT-4o）则能够生成 1\~4 张图片的 prompt。
### 简单调用指令：天气查询
@@ -126,11 +121,11 @@ Create images from a text-only prompt.
针对这一个工具，我们构建的测试组中包含了三个指令：
-| 指令编号 | 指令内容 | 基础 Tools Calling 调用 | 并发调用 | 复合指令跟随 |
-| --- | --- | --- | --- | --- |
-| 指令 ① | 告诉我杭州和北京的天气，先回答我好的 | 🟢 | 🟢 | 🟢 |
-| 指令 ② | 告诉我杭州和北京的天气 | 🟢 | 🟢 | - |
-| 指令 ③ | 告诉我杭州的天气 | 🟢 | - | - |
+| 指令编号 | 指令内容 | 基础 Tools Calling 调用 | 并发调用 | 复合指令跟随 |
+| ---- | ------------------ | ------------------- | ---- | ------ |
+| 指令 ① | 告诉我杭州和北京的天气，先回答我好的 | 🟢 | 🟢 | 🟢 |
+| 指令 ② | 告诉我杭州和北京的天气 | 🟢 | 🟢 | - |
+| 指令 ③ | 告诉我杭州的天气 | 🟢 | - | - |
上述三个指令的复杂度逐渐递减，我们可以通过这三个指令来测试模型对于简单指令的处理能力。
@@ -147,30 +142,26 @@ Create images from a text-only prompt.
-
根据我们实际的日常使用，工具调用往往会和普通文本生成结合在一起回答。例如比较经典的 Code Interpreter 插件，ChatGPT 往往会先回复一些代码生成的思路，然后再调用 Code Interpreter 插件生成代码。
这种情况下，我们需要模型能够正确地识别出用户的意图，然后调用对应的工具。
因此， 指令 ① 中的「告诉我杭州和北京的天气，先回答我好的」就是一个复合指令跟随的例子。前半句期望模型调用天气查询工具，后半句期望模型回答「好的」。并且理想的顺序应该是先回答「好的」，然后再调用天气查询工具。
+
-
-
并发工具调用（Parallel function calling）是指模型能够同时调用多个工具，或同时调用一个工具多次，这在对话中可以大大降低用户等待的时间，提升用户体验。
- 并发工具调用能力由 OpenAI 于 2023年11月率先提出，目前支持并发工具调用的模型并不算多，属于是 Tools Calling 的进阶能力。
+ 并发工具调用能力由 OpenAI 于 2023 年 11 月率先提出，目前支持并发工具调用的模型并不算多，属于是 Tools Calling 的进阶能力。
指令 ② 中的「告诉我杭州和北京的天气」就是一个期望执行并发调用的例子。理想的情况下，单个模型的返回应该存在两个工具的调用返回。
+
-
-
基础工具调用不必再赘述，这是 Tools Calling 的基础能力。
指令 ③ 中的「告诉我杭州的天气」就是最基本的工具调用的例子。
-
-
+
### 复杂调用指令：文生图
@@ -229,10 +220,10 @@ Create images from a text-only prompt.
针对这一个工具，我们构建的测试组中包含了两个指令：
-| 指令编号 | 指令内容 | 流式调用 | 复杂 Tools Calling 调用 | 并发调用 | 复合指令跟随 |
-| --- | --- | --- | --- | --- | --- |
-| 指令 ① | 我要画 3 幅画，第一幅画的主体为一只达芬奇风格的小狗，第二幅是毕加索风格的大雁，最后一幅是莫奈风格的狮子。每一幅都需要产出 2 个 prompts。请先说明你的构思，然后开始生成相应的图片。 | 🟢 | 🟢 | 🟢 | 🟢 |
-| 指令 ② | 画一只小狗 | 🟢 | 🟢 | - | - |
+| 指令编号 | 指令内容 | 流式调用 | 复杂 Tools Calling 调用 | 并发调用 | 复合指令跟随 |
+| ---- | ------------------------------------------------------------------------------------------------ | ---- | ------------------- | ---- | ------ |
+| 指令 ① | 我要画 3 幅画，第一幅画的主体为一只达芬奇风格的小狗，第二幅是毕加索风格的大雁，最后一幅是莫奈风格的狮子。每一幅都需要产出 2 个 prompts。请先说明你的构思，然后开始生成相应的图片。 | 🟢 | 🟢 | 🟢 | 🟢 |
+| 指令 ② | 画一只小狗 | 🟢 | 🟢 | - | - |
此外，由于文生图的 prompts 的生成时间较长，这一组指令也可以清晰地测试出模型的 API 是否支持流式 Tools Calling。
@@ -242,12 +233,13 @@ Create images from a text-only prompt.
+
+
-
+
+
+
diff --git a/docs/usage/tools-calling/anthropic.mdx b/docs/usage/tools-calling/anthropic.mdx
index a6ed47134b36a..37876a688467f 100644
--- a/docs/usage/tools-calling/anthropic.mdx
+++ b/docs/usage/tools-calling/anthropic.mdx
@@ -15,12 +15,12 @@ tags:
Overview of Anthropic Claude Series model Tools Calling capabilities:
-| Model | Support Tools Calling | Stream | Parallel | Simple Instruction Score | Complex Instruction |
-| --- | --- | --- | --- | --- | --- |
-| Claude 3.5 Sonnet | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟🌟 |
-| Claude 3 Opus | ✅ | ✅ | ❌ | 🌟 | ⛔️ |
-| Claude 3 Sonnet | ✅ | ✅ | ❌ | 🌟🌟 | ⛔️ |
-| Claude 3 Haiku | ✅ | ✅ | ❌ | 🌟🌟 | ⛔️ |
+| Model | Support Tools Calling | Stream | Parallel | Simple Instruction Score | Complex Instruction |
+| ----------------- | --------------------- | ------ | -------- | ------------------------ | ------------------- |
+| Claude 3.5 Sonnet | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟🌟 |
+| Claude 3 Opus | ✅ | ✅ | ❌ | 🌟 | ⛔️ |
+| Claude 3 Sonnet | ✅ | ✅ | ❌ | 🌟🌟 | ⛔️ |
+| Claude 3 Haiku | ✅ | ✅ | ❌ | 🌟🌟 | ⛔️ |
## Claude 3.5 Sonnet
@@ -30,18 +30,13 @@ Test Instruction: Instruction ①
-
+Tools Calling Raw Output:
-```yml
-
-```
-
+ ```yml
+ ```
### Complex Instruction Call: Literary Map
@@ -55,18 +50,13 @@ From the above video:
1. Sonnet 3.5 supports Stream Tools Calling and Parallel Tools Calling;
2. In Stream Tools Calling, it is observed that creating long sentences will cause a delay (as seen in the Tools Calling raw output `[chunk 40]` and `[chunk 41]` with a delay of 6s). Therefore, there will be a relatively long waiting time at the beginning stage of Tools Calling.
-
+Tools Calling Raw Output:
-```yml
-
-```
-
+ ```yml
+ ```
## Claude 3 Opus
@@ -83,14 +73,10 @@ From the above video:
2. Opus triggers Tools Calling twice, indicating that it does not support Parallel Tools Calling;
3. The raw output of Tools Calling shows that Opus also supports Stream Tools Calling.
-
+Tools Calling Raw Output:
-
### Complex Instruction Call: Literary Map
@@ -104,14 +90,10 @@ From the above video:
1. Combining with simple tasks, Opus will always output a `` tag, which significantly impacts the user experience;
2. Opus outputs the prompts field as a string instead of an array, causing an error and preventing the plugin from being called correctly.
-
+Tools Calling Raw Output:
-
## Claude 3 Sonnet
@@ -124,14 +106,10 @@ Test Instruction: Instruction ①
From the above video, it can be seen that Claude 3 Sonnet triggers Tools Calling twice, indicating that it does not support Parallel Tools Calling.
-
+Tools Calling Raw Output:
-
### Complex Instruction Call: Literary Map
@@ -142,14 +120,10 @@ Test Instruction: Instruction ②
From the above video, it can be seen that Sonnet 3 fails in the complex instruction call. The error is due to prompts being expected as an array but generated as a string.
-
+Tools Calling Raw Output:
-
## Claude 3 Haiku
@@ -161,10 +135,7 @@ From the above video:
1. Claude 3 Haiku triggers Tools Calling twice, indicating that it also does not support Parallel Tools Calling;
2. Haiku does not provide a good response and directly calls the tool;
-
+
### Complex Instruction Call: Literary Map
@@ -174,12 +145,8 @@ Test Instruction: Instruction ②
From the above video, it can be seen that Haiku 3 also fails in the complex instruction call. The error is the same as prompts generating a string instead of an array.
-
+Tools Calling Raw Output:
-
diff --git a/docs/usage/tools-calling/anthropic.zh-CN.mdx b/docs/usage/tools-calling/anthropic.zh-CN.mdx
index 95db4ff8214b9..c052feaaa69ec 100644
--- a/docs/usage/tools-calling/anthropic.zh-CN.mdx
+++ b/docs/usage/tools-calling/anthropic.zh-CN.mdx
@@ -15,12 +15,12 @@ tags:
Anthropic Claude 系列模型 Tools Calling 能力一览：
-| 模型 | 支持 Tools Calling | 流式 （Stream） | 并发（Parallel） | 简单指令得分 | 复杂指令 |
-| --- | --- | --- | --- | --- | --- |
-| Claude 3.5 Sonnet | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟🌟 |
-| Claude 3 Opus | ✅ | ✅ | ❌ | 🌟 | ⛔️ |
-| Claude 3 Sonnet | ✅ | ✅ | ❌ | 🌟🌟 | ⛔️ |
-| Claude 3 Haiku | ✅ | ✅ | ❌ | 🌟🌟 | ⛔️ |
+| 模型 | 支持 Tools Calling | 流式 （Stream） | 并发（Parallel） | 简单指令得分 | 复杂指令 |
+| ----------------- | ---------------- | ----------- | ------------ | ------ | ---- |
+| Claude 3.5 Sonnet | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟🌟 |
+| Claude 3 Opus | ✅ | ✅ | ❌ | 🌟 | ⛔️ |
+| Claude 3 Sonnet | ✅ | ✅ | ❌ | 🌟🌟 | ⛔️ |
+| Claude 3 Haiku | ✅ | ✅ | ❌ | 🌟🌟 | ⛔️ |
## Claude 3.5 Sonnet
@@ -30,18 +30,13 @@ Anthropic Claude 系列模型 Tools Calling 能力一览：
-
+Tools Calling 原始输出：
-```yml
-
-```
-
+ ```yml
+ ```
### 复杂调用指令：文生图
@@ -52,21 +47,16 @@ Anthropic Claude 系列模型 Tools Calling 能力一览：
从上述视频中可以看到：
-1. Sonnet 3.5 支持流式 Tools Calling 和 Parallel Tools Calling；
-2. 在流式 Tools Calling 时，表现出来的特征是在创建长句会等待住（详见 Tools Calling 原始输出 `[chunk 40]` 和 `[chunk 41]` 中间的耗时达到 6s）。所以相对来说会在 Tools Calling 的起始阶段有一个较长的等待时间。
+1. Sonnet 3.5 支持流式 Tools Calling 和 Parallel Tools Calling；
+2. 在流式 Tools Calling 时，表现出来的特征是在创建长句会等待住（详见 Tools Calling 原始输出 `[chunk 40]` 和 `[chunk 41]` 中间的耗时达到 6s）。所以相对来说会在 Tools Calling 的起始阶段有一个较长的等待时间。
-
+Tools Calling 原始输出：
-```yml
-
-```
-
+ ```yml
+ ```
## Claude 3 Opus
@@ -83,14 +73,10 @@ Anthropic Claude 系列模型 Tools Calling 能力一览：
2. Opus 会触发两次 Tools Calling，说明它并不支持 Parallel Tools Calling；
3. 从 Tools Calling 的原始输出来看， Opus 也是支持流式 Tools Calling 的
-
-
-
- Tools Calling 原始输出：
+
+
+ Tools Calling 原始输出：
### 复杂调用指令：文生图
@@ -104,14 +90,10 @@ Anthropic Claude 系列模型 Tools Calling 能力一览：
1. 结合简单任务， Opus 的工具调用一定会输出 `` 标签，这其实对体验影响非常大
2. Opus 输出的 prompts 字段是字符串，而不是数组，导致报错，无法正常调用插件。
-
+Tools Calling 原始输出：
-
## Claude 3 Sonnet
@@ -124,14 +106,10 @@ Anthropic Claude 系列模型 Tools Calling 能力一览：
从上述视频中可以看出，Claude 3 Sonnet 会调用两次 Tools Calling，说明它并不支持 Parallel Tools Calling。
-
+Tools Calling 原始输出：
-
### 复杂调用指令：文生图
@@ -142,14 +120,10 @@ Anthropic Claude 系列模型 Tools Calling 能力一览：
从上述视频中可以看到， Sonnet 3 在复杂指令调用下就失败了。报错原因是 prompts 原本预期为一个数组，但是生成的却是一个字符串。
-
+Tools Calling 原始输出：
-
## Claude 3 Haiku
@@ -161,10 +135,7 @@ Anthropic Claude 系列模型 Tools Calling 能力一览：
1. Claude 3 Haiku 会调用两次 Tools Calling，说明它也不支持 Parallel Tools Calling；
2. Haiku 并没有回答好的，也是直接调用的工具；
-
+
### 复杂调用指令：文生图
@@ -174,12 +145,8 @@ Anthropic Claude 系列模型 Tools Calling 能力一览：
从上述视频中可以看到， Haiku 3 在复杂指令调用下也是失败的。报错原因同样是 prompts 生成了字符串而不是数组。
-
+Tools Calling 原始输出：
-
diff --git a/docs/usage/tools-calling/google.mdx b/docs/usage/tools-calling/google.mdx
index 2df73266a826b..2699e7a2130e6 100644
--- a/docs/usage/tools-calling/google.mdx
+++ b/docs/usage/tools-calling/google.mdx
@@ -15,10 +15,10 @@ tags:
Overview of Google Gemini series model Tools Calling capabilities:
-| Model | Tools Calling Support | Streaming | Parallel | Simple Instruction Score | Complex Instruction |
-| --- | --- | --- | --- | --- | --- |
-| Gemini 1.5 Pro | ✅ | ❌ | ✅ | ⛔ | ⛔ |
-| Gemini 1.5 Flash | ❌ | ❌ | ❌ | ⛔ | ⛔ |
+| Model | Tools Calling Support | Streaming | Parallel | Simple Instruction Score | Complex Instruction |
+| ---------------- | --------------------- | --------- | -------- | ------------------------ | ------------------- |
+| Gemini 1.5 Pro | ✅ | ❌ | ✅ | ⛔ | ⛔ |
+| Gemini 1.5 Flash | ❌ | ❌ | ❌ | ⛔ | ⛔ |
Based on our actual tests, we strongly recommend not enabling plugins for Gemini because as of
@@ -35,38 +35,31 @@ Test Instruction: Instruction ①
In the json output from Gemini, the name is incorrect, so LobeChat cannot recognize which plugin it called. (In the input, the name of the weather plugin is `realtime-weather____fetchCurrentWeather`, while Gemini returns `weather____fetchCurrentWeather`).
-
+Original Tools Calling Output:
-```yml
-[stream start] 2024-7-7 17:53:25.647
-[chunk 0] 2024-7-7 17:53:25.654
-{"candidates":[{"content":{"parts":[{"text":"好的"}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":1,"totalTokenCount":96}}
-
-[chunk 1] 2024-7-7 17:53:26.288
-{"candidates":[{"content":{"parts":[{"text":"\n\n"}],"role":"model"},"finishReason":"STOP","index":0,"safetyRatings":[{"category":"HARM_CATEGORY_SEXUALLY_EXPLICIT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HATE_SPEECH","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HARASSMENT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_DANGEROUS_CONTENT","probability":"NEGLIGIBLE"}]}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":1,"totalTokenCount":96}}
+ ```yml
+ [stream start] 2024-7-7 17:53:25.647
+ [chunk 0] 2024-7-7 17:53:25.654
+ {"candidates":[{"content":{"parts":[{"text":"好的"}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":1,"totalTokenCount":96}}
-[chunk 2] 2024-7-7 17:53:26.336
-{"candidates":[{"content":{"parts":[{"functionCall":{"name":"weather____fetchCurrentWeather","args":{"city":"Hangzhou"}}},{"functionCall":{"name":"weather____fetchCurrentWeather","args":{"city":"Beijing"}}}],"role":"model"},"finishReasoSTOP","index":0,"safetyRatings":[{"category":"HARM_CATEGORY_SEXUALLY_EXPLICIT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HATE_SPEECH","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HARASSMENT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_DANGEROUS_CONTENT","probability":"NEGLIGIBLE"}]}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":79,"totalTokenCount":174}}
+ [chunk 1] 2024-7-7 17:53:26.288
+ {"candidates":[{"content":{"parts":[{"text":"\n\n"}],"role":"model"},"finishReason":"STOP","index":0,"safetyRatings":[{"category":"HARM_CATEGORY_SEXUALLY_EXPLICIT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HATE_SPEECH","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HARASSMENT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_DANGEROUS_CONTENT","probability":"NEGLIGIBLE"}]}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":1,"totalTokenCount":96}}
-[stream finished] total chunks: 3
-```
+ [chunk 2] 2024-7-7 17:53:26.336
+ {"candidates":[{"content":{"parts":[{"functionCall":{"name":"weather____fetchCurrentWeather","args":{"city":"Hangzhou"}}},{"functionCall":{"name":"weather____fetchCurrentWeather","args":{"city":"Beijing"}}}],"role":"model"},"finishReasoSTOP","index":0,"safetyRatings":[{"category":"HARM_CATEGORY_SEXUALLY_EXPLICIT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HATE_SPEECH","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HARASSMENT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_DANGEROUS_CONTENT","probability":"NEGLIGIBLE"}]}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":79,"totalTokenCount":174}}
+ [stream finished] total chunks: 3
+ ```
### Complex Instruction Call: Image Generation
Test Instruction: Instruction ②
-
+
When testing a set of complex instructions, Google throws an error directly:
diff --git a/docs/usage/tools-calling/google.zh-CN.mdx b/docs/usage/tools-calling/google.zh-CN.mdx
index fc3c78df22eab..35c54c5d9805a 100644
--- a/docs/usage/tools-calling/google.zh-CN.mdx
+++ b/docs/usage/tools-calling/google.zh-CN.mdx
@@ -15,13 +15,13 @@ tags:
Google Gemini 系列模型 Tools Calling 能力一览：
-| 模型 | 支持 Tools Calling | 流式 （Stream） | 并发（Parallel） | 简单指令得分 | 复杂指令 |
-| --- | --- | --- | --- | --- | --- |
-| Gemini 1.5 Pro | ✅ | ❌ | ✅ | ⛔ | ⛔ |
-| Gemini 1.5 Flash | ❌ | ❌ | ❌ | ⛔ | ⛔ |
+| 模型 | 支持 Tools Calling | 流式 （Stream） | 并发（Parallel） | 简单指令得分 | 复杂指令 |
+| ---------------- | ---------------- | ----------- | ------------ | ------ | ---- |
+| Gemini 1.5 Pro | ✅ | ❌ | ✅ | ⛔ | ⛔ |
+| Gemini 1.5 Flash | ❌ | ❌ | ❌ | ⛔ | ⛔ |
- 根据我们的的实际测试，强烈建议不要给 Gemini 开启插件，因为目前（截止2024.07.07）它的 Tools Calling
+ 根据我们的的实际测试，强烈建议不要给 Gemini 开启插件，因为目前（截止 2024.07.07）它的 Tools Calling
能力实在太烂了。
@@ -35,38 +35,31 @@ Google Gemini 系列模型 Tools Calling 能力一览：
Gemini 输出的 json 中，name 是错误的，因此 LobeChat 无法识别到它调用了什么插件。（入参中，天气插件的 name 为 `realtime-weather____fetchCurrentWeather`，而 Gemini 返回的是 `weather____fetchCurrentWeather`）。
-
+Tools Calling 原始输出：
-```yml
-[stream start] 2024-7-7 17:53:25.647
-[chunk 0] 2024-7-7 17:53:25.654
-{"candidates":[{"content":{"parts":[{"text":"好的"}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":1,"totalTokenCount":96}}
-
-[chunk 1] 2024-7-7 17:53:26.288
-{"candidates":[{"content":{"parts":[{"text":"\n\n"}],"role":"model"},"finishReason":"STOP","index":0,"safetyRatings":[{"category":"HARM_CATEGORY_SEXUALLY_EXPLICIT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HATE_SPEECH","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HARASSMENT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_DANGEROUS_CONTENT","probability":"NEGLIGIBLE"}]}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":1,"totalTokenCount":96}}
+ ```yml
+ [stream start] 2024-7-7 17:53:25.647
+ [chunk 0] 2024-7-7 17:53:25.654
+ {"candidates":[{"content":{"parts":[{"text":"好的"}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":1,"totalTokenCount":96}}
-[chunk 2] 2024-7-7 17:53:26.336
-{"candidates":[{"content":{"parts":[{"functionCall":{"name":"weather____fetchCurrentWeather","args":{"city":"杭州"}}},{"functionCall":{"name":"weather____fetchCurrentWeather","args":{"city":"北京"}}}],"role":"model"},"finishReasoSTOP","index":0,"safetyRatings":[{"category":"HARM_CATEGORY_SEXUALLY_EXPLICIT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HATE_SPEECH","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HARASSMENT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_DANGEROUS_CONTENT","probability":"NEGLIGIBLE"}]}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":79,"totalTokenCount":174}}
+ [chunk 1] 2024-7-7 17:53:26.288
+ {"candidates":[{"content":{"parts":[{"text":"\n\n"}],"role":"model"},"finishReason":"STOP","index":0,"safetyRatings":[{"category":"HARM_CATEGORY_SEXUALLY_EXPLICIT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HATE_SPEECH","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HARASSMENT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_DANGEROUS_CONTENT","probability":"NEGLIGIBLE"}]}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":1,"totalTokenCount":96}}
-[stream finished] total chunks: 3
-```
+ [chunk 2] 2024-7-7 17:53:26.336
+ {"candidates":[{"content":{"parts":[{"functionCall":{"name":"weather____fetchCurrentWeather","args":{"city":"杭州"}}},{"functionCall":{"name":"weather____fetchCurrentWeather","args":{"city":"北京"}}}],"role":"model"},"finishReasoSTOP","index":0,"safetyRatings":[{"category":"HARM_CATEGORY_SEXUALLY_EXPLICIT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HATE_SPEECH","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_HARASSMENT","probability":"NEGLIGIBLE"},{"category":"HARM_CATEGORY_DANGEROUS_CONTENT","probability":"NEGLIGIBLE"}]}],"usageMetadata":{"promptTokenCount":95,"candidatesTokenCount":79,"totalTokenCount":174}}
+ [stream finished] total chunks: 3
+ ```
### 复杂调用指令：文生图
测试指令：指令 ②
-
+
在测试复杂指令集时，Google 直接抛错：
diff --git a/docs/usage/tools-calling/groq.mdx b/docs/usage/tools-calling/groq.mdx
index 1333ed77b7e1e..99a1d0b166d2f 100644
--- a/docs/usage/tools-calling/groq.mdx
+++ b/docs/usage/tools-calling/groq.mdx
@@ -1 +1,10 @@
+---
+title: ''
+description: 学习如何有效管理待办事项，提高工作效率和组织能力。
+tags:
+ - 待办事项
+ - 工作效率
+ - 时间管理
+---
+
TODO
diff --git a/docs/usage/tools-calling/groq.zh-CN.mdx b/docs/usage/tools-calling/groq.zh-CN.mdx
index baabe3b422312..6fb22aab486ae 100644
--- a/docs/usage/tools-calling/groq.zh-CN.mdx
+++ b/docs/usage/tools-calling/groq.zh-CN.mdx
@@ -17,11 +17,11 @@ tags:
Groq 平台的模型 Tools Calling 能力一览：
-| 模型 | 支持 Tools Calling | 流式 （Stream） | 并发（Parallel） | 简单指令得分 | 复杂指令 |
-| ------------ | ------------------ | --------------- | ---------------- | ------------ | -------- |
-| LLAMA3 70B | ✅ | ❌ | ✅ | 🌟🌟 | 🌟🌟 |
-| LLAMA3 8B | ✅ | ❌ | ✅ | 🌟🌟 | 🌟 |
-| Mixtral-8x7B | ✅ | ❌ | ✅ | ⛔ | 🌟🌟 |
+| 模型 | 支持 Tools Calling | 流式 （Stream） | 并发（Parallel） | 简单指令得分 | 复杂指令 |
+| ------------ | ---------------- | ----------- | ------------ | ------ | ---- |
+| LLAMA3 70B | ✅ | ❌ | ✅ | 🌟🌟 | 🌟🌟 |
+| LLAMA3 8B | ✅ | ❌ | ✅ | 🌟🌟 | 🌟 |
+| Mixtral-8x7B | ✅ | ❌ | ✅ | ⛔ | 🌟🌟 |
## LLAMA3 70B
@@ -33,21 +33,17 @@ Groq 平台的模型 Tools Calling 能力一览：
从上述视频中可以看到 LLAMA3 70B 支持并发 Tools Calling，可以同时调用多次天气查询。
-
+Tools Calling 原始输出：
-```yml
-[no stream response] 2024-7-8 15:50:40.166
+ ```yml
+ [no stream response] 2024-7-8 15:50:40.166
-{"id":"chatcmpl-ec4b6c0b-1078-4f50-a39c-e58b3b1f9c31","object":"chat.completion","created":1720425030,"model":"llama3-70b-8192","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_v89g","type":"function","function":{"name":"realtime-weather____fetchCurrentWeather","arguments":"{\"city\":\"杭州\"}"}},{"id":"call_jxwk","type":"function","function":{"name":"realtime-weather____fetchCurrentWeather","arguments":"{\"city\":\"北京}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":969,"prompt_time":0.224209489,"completion_tokens":68,"completion_time":0.194285714,"total_tokens":1037,"total_time":0.418495203},"system_fingerprint":"fp_87cbfbbc4d","x_groq":{"id":"req_01j28n57x9e78a6bfbn9sdn139"}}
-
-```
+ {"id":"chatcmpl-ec4b6c0b-1078-4f50-a39c-e58b3b1f9c31","object":"chat.completion","created":1720425030,"model":"llama3-70b-8192","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_v89g","type":"function","function":{"name":"realtime-weather____fetchCurrentWeather","arguments":"{\"city\":\"杭州\"}"}},{"id":"call_jxwk","type":"function","function":{"name":"realtime-weather____fetchCurrentWeather","arguments":"{\"city\":\"北京}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":969,"prompt_time":0.224209489,"completion_tokens":68,"completion_time":0.194285714,"total_tokens":1037,"total_time":0.418495203},"system_fingerprint":"fp_87cbfbbc4d","x_groq":{"id":"req_01j28n57x9e78a6bfbn9sdn139"}}
+ ```
### 复杂调用指令：文生图
@@ -56,22 +52,18 @@ Groq 平台的模型 Tools Calling 能力一览：
-
+
-Tools Calling 原始输出：
-
-```yml
-[no stream response] 2024-7-8 18:0:34.811
+ Tools Calling 原始输出：
-{"id":"chatcmpl-e3b59ca9-1172-4ae2-96c7-3d6997a1f8a8","object":"chat.completion","created":1720432834,"model":"llama3-70b-8192","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_azm9","type":"function","function":{"name":"lobe-image-designer____text2image____builtin","arguments":"{\"prompts\":[\"A small, fluffy, and playful golden retriever puppy with a white patch on its forehead, sitting on a green grass field with a bright blue sky in the background, photo.\",\"A cute, little, brown and white Dalmatian puppy with a red collar, running around in a park with a sunny day, illustration.\",\"A tiny, grey and white Poodle puppy with a pink ribbon, sitting on a white couch with a few toys surrounding it, watercolor painting.\",\"A sweet, small, black and white Chihuahua puppy with a pink bow, lying on a soft, white blanket with a few stuffed animals nearby, oil painting.\"],\"quality\":\"standard\",\"seeds\":[],\"size\":\"1024x1024\",\"style\":\"vivid\"}"}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":2305,"prompt_time":3.027052298,"completion_tokens":246,"completion_time":0.702857143,"total_tokens":2551,"total_time":3.729909441},"system_fingerprint":"fp_7ab5f7e105","x_groq":{"id":"req_01j28wk2q0efvs22qatw7rd0ds"}}
+ ```yml
+ [no stream response] 2024-7-8 18:0:34.811
-POST /api/chat/groq 200 in 17462ms
-```
+ {"id":"chatcmpl-e3b59ca9-1172-4ae2-96c7-3d6997a1f8a8","object":"chat.completion","created":1720432834,"model":"llama3-70b-8192","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_azm9","type":"function","function":{"name":"lobe-image-designer____text2image____builtin","arguments":"{\"prompts\":[\"A small, fluffy, and playful golden retriever puppy with a white patch on its forehead, sitting on a green grass field with a bright blue sky in the background, photo.\",\"A cute, little, brown and white Dalmatian puppy with a red collar, running around in a park with a sunny day, illustration.\",\"A tiny, grey and white Poodle puppy with a pink ribbon, sitting on a white couch with a few toys surrounding it, watercolor painting.\",\"A sweet, small, black and white Chihuahua puppy with a pink bow, lying on a soft, white blanket with a few stuffed animals nearby, oil painting.\"],\"quality\":\"standard\",\"seeds\":[],\"size\":\"1024x1024\",\"style\":\"vivid\"}"}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":2305,"prompt_time":3.027052298,"completion_tokens":246,"completion_time":0.702857143,"total_tokens":2551,"total_time":3.729909441},"system_fingerprint":"fp_7ab5f7e105","x_groq":{"id":"req_01j28wk2q0efvs22qatw7rd0ds"}}
+ POST /api/chat/groq 200 in 17462ms
+ ```
## LLAMA3-8B
@@ -84,22 +76,18 @@ POST /api/chat/groq 200 in 17462ms
从上述视频中可以看到 LLAMA3-8B 对于天气插件可以正常调用，并获得正确的总结结果。但是它并没有完全 follow 我们的描述指令，没有回答「好的」。
-
+Tools Calling 原始输出：
-```yml
-[no stream response] 2024-7-9 11:33:16.920
+ ```yml
+ [no stream response] 2024-7-9 11:33:16.920
-{"id":"chatcmpl-f3672d59-e91d-4253-af1b-bfc4e0912085","object":"chat.completion","created":1720495996,"model":"llama3-8b-8192","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_rjtk","type":"function","function":{"name":"realtime-weather____fetchCurrentWeather","arguments":"{\"city\":\"杭州市\"}"}},{"id":"call_7pqh","type":"functi,"function":{"name":"realtime-weather____fetchCurrentWeather","arguments":"{\"city\":\"北京市\"}"}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":969,"ppt_time":0.145428625,"completion_tokens":128,"completion_time":0.101364747,"total_tokens":1097,"total_time":0.246793372},"system_fingerprint":"fp_33d61fdfc3","x_groq":{"id":"req_01j2artze1exz82nettf2h9066"}}
-
-POST /api/chat/groq 200 in 1649ms
-```
+ {"id":"chatcmpl-f3672d59-e91d-4253-af1b-bfc4e0912085","object":"chat.completion","created":1720495996,"model":"llama3-8b-8192","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_rjtk","type":"function","function":{"name":"realtime-weather____fetchCurrentWeather","arguments":"{\"city\":\"杭州市\"}"}},{"id":"call_7pqh","type":"functi,"function":{"name":"realtime-weather____fetchCurrentWeather","arguments":"{\"city\":\"北京市\"}"}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":969,"ppt_time":0.145428625,"completion_tokens":128,"completion_time":0.101364747,"total_tokens":1097,"total_time":0.246793372},"system_fingerprint":"fp_33d61fdfc3","x_groq":{"id":"req_01j2artze1exz82nettf2h9066"}}
+ POST /api/chat/groq 200 in 1649ms
+ ```
### 复杂调用指令：文生图
@@ -110,22 +98,18 @@ POST /api/chat/groq 200 in 1649ms
LLAMA3 8B 在 DallE 的输出场景下，只会输出 1 张图片，而不是像 LLAMA3 70B 一样输出 4 张，意味着在复杂 Tools 指令层面，能力和 GPT 3.5 Turbo 接近，不如 GPT 4。
-
+
-Tools Calling 原始输出：
-
-```yml
-[no stream response] 2024-7-9 11:58:27.40
+ Tools Calling 原始输出：
-{"id":"chatcmpl-3c38f4d2-3424-416c-9fb0-0969d2683959","object":"chat.completion","created":1720497506,"model":"llama3-8b-8192","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_k6xj","type":"function","function":{"name":"lobe-image-designer____text2image____builtin","arguments":"{\"prompts\":[\"Create a watercolor painting of a small white dog with a pink nose, wearing a red collar and sitting on a green grass. The dog's ears should be floppy and its fur should be curly.\"],\"quality\":\"standard\",\"seeds\":[],\"size\":\"1024x1024\",\"style\":\"natural\"}"}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":2282,"prompt_time":0.342335558,"completion_tokens":148,"completion_time":0.118023813,"total_tokens":2430,"total_time":0.460359371},"system_fingerprint":"fp_179b0f92c9","x_groq":{"id":"req_01j2at921tec8aymdq48czcw1y"}}
+ ```yml
+ [no stream response] 2024-7-9 11:58:27.40
-POST /api/chat/groq 200 in 2517ms
-```
+ {"id":"chatcmpl-3c38f4d2-3424-416c-9fb0-0969d2683959","object":"chat.completion","created":1720497506,"model":"llama3-8b-8192","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_k6xj","type":"function","function":{"name":"lobe-image-designer____text2image____builtin","arguments":"{\"prompts\":[\"Create a watercolor painting of a small white dog with a pink nose, wearing a red collar and sitting on a green grass. The dog's ears should be floppy and its fur should be curly.\"],\"quality\":\"standard\",\"seeds\":[],\"size\":\"1024x1024\",\"style\":\"natural\"}"}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":2282,"prompt_time":0.342335558,"completion_tokens":148,"completion_time":0.118023813,"total_tokens":2430,"total_time":0.460359371},"system_fingerprint":"fp_179b0f92c9","x_groq":{"id":"req_01j2at921tec8aymdq48czcw1y"}}
+ POST /api/chat/groq 200 in 2517ms
+ ```
## Mixtral-8x7B
@@ -138,23 +122,19 @@ POST /api/chat/groq 200 in 2517ms
从上述视频中可以看到 Mixtral-8x7B 对于天气插件的查询输出的参数有问题，导致无法正常调用插件。
-
+Tools Calling 原始输出：
-```yml
-
-[no stream response] 2024-7-8 22:18:19.682
+ ```yml
-{"id":"chatcmpl-9f89d669-5642-48be-b5cd-7a29756800c0","object":"chat.completion","created":1720448299,"model":"mixtral-8x7b-32768","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_719t","type":"function","function":{"name":"realtime-weather____fetchCurrentWeather","arguments":"{\"city\":\"Hangzhou,Beijing\"}"}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":1262,"prompt_time":0.116684046,"completion_tokens":102,"completion_time":0.163113006,"total_tokens":1364,"total_time":0.279797052},"system_fingerprint":"fp_c5f20b5bb1","x_groq":{"id":"req_01j29bbc8xen2s3thp9qen5bys"}}
+ [no stream response] 2024-7-8 22:18:19.682
-POST /api/chat/groq 200 in 4860ms
-```
+ {"id":"chatcmpl-9f89d669-5642-48be-b5cd-7a29756800c0","object":"chat.completion","created":1720448299,"model":"mixtral-8x7b-32768","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_719t","type":"function","function":{"name":"realtime-weather____fetchCurrentWeather","arguments":"{\"city\":\"Hangzhou,Beijing\"}"}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":1262,"prompt_time":0.116684046,"completion_tokens":102,"completion_time":0.163113006,"total_tokens":1364,"total_time":0.279797052},"system_fingerprint":"fp_c5f20b5bb1","x_groq":{"id":"req_01j29bbc8xen2s3thp9qen5bys"}}
+ POST /api/chat/groq 200 in 4860ms
+ ```
### 复杂调用指令：文生图
@@ -165,25 +145,19 @@ POST /api/chat/groq 200 in 4860ms
非常意外地，Mixtral 8x7B 在文生图的 Tools Calling 的场景下，居然可以正常出图，而且出图的能力基本和 LLAMA3 70B 持平。
-
-
+
+
+Tools Calling 原始输出：
-```yml
-[no stream response] 2024-7-9 12:6:38.11
-
-{"id":"chatcmpl-a90069f3-b119-41b9-b8d7-c7f97373529e","object":"chat.completion","created":1720497998,"model":"mixtral-8x7b-32768","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_hw8t","type":"function","function":{"name":"lobe-image-designer____text2image____builtin","arguments":"{\"prompts\":[\"A colorful and playful illustration of a friendly small dog, looking directly at the viewer with a wagging tail and bright, expressive eyes. The dog's fur is glossy and well-groomed, with a mix of black, white, and brown colors. The background is a lush, green park with a clear blue sky and a few fluffy clouds.\",\"An oil painting of a small, energetic dog in a sunlit park, chasing a frisbee. The dog is a mix of black and white fur, with a distinct brown spot above its left eye. The park features tall trees, providing dappled shade across the grassy landscape. The frisbee is bright orange and stands out against the natural surroundings.\",\"A realistic watercolor painting of a small, fluffy white dog curled up next to a warm fireplace during a cozy winter evening. The dog's eyes are closed in contentment, and a single red bow is tied around its neck. The background includes a plush armchair, a stack of books, and a softly lit room.\",\"A fun and engaging cartoon of a small dog sitting at a café table, enjoying a cup of coffee and a croissant. The dog has a expressive face and a blue scarf around its neck. The café has a vintage, 1920's style and a red awning, with a bustling city background.\"],\"quality\":\"standard\",\"size\":\"1024x1024\",\"style\":\"vivid\"}"}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":2920,"prompt_time":0.228639219,"completion_tokens":465,"completion_time":0.755757988,"total_tokens":3385,"total_time":0.984397207},"system_fingerprint":"fp_c5f20b5bb1","x_groq":{"id":"req_01j2atr155f0nv8rmfk448e2at"}}
+ ```yml
+ [no stream response] 2024-7-9 12:6:38.11
-POST /api/chat/groq 200 in 6216ms
+ {"id":"chatcmpl-a90069f3-b119-41b9-b8d7-c7f97373529e","object":"chat.completion","created":1720497998,"model":"mixtral-8x7b-32768","choices":[{"index":0,"message":{"role":"assistant","tool_calls":[{"id":"call_hw8t","type":"function","function":{"name":"lobe-image-designer____text2image____builtin","arguments":"{\"prompts\":[\"A colorful and playful illustration of a friendly small dog, looking directly at the viewer with a wagging tail and bright, expressive eyes. The dog's fur is glossy and well-groomed, with a mix of black, white, and brown colors. The background is a lush, green park with a clear blue sky and a few fluffy clouds.\",\"An oil painting of a small, energetic dog in a sunlit park, chasing a frisbee. The dog is a mix of black and white fur, with a distinct brown spot above its left eye. The park features tall trees, providing dappled shade across the grassy landscape. The frisbee is bright orange and stands out against the natural surroundings.\",\"A realistic watercolor painting of a small, fluffy white dog curled up next to a warm fireplace during a cozy winter evening. The dog's eyes are closed in contentment, and a single red bow is tied around its neck. The background includes a plush armchair, a stack of books, and a softly lit room.\",\"A fun and engaging cartoon of a small dog sitting at a café table, enjoying a cup of coffee and a croissant. The dog has a expressive face and a blue scarf around its neck. The café has a vintage, 1920's style and a red awning, with a bustling city background.\"],\"quality\":\"standard\",\"size\":\"1024x1024\",\"style\":\"vivid\"}"}}]},"logprobs":null,"finish_reason":"tool_calls"}],"usage":{"prompt_tokens":2920,"prompt_time":0.228639219,"completion_tokens":465,"completion_time":0.755757988,"total_tokens":3385,"total_time":0.984397207},"system_fingerprint":"fp_c5f20b5bb1","x_groq":{"id":"req_01j2atr155f0nv8rmfk448e2at"}}
-```
+ POST /api/chat/groq 200 in 6216ms
+ ```
diff --git a/docs/usage/tools-calling/moonshot.mdx b/docs/usage/tools-calling/moonshot.mdx
index 1333ed77b7e1e..422f796b92997 100644
--- a/docs/usage/tools-calling/moonshot.mdx
+++ b/docs/usage/tools-calling/moonshot.mdx
@@ -1 +1,10 @@
+---
+title: ''
+description: 学习如何有效管理待办事项，提高工作效率和生产力。
+tags:
+ - 待办事项
+ - 工作效率
+ - 生产力
+---
+
TODO
diff --git a/docs/usage/tools-calling/openai.mdx b/docs/usage/tools-calling/openai.mdx
index 98fde363d613c..01f3af90ef6ab 100644
--- a/docs/usage/tools-calling/openai.mdx
+++ b/docs/usage/tools-calling/openai.mdx
@@ -15,11 +15,11 @@ tags:
Overview of the Tool Calling capabilities of OpenAI GPT series models:
-| Model | Tool Calling Support | Streaming | Parallel | Simple Instruction Score | Complex Instruction Score |
-| --- | --- | --- | --- | --- | --- |
-| GPT-3.5-turbo | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟 |
-| GPT-4-turbo | ✅ | ✅ | ✅ | 🌟🌟 | 🌟🌟 |
-| GPT-4o | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟🌟 |
+| Model | Tool Calling Support | Streaming | Parallel | Simple Instruction Score | Complex Instruction Score |
+| ------------- | -------------------- | --------- | -------- | ------------------------ | ------------------------- |
+| GPT-3.5-turbo | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟 |
+| GPT-4-turbo | ✅ | ✅ | ✅ | 🌟🌟 | 🌟🌟 |
+| GPT-4o | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟🌟 |
For testing instructions, see [Tools Calling - Evaluation Task
@@ -34,14 +34,10 @@ Test Instruction: Instruction ①
-
+
-Streaming Tool Calling Raw Output:
-
+ Streaming Tool Calling Raw Output:
### Complex Instruction Call: Wenshengtu
@@ -50,14 +46,10 @@ Test Instruction: Instruction ②
-
+
-Streaming Tool Calling Raw Output:
-
+ Streaming Tool Calling Raw Output:
## GPT-4 Turbo
@@ -72,14 +64,10 @@ Of course, it is also possible that GPT-4 Turbo's model has more "autonomy" and
-
+
-Streaming Tool Calling Raw Output:
-
+ Streaming Tool Calling Raw Output:
### Complex Instruction Call: Wenshengtu
@@ -88,14 +76,10 @@ Test Instruction: Instruction ②
-
+
-Streaming Tool Calling Raw Output:
-
+ Streaming Tool Calling Raw Output:
## GPT-4o
@@ -108,14 +92,10 @@ Similar to GPT-3.5, GPT-4o performs very well in following compound instructions
-
+
- Streaming Tool Calling Raw Output:
-
+ Streaming Tool Calling Raw Output:
### Complex Instruction Call: Wenshengtu
@@ -124,16 +104,11 @@ Test Instruction: Instruction ②
-
+
- Streaming Tool Calling Raw Output:
-
-```yml
-
-```
+ Streaming Tool Calling Raw Output:
+ ```yml
+ ```
diff --git a/docs/usage/tools-calling/openai.zh-CN.mdx b/docs/usage/tools-calling/openai.zh-CN.mdx
index 8d401abaf2d8d..babd83068a825 100644
--- a/docs/usage/tools-calling/openai.zh-CN.mdx
+++ b/docs/usage/tools-calling/openai.zh-CN.mdx
@@ -15,11 +15,11 @@ tags:
OpenAI GPT 系列模型 Tool Calling 能力一览：
-| 模型 | 支持 Tool Calling | 流式 （Stream） | 并发（Parallel） | 简单指令得分 | 复杂指令 |
-| ------------- | ----------------- | --------------- | ---------------- | ------------ | -------- |
-| GPT-3.5-turbo | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟 |
-| GPT-4-turbo | ✅ | ✅ | ✅ | 🌟🌟 | 🌟🌟 |
-| GPT-4o | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟🌟 |
+| 模型 | 支持 Tool Calling | 流式 （Stream） | 并发（Parallel） | 简单指令得分 | 复杂指令 |
+| ------------- | --------------- | ----------- | ------------ | ------ | ---- |
+| GPT-3.5-turbo | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟 |
+| GPT-4-turbo | ✅ | ✅ | ✅ | 🌟🌟 | 🌟🌟 |
+| GPT-4o | ✅ | ✅ | ✅ | 🌟🌟🌟 | 🌟🌟 |
关于测试指令，详见 [工具调用 Tools Calling -
@@ -34,14 +34,10 @@ OpenAI GPT 系列模型 Tool Calling 能力一览：
-
+
-流式 Tool Calling 原始输出：
-
+ 流式 Tool Calling 原始输出：
### 复杂调用指令：文生图
@@ -50,14 +46,10 @@ OpenAI GPT 系列模型 Tool Calling 能力一览：
-
+
-流式 Tool Calling 原始输出：
-
+ 流式 Tool Calling 原始输出：
## GPT-4 Turbo
@@ -68,18 +60,14 @@ OpenAI GPT 系列模型 Tool Calling 能力一览：
GPT-4 Turbo 在调用 Tool Calling 时并没有像 GPT-3.5 Turbo 一样回复「好的」，且经过多次测试始终一样，因此在这一条复合指令的跟随中反而不如 GPT-3.5 Turbo，但剩余两项能力均不错。
-当然，也有可能是因为 GPT-4 Turbo 的模型更加有“自主意识”，认为不需要输出这一句“好的”。
+当然，也有可能是因为 GPT-4 Turbo 的模型更加有 “自主意识”，认为不需要输出这一句 “好的”。
-
+
-流式 Tool Calling 原始输出：
-
+ 流式 Tool Calling 原始输出：
### 复杂调用指令：文生图
@@ -88,14 +76,10 @@ GPT-4 Turbo 在调用 Tool Calling 时并没有像 GPT-3.5 Turbo 一样回复「
-
+
-流式 Tool Calling 原始输出：
-
+ 流式 Tool Calling 原始输出：
## GPT 4o
@@ -108,14 +92,10 @@ GPT-4o 和 3.5 一样，在简单调用指令中，能够达到非常不错的
-
+
- 流式 Tool Calling 原始输出：
-
+ 流式 Tool Calling 原始输出：
### 复杂调用指令：文生图
@@ -124,16 +104,11 @@ GPT-4o 和 3.5 一样，在简单调用指令中，能够达到非常不错的
-
+
- 流式 Tool Calling 原始输出：
-
-```yml
-
-```
+ 流式 Tool Calling 原始输出：
+ ```yml
+ ```
diff --git a/locales/ar/changelog.json b/locales/ar/changelog.json
index ff12b6c86033a..261158ccda261 100644
--- a/locales/ar/changelog.json
+++ b/locales/ar/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "عرض جميع سجلات التحديثات",
"description": "تابع الميزات الجديدة والتحسينات في {{appName}}",
"pagination": {
- "older": "عرض التغييرات السابقة",
- "prev": "الصفحة السابقة"
+ "next": "الصفحة التالية",
+ "older": "عرض التغييرات السابقة"
},
"readDetails": "اقرأ التفاصيل",
"title": "سجل التحديثات",
diff --git a/locales/ar/common.json b/locales/ar/common.json
index 8d3506f269fe7..43d0959d9f490 100644
--- a/locales/ar/common.json
+++ b/locales/ar/common.json
@@ -275,6 +275,7 @@
},
"temp": "مؤقت",
"terms": "شروط الخدمة",
+ "update": "تحديث",
"updateAgent": "تحديث معلومات الوكيل",
"upgradeVersion": {
"action": "ترقية",
@@ -285,6 +286,7 @@
"anonymousNickName": "مستخدم مجهول",
"billing": "إدارة الفواتير",
"cloud": "تجربة {{name}}",
+ "community": "نسخة المجتمع",
"data": "تخزين البيانات",
"defaultNickname": "مستخدم النسخة المجتمعية",
"discord": "الدعم المجتمعي",
@@ -294,7 +296,6 @@
"help": "مركز المساعدة",
"moveGuide": "تم نقل زر الإعدادات إلى هنا",
"plans": "خطط الاشتراك",
- "preview": "المعاينة",
"profile": "إدارة الحساب",
"setting": "إعدادات التطبيق",
"usages": "إحصاءات الاستخدام"
diff --git a/locales/ar/components.json b/locales/ar/components.json
index 5ca5b783a446a..4dc940bbc04d4 100644
--- a/locales/ar/components.json
+++ b/locales/ar/components.json
@@ -76,6 +76,7 @@
"custom": "نموذج مخصص، الإعداد الافتراضي يدعم الاستدعاء الوظيفي والتعرف البصري، يرجى التحقق من قدرة النموذج على القيام بذلك بناءً على الحالة الفعلية",
"file": "يدعم هذا النموذج قراءة وتعرف الملفات المرفوعة",
"functionCall": "يدعم هذا النموذج استدعاء الوظائف",
+ "reasoning": "يدعم هذا النموذج التفكير العميق",
"tokens": "يدعم هذا النموذج حتى {{tokens}} رمزًا في جلسة واحدة",
"vision": "يدعم هذا النموذج التعرف البصري"
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "لا توجد نماذج ممكن تمكينها، يرجى الانتقال إلى الإعدادات لتمكينها",
"provider": "مزود"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "بسبب قيود أمان المتصفح، تحتاج إلى تكوين CORS لـ Ollama لاستخدامه بشكل صحيح.",
+ "linux": {
+ "env": "أضف `Environment` تحت قسم [Service]، وأضف متغير البيئة OLLAMA_ORIGINS:",
+ "reboot": "أعد تحميل systemd وأعد تشغيل Ollama",
+ "systemd": "استخدم systemd لتحرير خدمة ollama:"
+ },
+ "macos": "يرجى فتح تطبيق «الطرفية» ولصق الأوامر التالية ثم الضغط على Enter للتنفيذ",
+ "reboot": "يرجى إعادة تشغيل خدمة Ollama بعد الانتهاء من التنفيذ",
+ "title": "تكوين Ollama للسماح بالوصول عبر النطاقات المتعددة",
+ "windows": "على نظام Windows، انقر على «لوحة التحكم»، ثم انتقل إلى تحرير متغيرات البيئة للنظام. أنشئ متغير بيئة جديد باسم «OLLAMA_ORIGINS» لقائمة المستخدم الخاصة بك، وقيمته هي *، ثم انقر على «موافق/تطبيق» لحفظ التغييرات."
+ },
+ "install": {
+ "description": "يرجى التأكد من أنك قد قمت بتشغيل Ollama، إذا لم تقم بتنزيل Ollama، يرجى زيارة الموقع الرسمي <1>للتنزيل",
+ "docker": "إذا كنت تفضل استخدام Docker، فإن Ollama يوفر أيضًا صورة Docker رسمية، يمكنك سحبها باستخدام الأمر التالي:",
+ "linux": {
+ "command": "قم بتثبيت باستخدام الأمر التالي:",
+ "manual": "أو يمكنك الرجوع إلى <1>دليل التثبيت اليدوي لنظام Linux للتثبيت بنفسك."
+ },
+ "title": "تثبيت وتشغيل تطبيق Ollama محليًا",
+ "windowsTab": "Windows (نسخة المعاينة)"
+ }
+ },
+ "Thinking": {
+ "thinking": "في حالة تفكير عميق...",
+ "thought": "لقد فكرت بعمق (استغرق الأمر {{duration}} ثانية)",
+ "thoughtWithDuration": "لقد فكرت بعمق"
}
}
diff --git a/locales/ar/discover.json b/locales/ar/discover.json
index a3ab309e5978a..c3d1dd550a028 100644
--- a/locales/ar/discover.json
+++ b/locales/ar/discover.json
@@ -126,6 +126,10 @@
"title": "جدة الموضوع"
},
"range": "نطاق",
+ "reasoning_effort": {
+ "desc": "تُستخدم هذه الإعدادات للتحكم في شدة التفكير التي يقوم بها النموذج قبل توليد الإجابات. الشدة المنخفضة تعطي الأولوية لسرعة الاستجابة وتوفر الرموز، بينما الشدة العالية توفر تفكيرًا أكثر اكتمالًا ولكنها تستهلك المزيد من الرموز وتقلل من سرعة الاستجابة. القيمة الافتراضية هي متوسطة، مما يوازن بين دقة التفكير وسرعة الاستجابة.",
+ "title": "شدة التفكير"
+ },
"temperature": {
"desc": "تؤثر هذه الإعدادات على تنوع استجابة النموذج. القيم المنخفضة تؤدي إلى استجابات أكثر توقعًا ونمطية، بينما القيم الأعلى تشجع على استجابات أكثر تنوعًا وغير شائعة. عندما تكون القيمة 0، يعطي النموذج نفس الاستجابة دائمًا لنفس المدخل.",
"title": "عشوائية"
diff --git a/locales/ar/modelProvider.json b/locales/ar/modelProvider.json
index 68c55561cdaaa..6459b3208efce 100644
--- a/locales/ar/modelProvider.json
+++ b/locales/ar/modelProvider.json
@@ -19,6 +19,24 @@
"title": "مفتاح API"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "إصدار واجهة برمجة التطبيقات Azure، يتبع تنسيق YYYY-MM-DD، راجع [الإصدار الأحدث](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "الحصول على القائمة",
+ "title": "إصدار واجهة برمجة التطبيقات Azure"
+ },
+ "endpoint": {
+ "desc": "ابحث عن نقطة نهاية استدلال نموذج Azure AI من نظرة عامة على مشروع Azure AI",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "نقطة نهاية Azure AI"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "ابحث عن مفتاح واجهة برمجة التطبيقات من نظرة عامة على مشروع Azure AI",
+ "placeholder": "مفتاح Azure",
+ "title": "المفتاح"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "أدخل AWS Access Key Id",
@@ -63,6 +81,46 @@
"title": "رقم حساب Cloudflare / عنوان URL API"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "يرجى إدخال مفتاح API الخاص بك",
+ "title": "مفتاح API"
+ },
+ "basicTitle": "المعلومات الأساسية",
+ "configTitle": "معلومات التكوين",
+ "confirm": "إنشاء جديد",
+ "createSuccess": "تم الإنشاء بنجاح",
+ "description": {
+ "placeholder": "نبذة عن مزود الخدمة (اختياري)",
+ "title": "نبذة عن مزود الخدمة"
+ },
+ "id": {
+ "desc": "معرف فريد لمزود الخدمة، لا يمكن تعديله بعد الإنشاء",
+ "format": "يمكن أن يحتوي فقط على أرقام، أحرف صغيرة، شرطات (-) وشرطات سفلية (_) ",
+ "placeholder": "يفضل أن يكون بالكامل بحروف صغيرة، مثل openai، لن يمكن تعديله بعد الإنشاء",
+ "required": "يرجى إدخال معرف المزود",
+ "title": "معرف المزود"
+ },
+ "logo": {
+ "required": "يرجى تحميل شعار المزود بشكل صحيح",
+ "title": "شعار المزود"
+ },
+ "name": {
+ "placeholder": "يرجى إدخال اسم العرض لمزود الخدمة",
+ "required": "يرجى إدخال اسم المزود",
+ "title": "اسم المزود"
+ },
+ "proxyUrl": {
+ "required": "يرجى إدخال عنوان الوكيل",
+ "title": "عنوان الوكيل"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "يرجى اختيار نوع SDK",
+ "title": "تنسيق الطلب"
+ },
+ "title": "إنشاء مزود AI مخصص"
+ },
"github": {
"personalAccessToken": {
"desc": "أدخل رمز الوصول الشخصي الخاص بك على Github، انقر [هنا](https://github.com/settings/tokens) لإنشاء واحد",
@@ -77,6 +135,23 @@
"title": "رمز HuggingFace"
}
},
+ "list": {
+ "title": {
+ "disabled": "مزود الخدمة غير مفعل",
+ "enabled": "مزود الخدمة مفعل"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "إضافة مزود خدمة مخصص",
+ "all": "الكل",
+ "list": {
+ "disabled": "غير مفعل",
+ "enabled": "مفعل"
+ },
+ "notFound": "لم يتم العثور على نتائج البحث",
+ "searchProviders": "البحث عن مزودين...",
+ "sort": "ترتيب مخصص"
+ },
"ollama": {
"checker": {
"desc": "اختبر ما إذا تم إدخال عنوان الوكيل بشكل صحيح",
@@ -94,33 +169,9 @@
"title": "جارٍ تنزيل النموذج {{model}} "
},
"endpoint": {
- "desc": "أدخل عنوان واجهة برمجة التطبيقات الخاص بـ Ollama، إذا لم يتم تحديده محليًا، يمكن تركه فارغًا",
+ "desc": "يجب أن تحتوي على http(s)://، يمكن تركها فارغة إذا لم يتم تحديدها محليًا",
"title": "عنوان وكيل الواجهة"
},
- "setup": {
- "cors": {
- "description": "بسبب قيود الأمان في المتصفح، يجب تكوين الوصول عبر المواقع المختلفة لـ Ollama لاستخدامه بشكل صحيح.",
- "linux": {
- "env": "في القسم [Service]، أضف `Environment` وأضف متغير البيئة OLLAMA_ORIGINS:",
- "reboot": "أعد تحميل systemd وأعد تشغيل Ollama",
- "systemd": "استدعاء تحرير خدمة ollama في systemd:"
- },
- "macos": "افتح تطبيق \"Terminal\" والصق الأمر التالي، ثم اضغط على Enter للتشغيل.",
- "reboot": "يرجى إعادة تشغيل خدمة Ollama بعد الانتهاء من التنفيذ",
- "title": "تكوين Ollama للسماح بالوصول عبر المواقع المختلفة",
- "windows": "على نظام Windows، انقر فوق \"لوحة التحكم\"، ثم ادخل إلى تحرير متغيرات البيئة النظامية. قم بإنشاء متغير بيئي بعنوان \"OLLAMA_ORIGINS\" لحساب المستخدم الخاص بك، واجعل قيمته * ثم انقر على \"موافق/تطبيق\" للحفظ."
- },
- "install": {
- "description": "يرجى التأكد من أنك قد قمت بتشغيل Ollama ، إذا لم تقم بتنزيل Ollama ، يرجى زيارة الموقع الرسمي <1>للتنزيل",
- "docker": "إذا كنت تفضل استخدام Docker، يوفر Ollama أيضًا صور Docker الرسمية، يمكنك سحبها باستخدام الأمر التالي:",
- "linux": {
- "command": "قم بتثبيته باستخدام الأمر التالي:",
- "manual": "أو يمكنك الرجوع إلى <1>دليل تثبيت Linux يدويًا للقيام بالتثبيت بنفسك."
- },
- "title": "تثبيت وتشغيل تطبيق Ollama محليًا",
- "windowsTab": "Windows (نسخة معاينة)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Cancel Download",
@@ -131,25 +182,145 @@
"title": "Download specified Ollama model"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "أدخل مفتاح الوصول من منصة بايدو تشيانفان",
- "placeholder": "مفتاح الوصول لتشيانفان",
- "title": "مفتاح الوصول"
+ "providerModels": {
+ "config": {
+ "aesGcm": "سيتم استخدام خوارزمية التشفير <1>AES-GCM لتشفير مفتاحك وعنوان الوكيل وما إلى ذلك",
+ "apiKey": {
+ "desc": "يرجى إدخال مفتاح API الخاص بك {{name}}",
+ "placeholder": "{{name}} مفتاح API",
+ "title": "مفتاح API"
+ },
+ "baseURL": {
+ "desc": "يجب أن يحتوي على http(s)://",
+ "invalid": "يرجى إدخال عنوان URL صالح",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "عنوان وكيل API"
+ },
+ "checker": {
+ "button": "تحقق",
+ "desc": "اختبار ما إذا كان مفتاح API وعنوان الوكيل قد تم إدخالهما بشكل صحيح",
+ "pass": "تم التحقق بنجاح",
+ "title": "اختبار الاتصال"
+ },
+ "fetchOnClient": {
+ "desc": "سيتم بدء طلب الجلسة مباشرة من المتصفح، مما قد يحسن سرعة الاستجابة",
+ "title": "استخدام وضع الطلب من العميل"
+ },
+ "helpDoc": "دليل التكوين",
+ "waitingForMore": "المزيد من النماذج قيد <1>التخطيط للإدماج، يرجى الانتظار"
},
- "checker": {
- "desc": "اختبر ما إذا كان مفتاح الوصول / مفتاح السر صحيحًا"
+ "createNew": {
+ "title": "إنشاء نموذج AI مخصص"
+ },
+ "item": {
+ "config": "تكوين النموذج",
+ "customModelCards": {
+ "addNew": "إنشاء وإضافة نموذج {{id}}",
+ "confirmDelete": "سيتم حذف هذا النموذج المخصص، ولن يمكن استعادته بعد الحذف، يرجى توخي الحذر."
+ },
+ "delete": {
+ "confirm": "هل تؤكد حذف النموذج {{displayName}}؟",
+ "success": "تم الحذف بنجاح",
+ "title": "حذف النموذج"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "الحقل المطلوب في Azure OpenAI",
+ "placeholder": "يرجى إدخال اسم نشر النموذج في Azure",
+ "title": "اسم نشر النموذج"
+ },
+ "deployName": {
+ "extra": "سيتم استخدام هذا الحقل كمعرف نموذج عند إرسال الطلب",
+ "placeholder": "يرجى إدخال اسم أو معرف النشر الفعلي للنموذج",
+ "title": "اسم نشر النموذج"
+ },
+ "displayName": {
+ "placeholder": "يرجى إدخال اسم العرض للنموذج، مثل ChatGPT، GPT-4، إلخ",
+ "title": "اسم عرض النموذج"
+ },
+ "files": {
+ "extra": "تنفيذ تحميل الملفات الحالي هو مجرد حل Hack، يقتصر على التجربة الذاتية. يرجى الانتظار حتى يتم تنفيذ القدرة الكاملة لتحميل الملفات لاحقًا",
+ "title": "دعم تحميل الملفات"
+ },
+ "functionCall": {
+ "extra": "هذا الإعداد سيفتح فقط قدرة النموذج على استخدام الأدوات، مما يسمح بإضافة مكونات إضافية من نوع الأدوات للنموذج. لكن ما إذا كان يمكن استخدام الأدوات فعليًا يعتمد تمامًا على النموذج نفسه، يرجى اختبار مدى قابليته للاستخدام",
+ "title": "دعم استخدام الأدوات"
+ },
+ "id": {
+ "extra": "لا يمكن تعديله بعد الإنشاء، سيتم استخدامه كمعرف نموذج عند استدعاء الذكاء الاصطناعي",
+ "placeholder": "يرجى إدخال معرف النموذج، مثل gpt-4o أو claude-3.5-sonnet",
+ "title": "معرف النموذج"
+ },
+ "modalTitle": "تكوين النموذج المخصص",
+ "reasoning": {
+ "extra": "هذا الإعداد سيفتح فقط قدرة النموذج على التفكير العميق، التأثير الفعلي يعتمد بالكامل على النموذج نفسه، يرجى اختبار ما إذا كان هذا النموذج يمتلك القدرة على التفكير العميق القابل للاستخدام",
+ "title": "يدعم التفكير العميق"
+ },
+ "tokens": {
+ "extra": "تعيين الحد الأقصى لعدد الرموز المدعومة من قبل النموذج",
+ "title": "أقصى نافذة سياق",
+ "unlimited": "غير محدود"
+ },
+ "vision": {
+ "extra": "سيؤدي هذا التكوين إلى فتح إعدادات تحميل الصور في التطبيق، ما إذا كان يدعم التعرف يعتمد بالكامل على النموذج نفسه، يرجى اختبار قابلية استخدام التعرف البصري لهذا النموذج بنفسك",
+ "title": "دعم التعرف البصري"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/صورة",
+ "inputCharts": "${{amount}}/M حرف",
+ "inputMinutes": "${{amount}}/دقيقة",
+ "inputTokens": "إدخال ${{amount}}/م",
+ "outputTokens": "إخراج ${{amount}}/م"
+ },
+ "releasedAt": "صدر في {{releasedAt}}"
},
- "secretKey": {
- "desc": "أدخل مفتاح السر من منصة بايدو تشيانفان",
- "placeholder": "مفتاح السر لتشيانفان",
- "title": "مفتاح السر"
+ "list": {
+ "addNew": "إضافة نموذج",
+ "disabled": "غير مفعل",
+ "disabledActions": {
+ "showMore": "عرض الكل"
+ },
+ "empty": {
+ "desc": "يرجى إنشاء نموذج مخصص أو سحب نموذج للبدء في الاستخدام",
+ "title": "لا توجد نماذج متاحة"
+ },
+ "enabled": "مفعل",
+ "enabledActions": {
+ "disableAll": "تعطيل الكل",
+ "enableAll": "تفعيل الكل",
+ "sort": "ترتيب النموذج حسب التخصيص"
+ },
+ "enabledEmpty": "لا توجد نماذج مفعلة، يرجى تفعيل النماذج المفضلة لديك من القائمة أدناه~",
+ "fetcher": {
+ "clear": "مسح النماذج المستخرجة",
+ "fetch": "الحصول على قائمة النماذج",
+ "fetching": "جارٍ الحصول على قائمة النماذج...",
+ "latestTime": "آخر تحديث: {{time}}",
+ "noLatestTime": "لم يتم الحصول على القائمة بعد"
+ },
+ "resetAll": {
+ "conform": "هل أنت متأكد من إعادة تعيين جميع التعديلات على النموذج الحالي؟ بعد إعادة التعيين، ستعود قائمة النماذج الحالية إلى الحالة الافتراضية",
+ "success": "تمت إعادة التعيين بنجاح",
+ "title": "إعادة تعيين جميع التعديلات"
+ },
+ "search": "ابحث عن نموذج...",
+ "searchResult": "تم العثور على {{count}} نموذج",
+ "title": "قائمة النماذج",
+ "total": "إجمالي {{count}} نموذج متاح"
},
- "unlock": {
- "customRegion": "منطقة الخدمة المخصصة",
- "description": "أدخل مفتاح الوصول / مفتاح السر لبدء المحادثة. التطبيق لن يسجل إعدادات المصادقة الخاصة بك",
- "title": "استخدام معلومات مصادقة وينشين يي يان المخصصة"
- }
+ "searchNotFound": "لم يتم العثور على نتائج البحث"
+ },
+ "sortModal": {
+ "success": "تم تحديث الترتيب بنجاح",
+ "title": "ترتيب مخصص",
+ "update": "تحديث"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "سيتم حذف مزود AI هذا، ولن يمكن استعادته بعد الحذف، هل تؤكد الحذف؟",
+ "deleteSuccess": "تم الحذف بنجاح",
+ "tooltip": "تحديث التكوين الأساسي للمزود",
+ "updateSuccess": "تم التحديث بنجاح"
},
"zeroone": {
"title": "01.AI الأشياء الصغرى"
diff --git a/locales/ar/models.json b/locales/ar/models.json
index 98593a2b58d2e..41c7ec2170e76 100644
--- a/locales/ar/models.json
+++ b/locales/ar/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K يركز على الأمان الدلالي والتوجيه المسؤول، مصمم خصيصًا لتطبيقات تتطلب مستوى عالٍ من الأمان في المحتوى، مما يضمن دقة وموثوقية تجربة المستخدم."
},
+ "360gpt2-o1": {
+ "description": "يستخدم 360gpt2-o1 البحث الشجري لبناء سلسلة التفكير، ويقدم آلية للتفكير العميق، ويستخدم التعلم المعزز للتدريب، مما يمنح النموذج القدرة على التفكير الذاتي وتصحيح الأخطاء."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro هو نموذج متقدم لمعالجة اللغة الطبيعية تم إطلاقه من قبل شركة 360، يتمتع بقدرات استثنائية في توليد وفهم النصوص، خاصة في مجالات التوليد والإبداع، ويستطيع التعامل مع مهام تحويل اللغة المعقدة وأداء الأدوار."
},
+ "360zhinao2-o1": {
+ "description": "يستخدم 360zhinao2-o1 البحث الشجري لبناء سلسلة التفكير، ويقدم آلية للتفكير النقدي، ويستخدم التعلم المعزز للتدريب، مما يمنح النموذج القدرة على التفكير الذاتي وتصحيح الأخطاء."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra هو أقوى إصدار في سلسلة نماذج Spark، حيث يعزز فهم النصوص وقدرات التلخيص مع تحسين روابط البحث عبر الإنترنت. إنه حل شامل يهدف إلى تعزيز إنتاجية المكتب والاستجابة الدقيقة للاحتياجات، ويعتبر منتجًا ذكيًا رائدًا في الصناعة."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "النموذج الأول محليًا، يتفوق على النماذج الرئيسية الأجنبية في المهام الصينية مثل المعرفة الموسوعية، النصوص الطويلة، والإبداع. كما يتمتع بقدرات متعددة الوسائط الرائدة في الصناعة، ويظهر أداءً ممتازًا في العديد من معايير التقييم الموثوقة."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "نموذج التقطير DeepSeek-R1 المستند إلى Qwen2.5-Math-1.5B، تم تحسين أداء الاستدلال من خلال التعلم المعزز وبيانات البداية الباردة، ويعيد نموذج المصدر فتح معايير المهام المتعددة."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "نموذج التقطير DeepSeek-R1 المستند إلى Qwen2.5-14B، تم تحسين أداء الاستدلال من خلال التعلم المعزز وبيانات البداية الباردة، ويعيد نموذج المصدر فتح معايير المهام المتعددة."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "تسلسل DeepSeek-R1 يحسن أداء الاستدلال من خلال التعلم المعزز وبيانات البداية الباردة، ويعيد نموذج المصدر فتح معايير المهام المتعددة، متجاوزًا مستوى OpenAI-o1-mini."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "نموذج التقطير DeepSeek-R1 المستند إلى Qwen2.5-Math-7B، تم تحسين أداء الاستدلال من خلال التعلم المعزز وبيانات البداية الباردة، ويعيد نموذج المصدر فتح معايير المهام المتعددة."
+ },
"Doubao-lite-128k": {
"description": "دو باو-لايت يوفر سرعة استجابة فائقة وقيمة جيدة للكلفة، ويقدم خيارات أكثر مرونة للعملاء في سيناريوهات مختلفة. يدعم الاستدلال والتنقيح بسعة سياق 128k."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "نموذج اللغة الكبير الرائد الذي طورته بايدو، والذي شهد ترقية شاملة في القدرات مقارنةً بـERNIE 3.5، ويستخدم على نطاق واسع في مجالات متعددة لمهام معقدة؛ يدعم الاتصال التلقائي بإضافات بحث بايدو لضمان تحديث معلومات الإجابة."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "نموذج اللغة الكبير الرائد من بايدو، يظهر أداءً ممتازًا في مجموعة واسعة من المهام المعقدة في مختلف المجالات؛ يدعم الاتصال التلقائي بمكونات بحث بايدو، مما يضمن تحديث معلومات الأسئلة والأجوبة. مقارنةً بـ ERNIE 4.0، يظهر أداءً أفضل."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "نموذج اللغة الكبير الرائد الذي طورته بايدو، والذي يظهر أداءً ممتازًا في مجالات متعددة، مما يجعله مناسبًا لمجموعة واسعة من المهام المعقدة؛ يدعم الاتصال التلقائي بمكونات البحث من بايدو، مما يضمن تحديث معلومات الأسئلة والأجوبة. مقارنة بـ ERNIE 4.0، يظهر أداءً أفضل."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct هو أحد أحدث نماذج اللغة الكبيرة التي أصدرتها Alibaba Cloud. يتمتع هذا النموذج بقدرات محسنة بشكل ملحوظ في مجالات الترميز والرياضيات. كما يوفر دعمًا للغات متعددة، تغطي أكثر من 29 لغة، بما في ذلك الصينية والإنجليزية. أظهر النموذج تحسينات ملحوظة في اتباع التعليمات، وفهم البيانات الهيكلية، وتوليد المخرجات الهيكلية (خاصة JSON)."
},
+ "MiniMax-Text-01": {
+ "description": "في سلسلة نماذج MiniMax-01، قمنا بإجراء ابتكارات جريئة: تم تنفيذ آلية الانتباه الخطي على نطاق واسع لأول مرة، لم يعد هيكل Transformer التقليدي هو الخيار الوحيد. يصل عدد معلمات هذا النموذج إلى 456 مليار، مع تنشيط واحد يصل إلى 45.9 مليار. الأداء الشامل للنموذج يتساوى مع النماذج الرائدة في الخارج، بينما يمكنه معالجة سياقات تصل إلى 4 ملايين توكن، وهو 32 مرة من GPT-4o و20 مرة من Claude-3.5-Sonnet."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO هو دمج متعدد النماذج مرن للغاية، يهدف إلى تقديم تجربة إبداعية ممتازة."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "أظهر InternVL2 أداءً رائعًا في مجموعة متنوعة من مهام اللغة البصرية، بما في ذلك فهم الوثائق والرسوم البيانية، وفهم النصوص في المشاهد، وOCR، وحل المشكلات العلمية والرياضية."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "أظهر InternVL2 أداءً رائعًا في مجموعة متنوعة من مهام اللغة البصرية، بما في ذلك فهم الوثائق والرسوم البيانية، وفهم النصوص في المشاهد، وOCR، وحل المشكلات العلمية والرياضية."
- },
"Phi-3-medium-128k-instruct": {
"description": "نموذج Phi-3-medium نفسه، ولكن مع حجم سياق أكبر لـ RAG أو التوجيه القليل."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat هو الإصدار مفتوح المصدر من نموذج GLM-4 الذي أطلقته Zhizhu AI. أظهر هذا النموذج أداءً ممتازًا في مجالات الدلالات، والرياضيات، والاستدلال، والشيفرة، والمعرفة. بالإضافة إلى دعم المحادثات متعددة الجولات، يتمتع GLM-4-9B-Chat أيضًا بميزات متقدمة مثل تصفح الويب، وتنفيذ الشيفرة، واستدعاء الأدوات المخصصة (Function Call)، والاستدلال على النصوص الطويلة. يدعم النموذج 26 لغة، بما في ذلك الصينية، والإنجليزية، واليابانية، والكورية، والألمانية. أظهر GLM-4-9B-Chat أداءً ممتازًا في العديد من اختبارات المعايير مثل AlignBench-v2 وMT-Bench وMMLU وC-Eval. يدعم النموذج طول سياق يصل إلى 128K، مما يجعله مناسبًا للأبحاث الأكاديمية والتطبيقات التجارية."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 هو نموذج استدلال مدفوع بالتعلم المعزز (RL)، يعالج مشكلات التكرار وقابلية القراءة في النموذج. قبل التعلم المعزز، أدخل DeepSeek-R1 بيانات بدء التشغيل الباردة، مما أدى إلى تحسين أداء الاستدلال. إنه يتفوق في المهام الرياضية، والبرمجة، والاستدلال مقارنةً بـ OpenAI-o1، وقد حسّن الأداء العام من خلال طرق تدريب مصممة بعناية."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 هو نموذج لغوي مختلط الخبراء (MoE) يحتوي على 6710 مليار معلمة، يستخدم الانتباه المتعدد الرؤوس (MLA) وهيكل DeepSeekMoE، ويجمع بين استراتيجيات توازن الحمل بدون خسائر مساعدة، مما يحسن كفاءة الاستدلال والتدريب. تم تدريبه مسبقًا على 14.8 تريليون توكن عالية الجودة، وتم إجراء تعديل دقيق تحت الإشراف والتعلم المعزز، مما يجعل DeepSeek-V3 يتفوق على نماذج مفتوحة المصدر الأخرى، ويقترب من النماذج المغلقة الرائدة."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma هو أحد نماذج Google المتقدمة والخفيفة الوزن من سلسلة النماذج المفتوحة. إنه نموذج لغوي كبير يعتمد على فك الشيفرة فقط، يدعم اللغة الإنجليزية، ويقدم أوزان مفتوحة، ومتغيرات مدربة مسبقًا، ومتغيرات معدلة وفقًا للتعليمات. نموذج Gemma مناسب لمجموعة متنوعة من مهام توليد النصوص، بما في ذلك الأسئلة والأجوبة، والتلخيص، والاستدلال. تم تدريب هذا النموذج 9B على 8 تريليون توكن. حجمه النسبي الصغير يجعله مناسبًا للنشر في بيئات ذات موارد محدودة، مثل أجهزة الكمبيوتر المحمولة، وأجهزة الكمبيوتر المكتبية، أو البنية التحتية السحابية الخاصة بك، مما يتيح لمزيد من الأشخاص الوصول إلى نماذج الذكاء الاصطناعي المتقدمة وتعزيز الابتكار."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 هو جزء من عائلة نماذج اللغة الكبيرة متعددة اللغات التي طورتها Meta، بما في ذلك متغيرات مدربة مسبقًا ومعدلة وفقًا للتعليمات بحجم 8B و70B و405B. تم تحسين هذا النموذج 8B وفقًا لمشاهدات المحادثات متعددة اللغات، وأظهر أداءً ممتازًا في العديد من اختبارات المعايير الصناعية. تم تدريب النموذج باستخدام أكثر من 15 تريليون توكن من البيانات العامة، واستخدم تقنيات مثل التعديل الخاضع للإشراف والتعلم المعزز من ردود الفعل البشرية لتحسين فائدة النموذج وأمانه. يدعم Llama 3.1 توليد النصوص وتوليد الشيفرة، مع تاريخ معرفة حتى ديسمبر 2023."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview هو نموذج معالجة اللغة الطبيعية المبتكر، قادر على معالجة مهام توليد الحوار وفهم السياق بشكل فعال."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview هو نموذج بحثي طورته فريق Qwen يركز على قدرات الاستدلال البصري، حيث يتمتع بميزة فريدة في فهم المشاهد المعقدة وحل المشكلات الرياضية المتعلقة بالرؤية."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview هو أحدث نموذج بحث تجريبي من Qwen، يركز على تعزيز قدرات الاستدلال للذكاء الاصطناعي. من خلال استكشاف آليات معقدة مثل خلط اللغة والاستدلال التكراري، تشمل المزايا الرئيسية القدرة القوية على التحليل الاستدلالي، والقدرات الرياضية والبرمجية. في الوقت نفسه، هناك أيضًا مشكلات في تبديل اللغة، ودورات الاستدلال، واعتبارات الأمان، واختلافات في القدرات الأخرى."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct هو أحدث إصدار من سلسلة نماذج اللغة الكبيرة المحددة للشيفرة التي أصدرتها Alibaba Cloud. تم تحسين هذا النموذج بشكل كبير في توليد الشيفرة، والاستدلال، وإصلاح الأخطاء، من خلال تدريب على 55 تريليون توكن."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math يركز على حل المشكلات في مجال الرياضيات، ويقدم إجابات احترافية للأسئلة الصعبة."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 هو أحدث سلسلة من نموذج Qwen، ويدعم سياقًا يصل إلى 128 ألف، مقارنةً بأفضل النماذج مفتوحة المصدر الحالية، يتفوق Qwen2-72B بشكل ملحوظ في فهم اللغة الطبيعية والمعرفة والترميز والرياضيات والقدرات متعددة اللغات."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 هو أحدث سلسلة من نموذج Qwen، قادر على التفوق على النماذج مفتوحة المصدر ذات الحجم المماثل أو حتى النماذج الأكبر حجمًا، حقق Qwen2 7B مزايا ملحوظة في عدة تقييمات، خاصة في فهم الترميز والصينية."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B هو نموذج قوي للغة البصرية، يدعم المعالجة متعددة الوسائط للصورة والنص، ويستطيع التعرف بدقة على محتوى الصورة وتوليد أوصاف أو إجابات ذات صلة."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct هو نموذج لغوي كبير يحتوي على 14 مليار معلمة، يتميز بأداء ممتاز، تم تحسينه لمشاهد اللغة الصينية واللغات المتعددة، ويدعم التطبيقات مثل الأسئلة الذكية وتوليد المحتوى."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct هو نموذج لغوي كبير يحتوي على 7 مليارات معلمة، يدعم الاتصال الوظيفي مع الأنظمة الخارجية بسلاسة، مما يعزز المرونة وقابلية التوسع بشكل كبير. تم تحسينه لمشاهد اللغة الصينية واللغات المتعددة، ويدعم التطبيقات مثل الأسئلة الذكية وتوليد المحتوى."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct هو نموذج تعليمات برمجة قائم على تدريب مسبق واسع النطاق، يتمتع بقدرة قوية على فهم وتوليد الشيفرات، مما يجعله فعالاً في معالجة مختلف مهام البرمجة، وخاصة كتابة الشيفرات الذكية، وتوليد السكربتات الآلية، وحل مشكلات البرمجة."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct هو نموذج لغوي كبير مصمم خصيصًا لتوليد الشيفرات، وفهم الشيفرات، ومشاهد التطوير الفعالة، مع استخدام حجم 32B من المعلمات الرائدة في الصناعة، مما يلبي احتياجات البرمجة المتنوعة."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "نموذج TeleMM هو نموذج كبير لفهم متعدد الوسائط تم تطويره ذاتيًا من قبل China Telecom، يمكنه معالجة مدخلات متعددة الوسائط مثل النصوص والصور، ويدعم وظائف مثل فهم الصور، وتحليل الرسوم البيانية، مما يوفر خدمات فهم متعددة الوسائط للمستخدمين. يمكن للنموذج التفاعل مع المستخدمين بطرق متعددة الوسائط، وفهم المحتوى المدخل بدقة، والإجابة على الأسئلة، والمساعدة في الإبداع، وتوفير معلومات متعددة الوسائط ودعم الإلهام بكفاءة. أظهر أداءً ممتازًا في المهام متعددة الوسائط مثل الإدراك الدقيق، والاستدلال المنطقي."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large هو أكبر نموذج MoE مفتوح المصدر في الصناعة، مع 389 مليار إجمالي عدد المعلمات و52 مليار عدد المعلمات النشطة."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct هو نموذج لغوي كبير تم تعديله وفقًا للتعليمات في سلسلة Qwen2، بحجم 72B. يعتمد هذا النموذج على بنية Transformer، ويستخدم تقنيات مثل دالة تنشيط SwiGLU، وتحويل QKV، والانتباه الجماعي. يمكنه معالجة المدخلات الكبيرة. أظهر النموذج أداءً ممتازًا في فهم اللغة، والتوليد، والقدرات متعددة اللغات، والترميز، والرياضيات، والاستدلال في العديد من اختبارات المعايير، متجاوزًا معظم النماذج مفتوحة المصدر."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct هو أحد أحدث نماذج اللغة الكبيرة التي أصدرتها Alibaba Cloud. يتمتع هذا النموذج بقدرات محسنة بشكل ملحوظ في مجالات الترميز والرياضيات. كما يوفر دعمًا للغات متعددة، تغطي أكثر من 29 لغة، بما في ذلك الصينية والإنجليزية. أظهر النموذج تحسينات ملحوظة في اتباع التعليمات، وفهم البيانات الهيكلية، وتوليد المخرجات الهيكلية (خاصة JSON)."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "محسن لمشاهد الحوار باللغة الصينية، يوفر قدرة توليد حوار سلس ومتوافق مع عادات التعبير الصينية."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "نموذج استدعاء الدوال مفتوح المصدر من Fireworks، يوفر قدرة تنفيذ تعليمات ممتازة وخصائص قابلة للتخصيص."
+ "abab7-chat-preview": {
+ "description": "تحسين كبير في قدرات النموذج مقارنة بسلسلة abab6.5 في النصوص الطويلة، الرياضيات، والكتابة."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Firefunction-v2 من شركة Fireworks هو نموذج استدعاء دوال عالي الأداء، تم تطويره بناءً على Llama-3، وتم تحسينه بشكل كبير، مناسب بشكل خاص لاستدعاء الدوال، والحوار، واتباع التعليمات."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 هو نموذج لغة كبير متقدم، تم تحسينه من خلال التعلم المعزز وبيانات البدء البارد، ويتميز بأداء ممتاز في الاستدلال، والرياضيات، والبرمجة."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b هو نموذج لغوي بصري، يمكنه استقبال المدخلات من الصور والنصوص، تم تدريبه على بيانات عالية الجودة، مناسب للمهام متعددة الوسائط."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "نموذج اللغة القوي من Deepseek، الذي يعتمد على مزيج من الخبراء (MoE)، بإجمالي عدد معلمات يبلغ 671 مليار، حيث يتم تفعيل 37 مليار معلمة لكل علامة."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "نموذج Llama 3 70B للتعليمات، مصمم للحوار متعدد اللغات وفهم اللغة الطبيعية، أداءه يتفوق على معظم النماذج المنافسة."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "نموذج Llama 3 70B للتعليمات (نسخة HF)، يتوافق مع نتائج التنفيذ الرسمية، مناسب لمهام اتباع التعليمات عالية الجودة."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "نموذج Llama 3 8B للتعليمات، تم تحسينه للحوار والمهام متعددة اللغات، يظهر أداءً ممتازًا وفعالًا."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "نموذج استدلال الصور المعدل من Meta ذو 11B معلمات. تم تحسين هذا النموذج للتعرف البصري، واستدلال الصور، ووصف الصور، والإجابة عن الأسئلة العامة المتعلقة بالصور. يستطيع النموذج فهم البيانات البصرية مثل الرسوم البيانية والرسوم، ويسد الفجوة بين الرؤية واللغة من خلال توليد أوصاف نصية لجزئيات الصور."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "نموذج التوجيه Llama 3.2 1B هو نموذج متعدد اللغات خفيف الوزن قدمته Meta. يهدف هذا النموذج إلى زيادة الكفاءة، مع تحسينات ملحوظة في التأخير والتكلفة مقارنة بالنماذج الأكبر. تشمل حالات الاستخدام النموذجية لهذا النموذج الاسترجاع والتلخيص."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "نموذج التوجيه Llama 3.2 3B هو نموذج متعدد اللغات خفيف الوزن قدمته Meta. يهدف هذا النموذج إلى زيادة الكفاءة، مع تحسينات ملحوظة في التأخير والتكلفة مقارنة بالنماذج الأكبر. تشمل حالات الاستخدام النموذجية لهذا النموذج الاستفسارات وإعادة كتابة الملاحظات والمساعدة في الكتابة."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "نموذج استدلال الصور المعدل من Meta ذو 90B معلمات. تم تحسين هذا النموذج للتعرف البصري، واستدلال الصور، ووصف الصور، والإجابة عن الأسئلة العامة المتعلقة بالصور. يستطيع النموذج فهم البيانات البصرية مثل الرسوم البيانية والرسوم، ويسد الفجوة بين الرؤية واللغة من خلال توليد أوصاف نصية لجزئيات الصور."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct هو الإصدار المحدث من Llama 3.1 70B في ديسمبر. تم تحسين هذا النموذج بناءً على Llama 3.1 70B (الذي تم إصداره في يوليو 2024) لتعزيز استدعاء الأدوات، ودعم النصوص متعددة اللغات، والقدرات الرياضية وبرمجة. لقد حقق هذا النموذج مستويات رائدة في الصناعة في الاستدلال، والرياضيات، واتباع التعليمات، ويستطيع تقديم أداء مشابه لـ 3.1 405B، مع مزايا ملحوظة في السرعة والتكلفة."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "نموذج بـ 24 مليار معلمة، يتمتع بقدرات متقدمة تعادل النماذج الأكبر حجماً."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "نموذج Mixtral MoE 8x22B للتعليمات، مع معلمات ضخمة وهيكل خبير متعدد، يدعم معالجة فعالة لمهام معقدة."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "نموذج Mixtral MoE 8x7B للتعليمات، يوفر هيكل خبير متعدد لتقديم تعليمات فعالة واتباعها."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "نموذج Mixtral MoE 8x7B للتعليمات (نسخة HF)، الأداء يتوافق مع التنفيذ الرسمي، مناسب لمجموعة متنوعة من سيناريوهات المهام الفعالة."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "نموذج MythoMax L2 13B، يجمع بين تقنيات الدمج الجديدة، بارع في السرد وأدوار الشخصيات."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "نموذج QwQ هو نموذج بحث تجريبي تم تطويره بواسطة فريق Qwen، يركز على تعزيز قدرات الاستدلال للذكاء الاصطناعي."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "الإصدار 72B من نموذج Qwen-VL هو نتيجة أحدث ابتكارات Alibaba، ويمثل ما يقرب من عام من الابتكار."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 هي سلسلة من نماذج اللغة التي طورتها مجموعة Qwen من علي بابا، تحتوي فقط على شريحة فك شفرات. توفر هذه النماذج أحجامًا مختلفة، بما في ذلك 0.5B، 1.5B، 3B، 7B، 14B، 32B و72B، وتأتي بنسخ أساسية (base) ونماذج توجيهية (instruct)."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct هو أحدث إصدار من سلسلة نماذج اللغة الكبيرة المحددة للشيفرة التي أصدرتها Alibaba Cloud. تم تحسين هذا النموذج بشكل كبير في توليد الشيفرة، والاستدلال، وإصلاح الأخطاء، من خلال تدريب على 55 تريليون توكن."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "نموذج StarCoder 15.5B، يدعم مهام البرمجة المتقدمة، مع تعزيز القدرة على التعامل مع لغات متعددة، مناسب لتوليد وفهم الشيفرات المعقدة."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "نموذج StarCoder 7B، تم تدريبه على أكثر من 80 لغة برمجة، يتمتع بقدرة ممتازة على ملء البرمجة وفهم السياق."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "نموذج Yi-Large، يتمتع بقدرة معالجة لغوية ممتازة، يمكن استخدامه في جميع أنواع مهام توليد وفهم اللغة."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus هو أقوى نموذج من Anthropic لمعالجة المهام المعقدة للغاية. يتميز بأداء ممتاز وذكاء وسلاسة وفهم."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku هو أسرع نموذج من الجيل التالي من Anthropic. مقارنةً بـ Claude 3 Haiku، تم تحسين Claude 3.5 Haiku في جميع المهارات، وتفوق في العديد من اختبارات الذكاء على النموذج الأكبر من الجيل السابق Claude 3 Opus."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet يقدم قدرات تتجاوز Opus وسرعة أكبر من Sonnet، مع الحفاظ على نفس السعر. يتميز Sonnet بمهارات خاصة في البرمجة وعلوم البيانات ومعالجة الصور والمهام الوكيلة."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet يوفر توازنًا مثاليًا بين الذكاء والسرعة لحمولات العمل المؤسسية. يقدم أقصى فائدة بسعر أقل، موثوق ومناسب للنشر على نطاق واسع."
},
- "code-raccoon-v1": {
- "description": "كود راكون هو مساعد ذكي لتطوير البرمجيات يعتمد على نموذج اللغة الكبير من SenseTime، يغطي مراحل تحليل متطلبات البرمجيات، وتصميم الهيكل، وكتابة الشيفرات، واختبار البرمجيات، لتلبية احتياجات المستخدمين في كتابة الشيفرات، وتعلم البرمجة، وغيرها من المتطلبات. يدعم كود راكون أكثر من 90 لغة برمجة رئيسية مثل Python وJava وJavaScript وC++ وGo وSQL، بالإضافة إلى IDEs الرئيسية مثل VS Code وIntelliJ IDEA. في التطبيقات العملية، يمكن أن يساعد كود راكون المطورين في زيادة كفاءة البرمجة بأكثر من 50%."
- },
"codegeex-4": {
"description": "CodeGeeX-4 هو مساعد برمجي قوي، يدعم مجموعة متنوعة من لغات البرمجة في الإجابة الذكية وإكمال الشيفرة، مما يعزز من كفاءة التطوير."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ هو نموذج لغوي كبير عالي الأداء، مصمم لمشاهد الأعمال الحقيقية والتطبيقات المعقدة."
},
+ "dall-e-2": {
+ "description": "النموذج الثاني من DALL·E، يدعم توليد صور أكثر واقعية ودقة، بدقة تعادل أربعة أضعاف الجيل الأول."
+ },
+ "dall-e-3": {
+ "description": "أحدث نموذج DALL·E، تم إصداره في نوفمبر 2023. يدعم توليد صور أكثر واقعية ودقة، مع قدرة أكبر على التعبير عن التفاصيل."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct يوفر قدرة معالجة تعليمات موثوقة، يدعم تطبيقات متعددة الصناعات."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 هو نموذج لغوي قوي وفعال من حيث التكلفة يعتمد على الخبراء المختلطين (MoE). تم تدريبه مسبقًا على مجموعة بيانات عالية الجودة تحتوي على 8.1 تريليون توكن، وتم تحسين قدراته من خلال التعديل الخاضع للإشراف (SFT) والتعلم المعزز (RL). مقارنةً بـ DeepSeek 67B، يوفر DeepSeek-V2 أداءً أقوى مع توفير 42.5% من تكاليف التدريب، وتقليل 93.3% من ذاكرة التخزين المؤقت KV، وزيادة الحد الأقصى لمعدل الإنتاج إلى 5.76 مرة. يدعم النموذج طول سياق يصل إلى 128k، ويظهر أداءً ممتازًا في اختبارات المعايير القياسية وتقييمات التوليد المفتوحة."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 هو نموذج استدلال مدفوع بالتعلم المعزز (RL) يعالج مشكلات التكرار وقابلية القراءة في النموذج. قبل استخدام RL، قدم DeepSeek-R1 بيانات بدء باردة، مما أدى إلى تحسين أداء الاستدلال. إنه يقدم أداءً مماثلاً لـ OpenAI-o1 في المهام الرياضية والبرمجية والاستدلال، وقد حسّن النتائج العامة من خلال طرق تدريب مصممة بعناية."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "نموذج التقطير DeepSeek-R1، تم تحسين أداء الاستدلال من خلال التعلم المعزز وبيانات البداية الباردة، ويعيد نموذج المصدر فتح معايير المهام المتعددة."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B هو نموذج تم تطويره بناءً على Llama-3.1-8B. تم ضبط هذا النموذج باستخدام عينات تم إنشاؤها بواسطة DeepSeek-R1، ويظهر قدرة استدلال ممتازة. حقق أداءً جيدًا في اختبارات المعايير، حيث حقق دقة 89.1% في MATH-500، وحقق معدل نجاح 50.4% في AIME 2024، وحصل على تقييم 1205 في CodeForces، مما يظهر قدرة قوية في الرياضيات والبرمجة كنموذج بحجم 8B."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "نموذج التقطير DeepSeek-R1، تم تحسين أداء الاستدلال من خلال التعلم المعزز وبيانات البداية الباردة، ويعيد نموذج المصدر فتح معايير المهام المتعددة."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "نموذج التقطير DeepSeek-R1، تم تحسين أداء الاستدلال من خلال التعلم المعزز وبيانات البداية الباردة، ويعيد نموذج المصدر فتح معايير المهام المتعددة."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B هو نموذج تم الحصول عليه من Qwen2.5-32B من خلال التقطير المعرفي. تم ضبط هذا النموذج باستخدام 800,000 عينة مختارة تم إنشاؤها بواسطة DeepSeek-R1، ويظهر أداءً ممتازًا في مجالات متعددة مثل الرياضيات، البرمجة، والاستدلال. حقق نتائج ممتازة في اختبارات المعايير مثل AIME 2024، MATH-500، وGPQA Diamond، حيث حقق دقة 94.3% في MATH-500، مما يظهر قدرة قوية في الاستدلال الرياضي."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B هو نموذج تم الحصول عليه من Qwen2.5-Math-7B من خلال التقطير المعرفي. تم ضبط هذا النموذج باستخدام 800,000 عينة مختارة تم إنشاؤها بواسطة DeepSeek-R1، ويظهر أداءً ممتازًا في الاستدلال. حقق نتائج ممتازة في اختبارات المعايير، حيث حقق دقة 92.8% في MATH-500، وحقق معدل نجاح 55.5% في AIME 2024، وحصل على تقييم 1189 في CodeForces، مما يظهر قدرة قوية في الرياضيات والبرمجة كنموذج بحجم 7B."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 يجمع بين الميزات الممتازة للإصدارات السابقة، ويعزز القدرات العامة والترميز."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 هو نموذج لغوي مختلط الخبراء (MoE) يحتوي على 6710 مليار معلمة، يستخدم انتباه متعدد الرؤوس (MLA) وبنية DeepSeekMoE، ويجمع بين استراتيجية توازن الحمل بدون خسارة مساعدة، مما يحسن كفاءة الاستدلال والتدريب. من خلال التدريب المسبق على 14.8 تريليون توكن عالي الجودة، وإجراء تعديلات إشرافية وتعلم معزز، يتفوق DeepSeek-V3 في الأداء على نماذج المصدر المفتوح الأخرى، ويقترب من النماذج المغلقة الرائدة."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B هو نموذج متقدم تم تدريبه للحوار المعقد."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "نموذج لغوي متقدم وفعال، بارع في الاستدلال، والرياضيات، والبرمجة."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 هو نموذج لغوي بصري مختلط الخبراء (MoE) تم تطويره بناءً على DeepSeekMoE-27B، يستخدم بنية MoE ذات تفعيل نادر، محققًا أداءً ممتازًا مع تفعيل 4.5 مليار معلمة فقط. يقدم هذا النموذج أداءً ممتازًا في مهام مثل الأسئلة البصرية، التعرف الضوئي على الأحرف، فهم الوثائق/الجداول/الرسوم البيانية، وتحديد المواقع البصرية."
+ },
"deepseek-chat": {
"description": "نموذج مفتوح المصدر الجديد الذي يجمع بين القدرات العامة وقدرات البرمجة، لا يحتفظ فقط بالقدرات الحوارية العامة لنموذج الدردشة الأصلي وقدرات معالجة الشيفرة القوية لنموذج Coder، بل يتماشى أيضًا بشكل أفضل مع تفضيلات البشر. بالإضافة إلى ذلك، حقق DeepSeek-V2.5 تحسينات كبيرة في مهام الكتابة، واتباع التعليمات، وغيرها من المجالات."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 هو نموذج شيفرة مفتوح المصدر من نوع خبير مختلط، يقدم أداءً ممتازًا في مهام الشيفرة، ويضاهي GPT4-Turbo."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 هو نموذج استدلال مدفوع بالتعلم المعزز (RL) يعالج مشكلات التكرار وقابلية القراءة في النموذج. قبل استخدام RL، قدم DeepSeek-R1 بيانات بدء باردة، مما أدى إلى تحسين أداء الاستدلال. إنه يقدم أداءً مماثلاً لـ OpenAI-o1 في المهام الرياضية والبرمجية والاستدلال، وقد حسّن النتائج العامة من خلال طرق تدريب مصممة بعناية."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 - النموذج الأكبر والأذكى في مجموعة DeepSeek - تم تقطيره إلى بنية Llama 70B. بناءً على اختبارات المعايير والتقييمات البشرية، يظهر هذا النموذج ذكاءً أكبر من Llama 70B الأصلي، خاصة في المهام التي تتطلب دقة رياضية وحقائق."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "نموذج DeepSeek-R1-Distill تم تطويره من خلال تقنية تقطير المعرفة، حيث تم تعديل عينات تم إنشاؤها بواسطة DeepSeek-R1 على نماذج مفتوحة المصدر مثل Qwen وLlama."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "نموذج DeepSeek-R1-Distill تم تطويره من خلال تقنية تقطير المعرفة، حيث تم تعديل عينات تم إنشاؤها بواسطة DeepSeek-R1 على نماذج مفتوحة المصدر مثل Qwen وLlama."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "نموذج DeepSeek-R1-Distill تم تطويره من خلال تقنية تقطير المعرفة، حيث تم تعديل عينات تم إنشاؤها بواسطة DeepSeek-R1 على نماذج مفتوحة المصدر مثل Qwen وLlama."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "نموذج DeepSeek-R1-Distill تم تطويره من خلال تقنية تقطير المعرفة، حيث تم تعديل عينات تم إنشاؤها بواسطة DeepSeek-R1 على نماذج مفتوحة المصدر مثل Qwen وLlama."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "نموذج DeepSeek-R1-Distill تم تطويره من خلال تقنية تقطير المعرفة، حيث تم تعديل عينات تم إنشاؤها بواسطة DeepSeek-R1 على نماذج مفتوحة المصدر مثل Qwen وLlama."
+ },
+ "deepseek-reasoner": {
+ "description": "نموذج الاستدلال الذي أطلقته DeepSeek. قبل تقديم الإجابة النهائية، يقوم النموذج أولاً بإخراج سلسلة من المحتوى الفكري لتحسين دقة الإجابة النهائية."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 هو نموذج لغوي فعال من نوع Mixture-of-Experts، مناسب لاحتياجات المعالجة الاقتصادية."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B هو نموذج تصميم الشيفرة لـ DeepSeek، يوفر قدرة توليد شيفرة قوية."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 هو نموذج MoE تم تطويره بواسطة شركة Hangzhou DeepSeek AI Technology Research Co.، Ltd، وقد حقق نتائج بارزة في العديد من التقييمات، ويحتل المرتبة الأولى بين نماذج المصدر المفتوح في القوائم الرئيسية. مقارنةً بنموذج V2.5، حقق V3 زيادة في سرعة التوليد بمقدار 3 مرات، مما يوفر تجربة استخدام أسرع وأكثر سلاسة للمستخدمين."
+ },
"deepseek/deepseek-chat": {
"description": "نموذج مفتوح المصدر جديد يجمع بين القدرات العامة وقدرات البرمجة، لا يحتفظ فقط بقدرات الحوار العامة لنموذج الدردشة الأصلي وقدرات معالجة الأكواد القوية لنموذج Coder، بل يتماشى أيضًا بشكل أفضل مع تفضيلات البشر. بالإضافة إلى ذلك، حقق DeepSeek-V2.5 تحسينات كبيرة في مهام الكتابة، واتباع التعليمات، وغيرها من المجالات."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 يعزز بشكل كبير من قدرة النموذج على الاستدلال في ظل وجود بيانات محدودة جدًا. قبل تقديم الإجابة النهائية، يقوم النموذج أولاً بإخراج سلسلة من التفكير لتحسين دقة الإجابة النهائية."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 يعزز بشكل كبير من قدرة النموذج على الاستدلال في ظل وجود بيانات محدودة جدًا. قبل تقديم الإجابة النهائية، يقوم النموذج أولاً بإخراج سلسلة من التفكير لتحسين دقة الإجابة النهائية."
+ },
"emohaa": {
"description": "Emohaa هو نموذج نفسي، يتمتع بقدرات استشارية متخصصة، يساعد المستخدمين في فهم القضايا العاطفية."
},
+ "ernie-3.5-128k": {
+ "description": "نموذج اللغة الكبير الرائد الذي طورته بايدو، يغطي كمية هائلة من البيانات باللغة الصينية والإنجليزية، ويتميز بقدرات عامة قوية، تلبي متطلبات معظم حالات الحوار، والإجابة، والتوليد، وتطبيقات المكونات الإضافية؛ يدعم الاتصال التلقائي بمكونات البحث من بايدو، مما يضمن تحديث معلومات الإجابة."
+ },
+ "ernie-3.5-8k": {
+ "description": "نموذج اللغة الكبير الرائد الذي طورته بايدو، يغطي كمية هائلة من البيانات باللغة الصينية والإنجليزية، ويتميز بقدرات عامة قوية، تلبي متطلبات معظم حالات الحوار، والإجابة، والتوليد، وتطبيقات المكونات الإضافية؛ يدعم الاتصال التلقائي بمكونات البحث من بايدو، مما يضمن تحديث معلومات الإجابة."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "نموذج اللغة الكبير الرائد الذي طورته بايدو، يغطي كمية هائلة من البيانات باللغة الصينية والإنجليزية، ويتميز بقدرات عامة قوية، تلبي متطلبات معظم حالات الحوار، والإجابة، والتوليد، وتطبيقات المكونات الإضافية؛ يدعم الاتصال التلقائي بمكونات البحث من بايدو، مما يضمن تحديث معلومات الإجابة."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "نموذج اللغة الكبير الرائد الذي طورته بايدو، والذي حقق ترقية شاملة في القدرات مقارنةً بـ ERNIE 3.5، ويستخدم على نطاق واسع في مشاهد المهام المعقدة في مختلف المجالات؛ يدعم الاتصال التلقائي بمكونات البحث من بايدو، مما يضمن تحديث معلومات الإجابة."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "نموذج اللغة الكبير الرائد الذي طورته بايدو، والذي حقق ترقية شاملة في القدرات مقارنةً بـ ERNIE 3.5، ويستخدم على نطاق واسع في مشاهد المهام المعقدة في مختلف المجالات؛ يدعم الاتصال التلقائي بمكونات البحث من بايدو، مما يضمن تحديث معلومات الإجابة."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "نموذج اللغة الكبير الرائد الذي طورته بايدو، والذي يظهر أداءً ممتازًا بشكل شامل، ويستخدم على نطاق واسع في مشاهد المهام المعقدة في مختلف المجالات؛ يدعم الاتصال التلقائي بمكونات البحث من بايدو، مما يضمن تحديث معلومات الإجابة. مقارنةً بـ ERNIE 4.0، يظهر أداءً أفضل."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "نموذج اللغة الكبير الرائد الذي طورته بايدو، والذي يظهر أداءً ممتازًا بشكل شامل، ويستخدم على نطاق واسع في مشاهد المهام المعقدة في مختلف المجالات؛ يدعم الاتصال التلقائي بمكونات البحث من بايدو، مما يضمن تحديث معلومات الإجابة. مقارنةً بـ ERNIE 4.0، يظهر أداءً أفضل."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "نموذج اللغة الكبير الرائد الذي طورته بايدو، والذي يظهر أداءً ممتازًا بشكل شامل، ويستخدم على نطاق واسع في مشاهد المهام المعقدة في مختلف المجالات؛ يدعم الاتصال التلقائي بمكونات البحث من بايدو، مما يضمن تحديث معلومات الإجابة. مقارنةً بـ ERNIE 4.0، يظهر أداءً أفضل."
+ },
+ "ernie-char-8k": {
+ "description": "نموذج اللغة الكبير المخصص الذي طورته بايدو، مناسب لتطبيقات مثل NPC في الألعاب، محادثات خدمة العملاء، وأدوار الحوار، حيث يتميز بأسلوب شخصيات واضح ومتسق، وقدرة قوية على اتباع التعليمات، وأداء استدلال ممتاز."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "نموذج اللغة الكبير المخصص الذي طورته بايدو، مناسب لتطبيقات مثل NPC في الألعاب، محادثات خدمة العملاء، وأدوار الحوار، حيث يتميز بأسلوب شخصيات واضح ومتسق، وقدرة قوية على اتباع التعليمات، وأداء استدلال ممتاز."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite هو نموذج اللغة الكبير الخفيف الذي طورته بايدو، يجمع بين أداء النموذج الممتاز وأداء الاستدلال، مناسب للاستخدام مع بطاقات تسريع الذكاء الاصطناعي ذات القدرة الحاسوبية المنخفضة."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "نموذج اللغة الكبير الخفيف الذي طورته بايدو، يجمع بين أداء النموذج الممتاز وأداء الاستدلال، ويظهر أداءً أفضل من ERNIE Lite، مناسب للاستخدام مع بطاقات تسريع الذكاء الاصطناعي ذات القدرة الحاسوبية المنخفضة."
+ },
+ "ernie-novel-8k": {
+ "description": "نموذج اللغة الكبير العام الذي طورته بايدو، يظهر مزايا واضحة في القدرة على كتابة روايات، ويمكن استخدامه أيضًا في مشاهد مثل المسرحيات القصيرة والأفلام."
+ },
+ "ernie-speed-128k": {
+ "description": "نموذج اللغة الكبير عالي الأداء الذي طورته بايدو، والذي تم إصداره في عام 2024، يتمتع بقدرات عامة ممتازة، مناسب كنموذج أساسي للتعديل، مما يساعد على معالجة مشكلات المشاهد المحددة بشكل أفضل، ويظهر أداءً ممتازًا في الاستدلال."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "نموذج اللغة الكبير عالي الأداء الذي طورته بايدو، والذي تم إصداره في عام 2024، يتمتع بقدرات عامة ممتازة، ويظهر أداءً أفضل من ERNIE Speed، مناسب كنموذج أساسي للتعديل، مما يساعد على معالجة مشكلات المشاهد المحددة بشكل أفضل، ويظهر أداءً ممتازًا في الاستدلال."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny هو نموذج اللغة الكبير عالي الأداء الذي طورته بايدو، وتكاليف النشر والتعديل هي الأدنى بين نماذج سلسلة Wenxin."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (تعديل) يوفر أداءً مستقرًا وقابلًا للتعديل، وهو الخيار المثالي لحلول المهام المعقدة."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro يدعم ما يصل إلى 2 مليون توكن، وهو الخيار المثالي للنماذج المتوسطة الحجم متعددة الوسائط، مناسب لدعم المهام المعقدة من جوانب متعددة."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash يقدم ميزات وتحسينات من الجيل التالي، بما في ذلك سرعة فائقة، واستخدام أدوات أصلية، وتوليد متعدد الوسائط، ونافذة سياق تصل إلى 1M توكن."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash يقدم ميزات وتحسينات من الجيل التالي، بما في ذلك سرعة فائقة، واستخدام أدوات أصلية، وتوليد متعدد الوسائط، ونافذة سياق تصل إلى 1M توكن."
+ },
"gemini-2.0-flash-exp": {
"description": "جيميناي 2.0 فلاش إكسب هو أحدث نموذج ذكاء اصطناعي متعدد الوسائط من جوجل، يتمتع بميزات الجيل القادم، وسرعة فائقة، واستدعاء أدوات أصلية، وتوليد متعدد الوسائط."
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "نموذج Gemini 2.0 Flash، تم تحسينه لأهداف التكلفة المنخفضة والكمون المنخفض."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Exp هو أحدث نموذج تجريبي متعدد الوسائط من Google، يتمتع بميزات الجيل التالي، وسرعة فائقة، واستدعاء أدوات أصلية، وتوليد متعدد الوسائط."
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Exp هو أحدث نموذج ذكاء اصطناعي متعدد الوسائط التجريبي من Google، يتميز بخصائص الجيل التالي، وسرعة فائقة، واستدعاء أدوات أصلية، وتوليد متعدد الوسائط."
},
- "gemini-exp-1114": {
- "description": "جيمني إكسب 1114 هو أحدث نموذج ذكاء اصطناعي متعدد الوسائط تجريبي من Google، يتميز بقدرة معالجة سريعة، ويدعم إدخالات النصوص والصور والفيديو، مما يجعله مناسبًا للتوسع الفعال في مهام متعددة."
- },
- "gemini-exp-1121": {
- "description": "جمني إكسب 1121 هو أحدث نموذج تجريبي متعدد الوسائط من جوجل، يتمتع بقدرة معالجة سريعة، ويدعم إدخال النصوص والصور والفيديو، مما يجعله مناسبًا للتوسع الفعال في مجموعة متنوعة من المهام."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental هو أحدث نموذج ذكاء اصطناعي متعدد الوسائط التجريبي من Google، مع تحسينات ملحوظة في الجودة مقارنة بالإصدارات السابقة، خاصة في المعرفة العالمية، والبرمجة، والسياقات الطويلة."
},
"gemini-exp-1206": {
"description": "جيميني إكسب 1206 هو أحدث نموذج ذكاء اصطناعي متعدد الوسائط من جوجل، مع تحسينات في الجودة مقارنةً بالإصدارات السابقة."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus يتمتع بقدرة على فهم محتوى الفيديو والصور المتعددة، مما يجعله مناسبًا للمهام متعددة الوسائط."
},
+ "glm-zero-preview": {
+ "description": "يمتلك GLM-Zero-Preview قدرة قوية على الاستدلال المعقد، ويظهر أداءً ممتازًا في مجالات الاستدلال المنطقي، والرياضيات، والبرمجة."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash يقدم ميزات وتحسينات من الجيل التالي، بما في ذلك سرعة فائقة، واستخدام أدوات أصلية، وتوليد متعدد الوسائط، ونافذة سياق تصل إلى 1M توكن."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental هو أحدث نموذج ذكاء اصطناعي متعدد الوسائط التجريبي من Google، مع تحسينات ملحوظة في الجودة مقارنة بالإصدارات السابقة، خاصة في المعرفة العالمية، والبرمجة، والسياقات الطويلة."
+ },
"google/gemini-flash-1.5": {
"description": "يقدم Gemini 1.5 Flash قدرات معالجة متعددة الوسائط محسّنة، مناسبة لمجموعة متنوعة من سيناريوهات المهام المعقدة."
},
"google/gemini-pro-1.5": {
"description": "يجمع Gemini 1.5 Pro بين أحدث تقنيات التحسين، مما يوفر قدرة معالجة بيانات متعددة الوسائط بشكل أكثر كفاءة."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 هو نموذج فعال أطلقته Google، يغطي مجموعة متنوعة من سيناريوهات التطبيقات من التطبيقات الصغيرة إلى معالجة البيانات المعقدة."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 تستمر في مفهوم التصميم الخفيف والفعال."
},
"google/gemma-2-2b-it": {
"description": "نموذج تحسين التعليمات الخفيف من Google"
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 هو نموذج فعال أطلقته Google، يغطي مجموعة متنوعة من سيناريوهات التطبيقات من التطبيقات الصغيرة إلى معالجة البيانات المعقدة."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 هو سلسلة نماذج نصية مفتوحة المصدر خفيفة الوزن من Google."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "نموذج GPT 3.5 Turbo، مناسب لمجموعة متنوعة من مهام توليد وفهم النصوص، يشير حاليًا إلى gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "نموذج GPT 3.5 Turbo 16k، نموذج توليد نصوص عالي السعة، مناسب للمهام المعقدة."
+ },
"gpt-3.5-turbo-instruct": {
"description": "نموذج GPT 3.5 Turbo، مناسب لمجموعة متنوعة من مهام توليد وفهم النصوص، يشير حاليًا إلى gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "تشات جي بي تي-4o هو نموذج ديناميكي يتم تحديثه في الوقت الفعلي للحفاظ على أحدث إصدار. يجمع بين الفهم اللغوي القوي وقدرة التوليد، مما يجعله مناسبًا لتطبيقات واسعة النطاق، بما في ذلك خدمة العملاء والتعليم والدعم الفني."
},
+ "gpt-4o-audio-preview": {
+ "description": "نموذج GPT-4o Audio، يدعم إدخال وإخراج الصوت."
+ },
"gpt-4o-mini": {
"description": "نموذج GPT-4o mini هو أحدث نموذج أطلقته OpenAI بعد GPT-4 Omni، ويدعم إدخال الصور والنصوص وإخراج النصوص. كأحد نماذجهم المتقدمة الصغيرة، فهو أرخص بكثير من النماذج الرائدة الأخرى في الآونة الأخيرة، وأرخص بأكثر من 60% من GPT-3.5 Turbo. يحتفظ بذكاء متقدم مع قيمة ممتازة. حصل GPT-4o mini على 82% في اختبار MMLU، وهو حاليًا يتفوق على GPT-4 في تفضيلات الدردشة."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "الإصدار المصغر الفوري من GPT-4o، يدعم إدخال وإخراج الصوت والنص في الوقت الحقيقي."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "الإصدار الفوري من GPT-4o، يدعم إدخال وإخراج الصوت والنص في الوقت الحقيقي."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "الإصدار الفوري من GPT-4o، يدعم إدخال وإخراج الصوت والنص في الوقت الحقيقي."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "الإصدار الفوري من GPT-4o، يدعم إدخال وإخراج الصوت والنص في الوقت الحقيقي."
+ },
"grok-2-1212": {
"description": "لقد تم تحسين هذا النموذج في الدقة، والامتثال للتعليمات، والقدرة على التعامل مع لغات متعددة."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "نموذج Hunyuan الأحدث من نوع MOE FunctionCall، تم تدريبه على بيانات FunctionCall عالية الجودة، مع نافذة سياق تصل إلى 32K، ويحتل مرتبة متقدمة في مؤشرات التقييم عبر عدة أبعاد."
},
+ "hunyuan-large": {
+ "description": "نموذج Hunyuan-large يحتوي على حوالي 389 مليار معلمة، مع حوالي 52 مليار معلمة نشطة، وهو أكبر نموذج MoE مفتوح المصدر في الصناعة من حيث حجم المعلمات وأفضلها من حيث الأداء."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "يتفوق في معالجة المهام الطويلة مثل تلخيص الوثائق والأسئلة والأجوبة المتعلقة بالوثائق، كما يمتلك القدرة على معالجة مهام إنشاء النصوص العامة. يظهر أداءً ممتازًا في تحليل وإنشاء النصوص الطويلة، مما يمكنه من التعامل بفعالية مع متطلبات معالجة المحتوى الطويل المعقد والمفصل."
+ },
"hunyuan-lite": {
"description": "تم الترقية إلى هيكل MOE، مع نافذة سياق تصل إلى 256k، متفوقًا على العديد من النماذج مفتوحة المصدر في تقييمات NLP، البرمجة، الرياضيات، والصناعات."
},
+ "hunyuan-lite-vision": {
+ "description": "نموذج مختلط حديث بقدرة 7 مليار معلمة، مع نافذة سياقية 32K، يدعم المحادثات متعددة الوسائط في السيناريوهات الصينية والإنجليزية، والتعرف على كائنات الصور، وفهم جداول الوثائق، والرياضيات متعددة الوسائط، ويتفوق في مؤشرات التقييم على نماذج المنافسة ذات 7 مليار معلمة في عدة أبعاد."
+ },
"hunyuan-pro": {
"description": "نموذج نصوص طويلة MOE-32K بحجم تريليون من المعلمات. يحقق مستوى رائد مطلق في مختلف المعايير، مع القدرة على التعامل مع التعليمات المعقدة والاستدلال، ويتميز بقدرات رياضية معقدة، ويدعم استدعاء الوظائف، مع تحسينات رئيسية في مجالات الترجمة متعددة اللغات، المالية، القانونية، والرعاية الصحية."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "يستخدم استراتيجية توجيه أفضل، مع تخفيف مشكلات التوازن في الحمل وتوافق الخبراء. في مجال النصوص الطويلة، تصل نسبة مؤشر البحث إلى 99.9%. MOE-256K يحقق اختراقًا إضافيًا في الطول والأداء، مما يوسع بشكل كبير طول المدخلات الممكنة."
},
+ "hunyuan-standard-vision": {
+ "description": "نموذج متعدد الوسائط حديث يدعم الإجابة بعدة لغات، مع توازن في القدرات بين الصينية والإنجليزية."
+ },
"hunyuan-turbo": {
"description": "نسخة المعاينة من الجيل الجديد من نموذج اللغة الكبير، يستخدم هيكل نموذج الخبراء المختلط (MoE) الجديد، مما يوفر كفاءة استدلال أسرع وأداء أقوى مقارنة بـ hunyuan-pro."
},
+ "hunyuan-turbo-20241120": {
+ "description": "الإصدار الثابت من hunyuan-turbo بتاريخ 20 نوفمبر 2024، وهو إصدار يقع بين hunyuan-turbo و hunyuan-turbo-latest."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "تحسينات في هذا الإصدار: توجيه البيانات، مما يعزز بشكل كبير قدرة النموذج على التعميم؛ تحسين كبير في القدرات الرياضية، البرمجية، وقدرات الاستدلال المنطقي؛ تحسين القدرات المتعلقة بفهم النصوص والكلمات؛ تحسين جودة إنشاء محتوى النص."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "تحسين تجربة شاملة، بما في ذلك فهم اللغة الطبيعية، إنشاء النصوص، الدردشة، الأسئلة والأجوبة المعرفية، الترجمة، والمجالات الأخرى؛ تعزيز الطابع الإنساني، وتحسين الذكاء العاطفي للنموذج؛ تعزيز قدرة النموذج على توضيح النوايا الغامضة؛ تحسين القدرة على معالجة الأسئلة المتعلقة بتحليل الكلمات؛ تحسين جودة الإبداع والتفاعل؛ تعزيز تجربة التفاعل المتعدد الجولات."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "نموذج اللغة البصرية الرائد من الجيل الجديد، يستخدم هيكل نموذج الخبراء المختلط (MoE) الجديد، مع تحسين شامل في القدرات المتعلقة بفهم النصوص والصور، وإنشاء المحتوى، والأسئلة والأجوبة المعرفية، والتحليل والاستدلال مقارنة بالنماذج السابقة."
+ },
"hunyuan-vision": {
"description": "نموذج Hunyuan الأحدث متعدد الوسائط، يدعم إدخال الصور والنصوص لتوليد محتوى نصي."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "سلسلة نماذجنا الأحدث، تتمتع بأداء استدلال ممتاز، تدعم طول سياق يصل إلى 1 مليون، بالإضافة إلى قدرة أقوى على اتباع التعليمات واستدعاء الأدوات."
},
+ "internlm3-latest": {
+ "description": "سلسلة نماذجنا الأحدث، تتمتع بأداء استدلال ممتاز، تتصدر نماذج المصدر المفتوح من نفس الفئة. تشير بشكل افتراضي إلى أحدث نماذج سلسلة InternLM3 التي تم إصدارها."
+ },
+ "jina-deepsearch-v1": {
+ "description": "البحث العميق يجمع بين البحث عبر الإنترنت، والقراءة، والاستدلال، مما يتيح إجراء تحقيق شامل. يمكنك اعتباره وكيلًا يتولى مهام البحث الخاصة بك - حيث يقوم بإجراء بحث واسع النطاق ويخضع لعدة تكرارات قبل تقديم الإجابة. تتضمن هذه العملية بحثًا مستمرًا، واستدلالًا، وحل المشكلات من زوايا متعددة. وهذا يختلف اختلافًا جوهريًا عن النماذج الكبيرة القياسية التي تولد الإجابات مباشرة من البيانات المدربة مسبقًا، وكذلك عن أنظمة RAG التقليدية التي تعتمد على البحث السطحي لمرة واحدة."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM هو نموذج لغوي تجريبي محدد المهام، تم تدريبه ليتماشى مع مبادئ علوم التعلم، يمكنه اتباع التعليمات النظامية في سيناريوهات التعليم والتعلم، ويعمل كمدرب خبير."
},
"lite": {
"description": "سبارك لايت هو نموذج لغوي كبير خفيف الوزن، يتميز بتأخير منخفض للغاية وكفاءة عالية في المعالجة، وهو مجاني تمامًا ومفتوح، ويدعم وظيفة البحث عبر الإنترنت في الوقت الحقيقي. تجعل خصائص استجابته السريعة منه مثاليًا لتطبيقات الاستدلال على الأجهزة ذات القدرة الحاسوبية المنخفضة وضبط النماذج، مما يوفر للمستخدمين قيمة ممتازة من حيث التكلفة وتجربة ذكية، خاصة في مجالات الأسئلة والأجوبة المعرفية، وتوليد المحتوى، وسيناريوهات البحث."
},
- "llama-3.1-70b-instruct": {
- "description": "نموذج Llama 3.1 70B للتعليمات، يتمتع بـ 70B من المعلمات، قادر على تقديم أداء ممتاز في مهام توليد النصوص الكبيرة والتعليمات."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B يوفر قدرة استدلال ذكائي أقوى، مناسب للتطبيقات المعقدة، يدعم معالجة حسابية ضخمة ويضمن الكفاءة والدقة."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B هو نموذج عالي الأداء، يوفر قدرة سريعة على توليد النصوص، مما يجعله مثاليًا لمجموعة من التطبيقات التي تتطلب كفاءة كبيرة وتكلفة فعالة."
},
- "llama-3.1-8b-instruct": {
- "description": "نموذج Llama 3.1 8B للتعليمات، يتمتع بـ 8B من المعلمات، يدعم تنفيذ مهام التعليمات بكفاءة، ويوفر قدرة ممتازة على توليد النصوص."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "نموذج Llama 3.1 Sonar Huge Online، يتمتع بـ 405B من المعلمات، يدعم طول سياق حوالي 127,000 علامة، مصمم لتطبيقات دردشة معقدة عبر الإنترنت."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "نموذج Llama 3.1 Sonar Large Chat، يتمتع بـ 70B من المعلمات، يدعم طول سياق حوالي 127,000 علامة، مناسب لمهام دردشة غير متصلة معقدة."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "نموذج Llama 3.1 Sonar Large Online، يتمتع بـ 70B من المعلمات، يدعم طول سياق حوالي 127,000 علامة، مناسب لمهام دردشة عالية السعة ومتنوعة."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "نموذج Llama 3.1 Sonar Small Chat، يتمتع بـ 8B من المعلمات، مصمم للدردشة غير المتصلة، يدعم طول سياق حوالي 127,000 علامة."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "نموذج Llama 3.1 Sonar Small Online، يتمتع بـ 8B من المعلمات، يدعم طول سياق حوالي 127,000 علامة، مصمم للدردشة عبر الإنترنت، قادر على معالجة تفاعلات نصية متنوعة بكفاءة."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 مصمم للتعامل مع المهام التي تجمع بين البيانات البصرية والنصية. يظهر أداءً ممتازًا في مهام وصف الصور والأسئلة البصرية، متجاوزًا الفجوة بين توليد اللغة والاستدلال البصري."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 هو النموذج الأكثر تقدمًا في سلسلة Llama، وهو نموذج لغوي مفتوح المصدر متعدد اللغات، يوفر تجربة أداء تنافس نموذج 405B بتكلفة منخفضة للغاية. يعتمد على هيكل Transformer، وتم تحسين فائدته وأمانه من خلال التعديل الدقيق تحت الإشراف (SFT) والتعلم المعزز من خلال التغذية الراجعة البشرية (RLHF). تم تحسين نسخة التعديل الخاصة به لتكون مثالية للحوار متعدد اللغات، حيث يتفوق في العديد من المعايير الصناعية على العديد من نماذج الدردشة المفتوحة والمغلقة. تاريخ انتهاء المعرفة هو ديسمبر 2023."
+ },
"llama-3.3-70b-versatile": {
"description": "ميتّا لاما 3.3 هو نموذج لغة كبير متعدد اللغات (LLM) يضم 70 مليار (إدخال نص/إخراج نص) من النموذج المدرب مسبقًا والمعدل وفقًا للتعليمات. تم تحسين نموذج لاما 3.3 المعدل وفقًا للتعليمات للاستخدامات الحوارية متعددة اللغات ويتفوق على العديد من النماذج المتاحة مفتوحة المصدر والمغلقة في المعايير الصناعية الشائعة."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "تم تصميم LLaMA 3.2 لمعالجة المهام التي تجمع بين البيانات البصرية والنصية. إنه يبرز في مهام وصف الصور والأسئلة البصرية، متجاوزًا الفجوة بين توليد اللغة واستدلال الرؤية."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 هو أحدث نموذج لغوي مفتوح المصدر متعدد اللغات من سلسلة Llama، يقدم تجربة مشابهة لأداء نموذج 405B بتكلفة منخفضة للغاية. يعتمد على هيكل Transformer، وتم تحسينه من خلال التعديل الإشرافي (SFT) والتعلم المعزز من خلال ردود الفعل البشرية (RLHF) لتعزيز الفائدة والأمان. تم تحسين نسخة التعديل الخاصة به للحوار متعدد اللغات، حيث يتفوق في العديد من المعايير الصناعية على العديد من نماذج الدردشة المفتوحة والمغلقة. تاريخ انتهاء المعرفة هو ديسمبر 2023."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "نموذج Meta Llama 3.3 متعدد اللغات (LLM) هو نموذج توليد تم تدريبه مسبقًا وضبطه على التعليمات في 70B (إدخال نص/إخراج نص). تم تحسين نموذج Llama 3.3 المعدل على التعليمات لحالات استخدام الحوار متعدد اللغات، ويتفوق على العديد من نماذج الدردشة المفتوحة والمغلقة المتاحة في المعايير الصناعية الشائعة."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "تم تصميم LLaMA 3.2 لمعالجة المهام التي تجمع بين البيانات البصرية والنصية. إنه يبرز في مهام وصف الصور والأسئلة البصرية، متجاوزًا الفجوة بين توليد اللغة واستدلال الرؤية."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "نموذج Llama 3.1 Turbo 405B يوفر دعمًا كبيرًا للسياق لمعالجة البيانات الكبيرة، ويظهر أداءً بارزًا في تطبيقات الذكاء الاصطناعي على نطاق واسع."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 هو نموذج رائد أطلقته Meta، يدعم ما يصل إلى 405B من المعلمات، ويمكن تطبيقه في مجالات المحادثات المعقدة، والترجمة متعددة اللغات، وتحليل البيانات."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B يوفر دعمًا فعالًا للحوار متعدد اللغات."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "تم تصميم LLaMA 3.2 لمعالجة المهام التي تجمع بين البيانات البصرية والنصية. إنه يتفوق في مهام وصف الصور والأسئلة البصرية، متجاوزًا الفجوة بين توليد اللغة والاستدلال البصري."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 هو النموذج الأكثر تقدمًا في سلسلة Llama، وهو نموذج لغوي مفتوح المصدر متعدد اللغات، يوفر تجربة أداء تنافس نموذج 405B بتكلفة منخفضة للغاية. يعتمد على هيكل Transformer، وتم تحسين فائدته وأمانه من خلال التعديل الدقيق تحت الإشراف (SFT) والتعلم المعزز من خلال التغذية الراجعة البشرية (RLHF). تم تحسين نسخة التعديل الخاصة به لتكون مثالية للحوار متعدد اللغات، حيث يتفوق في العديد من المعايير الصناعية على العديد من نماذج الدردشة المفتوحة والمغلقة. تاريخ انتهاء المعرفة هو ديسمبر 2023."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 هو النموذج الأكثر تقدمًا في سلسلة Llama، وهو نموذج لغوي مفتوح المصدر متعدد اللغات، يوفر تجربة أداء تنافس نموذج 405B بتكلفة منخفضة للغاية. يعتمد على هيكل Transformer، وتم تحسين فائدته وأمانه من خلال التعديل الدقيق تحت الإشراف (SFT) والتعلم المعزز من خلال التغذية الراجعة البشرية (RLHF). تم تحسين نسخة التعديل الخاصة به لتكون مثالية للحوار متعدد اللغات، حيث يتفوق في العديد من المعايير الصناعية على العديد من نماذج الدردشة المفتوحة والمغلقة. تاريخ انتهاء المعرفة هو ديسمبر 2023."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "نموذج Meta Llama 3.1 405B Instruct هو أكبر وأقوى نموذج في مجموعة نماذج Llama 3.1 Instruct، وهو نموذج متقدم للغاية لتوليد البيانات والحوار، ويمكن استخدامه كأساس للتدريب المستمر أو التخصيص في مجالات معينة. توفر Llama 3.1 نماذج لغوية كبيرة متعددة اللغات (LLMs) وهي مجموعة من النماذج المدربة مسبقًا والمعدلة وفقًا للتعليمات، بما في ذلك أحجام 8B و70B و405B (إدخال/إخراج نصي). تم تحسين نماذج النص المعدلة وفقًا للتعليمات (8B و70B و405B) لحالات الاستخدام الحوارية متعددة اللغات، وقد تفوقت في العديد من اختبارات المعايير الصناعية الشائعة على العديد من نماذج الدردشة مفتوحة المصدر المتاحة. تم تصميم Llama 3.1 للاستخدام التجاري والبحثي في عدة لغات. نماذج النص المعدلة وفقًا للتعليمات مناسبة للدردشة الشبيهة بالمساعد، بينما يمكن للنماذج المدربة مسبقًا التكيف مع مجموعة متنوعة من مهام توليد اللغة الطبيعية. تدعم نماذج Llama 3.1 أيضًا تحسين نماذج أخرى باستخدام مخرجاتها، بما في ذلك توليد البيانات الاصطناعية والتنقيح. Llama 3.1 هو نموذج لغوي ذاتي التكرار يستخدم بنية المحولات المحسّنة. تستخدم النسخ المعدلة التعلم المعزز مع التغذية الراجعة البشرية (RLHF) لتلبية تفضيلات البشر فيما يتعلق بالمساعدة والأمان."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 هو نموذج لغوي كبير مفتوح (LLM) موجه للمطورين والباحثين والشركات، يهدف إلى مساعدتهم في بناء وتجربة وتوسيع أفكارهم في الذكاء الاصطناعي بشكل مسؤول. كجزء من نظام الابتكار المجتمعي العالمي، فهو مثالي للأجهزة ذات القدرة الحاسوبية والموارد المحدودة، والأجهزة الطرفية، وأوقات التدريب الأسرع."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "نموذج لغوي متقدم، يدعم توليد البيانات الاصطناعية، وتقطير المعرفة، والاستدلال، مناسب للدردشة، والبرمجة، والمهام الخاصة."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "يمكنه تمكين المحادثات المعقدة، ويتميز بفهم سياقي ممتاز، وقدرات استدلال، وقدرة على توليد النصوص."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "نموذج متقدم من الطراز الأول، يتمتع بفهم اللغة، وقدرات استدلال ممتازة، وقدرة على توليد النصوص."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "نموذج متقدم للرؤية واللغة، بارع في إجراء استدلال عالي الجودة من الصور."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "نموذج لغوي صغير متقدم، يتمتع بفهم اللغة، وقدرات استدلال ممتازة، وقدرة على توليد النصوص."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "نموذج لغوي صغير متقدم، يتمتع بفهم اللغة، وقدرات استدلال ممتازة، وقدرة على توليد النصوص."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "نموذج متقدم للرؤية واللغة، بارع في إجراء استدلال عالي الجودة من الصور."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "نموذج لغوي متقدم، بارع في الاستدلال، والرياضيات، والمعرفة العامة، واستدعاء الدوال."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 هو نموذج لغوي تقدمه Microsoft AI، يتميز بأداء ممتاز في المحادثات المعقدة، واللغات المتعددة، والاستدلال، ومساعدات الذكاء."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K هو نموذج يتمتع بقدرة معالجة سياقات طويلة جدًا، مناسب لتوليد نصوص طويلة جدًا، يلبي احتياجات المهام المعقدة، قادر على معالجة ما يصل إلى 128,000 توكن، مما يجعله مثاليًا للبحث، والأكاديميات، وتوليد الوثائق الكبيرة."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "نموذج Kimi البصري (بما في ذلك moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview وغيرها) قادر على فهم محتوى الصور، بما في ذلك النصوص والألوان وأشكال الأجسام."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K يوفر قدرة معالجة سياقات متوسطة الطول، قادر على معالجة 32,768 توكن، مناسب بشكل خاص لتوليد مجموعة متنوعة من الوثائق الطويلة والحوار المعقد، ويستخدم في إنشاء المحتوى، وتوليد التقارير، وأنظمة الحوار."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "نموذج Kimi البصري (بما في ذلك moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview وغيرها) قادر على فهم محتوى الصور، بما في ذلك النصوص والألوان وأشكال الأجسام."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K مصمم خصيصًا لتوليد مهام النصوص القصيرة، يتمتع بأداء معالجة فعال، قادر على معالجة 8,192 توكن، مما يجعله مثاليًا للحوار القصير، والتدوين السريع، وتوليد المحتوى السريع."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "نموذج Kimi البصري (بما في ذلك moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview وغيرها) قادر على فهم محتوى الصور، بما في ذلك النصوص والألوان وأشكال الأجسام."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B هو إصدار مطور من Nous Hermes 2، ويحتوي على أحدث مجموعات البيانات المطورة داخليًا."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B هو نموذج لغوي كبير مُخصص من NVIDIA، يهدف إلى تحسين استجابة LLM لمساعدة استفسارات المستخدمين."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B هو نموذج لغوي كبير مخصص من NVIDIA، يهدف إلى تحسين استجابة LLM لمساعدة استفسارات المستخدمين. لقد أظهر النموذج أداءً ممتازًا في اختبارات المعايير مثل Arena Hard وAlpacaEval 2 LC وGPT-4-Turbo MT-Bench، حيث احتل المرتبة الأولى في جميع اختبارات المحاذاة التلقائية الثلاثة حتى 1 أكتوبر 2024. تم تدريب النموذج باستخدام RLHF (خاصة REINFORCE) وLlama-3.1-Nemotron-70B-Reward وHelpSteer2-Preference على أساس نموذج Llama-3.1-70B-Instruct."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "نموذج لغوي فريد، يقدم دقة وأداء لا مثيل لهما."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B هو نموذج لغوي كبير مخصص من NVIDIA، مصمم لتحسين فائدة الاستجابات التي يولدها LLM."
+ },
"o1": {
"description": "يركز على الاستدلال المتقدم وحل المشكلات المعقدة، بما في ذلك المهام الرياضية والعلمية. مثالي للتطبيقات التي تتطلب فهمًا عميقًا للسياق وإدارة سير العمل."
},
- "o1-2024-12-17": {
- "description": "o1 هو نموذج الاستدلال الجديد من OpenAI، يدعم إدخال النصوص والصور وإخراج النصوص، وهو مناسب للمهام المعقدة التي تتطلب معرفة عامة واسعة. يحتوي هذا النموذج على 200K من السياق وتاريخ انتهاء المعرفة في أكتوبر 2023."
- },
"o1-mini": {
"description": "o1-mini هو نموذج استدلال سريع وفعال من حيث التكلفة مصمم لتطبيقات البرمجة والرياضيات والعلوم. يحتوي هذا النموذج على 128K من السياق وتاريخ انتهاء المعرفة في أكتوبر 2023."
},
"o1-preview": {
"description": "o1 هو نموذج استدلال جديد من OpenAI، مناسب للمهام المعقدة التي تتطلب معرفة عامة واسعة. يحتوي هذا النموذج على 128K من السياق وتاريخ انتهاء المعرفة في أكتوبر 2023."
},
+ "o3-mini": {
+ "description": "o3-mini هو أحدث نموذج استدلال صغير لدينا، يقدم ذكاءً عالياً تحت نفس تكاليف التأخير والأداء مثل o1-mini."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba هو نموذج لغة Mamba 2 يركز على توليد الشيفرة، ويوفر دعمًا قويًا لمهام الشيفرة المتقدمة والاستدلال."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini هو أحدث نموذج من OpenAI تم إطلاقه بعد GPT-4 Omni، ويدعم إدخال النصوص والصور وإخراج النصوص. كأحد نماذجهم المتقدمة الصغيرة، فهو أرخص بكثير من النماذج الرائدة الأخرى في الآونة الأخيرة، وأرخص بأكثر من 60% من GPT-3.5 Turbo. يحتفظ بذكاء متقدم مع قيمة ممتازة. حصل GPT-4o mini على 82% في اختبار MMLU، وهو حاليًا يتفوق على GPT-4 في تفضيلات الدردشة."
},
- "openai/o1": {
- "description": "o1 هو نموذج الاستدلال الجديد من OpenAI، يدعم إدخال النصوص والصور وإخراج النصوص، وهو مناسب للمهام المعقدة التي تتطلب معرفة عامة واسعة. يحتوي هذا النموذج على 200K من السياق وتاريخ انتهاء المعرفة في أكتوبر 2023."
- },
"openai/o1-mini": {
"description": "o1-mini هو نموذج استدلال سريع وفعال من حيث التكلفة مصمم لتطبيقات البرمجة والرياضيات والعلوم. يحتوي هذا النموذج على 128K من السياق وتاريخ انتهاء المعرفة في أكتوبر 2023."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "سبارك برو 128K مزود بقدرة معالجة سياق كبيرة جدًا، قادر على معالجة ما يصل إلى 128K من معلومات السياق، مما يجعله مناسبًا بشكل خاص للتحليل الشامل ومعالجة الروابط المنطقية طويلة الأمد في المحتوى الطويل، ويمكنه تقديم منطق سلس ومتسق ودعم متنوع للاقتباسات في الاتصالات النصية المعقدة."
},
+ "qvq-72b-preview": {
+ "description": "نموذج QVQ هو نموذج بحث تجريبي تم تطويره بواسطة فريق Qwen، يركز على تعزيز قدرات الاستدلال البصري، خاصة في مجال الاستدلال الرياضي."
+ },
"qwen-coder-plus-latest": {
"description": "نموذج كود Qwen الشامل."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "نموذج اللغة البصرية الكبير Qwen. مقارنةً بالنسخة المحسّنة، تعزز مرة أخرى من قدرة الاستدلال البصري وقدرة اتباع التعليمات، مما يوفر مستوى أعلى من الإدراك البصري والمعرفة."
},
+ "qwen-vl-ocr-latest": {
+ "description": "نموذج OCR الخاص بـ Tongyi Qianwen هو نموذج استخراج النصوص، يركز على قدرة استخراج النصوص من أنواع الصور مثل الوثائق، الجداول، الأسئلة، والنصوص المكتوبة بخط اليد. يمكنه التعرف على عدة لغات، بما في ذلك: الصينية، الإنجليزية، الفرنسية، اليابانية، الكورية، الألمانية، الروسية، الإيطالية، الفيتنامية، والعربية."
+ },
"qwen-vl-plus-latest": {
"description": "نسخة محسّنة من نموذج اللغة البصرية الكبير Qwen. تعزز بشكل كبير من قدرة التعرف على التفاصيل وقدرة التعرف على النصوص، وتدعم دقة تصل إلى أكثر من مليون بكسل وأبعاد صور بأي نسبة عرض إلى ارتفاع."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 هو سلسلة جديدة من نماذج اللغة الكبيرة، تتمتع بقدرات فهم وتوليد أقوى."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "نموذج لغوي موجه للغة الصينية والإنجليزية، يستهدف مجالات اللغة، والبرمجة، والرياضيات، والاستدلال، وغيرها."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "نموذج لغوي متقدم، يدعم توليد الشيفرة، والاستدلال، والإصلاح، ويغطي لغات البرمجة الرئيسية."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "نموذج قوي للبرمجة متوسطة الحجم، يدعم طول سياق يصل إلى 32K، بارع في البرمجة متعددة اللغات."
+ },
"qwen2": {
"description": "Qwen2 هو نموذج لغوي كبير من الجيل الجديد من Alibaba، يدعم أداءً ممتازًا لتلبية احتياجات التطبيقات المتنوعة."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "نموذج Qwen 2.5 مفتوح المصدر بحجم 14B."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "نموذج بحجم 72B مفتوح المصدر من Tongyi Qianwen 2.5."
+ },
"qwen2.5-32b-instruct": {
"description": "نموذج Qwen 2.5 مفتوح المصدر بحجم 32B."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "نموذج Qwen-Math يتمتع بقدرات قوية في حل المسائل الرياضية."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "تحسين شامل في اتباع التعليمات، الرياضيات، حل المشكلات، والبرمجة، وزيادة قدرة التعرف على العناصر البصرية، يدعم تنسيقات متعددة لتحديد العناصر البصرية بدقة، ويدعم فهم ملفات الفيديو الطويلة (حتى 10 دقائق) وتحديد اللحظات الزمنية بدقة، قادر على فهم التسلسل الزمني والسرعة، يدعم التحكم في أنظمة التشغيل أو الوكلاء المحمولة بناءً على قدرات التحليل والتحديد، قوي في استخراج المعلومات الرئيسية وإخراج البيانات بتنسيق Json، هذه النسخة هي النسخة 72B، وهي الأقوى في هذه السلسلة."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "تحسين شامل في اتباع التعليمات، الرياضيات، حل المشكلات، والبرمجة، وزيادة قدرة التعرف على العناصر البصرية، يدعم تنسيقات متعددة لتحديد العناصر البصرية بدقة، ويدعم فهم ملفات الفيديو الطويلة (حتى 10 دقائق) وتحديد اللحظات الزمنية بدقة، قادر على فهم التسلسل الزمني والسرعة، يدعم التحكم في أنظمة التشغيل أو الوكلاء المحمولة بناءً على قدرات التحليل والتحديد، قوي في استخراج المعلومات الرئيسية وإخراج البيانات بتنسيق Json، هذه النسخة هي النسخة 72B، وهي الأقوى في هذه السلسلة."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 هو الجيل الجديد من نماذج اللغة الكبيرة من Alibaba، يدعم احتياجات التطبيقات المتنوعة بأداء ممتاز."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro هو نموذج LLM عالي الذكاء تم إطلاقه من قبل Upstage، يركز على قدرة اتباع التعليمات على وحدة معالجة الرسوميات الواحدة، وسجل IFEval فوق 80. حاليًا يدعم اللغة الإنجليزية، ومن المقرر إصدار النسخة الرسمية في نوفمبر 2024، مع توسيع دعم اللغات وطول السياق."
},
+ "sonar": {
+ "description": "منتج بحث خفيف الوزن يعتمد على سياق البحث، أسرع وأرخص من Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "منتج بحث متقدم يدعم سياق البحث، مع دعم للاستعلامات المتقدمة والمتابعة."
+ },
+ "sonar-reasoning": {
+ "description": "منتج API الجديد المدعوم من نموذج الاستدلال من DeepSeek."
+ },
"step-1-128k": {
"description": "يوفر توازنًا بين الأداء والتكلفة، مناسب لمجموعة متنوعة من السيناريوهات."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "يمتلك هذا النموذج قدرة قوية على فهم الفيديو."
},
+ "step-1o-vision-32k": {
+ "description": "يمتلك هذا النموذج قدرة قوية على فهم الصور. مقارنةً بسلسلة نماذج step-1v، فإنه يتمتع بأداء بصري أقوى."
+ },
"step-1v-32k": {
"description": "يدعم المدخلات البصرية، يعزز تجربة التفاعل متعدد الوسائط."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "يدعم تفاعلات سياق كبيرة، مناسب لمشاهد الحوار المعقدة."
},
+ "step-2-mini": {
+ "description": "نموذج كبير سريع يعتمد على بنية الانتباه الجديدة MFA، يحقق نتائج مشابهة لـ step1 بتكلفة منخفضة جداً، مع الحفاظ على قدرة أعلى على المعالجة وزمن استجابة أسرع. يمكنه التعامل مع المهام العامة، ويتميز بقدرات قوية في البرمجة."
+ },
+ "taichu2_mm": {
+ "description": "يجمع بين فهم الصور، ونقل المعرفة، والاستدلال المنطقي، ويظهر أداءً بارزًا في مجال الأسئلة والأجوبة النصية والصورية."
+ },
"taichu_llm": {
"description": "نموذج اللغة الكبير TaiChu يتمتع بقدرات قوية في فهم اللغة، بالإضافة إلى إنشاء النصوص، والإجابة على الأسئلة، وبرمجة الأكواد، والحسابات الرياضية، والاستدلال المنطقي، وتحليل المشاعر، وتلخيص النصوص. يجمع بشكل مبتكر بين التدريب المسبق على البيانات الضخمة والمعرفة الغنية من مصادر متعددة، من خلال تحسين تقنيات الخوارزميات باستمرار واستيعاب المعرفة الجديدة من البيانات النصية الضخمة، مما يحقق تطورًا مستمرًا في أداء النموذج. يوفر للمستخدمين معلومات وخدمات أكثر سهولة وتجربة أكثر ذكاءً."
},
+ "text-embedding-3-large": {
+ "description": "أقوى نموذج لتضمين النصوص، مناسب للمهام الإنجليزية وغير الإنجليزية."
+ },
+ "text-embedding-3-small": {
+ "description": "نموذج التضمين من الجيل الجديد، فعال واقتصادي، مناسب لاسترجاع المعرفة وتطبيقات RAG وغيرها."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) يوفر قدرة حسابية معززة من خلال استراتيجيات فعالة وهندسة نموذجية."
},
+ "tts-1": {
+ "description": "أحدث نموذج لتحويل النص إلى كلام، تم تحسينه للسرعة في السيناريوهات الحية."
+ },
+ "tts-1-hd": {
+ "description": "أحدث نموذج لتحويل النص إلى كلام، تم تحسينه للجودة."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) مناسب لمهام التعليمات الدقيقة، يوفر قدرة معالجة لغوية ممتازة."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet يرفع المعايير الصناعية، حيث يتفوق على نماذج المنافسين وClaude 3 Opus، ويظهر أداءً ممتازًا في تقييمات واسعة، مع سرعة وتكلفة تتناسب مع نماذجنا المتوسطة."
},
+ "whisper-1": {
+ "description": "نموذج التعرف على الصوت العام، يدعم التعرف على الصوت متعدد اللغات، والترجمة الصوتية، والتعرف على اللغات."
+ },
"wizardlm2": {
"description": "WizardLM 2 هو نموذج لغوي تقدمه Microsoft AI، يتميز بأداء ممتاز في الحوار المعقد، واللغات المتعددة، والاستدلال، والمساعدين الذكيين."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "نموذج لمهام الرؤية المعقدة، يوفر قدرة عالية على فهم وتحليل الصور."
+ },
+ "yi-vision-v2": {
+ "description": "نموذج مهام بصرية معقدة، يوفر فهمًا عالي الأداء وقدرات تحليلية بناءً على صور متعددة."
}
}
diff --git a/locales/ar/providers.json b/locales/ar/providers.json
index ccae39bdf081d..e6bb7fd6e4d38 100644
--- a/locales/ar/providers.json
+++ b/locales/ar/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "توفر Azure مجموعة متنوعة من نماذج الذكاء الاصطناعي المتقدمة، بما في ذلك GPT-3.5 وأحدث سلسلة GPT-4، تدعم أنواع بيانات متعددة ومهام معقدة، وتلتزم بحلول ذكاء اصطناعي آمنة وموثوقة ومستدامة."
},
+ "azureai": {
+ "description": "توفر Azure مجموعة متنوعة من نماذج الذكاء الاصطناعي المتقدمة، بما في ذلك GPT-3.5 وأحدث سلسلة GPT-4، تدعم أنواع البيانات المتعددة والمهام المعقدة، وتهدف إلى تقديم حلول ذكاء اصطناعي آمنة وموثوقة ومستدامة."
+ },
"baichuan": {
"description": "Baichuan Intelligence هي شركة تركز على تطوير نماذج الذكاء الاصطناعي الكبيرة، حيث تظهر نماذجها أداءً ممتازًا في المهام الصينية مثل الموسوعات المعرفية ومعالجة النصوص الطويلة والإبداع. تتفوق على النماذج الرئيسية الأجنبية. كما تتمتع Baichuan Intelligence بقدرات متعددة الوسائط رائدة في الصناعة، وقد أظهرت أداءً ممتازًا في العديد من التقييمات الموثوقة. تشمل نماذجها Baichuan 4 وBaichuan 3 Turbo وBaichuan 3 Turbo 128k، وكل منها مُحسّن لمشاهد تطبيق مختلفة، مما يوفر حلولًا فعالة من حيث التكلفة."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek هي شركة تركز على أبحاث وتطبيقات تقنيات الذكاء الاصطناعي، حيث يجمع نموذجها الأحدث DeepSeek-V2.5 بين قدرات الحوار العامة ومعالجة الشيفرات، وقد حقق تحسينات ملحوظة في محاذاة تفضيلات البشر، ومهام الكتابة، واتباع التعليمات."
},
+ "doubao": {
+ "description": "نموذج كبير تم تطويره داخليًا بواسطة بايت دانس. تم التحقق من صحته من خلال أكثر من 50 سيناريو عمل داخلي، مع استخدام يومي يتجاوز تريليون توكن، مما يتيح تقديم قدرات متعددة الأنماط، ويعمل على توفير تجربة عمل غنية للشركات من خلال نموذج عالي الجودة."
+ },
"fireworksai": {
"description": "Fireworks AI هي شركة رائدة في تقديم خدمات نماذج اللغة المتقدمة، تركز على استدعاء الوظائف والمعالجة متعددة الوسائط. نموذجها الأحدث Firefunction V2 مبني على Llama-3، مُحسّن لاستدعاء الوظائف، والحوار، واتباع التعليمات. يدعم نموذج اللغة البصرية FireLLaVA-13B إدخال الصور والنصوص المختلطة. تشمل النماذج البارزة الأخرى سلسلة Llama وسلسلة Mixtral، مما يوفر دعمًا فعالًا لاتباع التعليمات وتوليدها بلغات متعددة."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "منظمة مفتوحة المصدر مكرسة لأبحاث وتطوير أدوات النماذج الكبيرة. توفر منصة مفتوحة المصدر فعالة وسهلة الاستخدام لجميع مطوري الذكاء الاصطناعي، مما يجعل أحدث تقنيات النماذج الكبيرة والخوارزميات في متناول اليد."
},
+ "jina": {
+ "description": "تأسست Jina AI في عام 2020، وهي شركة رائدة في مجال الذكاء الاصطناعي للبحث. تحتوي منصتنا الأساسية للبحث على نماذج متجهة، ومعيدي ترتيب، ونماذج لغوية صغيرة، لمساعدة الشركات في بناء تطبيقات بحث موثوقة وعالية الجودة تعتمد على الذكاء الاصطناعي التوليدي ومتعددة الوسائط."
+ },
+ "lmstudio": {
+ "description": "LM Studio هو تطبيق سطح مكتب لتطوير وتجربة نماذج اللغة الكبيرة (LLMs) على جهاز الكمبيوتر الخاص بك."
+ },
"minimax": {
"description": "MiniMax هي شركة تكنولوجيا الذكاء الاصطناعي العامة التي تأسست في عام 2021، تكرس جهودها للتعاون مع المستخدمين في إنشاء الذكاء. طورت MiniMax نماذج كبيرة عامة من أوضاع مختلفة، بما في ذلك نموذج نصي MoE الذي يحتوي على تريليونات من المعلمات، ونموذج صوتي، ونموذج صور. وقد أطلقت تطبيقات مثل Conch AI."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI هي منصة تقدم خدمات API لمجموعة متنوعة من نماذج اللغة الكبيرة وتوليد الصور بالذكاء الاصطناعي، مرنة وموثوقة وفعالة من حيث التكلفة. تدعم أحدث النماذج مفتوحة المصدر مثل Llama3 وMistral، وتوفر حلول API شاملة وسهلة الاستخدام وقابلة للتوسع تلقائيًا لتطوير تطبيقات الذكاء الاصطناعي، مما يجعلها مناسبة لنمو الشركات الناشئة في مجال الذكاء الاصطناعي."
},
+ "nvidia": {
+ "description": "تقدم NVIDIA NIM™ حاويات يمكن استخدامها لاستضافة خدمات استدلال معززة بواسطة GPU، تدعم نشر نماذج الذكاء الاصطناعي المدربة مسبقًا والمخصصة على السحابة ومراكز البيانات وأجهزة الكمبيوتر الشخصية RTX™ ومحطات العمل."
+ },
"ollama": {
"description": "تغطي نماذج Ollama مجموعة واسعة من مجالات توليد الشيفرة، والعمليات الرياضية، ومعالجة اللغات المتعددة، والتفاعل الحواري، وتدعم احتياجات النشر على مستوى المؤسسات والتخصيص المحلي."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "أطلقت الأكاديمية الصينية للعلوم ومعهد ووهان للذكاء الاصطناعي نموذجًا جديدًا متعدد الوسائط، يدعم أسئلة وأجوبة متعددة الجولات، وإنشاء النصوص، وتوليد الصور، وفهم 3D، وتحليل الإشارات، ويغطي مجموعة شاملة من مهام الأسئلة والأجوبة، مع قدرات أقوى في الإدراك والفهم والإبداع، مما يوفر تجربة تفاعلية جديدة."
},
+ "tencentcloud": {
+ "description": "قدرة المحرك المعرفي الذري (LLM Knowledge Engine Atomic Power) هي قدرة كاملة للإجابة على الأسئلة مبنية على تطوير المحرك المعرفي، موجهة نحو الشركات والمطورين، وتوفر القدرة على تجميع وتطوير تطبيقات النماذج بشكل مرن. يمكنك من خلال مجموعة من القدرات الذرية تجميع خدمة النموذج الخاصة بك، واستدعاء خدمات تحليل الوثائق، والتقسيم، والتضمين، وإعادة الكتابة متعددة الجولات، لتخصيص أعمال الذكاء الاصطناعي الخاصة بالشركة."
+ },
"togetherai": {
"description": "تسعى Together AI لتحقيق أداء رائد من خلال نماذج الذكاء الاصطناعي المبتكرة، وتقدم مجموعة واسعة من القدرات المخصصة، بما في ذلك دعم التوسع السريع وعمليات النشر البديهية، لتلبية احتياجات الشركات المتنوعة."
},
"upstage": {
"description": "تتخصص Upstage في تطوير نماذج الذكاء الاصطناعي لتلبية احتياجات الأعمال المتنوعة، بما في ذلك Solar LLM وDocument AI، بهدف تحقيق الذكاء الاصطناعي العام (AGI) القائم على العمل. من خلال واجهة Chat API، يمكن إنشاء وكلاء حوار بسيطين، وتدعم استدعاء الوظائف، والترجمة، والتضمين، وتطبيقات المجالات المحددة."
},
+ "vllm": {
+ "description": "vLLM هو مكتبة سريعة وسهلة الاستخدام لاستدلال LLM والخدمات."
+ },
+ "volcengine": {
+ "description": "منصة تطوير خدمات النماذج الكبيرة التي أطلقتها بايت دانس، تقدم خدمات استدعاء نماذج غنية بالوظائف وآمنة وتنافسية من حيث الأسعار، كما توفر بيانات النماذج، والتعديل الدقيق، والاستدلال، والتقييم، وغيرها من الوظائف الشاملة، لضمان تطوير تطبيقات الذكاء الاصطناعي الخاصة بك بشكل كامل."
+ },
"wenxin": {
"description": "منصة تطوير وخدمات النماذج الكبيرة والتطبيقات الأصلية للذكاء الاصطناعي على مستوى المؤسسات، تقدم مجموعة شاملة وسهلة الاستخدام من أدوات تطوير النماذج الذكية التوليدية وأدوات تطوير التطبيقات على مدار العملية بأكملها."
},
diff --git a/locales/ar/setting.json b/locales/ar/setting.json
index 70310b5a8ecdb..ed21e99242be2 100644
--- a/locales/ar/setting.json
+++ b/locales/ar/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "تمكين الحد الأقصى للردود"
},
+ "enableReasoningEffort": {
+ "title": "تفعيل ضبط قوة الاستدلال"
+ },
"frequencyPenalty": {
- "desc": "كلما زادت القيمة، زاد احتمال تقليل تكرار الكلمات",
- "title": "عقوبة التكرار"
+ "desc": "كلما زادت القيمة، كانت المفردات أكثر تنوعًا؛ وكلما انخفضت القيمة، كانت المفردات أكثر بساطة ووضوحًا",
+ "title": "تنوع المفردات"
},
"maxTokens": {
"desc": "عدد الرموز الأقصى المستخدمة في التفاعل الواحد",
@@ -212,19 +215,31 @@
"desc": "{{provider}} نموذج",
"title": "النموذج"
},
+ "params": {
+ "title": "إعدادات متقدمة"
+ },
"presencePenalty": {
- "desc": "كلما زادت القيمة، زاد احتمال التوسع في مواضيع جديدة",
- "title": "جديد الحديث"
+ "desc": "كلما زادت القيمة، زادت الميل إلى استخدام تعبيرات مختلفة، مما يتجنب تكرار المفاهيم؛ وكلما انخفضت القيمة، زادت الميل إلى استخدام المفاهيم أو السرد المتكرر، مما يجعل التعبير أكثر اتساقًا",
+ "title": "تنوع التعبير"
+ },
+ "reasoningEffort": {
+ "desc": "كلما زادت القيمة، زادت قدرة الاستدلال، ولكن قد يؤدي ذلك إلى زيادة وقت الاستجابة واستهلاك التوكنات",
+ "options": {
+ "high": "عالي",
+ "low": "منخفض",
+ "medium": "متوسط"
+ },
+ "title": "قوة الاستدلال"
},
"temperature": {
- "desc": "كلما زادت القيمة، زادت الردود عشوائية أكثر",
- "title": "التباين",
- "titleWithValue": "التباين {{value}}"
+ "desc": "كلما زادت القيمة، كانت الإجابات أكثر إبداعًا وخيالًا؛ وكلما انخفضت القيمة، كانت الإجابات أكثر دقة",
+ "title": "مستوى الإبداع",
+ "warning": "إذا كانت قيمة مستوى الإبداع مرتفعة جدًا، قد تحتوي المخرجات على تشويش"
},
"title": "إعدادات النموذج",
"topP": {
- "desc": "مشابه للتباين ولكن لا يجب تغييره مع التباين",
- "title": "العينة الأساسية"
+ "desc": "عدد الاحتمالات التي يتم أخذها في الاعتبار، كلما زادت القيمة، زادت احتمالية قبول إجابات متعددة؛ وكلما انخفضت القيمة، زادت الميل لاختيار الإجابة الأكثر احتمالًا. لا يُنصح بتغييرها مع مستوى الإبداع",
+ "title": "مستوى الانفتاح الفكري"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "إعدادات عامة",
"experiment": "تجربة",
"llm": "نموذج اللغة",
+ "provider": "مزود خدمة الذكاء الاصطناعي",
"sync": "مزامنة السحابة",
"system-agent": "مساعد النظام",
"tts": "خدمة الكلام"
diff --git a/locales/bg-BG/changelog.json b/locales/bg-BG/changelog.json
index e7fcce524eb5a..08d26f798ffa9 100644
--- a/locales/bg-BG/changelog.json
+++ b/locales/bg-BG/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "Вижте всички актуализации",
"description": "Следете новите функции и подобрения на {{appName}}",
"pagination": {
- "older": "Преглед на историческите промени",
- "prev": "Предишна страница"
+ "next": "Следваща страница",
+ "older": "Преглед на историческите промени"
},
"readDetails": "Прочетете подробности",
"title": "Актуализации",
diff --git a/locales/bg-BG/common.json b/locales/bg-BG/common.json
index 49e4c8b2618b9..0d6cddc11161d 100644
--- a/locales/bg-BG/common.json
+++ b/locales/bg-BG/common.json
@@ -275,6 +275,7 @@
},
"temp": "Временен",
"terms": "Условия за ползване",
+ "update": "Актуализиране",
"updateAgent": "Актуализирай информацията за агента",
"upgradeVersion": {
"action": "Надстрой",
@@ -285,6 +286,7 @@
"anonymousNickName": "Анонимен потребител",
"billing": "Управление на сметките",
"cloud": "Изпробвайте {{name}}",
+ "community": "Общностна версия",
"data": "Съхранение на данни",
"defaultNickname": "Потребител на общността",
"discord": "Поддръжка на общността",
@@ -294,7 +296,6 @@
"help": "Център за помощ",
"moveGuide": "Бутонът за настройки е преместен тук",
"plans": "Планове за абонамент",
- "preview": "Преглед",
"profile": "Управление на профила",
"setting": "Настройки на приложението",
"usages": "Статистика за използване"
diff --git a/locales/bg-BG/components.json b/locales/bg-BG/components.json
index 43f732c61c3d6..8eb2cc27c3f68 100644
--- a/locales/bg-BG/components.json
+++ b/locales/bg-BG/components.json
@@ -76,6 +76,7 @@
"custom": "Потребителски модел, по подразбиране поддържа функционалност за функционални обаждания и визуално разпознаване, моля, потвърдете наличието на тези възможности спрямо реалните условия",
"file": "Този модел поддържа качване на файлове и разпознаване",
"functionCall": "Този модел поддържа функционални обаждания (Function Call)",
+ "reasoning": "Този модел поддържа дълбочинно мислене",
"tokens": "Този модел поддържа до {{tokens}} токена за една сесия",
"vision": "Този модел поддържа визуално разпознаване"
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "Няма активирани модели, моля, посетете настройките и ги активирайте",
"provider": "Доставчик"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "Поради ограниченията на сигурността на браузъра, трябва да конфигурирате крос-домейн достъп за Ollama, за да можете да го използвате нормално.",
+ "linux": {
+ "env": "Добавете `Environment` в секцията [Service] и добавете променливата на средата OLLAMA_ORIGINS:",
+ "reboot": "Презаредете systemd и рестартирайте Ollama",
+ "systemd": "Извикайте systemd, за да редактирате услугата ollama:"
+ },
+ "macos": "Моля, отворете приложението „Терминал“ и поставете следната команда, след което натиснете Enter, за да я изпълните",
+ "reboot": "Моля, рестартирайте услугата Ollama след завършване на изпълнението",
+ "title": "Конфигуриране на Ollama за разрешаване на крос-домейн достъп",
+ "windows": "На Windows, кликнете върху „Контролен панел“, за да редактирате системните променливи на средата. Създайте нова променлива на средата с име „OLLAMA_ORIGINS“ за вашия потребителски акаунт, със стойност * и кликнете „OK/Приложи“, за да запазите"
+ },
+ "install": {
+ "description": "Моля, уверете се, че сте стартирали Ollama. Ако не сте изтеглили Ollama, моля, посетете официалния сайт <1>за изтегляне",
+ "docker": "Ако предпочитате да използвате Docker, Ollama предлага и официален Docker образ, който можете да изтеглите с следната команда:",
+ "linux": {
+ "command": "Инсталирайте с следната команда:",
+ "manual": "Или можете да се запознаете с <1>Ръководството за ръчна инсталация на Linux и да инсталирате сами"
+ },
+ "title": "Инсталиране и стартиране на приложението Ollama локално",
+ "windowsTab": "Windows (предварителна версия)"
+ }
+ },
+ "Thinking": {
+ "thinking": "В процес на дълбочинно размисъл...",
+ "thought": "Дълбоко размислих (отне ми {{duration}} секунди)",
+ "thoughtWithDuration": "Дълбоко размислих"
}
}
diff --git a/locales/bg-BG/discover.json b/locales/bg-BG/discover.json
index b3e924c3341c8..a3cdd7736fcea 100644
--- a/locales/bg-BG/discover.json
+++ b/locales/bg-BG/discover.json
@@ -126,6 +126,10 @@
"title": "Свежест на темата"
},
"range": "Обхват",
+ "reasoning_effort": {
+ "desc": "Тази настройка контролира интензивността на разсъжденията на модела преди генерирането на отговор. Ниска интензивност приоритизира скоростта на отговор и спестява токени, докато висока интензивност предоставя по-пълни разсъждения, но изразходва повече токени и намалява скоростта на отговор. Стойността по подразбиране е средна, което балансира точността на разсъжденията и скоростта на отговор.",
+ "title": "Интензивност на разсъжденията"
+ },
"temperature": {
"desc": "Тази настройка влияе на разнообразието на отговорите на модела. По-ниски стойности водят до по-предсказуеми и типични отговори, докато по-високи стойности насърчават по-разнообразни и необичайни отговори. Когато стойността е 0, моделът винаги дава един и същ отговор на даден вход.",
"title": "Случайност"
diff --git a/locales/bg-BG/modelProvider.json b/locales/bg-BG/modelProvider.json
index a6fe0f16f2b4b..ee715c999453e 100644
--- a/locales/bg-BG/modelProvider.json
+++ b/locales/bg-BG/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API ключ"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Версия на API на Azure, следваща формата YYYY-MM-DD, вижте [най-новата версия](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "Вземи списък",
+ "title": "Версия на API на Azure"
+ },
+ "endpoint": {
+ "desc": "Намерете крайна точка за инференция на моделите на Azure AI в прегледа на проекта на Azure AI",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Крайна точка на Azure AI"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Намерете API ключа в прегледа на проекта на Azure AI",
+ "placeholder": "Ключ на Azure",
+ "title": "Ключ"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "Въведете AWS Access Key Id",
@@ -63,6 +81,46 @@
"title": "ID на Cloudflare / API адрес"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Моля, въведете вашия API ключ",
+ "title": "API ключ"
+ },
+ "basicTitle": "Основна информация",
+ "configTitle": "Конфигурационна информация",
+ "confirm": "Създаване",
+ "createSuccess": "Създаването е успешно",
+ "description": {
+ "placeholder": "Описание на доставчика (по избор)",
+ "title": "Описание на доставчика"
+ },
+ "id": {
+ "desc": "Уникален идентификатор за доставчика на услуги, който не може да бъде променян след създаването му",
+ "format": "Може да съдържа само цифри, малки букви, тирета (-) и долни черти (_) ",
+ "placeholder": "Препоръчително изцяло с малки букви, например openai, след създаването не може да се промени",
+ "required": "Моля, въведете ID на доставчика",
+ "title": "ID на доставчика"
+ },
+ "logo": {
+ "required": "Моля, качете правилното лого на доставчика",
+ "title": "Лого на доставчика"
+ },
+ "name": {
+ "placeholder": "Моля, въведете показваното име на доставчика",
+ "required": "Моля, въведете името на доставчика",
+ "title": "Име на доставчика"
+ },
+ "proxyUrl": {
+ "required": "Моля, въведете адреса на проксито",
+ "title": "Адрес на прокси"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Моля, изберете тип SDK",
+ "title": "Формат на запитването"
+ },
+ "title": "Създаване на персонализиран AI доставчик"
+ },
"github": {
"personalAccessToken": {
"desc": "Въведете вашия GitHub PAT, кликнете [тук](https://github.com/settings/tokens), за да създадете",
@@ -77,6 +135,23 @@
"title": "HuggingFace токен"
}
},
+ "list": {
+ "title": {
+ "disabled": "Неактивен доставчик",
+ "enabled": "Активен доставчик"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Добавяне на персонализиран доставчик",
+ "all": "Всички",
+ "list": {
+ "disabled": "Неактивиран",
+ "enabled": "Активиран"
+ },
+ "notFound": "Не са намерени резултати от търсенето",
+ "searchProviders": "Търсене на доставчици...",
+ "sort": "Персонализирано сортиране"
+ },
"ollama": {
"checker": {
"desc": "Тестване дали адресът на прокси е попълнен правилно",
@@ -94,33 +169,9 @@
"title": "Изтегляне на модел {{model}} "
},
"endpoint": {
- "desc": "Въведете адрес на Ollama интерфейсния прокси, оставете празно, ако локално не е указано специално",
+ "desc": "Трябва да съдържа http(s)://, местният адрес може да остане празен, ако не е зададен допълнително",
"title": "Адрес на прокси интерфейс"
},
- "setup": {
- "cors": {
- "description": "Заради ограниченията за сигурност в браузъра, трябва да конфигурирате кросдомейн за Ollama, за да работи правилно.",
- "linux": {
- "env": "Добавете `Environment` в раздела [Service], като добавите променливата на средата OLLAMA_ORIGINS:",
- "reboot": "Презаредете systemd и рестартирайте Ollama",
- "systemd": "Извикайте systemd за редактиране на услугата ollama:"
- },
- "macos": "Моля, отворете приложението „Терминал“ и поставете следната команда, след което натиснете Enter",
- "reboot": "Моля, рестартирайте услугата Ollama след приключване на изпълнението",
- "title": "Конфигуриране на Ollama за позволяване на кросдомейн достъп",
- "windows": "На Windows кликнете върху „Контролен панел“, влезте в редактиране на системните променливи. Създайте нова променлива на средата с име „OLLAMA_ORIGINS“, стойност * и кликнете „ОК/Приложи“, за да запазите промените"
- },
- "install": {
- "description": "Моля, потвърдете, че сте активирали Ollama. Ако не сте го изтеглили, моля посетете <1>официалния сайт на Ollama.",
- "docker": "Ако предпочитате да използвате Docker, Ollama предлага официален Docker образ, който можете да изтеглите с помощта на следната команда:",
- "linux": {
- "command": "Инсталирайте чрез следната команда:",
- "manual": "Или може да се обадите на <1>Ръководство за ръчна инсталация на Linux и да инсталирате ръчно"
- },
- "title": "Инсталиране и стартиране на приложението Ollama локално",
- "windowsTab": "Windows (преглед)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Отмяна на изтеглянето",
@@ -131,25 +182,145 @@
"title": "Изтегляне на зададения Ollama модел"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Въведете Access Key от платформата Baidu Qianfan",
- "placeholder": "Qianfan Access Key",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "Вашият ключ и адреса на прокси ще бъдат криптирани с <1>AES-GCM алгоритъм",
+ "apiKey": {
+ "desc": "Моля, въведете вашия {{name}} API ключ",
+ "placeholder": "{{name}} API ключ",
+ "title": "API ключ"
+ },
+ "baseURL": {
+ "desc": "Трябва да съдържа http(s)://",
+ "invalid": "Моля, въведете валиден URL",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "API адрес на прокси"
+ },
+ "checker": {
+ "button": "Проверка",
+ "desc": "Тест на API ключа и адреса на прокси за правилно попълване",
+ "pass": "Проверката е успешна",
+ "title": "Проверка на свързаност"
+ },
+ "fetchOnClient": {
+ "desc": "Режимът на клиентски запитвания ще инициира сесийни запитвания директно от браузъра, което може да ускори времето за отговор",
+ "title": "Използване на клиентски режим на запитвания"
+ },
+ "helpDoc": "Ръководство за конфигуриране",
+ "waitingForMore": "Още модели са в <1>планиране, моля, очаквайте"
},
- "checker": {
- "desc": "Тествайте дали AccessKey / SecretAccess е попълнен правилно"
+ "createNew": {
+ "title": "Създаване на персонализиран AI модел"
+ },
+ "item": {
+ "config": "Конфигуриране на модела",
+ "customModelCards": {
+ "addNew": "Създаване и добавяне на модел {{id}}",
+ "confirmDelete": "Ще изтриете този персонализиран модел, след изтриването няма да може да бъде възстановен, моля, действайте внимателно."
+ },
+ "delete": {
+ "confirm": "Потвърдете ли, че искате да изтриете модела {{displayName}}?",
+ "success": "Изтриването е успешно",
+ "title": "Изтриване на модел"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Полето, което действително се изисква в Azure OpenAI",
+ "placeholder": "Моля, въведете името на модела за разполагане в Azure",
+ "title": "Име на разполагане на модела"
+ },
+ "deployName": {
+ "extra": "Това поле ще бъде използвано като ID на модела при изпращане на заявката",
+ "placeholder": "Моля, въведете действителното име или ID на разположението на модела",
+ "title": "Име на разположение на модела"
+ },
+ "displayName": {
+ "placeholder": "Моля, въведете показваното име на модела, например ChatGPT, GPT-4 и др.",
+ "title": "Показвано име на модела"
+ },
+ "files": {
+ "extra": "Текущата функция за качване на файлове е само един хак, само за опити. Пълната функционалност за качване на файлове ще бъде реализирана по-късно.",
+ "title": "Поддръжка на качване на файлове"
+ },
+ "functionCall": {
+ "extra": "Тази конфигурация ще активира само способността на модела да използва инструменти, което позволява добавянето на плъгини от клас инструменти. Но дали наистина ще се поддържа използването на инструменти зависи изцяло от самия модел, моля, тествайте неговата наличност",
+ "title": "Поддръжка на използването на инструменти"
+ },
+ "id": {
+ "extra": "След създаването не може да бъде променян, ще се използва като идентификатор на модела при извикване на AI",
+ "placeholder": "Моля, въведете идентификатор на модела, например gpt-4o или claude-3.5-sonnet",
+ "title": "ID на модела"
+ },
+ "modalTitle": "Конфигурация на персонализиран модел",
+ "reasoning": {
+ "extra": "Тази конфигурация ще активира само способността на модела за дълбоко мислене, конкретният ефект зависи изцяло от самия модел, моля, тествайте сами дали моделът притежава налична способност за дълбоко мислене",
+ "title": "Поддръжка на дълбоко мислене"
+ },
+ "tokens": {
+ "extra": "Настройте максималния брой токени, поддържани от модела",
+ "title": "Максимален контекстуален прозорец",
+ "unlimited": "Без ограничения"
+ },
+ "vision": {
+ "extra": "Тази конфигурация ще активира само конфигурацията за качване на изображения в приложението, дали поддържа разпознаване зависи изцяло от самия модел, моля, тествайте наличността на визуалната разпознаваемост на този модел.",
+ "title": "Поддръжка на визуално разпознаване"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/изображение",
+ "inputCharts": "${{amount}}/M символи",
+ "inputMinutes": "${{amount}}/минути",
+ "inputTokens": "Входящи ${{amount}}/М",
+ "outputTokens": "Изходящи ${{amount}}/М"
+ },
+ "releasedAt": "Пуснато на {{releasedAt}}"
},
- "secretKey": {
- "desc": "Въведете Secret Key от платформата Baidu Qianfan",
- "placeholder": "Qianfan Secret Key",
- "title": "Secret Key"
+ "list": {
+ "addNew": "Добавяне на модел",
+ "disabled": "Неактивен",
+ "disabledActions": {
+ "showMore": "Покажи всичко"
+ },
+ "empty": {
+ "desc": "Моля, създайте персонализиран модел или изтеглете модел, за да започнете да го използвате",
+ "title": "Няма налични модели"
+ },
+ "enabled": "Активен",
+ "enabledActions": {
+ "disableAll": "Деактивирай всичко",
+ "enableAll": "Активирай всичко",
+ "sort": "Персонализиране на подредбата на моделите"
+ },
+ "enabledEmpty": "Няма активни модели, моля активирайте желаните модели от списъка по-долу~",
+ "fetcher": {
+ "clear": "Изчисти получените модели",
+ "fetch": "Получаване на списък с модели",
+ "fetching": "Получаване на списък с модели...",
+ "latestTime": "Последно обновление: {{time}}",
+ "noLatestTime": "Все още не е получен списък"
+ },
+ "resetAll": {
+ "conform": "Потвърдете ли, че искате да нулирате всички промени в текущия модел? След нулирането списъкът с текущи модели ще се върне в първоначалното си състояние",
+ "success": "Успешно нулирано",
+ "title": "Нулиране на всички промени"
+ },
+ "search": "Търсене на модели...",
+ "searchResult": "Намерени са {{count}} модела",
+ "title": "Списък с модели",
+ "total": "Общо {{count}} налични модела"
},
- "unlock": {
- "customRegion": "Персонализиран регион на услугата",
- "description": "Въведете вашия AccessKey / SecretKey, за да започнете сесия. Приложението няма да запомня вашите конфигурации за удостоверяване",
- "title": "Използвайте персонализирана информация за удостоверяване на Wenxin Yiyan"
- }
+ "searchNotFound": "Не са намерени резултати от търсенето"
+ },
+ "sortModal": {
+ "success": "Сортирането е успешно обновено",
+ "title": "Персонализирано сортиране",
+ "update": "Актуализиране"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "Ще изтриете този AI доставчик, след изтриването няма да може да бъде възстановен, потвърдете ли, че искате да изтриете?",
+ "deleteSuccess": "Изтриването е успешно",
+ "tooltip": "Актуализиране на основната конфигурация на доставчика",
+ "updateSuccess": "Актуализацията е успешна"
},
"zeroone": {
"title": "01.AI Зероуан Всичко"
diff --git a/locales/bg-BG/models.json b/locales/bg-BG/models.json
index 35dd9d258b998..238079777e486 100644
--- a/locales/bg-BG/models.json
+++ b/locales/bg-BG/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K акцентира на семантичната безопасност и отговорността, проектиран специално за приложения с високи изисквания за безопасност на съдържанието, осигурявайки точност и стабилност на потребителското изживяване."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 използва дървесно търсене за изграждане на вериги от мисли и въвежда механизъм за размисъл, обучен чрез подсилено учене, моделът притежава способността за саморазмисъл и корекция на грешки."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro е усъвършенстван модел за обработка на естествен език, пуснат от компания 360, с изключителни способности за генериране и разбиране на текст, особено в областта на генерирането и творчеството, способен да обработва сложни езикови трансформации и ролеви игри."
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1 използва дървесно търсене за изграждане на мисловни вериги и въвежда механизъм за саморазмисъл, обучавайки се чрез подсилено учене, моделът притежава способността за саморазмисъл и корекция на грешки."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra е най-мощната версия в серията Starfire, която подобрява разбирането и обобщаването на текстовото съдържание, докато надгражда свързаните търсения. Това е всестранно решение за повишаване на производителността в офиса и точно отговаряне на нуждите, водещо в индустрията интелигентно решение."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "Моделът е лидер в страната по способности, надминавайки чуждестранните основни модели в задачи на китайски език, като знания, дълги текстове и генериране на творби. Също така притежава водещи в индустрията мултимодални способности и отлични резултати в множество авторитетни оценки."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "DeepSeek-R1 дестилиран модел, базиран на Qwen2.5-Math-1.5B, оптимизира производителността на разсъжденията чрез подсилено учене и данни за студен старт, отворен модел, който обновява многозадачния стандарт."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "DeepSeek-R1 дестилиран модел, базиран на Qwen2.5-14B, оптимизира производителността на разсъжденията чрез подсилено учене и данни за студен старт, отворен модел, който обновява многозадачния стандарт."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "Серията DeepSeek-R1 оптимизира производителността на разсъжденията чрез подсилено учене и данни за студен старт, отворен модел, който обновява многозадачния стандарт, надминавайки нивото на OpenAI-o1-mini."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1 дестилиран модел, базиран на Qwen2.5-Math-7B, оптимизира производителността на разсъжденията чрез подсилено учене и данни за студен старт, отворен модел, който обновява многозадачния стандарт."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite предлага изключителна скорост на отговор и по-добра цена, предоставяйки на клиентите гъвкави опции за различни сценарии. Поддържа извеждане и фин настройка на контекстов прозорец от 128k."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Флагманският модел на Baidu за изключително големи езикови модели, разработен самостоятелно, е напълно обновен в сравнение с ERNIE 3.5 и е широко приложим в сложни задачи в различни области; поддържа автоматично свързване с плъгина за търсене на Baidu, осигурявайки актуалност на информацията за отговори."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Флагманският модел на Baidu, изграден на собствена технология, с изключителни резултати и широко приложение в сложни задачи в различни области; поддържа автоматично свързване с плъгини за търсене на Baidu, осигурявайки актуалност на информацията за отговори. В сравнение с ERNIE 4.0, показва по-добра производителност."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "Патентованият флагмански модул на Baidu, изключително мащабен езиков модел, показващ отлични резултати и широко приложение в сложни сценарии. Поддържа автоматично свързване с плъгини на Baidu Search, гарантирайки актуалността на информацията. В сравнение с ERNIE 4.0, той представя по-добри резултати."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct е един от най-новите големи езикови модели, публикувани от Alibaba Cloud. Този 7B модел показва значителни подобрения в областите на кодирането и математиката. Моделът предлага многоезична поддръжка, обхващаща над 29 езика, включително китайски, английски и др. Моделът показва значителни подобрения в следването на инструкции, разбирането на структурирани данни и генерирането на структурирани изходи (особено JSON)."
},
+ "MiniMax-Text-01": {
+ "description": "В серията модели MiniMax-01 направихме смели иновации: за първи път реализирахме мащабно линейно внимание, традиционната архитектура на Transformer вече не е единственият избор. Параметрите на този модел достигат 4560 милиарда, с единична активация от 45.9 милиарда. Общата производителност на модела е на нивото на водещите модели в чужбина, като същевременно ефективно обработва глобалния контекст от 4 милиона токена, което е 32 пъти повече от GPT-4o и 20 пъти повече от Claude-3.5-Sonnet."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO е високо гъвкава многомоделна комбинация, предназначена да предостави изключителен креативен опит."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 демонстрира изключителни резултати в различни визуално-языкови задачи, включително разбиране на документи и графики, разбиране на текст в сцени, OCR, решаване на научни и математически проблеми."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 демонстрира изключителни резултати в различни визуално-языкови задачи, включително разбиране на документи и графики, разбиране на текст в сцени, OCR, решаване на научни и математически проблеми."
- },
"Phi-3-medium-128k-instruct": {
"description": "Същият модел Phi-3-medium, но с по-голям размер на контекста за RAG или малко подканване."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat е отворената версия на предварително обучен модел от серията GLM-4, пусната от Zhizhu AI. Моделът показва отлични резултати в семантика, математика, разсъждения, код и знания. Освен че поддържа многократни разговори, GLM-4-9B-Chat предлага и напреднали функции като уеб браузинг, изпълнение на код, извикване на персонализирани инструменти (Function Call) и разсъждения с дълги текстове. Моделът поддържа 26 езика, включително китайски, английски, японски, корейски и немски. В множество бенчмаркове, GLM-4-9B-Chat показва отлична производителност, като AlignBench-v2, MT-Bench, MMLU и C-Eval. Моделът поддържа максимална контекстна дължина от 128K, подходящ за академични изследвания и търговски приложения."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 е модел за инференция, управляван от обучение с подсилване (RL), който решава проблемите с повторяемостта и четимостта в моделите. Преди RL, DeepSeek-R1 въвежда данни за студен старт, за да оптимизира допълнително производителността на инференцията. Той показва сравними резултати с OpenAI-o1 в математически, кодови и инференционни задачи и подобрява общата ефективност чрез внимателно проектирани методи на обучение."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 е модел на езика с 6710 милиарда параметри, който използва архитектура на смесени експерти (MoE) с много глави на потенциално внимание (MLA) и стратегия за баланс на натоварването без помощни загуби, оптимизираща производителността на инференцията и обучението. Чрез предварително обучение на 14.8 трилиона висококачествени токени и последващо супервизирано фино настройване и обучение с подсилване, DeepSeek-V3 надминава производителността на други отворени модели и е близо до водещите затворени модели."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma е един от най-новите леки, авангардни отворени модели, разработени от Google. Това е голям езиков модел с един декодер, който поддържа английски и предлага отворени тегла, предварително обучени варианти и варианти с фино настройване на инструкции. Моделът Gemma е подходящ за различни задачи по генериране на текст, включително въпроси и отговори, резюмиране и разсъждения. Този 9B модел е обучен с 8 трилиона токена. Неговият относително малък размер позволява внедряване в среди с ограничени ресурси, като лаптопи, настолни компютри или собствена облачна инфраструктура, което позволява на повече хора да имат достъп до авангардни AI модели и да насърчават иновации."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 е семейство от многоезични големи езикови модели, разработени от Meta, включващо предварително обучени и модели с фино настройване с параметри 8B, 70B и 405B. Този 8B модел с фино настройване на инструкции е оптимизиран за многоезични разговорни сценарии и показва отлични резултати в множество индустриални бенчмаркове. Моделът е обучен с над 15 трилиона токена от публични данни и използва технологии като наблюдавано фино настройване и обучение с човешка обратна връзка, за да подобри полезността и безопасността на модела. Llama 3.1 поддържа генериране на текст и генериране на код, с дата на прекратяване на знанията до декември 2023 г."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview е иновативен модел за обработка на естествен език, способен да обработва ефективно сложни задачи за генериране на диалог и разбиране на контекста."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview е изследователски модел, разработен от екипа на Qwen, който се фокусира върху визуалните способности за извеждане и притежава уникални предимства в разбирането на сложни сцени и решаването на визуално свързани математически проблеми."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview е най-новият експериментален изследователски модел на Qwen, който се фокусира върху подобряване на AI разсъдъчните способности. Чрез изследване на сложни механизми като езикови смеси и рекурсивно разсъждение, основните предимства включват мощни аналитични способности, математически и програмистки умения. В същото време съществуват проблеми с езиковото превключване, цикли на разсъждение, съображения за безопасност и разлики в други способности."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct е най-новата версия на серията големи езикови модели, специфични за код, публикувана от Alibaba Cloud. Моделът значително подобрява способностите за генериране на код, разсъждения и корекции, след като е обучен с 55 трилиона токена на базата на Qwen2.5. Той не само подобрява кодовите умения, но и запазва предимствата в математиката и общите способности. Моделът предоставя по-пълна основа за практическите приложения като кодови интелигентни агенти."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math се фокусира върху решаването на математически проблеми, предоставяйки професионални отговори на трудни задачи."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 е най-новата серия на модела Qwen, поддържаща 128k контекст. В сравнение с текущите най-добри отворени модели, Qwen2-72B значително надминава водещите модели в области като разбиране на естествен език, знания, код, математика и многоезичност."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 е най-новата серия на модела Qwen, способен да надмине оптималните отворени модели с равен размер или дори по-големи модели. Qwen2 7B постига значителни предимства в множество тестове, особено в разбирането на код и китайския език."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B е мощен визуален езиков модел, който поддържа многомодално обработване на изображения и текст, способен точно да разпознава съдържанието на изображения и да генерира свързани описания или отговори."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct е голям езиков модел с 14 милиарда параметри, с отлично представяне, оптимизиран за китайски и многоезични сценарии, поддържа интелигентни въпроси и отговори, генериране на съдържание и други приложения."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct е голям езиков модел с 7 милиарда параметри, който поддържа безпроблемно взаимодействие с функции и външни системи, значително увеличавайки гъвкавостта и разширяемостта. Оптимизиран за китайски и многоезични сценарии, поддържа интелигентни въпроси и отговори, генериране на съдържание и други приложения."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct е модел за програмиране, базиран на мащабно предварително обучение, с мощни способности за разбиране и генериране на код, способен ефективно да обработва различни програмни задачи, особено подходящ за интелигентно писане на код, автоматично генериране на скриптове и отговори на програмни въпроси."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct е голям езиков модел, проектиран специално за генериране на код, разбиране на код и ефективни сценарии за разработка, с водеща в индустрията параметрична стойност от 32B, способен да отговори на разнообразни програмни нужди."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "TeleMM е многомодален голям модел, разработен от China Telecom, способен да обработва текст, изображения и други видове входни данни, поддържащ функции като разбиране на изображения и анализ на графики, предоставяйки услуги за разбиране на потребителите в различни модалности. Моделът може да взаимодейства с потребителите в многомодални сценарии, точно разбирайки входното съдържание, отговаряйки на въпроси, помагайки в творчеството и ефективно предоставяйки многомодална информация и вдъхновение. Моделът показва отлични резултати в задачи с фина перцепция и логическо разсъждение."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large е най-голямата отворена трансформаторна архитектура MoE в индустрията, с общо 3890 милиарда параметри и 52 милиарда активни параметри."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct е голям езиков модел с параметри 72B от серията Qwen2, специално настроен за инструкции. Моделът е базиран на архитектурата Transformer и използва технологии като SwiGLU активационна функция, QKV отклонение за внимание и групова внимание. Той може да обработва големи входни данни. Моделът показва отлични резултати в множество бенчмаркове за разбиране на езика, генериране, многоезични способности, кодиране, математика и разсъждения, надминавайки повечето отворени модели и показвайки конкурентоспособност на определени задачи в сравнение с патентовани модели."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct е един от най-новите големи езикови модели, публикувани от Alibaba Cloud. Този 72B модел показва значителни подобрения в областите на кодирането и математиката. Моделът предлага многоезична поддръжка, обхващаща над 29 езика, включително китайски, английски и др. Моделът показва значителни подобрения в следването на инструкции, разбирането на структурирани данни и генерирането на структурирани изходи (особено JSON)."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Оптимизирана за диалогови сценарии на китайски, предлагаща плавно и съответстващо на китайските изразни навици генериране на диалози."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Fireworks отворен модел за извикване на функции, предлагащ отлични способности за изпълнение на инструкции и отворени, персонализируеми характеристики."
+ "abab7-chat-preview": {
+ "description": "В сравнение с моделите от серията abab6.5, този модел предлага значително подобрение в способностите за работа с дълги текстове, математика и писане."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Fireworks компанията представя Firefunction-v2, модел за извикване на функции с изключителна производителност, разработен на базата на Llama-3 и оптимизиран за функции, диалози и следване на инструкции."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 е авангарден голям езиков модел, оптимизиран чрез подсилено обучение и данни за студен старт, с отлични способности в разсъжденията, математиката и програмирането."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b е визуален езиков модел, който може да приема изображения и текстови входове, обучен с висококачествени данни, подходящ за мултимодални задачи."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Мощен езиков модел Mixture-of-Experts (MoE) от Deepseek, с общ брой параметри 671B, активиращи 37B параметри на всеки токен."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Llama 3 70B модел за инструкции, специално оптимизиран за многоезични диалози и разбиране на естествен език, с производителност, превъзхождаща повечето конкурентни модели."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Llama 3 70B модел за инструкции (HF версия), с резултати, съвпадащи с официалната реализация, подходящ за висококачествени задачи за следване на инструкции."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Llama 3 8B модел за инструкции, оптимизиран за диалози и многоезични задачи, с изключителна производителност и ефективност."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Моделът за разсъждение по изображения с 11B параметри на Meta е оптимизиран за визуално разпознаване, разсъждение по изображения, описание на изображения и отговаряне на общи въпроси относно изображения. Моделът може да разбира визуални данни, като графики и таблици, и свързва визуалните данни с текстовите описания на детайлите на изображенията."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Моделът Llama 3.2 1B е лека многоезична разработка от Meta. Този модел е проектиран да подобри ефективността, предоставяйки значителни подобрения в забавянето и разходите в сравнение с по-големи модели. Примерни случаи на ползване включват извличане и обобщение."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Моделът Llama 3.2 3B е лека многоезична разработка от Meta. Този модел е проектиран да подобри ефективността, предоставяйки значителни подобрения в забавянето и разходите в сравнение с по-големи модели. Примерни случаи на ползване включват заявки, пренаписване на подканвания и подпомагане на писането."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Моделът за разсъждение по изображения с 90B параметри на Meta е оптимизиран за визуално разпознаване, разсъждение по изображения, описание на изображения и отговаряне на общи въпроси относно изображения. Моделът може да разбира визуални данни, като графики и таблици, и свързва визуалните данни с текстовите описания на детайлите на изображенията."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct е актуализирана версия на Llama 3.1 70B от декември. Този модел е подобрен на базата на Llama 3.1 70B (пуснат през юли 2024 г.), с подобрени възможности за извикване на инструменти, поддръжка на многоезичен текст, математика и програмиране. Моделът постига водещи в индустрията резултати в области като разсъждение, математика и следване на инструкции, и предлага производителност, подобна на 3.1 405B, с значителни предимства в скоростта и разходите."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "Модел с 24B параметри, предлагащ водещи в индустрията способности, сравними с по-големите модели."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Mixtral MoE 8x22B модел за инструкции, с голям брой параметри и архитектура с множество експерти, осигуряваща всестранна поддръжка за ефективна обработка на сложни задачи."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Mixtral MoE 8x7B модел за инструкции, архитектура с множество експерти, предлагаща ефективно следване и изпълнение на инструкции."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Mixtral MoE 8x7B модел за инструкции (HF версия), с производителност, съвпадаща с официалната реализация, подходящ за множество ефективни сценарии."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "MythoMax L2 13B модел, комбиниращ новаторски технологии за интеграция, специализиран в разказване на истории и ролеви игри."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "QwQ моделът е експериментален изследователски модел, разработен от екипа на Qwen, който се фокусира върху подобряване на AI разсъдъчните способности."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "72B версия на модела Qwen-VL е последната итерация на Alibaba, представляваща иновации от последната година."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 е серия от езикови модели, разработени от екипа на Alibaba Cloud Qwen, които съдържат само декодери. Тези модели предлагат различни размери, включително 0.5B, 1.5B, 3B, 7B, 14B, 32B и 72B, и разполагат с базови (base) и инструкти (instruct) варианти."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct е най-новата версия на серията големи езикови модели, специфични за код, публикувана от Alibaba Cloud. Моделът значително подобрява способностите за генериране на код, разсъждения и корекции, след като е обучен с 55 трилиона токена на базата на Qwen2.5. Той не само подобрява кодовите умения, но и запазва предимствата в математиката и общите способности. Моделът предоставя по-пълна основа за практическите приложения като кодови интелигентни агенти."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "StarCoder 15.5B модел, поддържащ напреднали програмни задачи, с подобрени многоезични способности, подходящ за сложна генерация и разбиране на код."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "StarCoder 7B модел, обучен за над 80 програмни езика, с отлични способности за попълване на код и разбиране на контекста."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Yi-Large модел, предлагащ изключителни способности за многоезична обработка, подходящ за различни задачи по генериране и разбиране на език."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus е най-мощният модел на Anthropic, предназначен за обработка на изключително сложни задачи. Той се отличава с изключителна производителност, интелигентност, гладкост и разбиране."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku е най-бързият следващ модел на Anthropic. В сравнение с Claude 3 Haiku, Claude 3.5 Haiku показва подобрения в различни умения и надминава предишното поколение най-голям модел Claude 3 Opus в много интелектуални бенчмаркове."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet предлага способности, надхвърлящи Opus, и по-бърза скорост в сравнение с Sonnet, като същевременно запазва същата цена. Sonnet е особено силен в програмирането, науката за данни, визуалната обработка и агентските задачи."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet предлага идеален баланс между интелигентност и скорост за корпоративни работни натоварвания. Той предлага максимална полезност на по-ниска цена, надежден и подходящ за мащабно внедряване."
},
- "code-raccoon-v1": {
- "description": "Кодово енотче е софтуерен интелигентен помощник за разработка, базиран на голям езиков модел на SenseTime, обхващащ етапите на анализ на софтуерни изисквания, проектиране на архитектура, писане на код и софтуерно тестване, отговарящ на нуждите на потребителите за писане на код и обучение по програмиране. Кодово енотче поддържа над 90 основни програмни езика, включително Python, Java, JavaScript, C++, Go, SQL и основни IDE като VS Code и IntelliJ IDEA. В практическото приложение, Кодово енотче може да помогне на разработчиците да увеличат ефективността на програмирането с над 50%."
- },
"codegeex-4": {
"description": "CodeGeeX-4 е мощен AI помощник за програмиране, който поддържа интелигентни въпроси и отговори и автоматично допълване на код за различни програмни езици, повишавайки ефективността на разработката."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ е високопроизводителен голям езиков модел, проектиран за реални бизнес сценарии и сложни приложения."
},
+ "dall-e-2": {
+ "description": "Второ поколение модел DALL·E, поддържащ по-реалистично и точно генериране на изображения, с резолюция 4 пъти по-висока от първото поколение."
+ },
+ "dall-e-3": {
+ "description": "Най-новият модел DALL·E, пуснат през ноември 2023 г. Поддържа по-реалистично и точно генериране на изображения с по-силна детайлност."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct предлага висока надеждност в обработката на инструкции, поддържаща приложения в множество индустрии."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 е мощен и икономически ефективен хибриден експертен (MoE) езиков модел. Той е предварително обучен на висококачествен корпус от 81 трилиона токена и е допълнително подобрен чрез наблюдавано фино настройване (SFT) и обучение с подсилване (RL). В сравнение с DeepSeek 67B, DeepSeek-V2 постига по-добра производителност, спестявайки 42.5% от разходите за обучение, намалявайки KV кеша с 93.3% и увеличавайки максималната производителност на генериране с 5.76 пъти. Моделът поддържа контекстна дължина от 128k и показва отлични резултати в стандартни бенчмаркове и оценки на отворено генериране."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 е модел за извеждане, управляван от подсилено обучение (RL), който решава проблемите с повторяемостта и четимостта в модела. Преди RL, DeepSeek-R1 въвежда данни за студен старт, за да оптимизира допълнително производителността на извеждане. Той показва сравнима производителност с OpenAI-o1 в математически, кодови и извеждащи задачи и подобрява общите резултати чрез внимателно проектирани методи на обучение."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "DeepSeek-R1 дестилиран модел, оптимизира производителността на разсъжденията чрез подсилено учене и данни за студен старт, отворен модел, който обновява многозадачния стандарт."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B е дестилиран модел, базиран на Llama-3.1-8B. Този модел е финализиран с примери, генерирани от DeepSeek-R1, и показва отлична производителност на разсъжденията. Той постига добри резултати в множество бенчмаркове, включително 89.1% точност в MATH-500, 50.4% успеваемост в AIME 2024 и 1205 точки в CodeForces, демонстрирайки силни способности за математика и програмиране."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "DeepSeek-R1 дестилиран модел, оптимизира производителността на разсъжденията чрез подсилено учене и данни за студен старт, отворен модел, който обновява многозадачния стандарт."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "DeepSeek-R1 дестилиран модел, оптимизира производителността на разсъжденията чрез подсилено учене и данни за студен старт, отворен модел, който обновява многозадачния стандарт."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B е модел, получен чрез знание дестилация на Qwen2.5-32B. Този модел е финализиран с 800 000 избрани примера, генерирани от DeepSeek-R1, и показва изключителна производителност в множество области, включително математика, програмиране и разсъждения. Той постига отлични резултати в множество бенчмаркове, включително 94.3% точност в MATH-500, демонстрирайки силни способности за математическо разсъждение."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B е модел, получен чрез знание дестилация на Qwen2.5-Math-7B. Този модел е финализиран с 800 000 избрани примера, генерирани от DeepSeek-R1, и показва отлична производителност на разсъжденията. Той постига отлични резултати в множество бенчмаркове, включително 92.8% точност в MATH-500, 55.5% успеваемост в AIME 2024 и 1189 точки в CodeForces, демонстрирайки силни способности за математика и програмиране."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 обединява отличителните характеристики на предишните версии, подобрявайки общите и кодиращите способности."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 е езиков модел с 6710 милиарда параметри, базиран на смесени експерти (MoE), който използва многоглаво потенциално внимание (MLA) и архитектурата DeepSeekMoE, комбинирайки стратегии за баланс на натоварването без помощни загуби, за да оптимизира производителността на извеждане и обучение. Чрез предварително обучение на 14.8 трилиона висококачествени токени и последващо наблюдавано фино настройване и подсилено обучение, DeepSeek-V3 надминава производителността на други отворени модели и се приближава до водещите затворени модели."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B е напреднал модел, обучен за диалози с висока сложност."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "Най-съвременен ефективен LLM, специализиран в разсъждения, математика и програмиране."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 е визуален езиков модел, разработен на базата на DeepSeekMoE-27B, който използва архитектура на смесени експерти (MoE) с рядка активация, постигайки изключителна производителност с активирани само 4.5B параметри. Моделът показва отлични резултати в множество задачи, включително визуални въпроси и отговори, оптично разпознаване на символи, разбиране на документи/таблици/графики и визуална локализация."
+ },
"deepseek-chat": {
"description": "Новооткритият отворен модел, който съчетава общи и кодови способности, не само запазва общата диалогова способност на оригиналния Chat модел и мощната способност за обработка на код на Coder модела, но също така по-добре се съгласува с човешките предпочитания. Освен това, DeepSeek-V2.5 постигна значителни подобрения в писателските задачи, следването на инструкции и много други области."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 е отворен хибриден експертен кодов модел, който се представя отлично в кодовите задачи, сравним с GPT4-Turbo."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 е модел за извеждане, управляван от подсилено обучение (RL), който решава проблемите с повторяемостта и четимостта в модела. Преди RL, DeepSeek-R1 въвежда данни за студен старт, за да оптимизира допълнително производителността на извеждане. Той показва сравнима производителност с OpenAI-o1 в математически, кодови и извеждащи задачи и подобрява общите резултати чрез внимателно проектирани методи на обучение."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 - по-голям и по-интелигентен модел в комплекта DeepSeek - е дестилиран в архитектурата Llama 70B. На базата на бенчмаркове и човешка оценка, този модел е по-интелигентен от оригиналния Llama 70B, особено в задачи, изискващи математическа и фактическа точност."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "Моделите от серията DeepSeek-R1-Distill са получени чрез техника на знание дестилация, като се фино настройват образците, генерирани от DeepSeek-R1, спрямо отворени модели като Qwen и Llama."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "Моделите от серията DeepSeek-R1-Distill са получени чрез техника на знание дестилация, като се фино настройват образците, генерирани от DeepSeek-R1, спрямо отворени модели като Qwen и Llama."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "Моделите от серията DeepSeek-R1-Distill са получени чрез техника на знание дестилация, като се фино настройват образците, генерирани от DeepSeek-R1, спрямо отворени модели като Qwen и Llama."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "Моделите от серията DeepSeek-R1-Distill са получени чрез техника на знание дестилация, като се фино настройват образците, генерирани от DeepSeek-R1, спрямо отворени модели като Qwen и Llama."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "Моделите от серията DeepSeek-R1-Distill са получени чрез техника на знание дестилация, като се фино настройват образците, генерирани от DeepSeek-R1, спрямо отворени модели като Qwen и Llama."
+ },
+ "deepseek-reasoner": {
+ "description": "Модел за извеждане, разработен от DeepSeek. Преди да предостави окончателния отговор, моделът първо извежда част от веригата на мислене, за да повиши точността на крайния отговор."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 е ефективен модел на Mixture-of-Experts, подходящ за икономически ефективни нужди от обработка."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B е кодовият модел на DeepSeek, предоставящ мощни способности за генериране на код."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 е MoE модел, разработен от Hangzhou DeepSeek AI Technology Research Co., Ltd., с отлични резултати в множество тестове, заемащ първото място в основните класации на отворените модели. V3 постига 3-кратно увеличение на скоростта на генериране в сравнение с V2.5, предоставяйки на потребителите по-бързо и гладко изживяване."
+ },
"deepseek/deepseek-chat": {
"description": "Новооткритият отворен модел, който съчетава общи и кодови способности, не само запазва общата диалогова способност на оригиналния Chat модел и мощната способност за обработка на код на Coder модела, но също така по-добре се съобразява с човешките предпочитания. Освен това, DeepSeek-V2.5 постигна значителни подобрения в задачи по писане, следване на инструкции и много други."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 значително подобри способността на модела за разсъждение при наличието на много малко маркирани данни. Преди да предостави окончателния отговор, моделът първо ще изведе част от съдържанието на веригата на мислене, за да повиши точността на окончателния отговор."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 значително подобри способността на модела за разсъждение при наличието на много малко маркирани данни. Преди да предостави окончателния отговор, моделът първо ще изведе част от съдържанието на веригата на мислене, за да повиши точността на окончателния отговор."
+ },
"emohaa": {
"description": "Emohaa е психологически модел с професионални консултантски способности, помагащ на потребителите да разберат емоционалните проблеми."
},
+ "ernie-3.5-128k": {
+ "description": "Флагманският голям езиков модел, разработен от Baidu, обхваща огромно количество китайски и английски текстове, притежаващ силни общи способности, способен да отговори на повечето изисквания за диалогови въпроси и отговори, генериране на съдържание и приложения на плъгини; поддържа автоматично свързване с плъгина за търсене на Baidu, осигурявайки актуалност на информацията."
+ },
+ "ernie-3.5-8k": {
+ "description": "Флагманският голям езиков модел, разработен от Baidu, обхваща огромно количество китайски и английски текстове, притежаващ силни общи способности, способен да отговори на повечето изисквания за диалогови въпроси и отговори, генериране на съдържание и приложения на плъгини; поддържа автоматично свързване с плъгина за търсене на Baidu, осигурявайки актуалност на информацията."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "Флагманският голям езиков модел, разработен от Baidu, обхваща огромно количество китайски и английски текстове, притежаващ силни общи способности, способен да отговори на повечето изисквания за диалогови въпроси и отговори, генериране на съдържание и приложения на плъгини; поддържа автоматично свързване с плъгина за търсене на Baidu, осигурявайки актуалност на информацията."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "Флагманският голям езиков модел, разработен от Baidu, с изключителни подобрения в сравнение с ERNIE 3.5, широко приложим в сложни задачи в различни области; поддържа автоматично свързване с плъгина за търсене на Baidu, осигурявайки актуалност на информацията."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "Флагманският голям езиков модел, разработен от Baidu, с изключителни подобрения в сравнение с ERNIE 3.5, широко приложим в сложни задачи в различни области; поддържа автоматично свързване с плъгина за търсене на Baidu, осигурявайки актуалност на информацията."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "Флагманският голям езиков модел, разработен от Baidu, с отлични общи резултати, широко приложим в сложни задачи в различни области; поддържа автоматично свързване с плъгина за търсене на Baidu, осигурявайки актуалност на информацията. В сравнение с ERNIE 4.0, показва по-добри резултати."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "Флагманският голям езиков модел, разработен от Baidu, с отлични общи резултати, широко приложим в сложни задачи в различни области; поддържа автоматично свързване с плъгина за търсене на Baidu, осигурявайки актуалност на информацията. В сравнение с ERNIE 4.0, показва по-добри резултати."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "Флагманският голям езиков модел, разработен от Baidu, с отлични общи резултати, широко приложим в сложни задачи в различни области; поддържа автоматично свързване с плъгина за търсене на Baidu, осигурявайки актуалност на информацията. В сравнение с ERNIE 4.0, показва по-добри резултати."
+ },
+ "ernie-char-8k": {
+ "description": "Специализиран голям езиков модел, разработен от Baidu, подходящ за приложения като NPC в игри, диалози на клиентска поддръжка и ролеви игри, с по-изразителен и последователен стил на персонажите, по-силна способност за следване на инструкции и по-добра производителност на разсъжденията."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Специализиран голям езиков модел, разработен от Baidu, подходящ за приложения като NPC в игри, диалози на клиентска поддръжка и ролеви игри, с по-изразителен и последователен стил на персонажите, по-силна способност за следване на инструкции и по-добра производителност на разсъжденията."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite е лек голям езиков модел, разработен от Baidu, който съчетава отлични резултати с производителност на разсъжденията, подходящ за използване с AI ускорителни карти с ниска изчислителна мощ."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Лек голям езиков модел, разработен от Baidu, който съчетава отлични резултати с производителност на разсъжденията, с по-добри резултати в сравнение с ERNIE Lite, подходящ за използване с AI ускорителни карти с ниска изчислителна мощ."
+ },
+ "ernie-novel-8k": {
+ "description": "Общ голям езиков модел, разработен от Baidu, с очевидни предимства в продължаването на разкази, подходящ и за кратки пиеси и филми."
+ },
+ "ernie-speed-128k": {
+ "description": "Най-новият високопроизводителен голям езиков модел, разработен от Baidu през 2024 г., с отлични общи способности, подходящ за финализиране на специфични проблеми, с отлична производителност на разсъжденията."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "Най-новият високопроизводителен голям езиков модел, разработен от Baidu през 2024 г., с отлични общи способности, с по-добри резултати в сравнение с ERNIE Speed, подходящ за финализиране на специфични проблеми, с отлична производителност на разсъжденията."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny е модел с изключителна производителност, разработен от Baidu, с най-ниски разходи за внедряване и фина настройка сред моделите от серията Wenxin."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Тунинг) предлага стабилна и настройваема производителност, идеален избор за решения на сложни задачи."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro поддържа до 2 милиона токена и е идеален избор за среден многомодален модел, подходящ за многостранна поддръжка на сложни задачи."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash предлага следващо поколение функции и подобрения, включително изключителна скорост, нативна употреба на инструменти, многомодално генериране и контекстен прозорец от 1M токена."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash предлага следващо поколение функции и подобрения, включително изключителна скорост, нативна употреба на инструменти, многомодално генериране и контекстен прозорец от 1M токена."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp е най-новият експериментален мултимодален AI модел на Google, предлагащ следващо поколение функции, изключителна скорост, нативно извикване на инструменти и мултимодално генериране."
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "Модел на Gemini 2.0 Flash, оптимизиран за икономичност и ниска латентност."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Exp е най-новият експериментален многомодален AI модел на Google, с ново поколение функции, изключителна скорост, нативно извикване на инструменти и многомодално генериране."
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Exp е най-новият експериментален мултимодален AI модел на Google, с ново поколение функции, изключителна скорост, нативни инструменти за извикване и мултимодално генериране."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 е най-новият експериментален многомодален AI модел на Google, който предлага бърза обработка и поддържа вход от текст, изображения и видео, подходящ за ефективно разширение на множество задачи."
- },
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 е най-новият експериментален мултимодален AI модел на Google, който предлага бърза обработка и поддържа текстови, изображенчески и видео входове, подходящ за ефективно разширяване на множество задачи."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental е най-новият експериментален многомодален AI модел на Google, който предлага значително подобрение в качеството в сравнение с предишните версии, особено по отношение на световни знания, код и дълги контексти."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 е най-новият експериментален мултимодален AI модел на Google, който предлага значително подобрение в качеството в сравнение с предишни версии."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus разполага с разбиране на видео съдържание и множество изображения, подходящ за мултимодални задачи."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview притежава мощни способности за сложни разсъждения, показвайки отлични резултати в логическото разсъждение, математиката и програмирането."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash предлага следващо поколение функции и подобрения, включително изключителна скорост, нативна употреба на инструменти, многомодално генериране и контекстен прозорец от 1M токена."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental е най-новият експериментален многомодален AI модел на Google, който предлага значително подобрение в качеството в сравнение с предишните версии, особено по отношение на световни знания, код и дълги контексти."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash предлага оптимизирани мултимодални обработващи способности, подходящи за различни сложни задачи."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro комбинира най-новите оптимизационни технологии, предоставяйки по-ефективна обработка на мултимодални данни."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 е ефективен модел, представен от Google, обхващащ множество приложения от малки приложения до сложна обработка на данни."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 продължава концепцията за лекота и ефективност."
},
"google/gemma-2-2b-it": {
"description": "Лек модел за настройка на инструкции от Google."
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 е ефективен модел, представен от Google, обхващащ множество приложения от малки приложения до сложна обработка на данни."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 е серия от леки отворени текстови модели на Google."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo, подходящ за различни задачи по генериране и разбиране на текст, в момента сочи към gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, модел за генериране на текст с висока капацитет, подходящ за сложни задачи."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo, подходящ за различни задачи по генериране и разбиране на текст, в момента сочи към gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o е динамичен модел, който се актуализира в реално време, за да поддържа най-новата версия. Той съчетава мощно разбиране и генериране на език и е подходящ за мащабни приложения, включително обслужване на клиенти, образование и техническа поддръжка."
},
+ "gpt-4o-audio-preview": {
+ "description": "Модел GPT-4o Audio, поддържащ вход и изход на аудио."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini е най-новият модел на OpenAI, след GPT-4 Omni, който поддържа текстово и визуално въвеждане и генерира текст. Като най-напредналият им малък модел, той е значително по-евтин от другите нови модели и е с над 60% по-евтин от GPT-3.5 Turbo. Запазва най-съвременната интелигентност, като същевременно предлага значителна стойност за парите. GPT-4o mini получи 82% на теста MMLU и в момента е с по-висок рейтинг от GPT-4 по предпочитания за чат."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "Реален вариант на GPT-4o-mini, поддържащ вход и изход на аудио и текст в реално време."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "Реален вариант на GPT-4o, поддържащ вход и изход на аудио и текст в реално време."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "Реален вариант на GPT-4o, поддържащ вход и изход на аудио и текст в реално време."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "Реален вариант на GPT-4o, поддържащ вход и изход на аудио и текст в реално време."
+ },
"grok-2-1212": {
"description": "Този модел е подобрен по отношение на точност, спазване на инструкции и многоезични способности."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "Най-новият модел на HunYuan с MOE архитектура за извикване на функции, обучен с висококачествени данни за извикване на функции, с контекстен прозорец от 32K, водещ в множество измерения на оценъчните показатели."
},
+ "hunyuan-large": {
+ "description": "Моделът Hunyuan-large има общ брой параметри около 389B, активни параметри около 52B, и е най-голямият и най-добър в индустрията отворен MoE модел с архитектура Transformer."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Специализира в обработката на дълги текстови задачи, като резюмета на документи и отговори на въпроси, и също така притежава способността да обработва общи текстови генериращи задачи. Показва отлични резултати в анализа и генерирането на дълги текстове, ефективно справяйки се с комплексни и подробни изисквания за обработка на дълги текстове."
+ },
"hunyuan-lite": {
"description": "Актуализиран до MOE структура, контекстният прозорец е 256k, водещ в множество оценъчни набори в NLP, код, математика и индустрия, пред много от отворените модели."
},
+ "hunyuan-lite-vision": {
+ "description": "Най-новият 7B мултимодален модел на Hunyuan, с контекстен прозорец от 32K, поддържа мултимодални разговори на китайски и английски, разпознаване на обекти в изображения, разбиране на документи и таблици, мултимодална математика и др., с показатели, които надвишават 7B конкурентни модели в множество измерения."
+ },
"hunyuan-pro": {
"description": "Модел с параметри от триллион MOE-32K за дълги текстове. Постига абсолютни водещи нива в различни бенчмаркове, с комплексни инструкции и разсъждения, притежаващи сложни математически способности, поддържа функция за извикване, с акцент върху оптимизацията в области като многоезичен превод, финанси, право и медицина."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Използва по-добра стратегия за маршрутизиране, като същевременно облекчава проблемите с балансирането на натоварването и сближаването на експертите. За дълги текстове, показателят за откритие достига 99.9%. MOE-256K прави допълнителен пробив в дължината и ефективността, значително разширявайки допустимата дължина на входа."
},
+ "hunyuan-standard-vision": {
+ "description": "Най-новият мултимодален модел на Hunyuan, поддържащ отговори на множество езици, с балансирани способности на китайски и английски."
+ },
"hunyuan-turbo": {
"description": "Предварителна версия на новото поколение голям езиков модел на HunYuan, използваща нова структура на смесен експертен модел (MoE), с по-бърза скорост на извеждане и по-силни резултати в сравнение с hunyuan-pro."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Фиксирана версия на hunyuan-turbo от 20 ноември 2024 г., която е между hunyuan-turbo и hunyuan-turbo-latest."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "Оптимизация в тази версия: скалиране на данни и инструкции, значително повишаване на общата генерализационна способност на модела; значително повишаване на математическите, кодовите и логическите способности; оптимизиране на свързаните с разбирането на текста и думите способности; оптимизиране на качеството на генерираното съдържание при създаване на текст."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "Оптимизация на общото изживяване, включително разбиране на NLP, създаване на текст, разговори, отговори на въпроси, превод и специфични области; повишаване на хуманността, оптимизиране на емоционалната интелигентност на модела; подобряване на способността на модела да изяснява при неясни намерения; повишаване на способността за обработка на въпроси, свързани с анализ на думи; подобряване на качеството и интерактивността на създаването; подобряване на многократното изживяване."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "Новото поколение визуално езиково флагманско голямо модел на Hunyuan, използващо нова структура на смесен експертен модел (MoE), с цялостно подобрение на способностите за основно разпознаване, създаване на съдържание, отговори на въпроси и анализ и разсъждение в сравнение с предишното поколение модели."
+ },
"hunyuan-vision": {
"description": "Най-новият мултимодален модел на HunYuan, поддържащ генериране на текстово съдържание от изображения и текстови входове."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "Нашата най-нова серия модели с изключителни способности за извеждане, поддържаща контекстна дължина от 1M и по-силни способности за следване на инструкции и извикване на инструменти."
},
+ "internlm3-latest": {
+ "description": "Нашата най-нова серия модели с изключителна производителност на разсъжденията, водеща в категорията на отворените модели. По подразбиране сочи към най-ново публикуваната серия модели InternLM3."
+ },
+ "jina-deepsearch-v1": {
+ "description": "Дълбокото търсене комбинира интернет търсене, четене и разсъждение, за да извърши обширно разследване. Можете да го разглеждате като агент, който приема вашата изследователска задача - той ще извърши широко търсене и ще премине през множество итерации, преди да предостави отговор. Този процес включва непрекъснато изследване, разсъждение и решаване на проблеми от различни ъгли. Това е коренно различно от стандартните големи модели, които генерират отговори директно от предварително обучени данни, и от традиционните RAG системи, които разчитат на еднократни повърхностни търсения."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM е експериментален езиков модел, специфичен за задачи, обучен да отговаря на принципите на научното обучение, способен да следва системни инструкции в учебни и обучителни сценарии, да действа като експертен ментор и др."
},
"lite": {
"description": "Spark Lite е лек модел на голям език, с изключително ниска латентност и ефективна обработка, напълно безплатен и отворен, поддържащ функции за онлайн търсене в реално време. Неговите бързи отговори го правят отличен за приложения на нискомощни устройства и фина настройка на модели, предоставяйки на потребителите отлична рентабилност и интелигентно изживяване, особено в контекста на въпроси и отговори, генериране на съдържание и търсене."
},
- "llama-3.1-70b-instruct": {
- "description": "Llama 3.1 70B Instruct модел, с 70B параметри, способен да предоставя изключителна производителност в задачи за генериране на текст и инструкции."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B предлага по-мощни способности за разсъждение на AI, подходящи за сложни приложения, поддържащи множество изчислителни обработки и осигуряващи ефективност и точност."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B е модел с висока производителност, предлагащ бързи способности за генериране на текст, особено подходящ за приложения, изискващи мащабна ефективност и икономичност."
},
- "llama-3.1-8b-instruct": {
- "description": "Llama 3.1 8B Instruct модел, с 8B параметри, поддържащ ефективно изпълнение на задачи с визуални указания, предлагащ качествени способности за генериране на текст."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Llama 3.1 Sonar Huge Online модел, с 405B параметри, поддържащ контекстова дължина от около 127,000 маркера, проектиран за сложни онлайн чат приложения."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Llama 3.1 Sonar Large Chat модел, с 70B параметри, поддържащ контекстова дължина от около 127,000 маркера, подходящ за сложни офлайн чат задачи."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Llama 3.1 Sonar Large Online модел, с 70B параметри, поддържащ контекстова дължина от около 127,000 маркера, подходящ за задачи с висока капацитет и разнообразие в чата."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Llama 3.1 Sonar Small Chat модел, с 8B параметри, проектиран за офлайн чат, поддържащ контекстова дължина от около 127,000 маркера."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Llama 3.1 Sonar Small Online модел, с 8B параметри, поддържащ контекстова дължина от около 127,000 маркера, проектиран за онлайн чат, способен да обработва ефективно различни текстови взаимодействия."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 е проектиран да обработва задачи, свързващи визуални и текстови данни. Той показва отлични резултати в задачи като описание на изображения и визуални въпроси и отговори, преодолявайки пропастта между генерирането на език и визуалното разсъждение."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 е най-напредналият многоезичен отворен езиков модел от серията Llama, който предлага производителност, сравнима с 405B моделите, на изключително ниска цена. Базиран на структурата Transformer и подобрен чрез супервизирано фино настройване (SFT) и обучение с човешка обратна връзка (RLHF) за повишаване на полезността и безопасността. Неговата версия, оптимизирана за инструкции, е специално проектирана за многоезични диалози и показва по-добри резултати от много от отворените и затворените чат модели в множество индустриални бенчмаркове. Краен срок за знания: декември 2023."
+ },
"llama-3.3-70b-versatile": {
"description": "Meta Llama 3.3 е многоезичен модел за генерация на език (LLM) с 70B (вход/изход на текст), който е предварително обучен и е пригоден за указания. Чистият текстов модел на Llama 3.3 е оптимизиран за многоезични диалогови случаи и надминава много налични отворени и затворени чат модели на стандартни индустриални тестове."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 е проектирана да обработва задачи, комбиниращи визуални и текстови данни. Тя демонстрира отлични резултати в задачи като описание на изображения и визуални въпроси и отговори, преодолявайки пропастта между генерирането на езици и визуалното разсъждение."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 е най-напредналият многоезичен отворен голям езиков модел от серията Llama, предлагащ производителност, сравнима с 405B моделите на изключително ниска цена. Базиран на структурата Transformer и подобрен чрез супервизирано фино настройване (SFT) и обучение с човешка обратна връзка (RLHF) за повишаване на полезността и безопасността. Неговата версия за оптимизация на инструкции е специално проектирана за многоезични диалози и показва по-добри резултати от много от отворените и затворените чат модели в множество индустриални бенчмаркове. Краен срок за знания: декември 2023 г."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Meta Llama 3.3 многоезичен голям езиков модел (LLM) е предварително обучен и коригиран за инструкции в 70B (текстов вход/текстов изход). Моделът Llama 3.3, коригиран за инструкции, е оптимизиран за многоезични диалогови случаи и превъзхожда много налични отворени и затворени чат модели на общи индустриални бенчмаркове."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 е проектирана да обработва задачи, комбиниращи визуални и текстови данни. Тя демонстрира отлични резултати в задачи като описание на изображения и визуални въпроси и отговори, преодолявайки пропастта между генерирането на езици и визуалното разсъждение."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "405B Llama 3.1 Turbo моделът предлага огромна контекстова поддръжка за обработка на големи данни, с изключителна производителност в приложения с изкуствен интелект с много голям мащаб."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 е водещ модел, представен от Meta, поддържащ до 405B параметри, подходящ за сложни разговори, многоезичен превод и анализ на данни."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B предлага ефективна поддръжка за многоезични диалози."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 е проектиран да обработва задачи, свързващи визуални и текстови данни. Той показва отлични резултати в задачи като описание на изображения и визуални въпроси, преодолявайки пропастта между генерирането на език и визуалното разсъждение."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 е най-напредналият многоезичен отворен езиков модел от серията Llama, който предлага производителност, сравнима с 405B моделите, на изключително ниска цена. Базиран на структурата Transformer и подобрен чрез супервизирано фино настройване (SFT) и обучение с човешка обратна връзка (RLHF) за повишаване на полезността и безопасността. Неговата версия, оптимизирана за инструкции, е специално проектирана за многоезични диалози и показва по-добри резултати от много от отворените и затворените чат модели в множество индустриални бенчмаркове. Краен срок за знания: декември 2023."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 е най-напредналият многоезичен отворен езиков модел от серията Llama, който предлага производителност, сравнима с 405B моделите, на изключително ниска цена. Базиран на структурата Transformer и подобрен чрез супервизирано фино настройване (SFT) и обучение с човешка обратна връзка (RLHF) за повишаване на полезността и безопасността. Неговата версия, оптимизирана за инструкции, е специално проектирана за многоезични диалози и показва по-добри резултати от много от отворените и затворените чат модели в множество индустриални бенчмаркове. Краен срок за знания: декември 2023."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct е най-голямата и най-мощната версия на модела Llama 3.1 Instruct. Това е високо напреднал модел за диалогово разсъждение и генериране на синтетични данни, който може да се използва и като основа за професионално продължително предварително обучение или фино настройване в специфични области. Многоезичният голям езиков модел (LLMs), предоставен от Llama 3.1, е набор от предварително обучени, коригирани по инструкции генеративни модели, включително размери 8B, 70B и 405B (текстов вход/изход). Текстовите модели, коригирани по инструкции (8B, 70B, 405B), са оптимизирани за многоезични диалогови случаи и надминават много налични отворени чат модели в общи индустриални бенчмаркове. Llama 3.1 е проектиран за търговски и изследователски цели на множество езици. Моделите, коригирани по инструкции, са подходящи за чатове, подобни на асистенти, докато предварително обучените модели могат да се адаптират към различни задачи за генериране на естествен език. Моделите на Llama 3.1 също поддържат използването на изхода на модела за подобряване на други модели, включително генериране на синтетични данни и рафиниране. Llama 3.1 е саморегресивен езиков модел, използващ оптимизирана трансформаторна архитектура. Коригираните версии използват супервизирано фино настройване (SFT) и обучение с човешка обратна връзка (RLHF), за да отговорят на предпочитанията на хората за полезност и безопасност."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 е отворен голям езиков модел (LLM), насочен към разработчици, изследователи и предприятия, предназначен да им помогне да изградят, експериментират и отговорно разширят своите идеи за генеративен ИИ. Като част от основната система на глобалната общност за иновации, той е особено подходящ за устройства с ограничени изчислителни ресурси и по-бързо време за обучение."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "Напреднал LLM, поддържащ генериране на синтетични данни, дестилация на знания и разсъждение, подходящ за чатботове, програмиране и специфични задачи."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "Улеснява сложни разговори, с изключителни способности за разбиране на контекста, разсъждение и генериране на текст."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "Напреднал, водещ модел с разбиране на езика, изключителни способности за разсъждение и генериране на текст."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "Водещ визуално-езиков модел, специализиран в извършване на висококачествени разсъждения от изображения."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "Напреднал, водещ малък езиков модел с разбиране на езика, изключителни способности за разсъждение и генериране на текст."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "Напреднал, водещ малък езиков модел с разбиране на езика, изключителни способности за разсъждение и генериране на текст."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "Водещ визуално-езиков модел, специализиран в извършване на висококачествени разсъждения от изображения."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "Напреднал LLM, специализиран в разсъждения, математика, общи познания и извикване на функции."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 е езиков модел, предоставен от Microsoft AI, който показва особени способности в сложни разговори, многоезичност, разсъждения и интелигентни асистенти."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K е модел с изключителна способност за обработка на дълги контексти, подходящ за генериране на много дълги текстове, отговарящи на сложни изисквания за генериране, способен да обработва до 128,000 токена, особено подходящ за научни изследвания, академични и генериране на големи документи."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Визуалният модел Kimi (включително moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview и др.) може да разбира съдържанието на изображения, включително текст в изображения, цветове и форми на обекти."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K предлага средна дължина на контекста, способен да обработва 32,768 токена, особено подходящ за генериране на различни дълги документи и сложни диалози, използван в области като създаване на съдържание, генериране на отчети и диалогови системи."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Визуалният модел Kimi (включително moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview и др.) може да разбира съдържанието на изображения, включително текст в изображения, цветове и форми на обекти."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K е проектиран за генериране на кратки текстови задачи, с ефективна производителност, способен да обработва 8,192 токена, особено подходящ за кратки диалози, бележки и бързо генериране на съдържание."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Визуалният модел Kimi (включително moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview и др.) може да разбира съдържанието на изображения, включително текст в изображения, цветове и форми на обекти."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B е обновена версия на Nous Hermes 2, включваща най-новите вътрешно разработени набори от данни."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B е голям езиков модел, персонализиран от NVIDIA с цел подобряване на отговорите на потребителските запитвания."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B е голям езиков модел, персонализиран от NVIDIA, предназначен да увеличи полезността на отговорите, генерирани от LLM на потребителските запитвания. Моделът показва отлични резултати в бенчмаркове като Arena Hard, AlpacaEval 2 LC и GPT-4-Turbo MT-Bench, като заема първо място в трите автоматизирани теста за подравняване към 1 октомври 2024 г. Моделът е обучен с RLHF (по-специално REINFORCE), Llama-3.1-Nemotron-70B-Reward и HelpSteer2-Preference подсказки на базата на Llama-3.1-70B-Instruct модела."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "Уникален езиков модел, предлагащ ненадмината точност и ефективност."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct е персонализиран голям езиков модел на NVIDIA, предназначен да подобри полезността на отговорите, генерирани от LLM."
+ },
"o1": {
"description": "Фокусиран върху напреднали изводи и решаване на сложни проблеми, включително математически и научни задачи. Изключително подходящ за приложения, изискващи дълбочинно разбиране на контекста и управление на работни потоци."
},
- "o1-2024-12-17": {
- "description": "o1 е новият модел за извод на OpenAI, който поддържа вход и изход на текст и изображения, подходящ за сложни задачи, изискващи обширни общи знания. Моделът разполага с контекст от 200K и крайна дата на знанията от октомври 2023 г."
- },
"o1-mini": {
"description": "o1-mini е бърз и икономичен модел за изводи, проектиран за приложения в програмирането, математиката и науката. Моделът разполага с контекст от 128K и дата на знание до октомври 2023."
},
"o1-preview": {
"description": "o1 е новият модел за изводи на OpenAI, подходящ за сложни задачи, изискващи обширни общи знания. Моделът разполага с контекст от 128K и дата на знание до октомври 2023."
},
+ "o3-mini": {
+ "description": "o3-mini е нашият най-нов малък модел за инференция, който предлага висока интелигентност при същите разходи и цели за закъснение като o1-mini."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba е модел на езика Mamba 2, специализиран в генерирането на код, предоставящ мощна поддръжка за напреднали кодови и разсъждателни задачи."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini е най-новият модел на OpenAI, пуснат след GPT-4 Omni, който поддържа вход и изход на текст и изображения. Като най-напредналият им малък модел, той е значително по-евтин от другите нови модели и е с над 60% по-евтин от GPT-3.5 Turbo. Запазва най-съвременната интелигентност, като предлага значителна стойност за парите. GPT-4o mini получи 82% на теста MMLU и в момента е с по-висок рейтинг от GPT-4 в предпочитанията за чат."
},
- "openai/o1": {
- "description": "o1 е новият модел за извод на OpenAI, който поддържа вход и изход на текст и изображения, подходящ за сложни задачи, изискващи обширни общи знания. Моделът разполага с контекст от 200K и крайна дата на знанията от октомври 2023 г."
- },
"openai/o1-mini": {
"description": "o1-mini е бърз и икономичен модел за изводи, проектиран за приложения в програмирането, математиката и науката. Моделът разполага с контекст от 128K и дата на знание до октомври 2023."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K е конфигуриран с изключителна способност за обработка на контекст, способен да обработва до 128K контекстна информация, особено подходящ за дълги текстове, изискващи цялостен анализ и дългосрочна логическа свързаност, предоставяйки гладка и последователна логика и разнообразна поддръжка на цитати в сложни текстови комуникации."
},
+ "qvq-72b-preview": {
+ "description": "QVQ моделът е експериментален изследователски модел, разработен от екипа на Qwen, фокусиран върху повишаване на визуалните способности за разсъждение, особено в областта на математическото разсъждение."
+ },
"qwen-coder-plus-latest": {
"description": "Модел за кодиране Qwen с общо предназначение."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Qwen-VL Max е модел за визуален език с изключително голям мащаб. В сравнение с подобрената версия, той отново подобрява способността за визуално разсъждение и следване на инструкции, предоставяйки по-високо ниво на визуално възприятие и познание."
},
+ "qwen-vl-ocr-latest": {
+ "description": "Qwen OCR е специализиран модел за извличане на текст, фокусиран върху способността за извличане на текст от изображения на документи, таблици, тестови въпроси, ръкописен текст и др. Той може да разпознава множество езици, включително: китайски, английски, френски, японски, корейски, немски, руски, италиански, виетнамски и арабски."
+ },
"qwen-vl-plus-latest": {
"description": "Моделят за визуален език Qwen-VL Plus е подобрена версия с голям мащаб. Значително подобрява способността за разпознаване на детайли и текст, поддържа резолюция над милион пиксела и изображения с произволно съотношение на страните."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 е нова серия от големи езикови модели с по-силни способности за разбиране и генериране."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "LLM, насочен към китайски и английски, за области като език, програмиране, математика и разсъждение."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "Напреднал LLM, поддържащ генериране на код, разсъждение и корекции, обхващащ основните програмни езици."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "Мощен среден модел за код, поддържащ 32K дължина на контекста, специализиран в многоезично програмиране."
+ },
"qwen2": {
"description": "Qwen2 е новото поколение голям езиков модел на Alibaba, предлагащ отлична производителност за разнообразни приложения."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "Модел с мащаб 14B, отворен за обществеността от Qwen 2.5."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "Qwen2.5 е отворен модел с мащаб 72B."
+ },
"qwen2.5-32b-instruct": {
"description": "Модел с мащаб 32B, отворен за обществеността от Qwen 2.5."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Моделът Qwen-Math притежава силни способности за решаване на математически задачи."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "Подобрение на следването на инструкции, математика, решаване на проблеми и код, повишаване на способността за разпознаване на обекти, поддържа директно точно локализиране на визуални елементи в различни формати, поддържа разбиране на дълги видео файлове (до 10 минути) и локализиране на събития в секунда, може да разбира времеви последователности и скорости, базирано на способности за анализ и локализация, поддържа управление на OS или Mobile агенти, силна способност за извличане на ключова информация и изход в JSON формат, тази версия е 72B, най-силната версия в серията."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "Подобрение на следването на инструкции, математика, решаване на проблеми и код, повишаване на способността за разпознаване на обекти, поддържа директно точно локализиране на визуални елементи в различни формати, поддържа разбиране на дълги видео файлове (до 10 минути) и локализиране на събития в секунда, може да разбира времеви последователности и скорости, базирано на способности за анализ и локализация, поддържа управление на OS или Mobile агенти, силна способност за извличане на ключова информация и изход в JSON формат, тази версия е 72B, най-силната версия в серията."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 е новото поколение мащабен езиков модел на Alibaba, който предлага отлична производителност, за да отговори на разнообразни приложни нужди."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro е високоинтелигентен LLM, пуснат от Upstage, фокусиран върху способността за следване на инструкции с един GPU, с IFEval оценка над 80. В момента поддържа английски, а официалната версия е планирана за пускане през ноември 2024 г., с разширена поддръжка на езици и дължина на контекста."
},
+ "sonar": {
+ "description": "Лек продукт за търсене, базиран на контекст на търсене, по-бърз и по-евтин от Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "Разширен продукт за търсене, който поддържа контекст на търсене, напреднали запитвания и проследяване."
+ },
+ "sonar-reasoning": {
+ "description": "Нови API продукти, поддържани от модела за разсъждение на DeepSeek."
+ },
"step-1-128k": {
"description": "Баланс между производителност и разходи, подходящ за общи сценарии."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "Този модел разполага с мощни способности за разбиране на видео."
},
+ "step-1o-vision-32k": {
+ "description": "Този модел разполага с мощни способности за разбиране на изображения. В сравнение с моделите от серията step-1v, предлага по-силна визуална производителност."
+ },
"step-1v-32k": {
"description": "Поддържа визуални входове, подобряваща мултимодалното взаимодействие."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Поддържа взаимодействия с голям мащаб на контекста, подходящи за сложни диалогови сценарии."
},
+ "step-2-mini": {
+ "description": "Модел с бърза производителност, базиран на новото поколение собствена архитектура Attention MFA, който постига резултати, подобни на step1 с много ниски разходи, като същевременно поддържа по-висока производителност и по-бързо време за отговор. Може да обработва общи задачи и притежава специализирани умения в кодирането."
+ },
+ "taichu2_mm": {
+ "description": "Сливайки способности за разбиране на изображения, прехвърляне на знания и логическо обяснение, моделът показва отлични резултати в областта на въпросите и отговорите на текст и изображения."
+ },
"taichu_llm": {
"description": "Моделът на езика TaiChu е с изключителни способности за разбиране на езика, текстово генериране, отговори на знания, програмиране, математически изчисления, логическо разсъждение, анализ на емоции, резюмиране на текст и др. Иновативно комбинира предварително обучение с големи данни и разнообразни източници на знания, чрез непрекъснато усъвършенстване на алгоритмичните технологии и усвояване на нови знания от масивни текстови данни, за да осигури на потребителите по-удобна информация и услуги, както и по-интелигентно изживяване."
},
+ "text-embedding-3-large": {
+ "description": "Най-мощният модел за векторизация, подходящ за английски и неанглийски задачи."
+ },
+ "text-embedding-3-small": {
+ "description": "Ефективен и икономичен ново поколение модел за вграждане, подходящ за извличане на знания, RAG приложения и други сценарии."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) предлага подобрена изчислителна мощ чрез ефективни стратегии и архитектура на модела."
},
+ "tts-1": {
+ "description": "Най-новият модел за текст в реч, оптимизиран за скорост в реални сценарии."
+ },
+ "tts-1-hd": {
+ "description": "Най-новият модел за текст в реч, оптимизиран за качество."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) е подходящ за прецизни задачи с инструкции, предлагащи отлични способности за обработка на език."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet повишава индустриалните стандарти, с производителност, надминаваща конкурентните модели и Claude 3 Opus, показвайки отлични резултати в широк спектър от оценки, като същевременно предлага скорост и разходи, сравними с нашите модели от средно ниво."
},
+ "whisper-1": {
+ "description": "Универсален модел за разпознаване на реч, поддържащ многоезично разпознаване на реч, превод на реч и разпознаване на езици."
+ },
"wizardlm2": {
"description": "WizardLM 2 е езиков модел, предоставен от Microsoft AI, който се отличава в сложни диалози, многоезичност, разсъждение и интелигентни асистенти."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Модел за сложни визуални задачи, предлагащ висока производителност за разбиране и анализ на изображения."
+ },
+ "yi-vision-v2": {
+ "description": "Модел за сложни визуални задачи, предлагащ висока производителност в разбирането и анализа на базата на множество изображения."
}
}
diff --git a/locales/bg-BG/providers.json b/locales/bg-BG/providers.json
index 9c4f453e6af31..c61bf62622140 100644
--- a/locales/bg-BG/providers.json
+++ b/locales/bg-BG/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure предлага разнообразие от напреднали AI модели, включително GPT-3.5 и най-новата серия GPT-4, поддържащи различни типове данни и сложни задачи, с акцент върху безопасни, надеждни и устойчиви AI решения."
},
+ "azureai": {
+ "description": "Azure предлага множество напреднали AI модели, включително GPT-3.5 и най-новата серия GPT-4, които поддържат различни типове данни и сложни задачи, ангажирани с безопасни, надеждни и устойчиви AI решения."
+ },
"baichuan": {
"description": "Baichuan Intelligence е компания, специализирана в разработката на големи модели за изкуствен интелект, чийто модели показват отлични резултати в китайски задачи, свързани с енциклопедии, обработка на дълги текстове и генериране на съдържание, надминавайки основните чуждестранни модели. Baichuan Intelligence също така притежава индустриално водещи мултимодални способности, показвайки отлични резултати в множество авторитетни оценки. Моделите им включват Baichuan 4, Baichuan 3 Turbo и Baichuan 3 Turbo 128k, оптимизирани за различни приложения, предлагащи решения с висока цена-качество."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek е компания, специализирана в изследвания и приложения на технологии за изкуствен интелект, чийто най-нов модел DeepSeek-V2.5 комбинира способности за общи диалози и обработка на код, постигайки значителни подобрения в съответствието с човешките предпочитания, писателските задачи и следването на инструкции."
},
+ "doubao": {
+ "description": "Модел, разработен от ByteDance. Проверен в над 50 бизнес сценария в рамките на компанията, с ежедневна употреба на триллиони токени, който продължава да се усъвършенства, предоставяйки множество модални възможности и създавайки богато бизнес изживяване с висококачествени модели."
+ },
"fireworksai": {
"description": "Fireworks AI е водещ доставчик на напреднали езикови модели, фокусирайки се върху извикване на функции и мултимодална обработка. Най-новият им модел Firefunction V2, базиран на Llama-3, е оптимизиран за извикване на функции, диалози и следване на инструкции. Визуалният езиков модел FireLLaVA-13B поддържа смесени входове от изображения и текст. Други забележителни модели включват серията Llama и серията Mixtral, предлагащи ефективна поддръжка за многоезично следване на инструкции и генериране."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "Отворена организация, посветена на изследването и разработването на инструменти за големи модели. Предоставя на всички AI разработчици ефективна и лесна за използване отворена платформа, която прави най-съвременните технологии и алгоритми за големи модели достъпни."
},
+ "jina": {
+ "description": "Jina AI е основана през 2020 г. и е водеща компания в областта на търсенето с AI. Нашата платформа за търсене включва векторни модели, реорганизатори и малки езикови модели, които помагат на предприятията да изградят надеждни и висококачествени генеративни AI и мултимодални приложения за търсене."
+ },
+ "lmstudio": {
+ "description": "LM Studio е настолно приложение за разработка и експериментиране с LLMs на вашия компютър."
+ },
"minimax": {
"description": "MiniMax е компания за универсален изкуствен интелект, основана през 2021 г., която се стреми да създаде интелигентност заедно с потребителите. MiniMax е разработила различни универсални големи модели, включително текстови модели с трилйон параметри, модели за глас и модели за изображения. Също така е пуснала приложения като Conch AI."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI е платформа, предлагаща API услуги за множество големи езикови модели и генериране на AI изображения, гъвкава, надеждна и икономически ефективна. Поддържа най-новите отворени модели, като Llama3 и Mistral, и предлага цялостни, потребителски приятелски и автоматично разширяеми API решения за разработка на генеративни AI приложения, подходящи за бързото развитие на AI стартъпи."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ предлага контейнери, които могат да се използват за самостоятелно хоствани GPU ускорени инференционни микросервизи, поддържащи разгръщането на предварително обучени и персонализирани AI модели в облака, центрове за данни, RTX™ AI персонални компютри и работни станции."
+ },
"ollama": {
"description": "Моделите, предоставени от Ollama, обхващат широк спектър от области, включително генериране на код, математически операции, многоезично обработване и диалогова интеракция, отговарящи на разнообразните нужди на предприятията и локализирани внедрявания."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "Институтът по автоматизация на Китайската академия на науките и Институтът по изкуствен интелект в Ухан представят ново поколение мултимодални големи модели, поддържащи многократни въпроси и отговори, текстово създаване, генериране на изображения, 3D разбиране, анализ на сигнали и др., с по-силни способности за познание, разбиране и създаване, предоставяйки ново взаимодействие."
},
+ "tencentcloud": {
+ "description": "Атомни способности на знаниевия двигател (LLM Knowledge Engine Atomic Power) са базирани на разработката на знаниевия двигател и предлагат пълна верига от способности за въпроси и отговори, насочени към предприятия и разработчици, предоставяйки гъвкави възможности за изграждане и разработка на моделни приложения. Можете да изградите собствена моделна услуга чрез множество атомни способности, като използвате услуги за анализ на документи, разделяне, вграждане, многократни пренаписвания и др., за да персонализирате AI бизнеса, специфичен за вашето предприятие."
+ },
"togetherai": {
"description": "Together AI се стреми да постигне водеща производителност чрез иновационни AI модели, предлагащи широки възможности за персонализация, включително бърза поддръжка за разширяване и интуитивни процеси на внедряване, отговарящи на разнообразните нужди на предприятията."
},
"upstage": {
"description": "Upstage се фокусира върху разработването на AI модели за различни бизнес нужди, включително Solar LLM и документен AI, с цел постигане на човешки универсален интелект (AGI). Създава прости диалогови агенти чрез Chat API и поддържа извикване на функции, превод, вграждане и специфични приложения."
},
+ "vllm": {
+ "description": "vLLM е бърза и лесна за използване библиотека за LLM инференция и услуги."
+ },
+ "volcengine": {
+ "description": "Платформа за разработка на услуги с големи модели, пусната от ByteDance, предлагаща богати на функции, безопасни и конкурентни по цена услуги за извикване на модели. Освен това предоставя край до край функции като данни за модели, фина настройка, инференция и оценка, за да осигури всестранна подкрепа за разработката на вашите AI приложения."
+ },
"wenxin": {
"description": "Платформа за разработка и услуги на корпоративно ниво, предлагаща цялостно решение за разработка на генеративни модели на изкуствен интелект и приложения, с най-пълния и лесен за използване инструментариум за целия процес на разработка на модели и приложения."
},
diff --git a/locales/bg-BG/setting.json b/locales/bg-BG/setting.json
index bd1cf469bdcaa..a6863667fbac1 100644
--- a/locales/bg-BG/setting.json
+++ b/locales/bg-BG/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Активиране на ограничението за максимален брой токени"
},
+ "enableReasoningEffort": {
+ "title": "Активиране на настройките за интензивност на разсъжденията"
+ },
"frequencyPenalty": {
- "desc": "Колкото по-висока е стойността, толкова по-вероятно е да се намалят повтарящите се думи",
- "title": "Наказание за честота"
+ "desc": "Колкото по-голяма е стойността, толкова по-богат и разнообразен е речникът; колкото по-ниска е стойността, толкова по-прост и обикновен е речникът.",
+ "title": "Богатство на речника"
},
"maxTokens": {
"desc": "Максималният брой токени, използвани за всяко взаимодействие",
@@ -212,19 +215,31 @@
"desc": "{{provider}} модел",
"title": "Модел"
},
+ "params": {
+ "title": "Разширени параметри"
+ },
"presencePenalty": {
- "desc": "Колкото по-висока е стойността, толкова по-вероятно е да се разшири до нови теми",
- "title": "Свежест на темата"
+ "desc": "Колкото по-голяма е стойността, толкова по-склонен е към различни изрази, избягвайки повторение на концепции; колкото по-ниска е стойността, толкова по-склонен е да използва повторение на концепции или разкази, изразявайки по-голяма последователност.",
+ "title": "Разнообразие на изразите"
+ },
+ "reasoningEffort": {
+ "desc": "Колкото по-висока е стойността, толкова по-силна е способността за разсъждение, но може да увеличи времето за отговор и консумацията на токени",
+ "options": {
+ "high": "Висока",
+ "low": "Ниска",
+ "medium": "Средна"
+ },
+ "title": "Интензивност на разсъжденията"
},
"temperature": {
- "desc": "Колкото по-висока е стойността, толкова по-случаен е отговорът",
- "title": "Случайност",
- "titleWithValue": "Случайност {{value}}"
+ "desc": "Колкото по-голямо е числото, толкова по-креативни и въображаеми са отговорите; колкото по-малко е числото, толкова по-строги са отговорите",
+ "title": "Креативна активност",
+ "warning": "Ако стойността на креативната активност е твърде голяма, изходът може да съдържа грешки"
},
"title": "Настройки на модела",
"topP": {
- "desc": "Подобно на случайността, но не се променя заедно със случайността",
- "title": "Top P вземане на проби"
+ "desc": "Колко възможности да се вземат предвид, по-голямата стойност приема повече възможни отговори; по-малката стойност предпочита най-вероятния отговор. Не се препоръчва да се променя заедно с креативната активност",
+ "title": "Отвореност на мисленето"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Общи настройки",
"experiment": "Експеримент",
"llm": "Езиков модел",
+ "provider": "AI доставчик",
"sync": "Синхронизиране в облака",
"system-agent": "Системен асистент",
"tts": "Текст към реч"
diff --git a/locales/de-DE/changelog.json b/locales/de-DE/changelog.json
index e397bb9fdb41f..dab201754545c 100644
--- a/locales/de-DE/changelog.json
+++ b/locales/de-DE/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "Alle Änderungsprotokolle anzeigen",
"description": "Verfolgen Sie die neuen Funktionen und Verbesserungen von {{appName}} kontinuierlich",
"pagination": {
- "older": "Ältere Änderungen anzeigen",
- "prev": "Vorherige Seite"
+ "next": "Nächste Seite",
+ "older": "Ältere Änderungen anzeigen"
},
"readDetails": "Details lesen",
"title": "Änderungsprotokoll",
diff --git a/locales/de-DE/common.json b/locales/de-DE/common.json
index 6fb1994d268ff..702242bae4f78 100644
--- a/locales/de-DE/common.json
+++ b/locales/de-DE/common.json
@@ -275,6 +275,7 @@
},
"temp": "Temporär",
"terms": "Nutzungsbedingungen",
+ "update": "Aktualisieren",
"updateAgent": "Assistentenprofil aktualisieren",
"upgradeVersion": {
"action": "Aktualisieren",
@@ -285,6 +286,7 @@
"anonymousNickName": "Anonymer Benutzer",
"billing": "Abrechnung verwalten",
"cloud": "Erleben Sie {{name}}",
+ "community": "Gemeinschaftsversion",
"data": "Daten speichern",
"defaultNickname": "Community User",
"discord": "Community-Support",
@@ -294,7 +296,6 @@
"help": "Hilfezentrum",
"moveGuide": "Die Einstellungen wurden hierher verschoben.",
"plans": "Abonnementpläne",
- "preview": "Vorschau",
"profile": "Kontoverwaltung",
"setting": "App-Einstellungen",
"usages": "Nutzungsstatistiken"
diff --git a/locales/de-DE/components.json b/locales/de-DE/components.json
index e7dd0cbca3ffb..319b1d613b18e 100644
--- a/locales/de-DE/components.json
+++ b/locales/de-DE/components.json
@@ -76,6 +76,7 @@
"custom": "Benutzerdefiniertes Modell, standardmäßig unterstützt es sowohl Funktionsaufrufe als auch visuelle Erkennung. Bitte überprüfen Sie die Verfügbarkeit dieser Fähigkeiten basierend auf den tatsächlichen Gegebenheiten.",
"file": "Dieses Modell unterstützt das Hochladen von Dateien und deren Erkennung.",
"functionCall": "Dieses Modell unterstützt Funktionsaufrufe.",
+ "reasoning": "Dieses Modell unterstützt tiefes Denken",
"tokens": "Dieses Modell unterstützt maximal {{tokens}} Tokens pro Sitzung.",
"vision": "Dieses Modell unterstützt die visuelle Erkennung."
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "Kein aktiviertes Modell. Bitte gehen Sie zu den Einstellungen, um es zu aktivieren.",
"provider": "Anbieter"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "Aufgrund von Sicherheitsbeschränkungen im Browser müssen Sie CORS für Ollama konfigurieren, um es ordnungsgemäß nutzen zu können.",
+ "linux": {
+ "env": "Fügen Sie im Abschnitt [Service] `Environment` hinzu und setzen Sie die Umgebungsvariable OLLAMA_ORIGINS:",
+ "reboot": "Laden Sie systemd neu und starten Sie Ollama neu",
+ "systemd": "Rufen Sie systemd auf, um den ollama-Dienst zu bearbeiten:"
+ },
+ "macos": "Bitte öffnen Sie die „Terminal“-Anwendung, fügen Sie die folgenden Befehle ein und drücken Sie die Eingabetaste, um sie auszuführen",
+ "reboot": "Bitte starten Sie den Ollama-Dienst nach Abschluss der Ausführung neu",
+ "title": "Konfigurieren Sie Ollama für den CORS-Zugriff",
+ "windows": "Klicken Sie unter Windows auf „Systemsteuerung“ und gehen Sie zu den Systemeigenschaften. Erstellen Sie eine neue Umgebungsvariable mit dem Namen „OLLAMA_ORIGINS“ für Ihr Benutzerkonto, setzen Sie den Wert auf * und klicken Sie auf „OK/Übernehmen“, um zu speichern"
+ },
+ "install": {
+ "description": "Bitte stellen Sie sicher, dass Sie Ollama gestartet haben. Wenn Sie Ollama nicht heruntergeladen haben, besuchen Sie die offizielle Website <1>zum Herunterladen",
+ "docker": "Wenn Sie lieber Docker verwenden möchten, bietet Ollama auch offizielle Docker-Images an, die Sie mit dem folgenden Befehl herunterladen können:",
+ "linux": {
+ "command": "Installieren Sie es mit dem folgenden Befehl:",
+ "manual": "Alternativ können Sie auch die <1>Linux-Handbuchinstallation zur Selbstinstallation konsultieren"
+ },
+ "title": "Ollama-Anwendung lokal installieren und starten",
+ "windowsTab": "Windows (Vorschau)"
+ }
+ },
+ "Thinking": {
+ "thinking": "Tiefes Nachdenken...",
+ "thought": "Tiefgründig nachgedacht (Dauer: {{duration}} Sekunden)",
+ "thoughtWithDuration": "Tiefgründig nachgedacht"
}
}
diff --git a/locales/de-DE/discover.json b/locales/de-DE/discover.json
index 4102d2df10bd9..789c094d43086 100644
--- a/locales/de-DE/discover.json
+++ b/locales/de-DE/discover.json
@@ -126,6 +126,10 @@
"title": "Themenfrische"
},
"range": "Bereich",
+ "reasoning_effort": {
+ "desc": "Diese Einstellung steuert die Intensität des Denkprozesses des Modells, bevor es eine Antwort generiert. Eine niedrige Intensität priorisiert die Geschwindigkeit der Antwort und spart Token, während eine hohe Intensität eine umfassendere Argumentation bietet, jedoch mehr Token verbraucht und die Antwortgeschwindigkeit verringert. Der Standardwert ist mittel, um eine Balance zwischen Genauigkeit des Denkens und Antwortgeschwindigkeit zu gewährleisten.",
+ "title": "Denkintensität"
+ },
"temperature": {
"desc": "Diese Einstellung beeinflusst die Vielfalt der Antworten des Modells. Niedrigere Werte führen zu vorhersehbareren und typischen Antworten, während höhere Werte zu vielfältigeren und weniger häufigen Antworten anregen. Wenn der Wert auf 0 gesetzt wird, gibt das Modell für einen bestimmten Input immer die gleiche Antwort.",
"title": "Zufälligkeit"
diff --git a/locales/de-DE/modelProvider.json b/locales/de-DE/modelProvider.json
index b2ab1569b0de9..1e0c7e4f484cb 100644
--- a/locales/de-DE/modelProvider.json
+++ b/locales/de-DE/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API Key"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "API-Version von Azure, im Format YYYY-MM-DD, siehe [aktuelle Version](https://learn.microsoft.com/de-de/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "Liste abrufen",
+ "title": "Azure API-Version"
+ },
+ "endpoint": {
+ "desc": "Finden Sie den Endpunkt für die Modellinferenz von Azure AI im Überblick über das Azure AI-Projekt",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Azure AI-Endpunkt"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Finden Sie den API-Schlüssel im Überblick über das Azure AI-Projekt",
+ "placeholder": "Azure-Schlüssel",
+ "title": "Schlüssel"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "Geben Sie Ihre AWS Access Key Id ein",
@@ -63,6 +81,46 @@
"title": "Cloudflare-Kundenkennung / API-Adresse"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Bitte geben Sie Ihren API-Schlüssel ein",
+ "title": "API-Schlüssel"
+ },
+ "basicTitle": "Grundinformationen",
+ "configTitle": "Konfigurationsinformationen",
+ "confirm": "Neu erstellen",
+ "createSuccess": "Erstellung erfolgreich",
+ "description": {
+ "placeholder": "Beschreibung des Anbieters (optional)",
+ "title": "Beschreibung des Anbieters"
+ },
+ "id": {
+ "desc": "Eindeutige Kennung des Anbieters, die nach der Erstellung nicht mehr geändert werden kann",
+ "format": "Darf nur aus Zahlen, Kleinbuchstaben, Bindestrichen (-) und Unterstrichen (_) bestehen",
+ "placeholder": "Empfohlen in Kleinbuchstaben, z.B. openai, nach der Erstellung nicht mehr änderbar",
+ "required": "Bitte geben Sie die Anbieter-ID ein",
+ "title": "Anbieter-ID"
+ },
+ "logo": {
+ "required": "Bitte laden Sie das korrekte Anbieter-Logo hoch",
+ "title": "Anbieter-Logo"
+ },
+ "name": {
+ "placeholder": "Bitte geben Sie den angezeigten Namen des Anbieters ein",
+ "required": "Bitte geben Sie den Namen des Anbieters ein",
+ "title": "Name des Anbieters"
+ },
+ "proxyUrl": {
+ "required": "Bitte geben Sie die Proxy-Adresse ein",
+ "title": "Proxy-Adresse"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Bitte wählen Sie den SDK-Typ aus",
+ "title": "Anforderungsformat"
+ },
+ "title": "Erstellen Sie einen benutzerdefinierten AI-Anbieter"
+ },
"github": {
"personalAccessToken": {
"desc": "Geben Sie Ihr GitHub-PAT ein und klicken Sie [hier](https://github.com/settings/tokens), um eines zu erstellen.",
@@ -77,6 +135,23 @@
"title": "HuggingFace-Token"
}
},
+ "list": {
+ "title": {
+ "disabled": "Dienstanbieter nicht aktiviert",
+ "enabled": "Dienstanbieter aktiviert"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Benutzerdefinierten Anbieter hinzufügen",
+ "all": "Alle",
+ "list": {
+ "disabled": "Nicht aktiviert",
+ "enabled": "Aktiviert"
+ },
+ "notFound": "Keine Suchergebnisse gefunden",
+ "searchProviders": "Anbieter suchen...",
+ "sort": "Benutzerdefinierte Sortierung"
+ },
"ollama": {
"checker": {
"desc": "Testen Sie, ob die Proxy-Adresse korrekt eingetragen wurde",
@@ -94,33 +169,9 @@
"title": "Lade Modell {{model}} herunter"
},
"endpoint": {
- "desc": "Geben Sie die Proxy-Adresse der Ollama-Schnittstelle ein, leer lassen, wenn lokal nicht spezifiziert",
+ "desc": "Muss http(s):// enthalten, kann leer gelassen werden, wenn lokal nicht zusätzlich angegeben.",
"title": "Schnittstellen-Proxy-Adresse"
},
- "setup": {
- "cors": {
- "description": "Aufgrund von Browser-Sicherheitsbeschränkungen müssen Sie die CORS-Einstellungen für Ollama konfigurieren, um es ordnungsgemäß zu verwenden.",
- "linux": {
- "env": "Fügen Sie unter [Service] `Environment` hinzu und setzen Sie die Umgebungsvariable OLLAMA_ORIGINS:",
- "reboot": "Systemd neu laden und Ollama neu starten",
- "systemd": "Rufen Sie systemd auf, um den Ollama-Dienst zu bearbeiten:"
- },
- "macos": "Öffnen Sie das Terminal und fügen Sie den folgenden Befehl ein, um fortzufahren.",
- "reboot": "Starten Sie den Ollama-Dienst nach Abschluss der Ausführung neu.",
- "title": "Konfigurieren Sie Ollama für den Zugriff über CORS",
- "windows": "Klicken Sie auf Windows auf 'Systemsteuerung', um die Systemumgebungsvariablen zu bearbeiten. Erstellen Sie eine Umgebungsvariable namens 'OLLAMA_ORIGINS' für Ihr Benutzerkonto mit dem Wert '*', und klicken Sie auf 'OK/Anwenden', um zu speichern."
- },
- "install": {
- "description": "Stelle sicher, dass du Ollama aktiviert hast. Wenn du Ollama noch nicht heruntergeladen hast, besuche die offizielle Website, um es <1>herunterzuladen.",
- "docker": "Wenn Sie Docker bevorzugen, bietet Ollama auch offizielle Docker-Images an. Sie können sie mit dem folgenden Befehl abrufen:",
- "linux": {
- "command": "Installieren Sie mit dem folgenden Befehl:",
- "manual": "Alternativ können Sie die <1>Linux-Installationsanleitung für die manuelle Installation verwenden."
- },
- "title": "Installieren und starten Sie die lokale Ollama-Anwendung",
- "windowsTab": "Windows (Vorschau)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Cancel Download",
@@ -131,25 +182,145 @@
"title": "Download specified Ollama model"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Geben Sie den Access Key der Baidu Qianfan-Plattform ein",
- "placeholder": "Qianfan Access Key",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "Ihr Schlüssel und die Proxy-Adresse werden mit dem <1>AES-GCM-Verschlüsselungsalgorithmus verschlüsselt",
+ "apiKey": {
+ "desc": "Bitte geben Sie Ihren {{name}} API-Schlüssel ein",
+ "placeholder": "{{name}} API-Schlüssel",
+ "title": "API-Schlüssel"
+ },
+ "baseURL": {
+ "desc": "Muss http(s):// enthalten",
+ "invalid": "Bitte geben Sie eine gültige URL ein",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "API-Proxy-Adresse"
+ },
+ "checker": {
+ "button": "Überprüfen",
+ "desc": "Testen Sie, ob der API-Schlüssel und die Proxy-Adresse korrekt eingegeben wurden",
+ "pass": "Überprüfung bestanden",
+ "title": "Verbindungsprüfung"
+ },
+ "fetchOnClient": {
+ "desc": "Der Client-Anforderungsmodus initiiert die Sitzung direkt aus dem Browser, was die Reaktionsgeschwindigkeit erhöhen kann",
+ "title": "Client-Anforderungsmodus verwenden"
+ },
+ "helpDoc": "Konfigurationsanleitung",
+ "waitingForMore": "Weitere Modelle werden <1>geplant, bitte warten Sie"
},
- "checker": {
- "desc": "Überprüfen Sie, ob der AccessKey / SecretAccess korrekt eingegeben wurde"
+ "createNew": {
+ "title": "Erstellen Sie ein benutzerdefiniertes AI-Modell"
+ },
+ "item": {
+ "config": "Modell konfigurieren",
+ "customModelCards": {
+ "addNew": "Erstellen und hinzufügen {{id}} Modell",
+ "confirmDelete": "Das benutzerdefinierte Modell wird gelöscht, nach dem Löschen kann es nicht wiederhergestellt werden. Bitte vorsichtig vorgehen."
+ },
+ "delete": {
+ "confirm": "Bestätigen Sie das Löschen des Modells {{displayName}}?",
+ "success": "Löschung erfolgreich",
+ "title": "Modell löschen"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Feld, das in Azure OpenAI tatsächlich angefordert wird",
+ "placeholder": "Bitte geben Sie den Modellbereitstellungsnamen in Azure ein",
+ "title": "Modellbereitstellungsname"
+ },
+ "deployName": {
+ "extra": "Dieses Feld wird als Modell-ID gesendet, wenn die Anfrage gesendet wird",
+ "placeholder": "Bitte geben Sie den tatsächlichen Namen oder die ID des bereitgestellten Modells ein",
+ "title": "Modellbereitstellungsname"
+ },
+ "displayName": {
+ "placeholder": "Bitte geben Sie den angezeigten Namen des Modells ein, z.B. ChatGPT, GPT-4 usw.",
+ "title": "Anzeigename des Modells"
+ },
+ "files": {
+ "extra": "Der aktuelle Datei-Upload ist nur eine Hack-Lösung und nur für eigene Versuche gedacht. Warten Sie auf die vollständige Datei-Upload-Funktionalität.",
+ "title": "Datei-Upload unterstützen"
+ },
+ "functionCall": {
+ "extra": "Diese Konfiguration aktiviert nur die Fähigkeit des Modells, Werkzeuge zu verwenden, und ermöglicht es, pluginartige Werkzeuge hinzuzufügen. Ob das Modell tatsächlich in der Lage ist, Werkzeuge zu verwenden, hängt jedoch vollständig vom Modell selbst ab. Bitte testen Sie die Verwendbarkeit selbst.",
+ "title": "Unterstützung der Werkzeugnutzung"
+ },
+ "id": {
+ "extra": "Nach der Erstellung nicht mehr änderbar, wird als Modell-ID verwendet, wenn AI aufgerufen wird",
+ "placeholder": "Bitte Modell-ID eingeben, z. B. gpt-4o oder claude-3.5-sonnet",
+ "title": "Modell-ID"
+ },
+ "modalTitle": "Benutzerdefinierte Modellkonfiguration",
+ "reasoning": {
+ "extra": "Diese Konfiguration aktiviert nur die Fähigkeit des Modells zu tiefem Denken. Die tatsächlichen Ergebnisse hängen vollständig vom Modell selbst ab. Bitte testen Sie selbst, ob das Modell über die Fähigkeit zum tiefen Denken verfügt.",
+ "title": "Unterstützung für tiefes Denken"
+ },
+ "tokens": {
+ "extra": "Maximale Token-Anzahl für das Modell festlegen",
+ "title": "Maximales Kontextfenster",
+ "unlimited": "Unbegrenzt"
+ },
+ "vision": {
+ "extra": "Diese Konfiguration aktiviert nur die Bild-Upload-Funktionalität in der Anwendung. Ob die Erkennung unterstützt wird, hängt vollständig vom Modell selbst ab. Bitte testen Sie die Verwendbarkeit der visuellen Erkennungsfähigkeit des Modells selbst.",
+ "title": "Visuelle Erkennung unterstützen"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/Bild",
+ "inputCharts": "${{amount}}/M Zeichen",
+ "inputMinutes": "${{amount}}/Minuten",
+ "inputTokens": "Eingabe ${{amount}}/M",
+ "outputTokens": "Ausgabe ${{amount}}/M"
+ },
+ "releasedAt": "Veröffentlicht am {{releasedAt}}"
},
- "secretKey": {
- "desc": "Geben Sie den Secret Key der Baidu Qianfan-Plattform ein",
- "placeholder": "Qianfan Secret Key",
- "title": "Secret Key"
+ "list": {
+ "addNew": "Modell hinzufügen",
+ "disabled": "Nicht aktiviert",
+ "disabledActions": {
+ "showMore": "Alle anzeigen"
+ },
+ "empty": {
+ "desc": "Bitte erstellen Sie ein benutzerdefiniertes Modell oder ziehen Sie ein Modell, um zu beginnen.",
+ "title": "Keine verfügbaren Modelle"
+ },
+ "enabled": "Aktiviert",
+ "enabledActions": {
+ "disableAll": "Alle deaktivieren",
+ "enableAll": "Alle aktivieren",
+ "sort": "Benutzerdefinierte Modellreihenfolge"
+ },
+ "enabledEmpty": "Keine aktivierten Modelle vorhanden, bitte aktivieren Sie Ihre bevorzugten Modelle aus der Liste unten~",
+ "fetcher": {
+ "clear": "Abgerufene Modelle löschen",
+ "fetch": "Modellliste abrufen",
+ "fetching": "Modellliste wird abgerufen...",
+ "latestTime": "Letzte Aktualisierung: {{time}}",
+ "noLatestTime": "Liste wurde noch nicht abgerufen"
+ },
+ "resetAll": {
+ "conform": "Möchten Sie alle Änderungen am aktuellen Modell wirklich zurücksetzen? Nach dem Zurücksetzen wird die aktuelle Modellliste auf den Standardzustand zurückgesetzt.",
+ "success": "Zurücksetzen erfolgreich",
+ "title": "Alle Änderungen zurücksetzen"
+ },
+ "search": "Modelle suchen...",
+ "searchResult": "{{count}} Modelle gefunden",
+ "title": "Modellliste",
+ "total": "Insgesamt {{count}} verfügbare Modelle"
},
- "unlock": {
- "customRegion": "Benutzerdefinierte Service-Region",
- "description": "Geben Sie Ihren AccessKey / SecretKey ein, um die Sitzung zu starten. Die Anwendung speichert Ihre Authentifizierungseinstellungen nicht.",
- "title": "Verwendung benutzerdefinierter Wenxin-Yiyan-Authentifizierungsinformationen"
- }
+ "searchNotFound": "Keine Suchergebnisse gefunden"
+ },
+ "sortModal": {
+ "success": "Sortierung erfolgreich aktualisiert",
+ "title": "Benutzerdefinierte Sortierung",
+ "update": "Aktualisieren"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "Der AI-Anbieter wird gelöscht, nach dem Löschen kann er nicht wiederhergestellt werden. Bestätigen Sie, ob Sie löschen möchten?",
+ "deleteSuccess": "Löschung erfolgreich",
+ "tooltip": "Aktualisieren Sie die grundlegenden Anbieterinformationen",
+ "updateSuccess": "Aktualisierung erfolgreich"
},
"zeroone": {
"title": "01.AI Alles und Nichts"
diff --git a/locales/de-DE/models.json b/locales/de-DE/models.json
index 6dd2cf56d0e3e..b5f45e6aeba78 100644
--- a/locales/de-DE/models.json
+++ b/locales/de-DE/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K betont semantische Sicherheit und verantwortungsbewusste Ausrichtung, speziell für Anwendungen mit hohen Anforderungen an die Inhaltssicherheit konzipiert, um die Genauigkeit und Robustheit der Benutzererfahrung zu gewährleisten."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 verwendet Baumsuche zur Konstruktion von Denkketten und führt einen Reflexionsmechanismus ein, der durch verstärkendes Lernen trainiert wird. Das Modell verfügt über die Fähigkeit zur Selbstreflexion und Fehlerkorrektur."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro ist ein fortschrittliches Modell zur Verarbeitung natürlicher Sprache, das von der 360 Company entwickelt wurde und über außergewöhnliche Textgenerierungs- und Verständnisfähigkeiten verfügt, insbesondere im Bereich der Generierung und Kreativität, und in der Lage ist, komplexe Sprachumwandlungs- und Rollendarstellungsaufgaben zu bewältigen."
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1 verwendet Baumsuche zur Konstruktion von Denkketten und führt einen Reflexionsmechanismus ein, der durch verstärkendes Lernen trainiert wird. Das Modell verfügt über die Fähigkeit zur Selbstreflexion und Fehlerkorrektur."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra ist die leistungsstärkste Version der Spark-Großmodellreihe, die die Online-Suchverbindung aktualisiert und die Fähigkeit zur Textverständnis und -zusammenfassung verbessert. Es ist eine umfassende Lösung zur Steigerung der Büroproduktivität und zur genauen Reaktion auf Anforderungen und ein führendes intelligentes Produkt in der Branche."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "Das Modell hat die höchste Leistungsfähigkeit im Inland und übertrifft ausländische Mainstream-Modelle in Aufgaben wie Wissensdatenbanken, langen Texten und kreativen Generierungen auf Chinesisch. Es verfügt auch über branchenführende multimodale Fähigkeiten und zeigt in mehreren anerkannten Bewertungsbenchmarks hervorragende Leistungen."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Das DeepSeek-R1-Distill-Modell basiert auf Qwen2.5-Math-1.5B und optimiert die Inferenzleistung durch verstärkendes Lernen und Kaltstartdaten. Das Open-Source-Modell setzt neue Maßstäbe für Multitasking."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Das DeepSeek-R1-Distill-Modell basiert auf Qwen2.5-14B und optimiert die Inferenzleistung durch verstärkendes Lernen und Kaltstartdaten. Das Open-Source-Modell setzt neue Maßstäbe für Multitasking."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "Die DeepSeek-R1-Serie optimiert die Inferenzleistung durch verstärkendes Lernen und Kaltstartdaten, das Open-Source-Modell setzt neue Maßstäbe für Multitasking und übertrifft das Niveau von OpenAI-o1-mini."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "Das DeepSeek-R1-Distill-Modell basiert auf Qwen2.5-Math-7B und optimiert die Inferenzleistung durch verstärkendes Lernen und Kaltstartdaten. Das Open-Source-Modell setzt neue Maßstäbe für Multitasking."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite bietet eine extrem hohe Reaktionsgeschwindigkeit und ein hervorragendes Preis-Leistungs-Verhältnis und bietet den Kunden flexiblere Optionen für verschiedene Szenarien. Es unterstützt Schlussfolgerungen und Feinabstimmungen mit einem 128k-Kontextfenster."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Das von Baidu entwickelte Flaggschiff-Modell für ultra-große Sprachverarbeitung, das im Vergleich zu ERNIE 3.5 eine umfassende Verbesserung der Modellfähigkeiten erreicht hat und sich breit für komplexe Aufgaben in verschiedenen Bereichen eignet; unterstützt die automatische Anbindung an das Baidu-Such-Plugin, um die Aktualität der Antwortinformationen zu gewährleisten."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Das von Baidu entwickelte Flaggschiff-Modell für groß angelegte Sprachverarbeitung zeigt hervorragende Gesamtergebnisse und ist weit verbreitet in komplexen Aufgabenbereichen anwendbar; unterstützt die automatische Anbindung an das Baidu-Suchplugin, um die Aktualität der Frage-Antwort-Informationen zu gewährleisten. Im Vergleich zu ERNIE 4.0 zeigt es eine bessere Leistungsfähigkeit."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "Baidus selbstentwickeltes Flaggschiff-Modell für großflächige Sprachverarbeitung, das in vielen komplexen Aufgaben hervorragende Ergebnisse zeigt und umfassend in verschiedenen Bereichen eingesetzt werden kann; unterstützt die automatische Anbindung an Baidu-Suchplugins, um die Aktualität von Antwortinformationen zu gewährleisten. Im Vergleich zu ERNIE 4.0 hat es eine bessere Leistung."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct ist eines der neuesten großen Sprachmodelle, die von Alibaba Cloud veröffentlicht wurden. Dieses 7B-Modell hat signifikante Verbesserungen in den Bereichen Codierung und Mathematik. Das Modell bietet auch mehrsprachige Unterstützung und deckt über 29 Sprachen ab, einschließlich Chinesisch und Englisch. Es zeigt signifikante Verbesserungen in der Befolgung von Anweisungen, im Verständnis strukturierter Daten und in der Generierung strukturierter Ausgaben (insbesondere JSON)."
},
+ "MiniMax-Text-01": {
+ "description": "In der MiniMax-01-Serie haben wir mutige Innovationen vorgenommen: Erstmals wurde die lineare Aufmerksamkeitsmechanismus in großem Maßstab implementiert, sodass die traditionelle Transformer-Architektur nicht mehr die einzige Wahl ist. Dieses Modell hat eine Parameteranzahl von bis zu 456 Milliarden, wobei eine Aktivierung 45,9 Milliarden beträgt. Die Gesamtleistung des Modells kann mit den besten Modellen im Ausland mithalten und kann gleichzeitig effizient den weltweit längsten Kontext von 4 Millionen Tokens verarbeiten, was 32-mal so viel wie GPT-4o und 20-mal so viel wie Claude-3.5-Sonnet ist."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO ist eine hochflexible Multi-Modell-Kombination, die darauf abzielt, außergewöhnliche kreative Erlebnisse zu bieten."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 zeigt herausragende Leistungen in verschiedenen visuellen Sprachaufgaben, einschließlich Dokumenten- und Diagrammverständnis, Szenentexterkennung, OCR, wissenschaftlicher und mathematischer Problemlösung."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 zeigt herausragende Leistungen in verschiedenen visuellen Sprachaufgaben, einschließlich Dokumenten- und Diagrammverständnis, Szenentexterkennung, OCR, wissenschaftlicher und mathematischer Problemlösung."
- },
"Phi-3-medium-128k-instruct": {
"description": "Das gleiche Phi-3-medium-Modell, jedoch mit einer größeren Kontextgröße für RAG oder Few-Shot-Prompting."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat ist die Open-Source-Version des GLM-4-Modells, das von Zhizhu AI eingeführt wurde. Dieses Modell zeigt hervorragende Leistungen in den Bereichen Semantik, Mathematik, Inferenz, Code und Wissen. Neben der Unterstützung für mehrstufige Dialoge bietet GLM-4-9B-Chat auch fortgeschrittene Funktionen wie Web-Browsing, Code-Ausführung, benutzerdefinierte Tool-Aufrufe (Function Call) und langes Textverständnis. Das Modell unterstützt 26 Sprachen, darunter Chinesisch, Englisch, Japanisch, Koreanisch und Deutsch. In mehreren Benchmark-Tests zeigt GLM-4-9B-Chat hervorragende Leistungen, wie AlignBench-v2, MT-Bench, MMLU und C-Eval. Das Modell unterstützt eine maximale Kontextlänge von 128K und ist für akademische Forschung und kommerzielle Anwendungen geeignet."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 ist ein durch verstärkendes Lernen (RL) gesteuertes Inferenzmodell, das Probleme mit Wiederholungen und Lesbarkeit im Modell löst. Vor dem RL führte DeepSeek-R1 Kaltstartdaten ein, um die Inferenzleistung weiter zu optimieren. Es zeigt in mathematischen, programmierbezogenen und Inferenzaufgaben eine vergleichbare Leistung zu OpenAI-o1 und verbessert die Gesamtleistung durch sorgfältig gestaltete Trainingsmethoden."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 ist ein hybrides Experten (MoE) Sprachmodell mit 6710 Milliarden Parametern, das eine Multi-Head-Latente-Attention (MLA) und DeepSeekMoE-Architektur verwendet, kombiniert mit einer Lastenausgleichsstrategie ohne Hilfskosten, um die Inferenz- und Trainingseffizienz zu optimieren. Durch das Pre-Training auf 14,8 Billionen hochwertigen Tokens und anschließende überwachte Feinabstimmung und verstärktes Lernen übertrifft DeepSeek-V3 in der Leistung andere Open-Source-Modelle und nähert sich führenden geschlossenen Modellen."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma ist eines der leichtgewichtigen, hochmodernen offenen Modellserien, die von Google entwickelt wurden. Es handelt sich um ein großes Sprachmodell mit nur Decoder, das Englisch unterstützt und offene Gewichte, vortrainierte Varianten und anweisungsfeinabgestimmte Varianten bietet. Das Gemma-Modell eignet sich für verschiedene Textgenerierungsaufgaben, einschließlich Fragen und Antworten, Zusammenfassungen und Inferenz. Dieses 9B-Modell wurde mit 80 Billionen Tokens trainiert. Seine relativ kleine Größe ermöglicht es, in ressourcenbeschränkten Umgebungen wie Laptops, Desktop-Computern oder Ihrer eigenen Cloud-Infrastruktur bereitgestellt zu werden, wodurch mehr Menschen Zugang zu modernsten KI-Modellen erhalten und Innovationen gefördert werden."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 ist eine Familie von mehrsprachigen großen Sprachmodellen, die von Meta entwickelt wurden und vortrainierte sowie anweisungsfeinabgestimmte Varianten mit 8B, 70B und 405B Parametern umfasst. Dieses 8B-Anweisungsfeinabgestimmte Modell wurde für mehrsprachige Dialogszenarien optimiert und zeigt in mehreren Branchen-Benchmark-Tests hervorragende Leistungen. Das Modelltraining verwendete über 150 Billionen Tokens aus öffentlichen Daten und nutzte Techniken wie überwachte Feinabstimmung und verstärkendes Lernen mit menschlichem Feedback, um die Nützlichkeit und Sicherheit des Modells zu verbessern. Llama 3.1 unterstützt Text- und Codegenerierung, mit einem Wissensstichtag von Dezember 2023."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview ist ein innovatives Modell für die Verarbeitung natürlicher Sprache, das komplexe Aufgaben der Dialoggenerierung und des Kontextverständnisses effizient bewältigen kann."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview ist ein forschungsorientiertes Modell, das vom Qwen-Team entwickelt wurde und sich auf visuelle Inferenzfähigkeiten konzentriert. Es hat einzigartige Vorteile beim Verständnis komplexer Szenen und der Lösung visuell verwandter mathematischer Probleme."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview ist das neueste experimentelle Forschungsmodell von Qwen, das sich auf die Verbesserung der KI-Inferenzfähigkeiten konzentriert. Durch die Erforschung komplexer Mechanismen wie Sprachmischung und rekursive Inferenz bietet es Hauptvorteile wie starke Analysefähigkeiten, mathematische und Programmierfähigkeiten. Gleichzeitig gibt es Herausforderungen wie Sprachwechsel, Inferenzzyklen, Sicherheitsüberlegungen und Unterschiede in anderen Fähigkeiten."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct ist die neueste Version der von Alibaba Cloud veröffentlichten Reihe von code-spezifischen großen Sprachmodellen. Dieses Modell basiert auf Qwen2.5 und wurde mit 55 Billionen Tokens trainiert, um die Fähigkeiten zur Codegenerierung, Inferenz und Fehlerbehebung erheblich zu verbessern. Es verbessert nicht nur die Codierungsfähigkeiten, sondern bewahrt auch die Vorteile in Mathematik und allgemeinen Fähigkeiten. Das Modell bietet eine umfassendere Grundlage für praktische Anwendungen wie Code-Agenten."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math konzentriert sich auf die Problemlösung im Bereich Mathematik und bietet professionelle Lösungen für schwierige Aufgaben."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 ist die neueste Reihe des Qwen-Modells, das 128k Kontext unterstützt. Im Vergleich zu den derzeit besten Open-Source-Modellen übertrifft Qwen2-72B in den Bereichen natürliche Sprachverständnis, Wissen, Code, Mathematik und Mehrsprachigkeit deutlich die führenden Modelle."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 ist die neueste Reihe des Qwen-Modells, das in der Lage ist, die besten Open-Source-Modelle ähnlicher Größe oder sogar größerer Modelle zu übertreffen. Qwen2 7B hat in mehreren Bewertungen signifikante Vorteile erzielt, insbesondere im Bereich Code und Verständnis der chinesischen Sprache."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B ist ein leistungsstarkes visuelles Sprachmodell, das multimodale Verarbeitung von Bildern und Text unterstützt und in der Lage ist, Bildinhalte präzise zu erkennen und relevante Beschreibungen oder Antworten zu generieren."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct ist ein großes Sprachmodell mit 14 Milliarden Parametern, das hervorragende Leistungen bietet, für chinesische und mehrsprachige Szenarien optimiert ist und Anwendungen wie intelligente Fragen und Antworten sowie Inhaltserstellung unterstützt."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct ist ein großes Sprachmodell mit 7 Milliarden Parametern, das Funktionsaufrufe unterstützt und nahtlos mit externen Systemen interagiert, was die Flexibilität und Skalierbarkeit erheblich erhöht. Es ist für chinesische und mehrsprachige Szenarien optimiert und unterstützt Anwendungen wie intelligente Fragen und Antworten sowie Inhaltserstellung."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct ist ein auf großflächigem Pre-Training basierendes Programmiermodell, das über starke Fähigkeiten zur Codeverstehung und -generierung verfügt und effizient verschiedene Programmieraufgaben bearbeiten kann. Es eignet sich besonders gut für intelligente Codeerstellung, automatisierte Skripterstellung und die Beantwortung von Programmierfragen."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct ist ein großes Sprachmodell, das speziell für die Codegenerierung, das Verständnis von Code und effiziente Entwicklungsszenarien entwickelt wurde. Es verwendet eine branchenführende Parametergröße von 32B und kann vielfältige Programmieranforderungen erfüllen."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "Das TeleMM-Modell ist ein multimodales Großmodell, das von China Telecom entwickelt wurde und in der Lage ist, Texte, Bilder und andere Modalitäten zu verarbeiten. Es unterstützt Funktionen wie Bildverständnis und Diagrammanalyse und bietet Benutzern multimodale Verständnisdienste. Das Modell kann mit Benutzern multimodal interagieren, den Eingabeinhalt genau verstehen, Fragen beantworten, bei der Erstellung helfen und effizient multimodale Informationen und Inspirationsunterstützung bereitstellen. Es zeigt hervorragende Leistungen in multimodalen Aufgaben wie feinkörniger Wahrnehmung und logischem Schlussfolgern."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large ist das größte Open-Source-Transformer-Architektur MoE-Modell der Branche mit insgesamt 389 Milliarden Parametern und 52 Milliarden aktiven Parametern."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct ist das anweisungsfeinabgestimmte große Sprachmodell der Qwen2-Serie mit einer Parametergröße von 72B. Dieses Modell basiert auf der Transformer-Architektur und verwendet Technologien wie die SwiGLU-Aktivierungsfunktion, QKV-Offsets und gruppierte Abfrageaufmerksamkeit. Es kann große Eingaben verarbeiten. Das Modell zeigt hervorragende Leistungen in der Sprachverständnis, -generierung, Mehrsprachigkeit, Codierung, Mathematik und Inferenz in mehreren Benchmark-Tests und übertrifft die meisten Open-Source-Modelle und zeigt in bestimmten Aufgaben eine vergleichbare Wettbewerbsfähigkeit mit proprietären Modellen."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct ist eines der neuesten großen Sprachmodelle, die von Alibaba Cloud veröffentlicht wurden. Dieses 72B-Modell hat signifikante Verbesserungen in den Bereichen Codierung und Mathematik. Das Modell bietet auch mehrsprachige Unterstützung und deckt über 29 Sprachen ab, einschließlich Chinesisch und Englisch. Es zeigt signifikante Verbesserungen in der Befolgung von Anweisungen, im Verständnis strukturierter Daten und in der Generierung strukturierter Ausgaben (insbesondere JSON)."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Für chinesische Charakterdialoge optimiert, bietet es flüssige und den chinesischen Ausdrucksgewohnheiten entsprechende Dialoggenerierung."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Das Open-Source-Funktionsaufrufmodell von Fireworks bietet hervorragende Anweisungsdurchführungsfähigkeiten und anpassbare Funktionen."
+ "abab7-chat-preview": {
+ "description": "Im Vergleich zur abab6.5-Modellreihe gibt es erhebliche Verbesserungen in den Fähigkeiten bei langen Texten, Mathematik und Schreiben."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Das neueste Firefunction-v2 von Fireworks ist ein leistungsstarkes Funktionsaufrufmodell, das auf Llama-3 basiert und durch zahlreiche Optimierungen besonders für Funktionsaufrufe, Dialoge und Anweisungsverfolgung geeignet ist."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 ist ein hochmodernes großes Sprachmodell, das durch verstärktes Lernen und Optimierung mit Kaltstartdaten hervorragende Leistungen in Inferenz, Mathematik und Programmierung bietet."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b ist ein visuelles Sprachmodell, das sowohl Bild- als auch Texteingaben verarbeiten kann und für multimodale Aufgaben geeignet ist, nachdem es mit hochwertigen Daten trainiert wurde."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Ein leistungsstarkes Mixture-of-Experts (MoE) Sprachmodell von Deepseek mit insgesamt 671B Parametern, wobei 37B Parameter pro Token aktiviert werden."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Das Llama 3 70B Instruct-Modell ist speziell für mehrsprachige Dialoge und natürliche Sprachverständnis optimiert und übertrifft die meisten Wettbewerbsmodelle."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Das Llama 3 70B Instruct-Modell (HF-Version) entspricht den offiziellen Ergebnissen und eignet sich für hochwertige Anweisungsverfolgungsaufgaben."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Das Llama 3 8B Instruct-Modell ist für Dialoge und mehrsprachige Aufgaben optimiert und bietet hervorragende und effiziente Leistungen."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Meta's 11B Parameter instruct-Modell für Bildverarbeitung. Dieses Modell ist optimiert für visuelle Erkennung, Bildverarbeitung, Bildbeschreibung und die Beantwortung allgemeiner Fragen zu Bildern. Es kann visuelle Daten wie Diagramme und Grafiken verstehen und schließt die Lücke zwischen visuellen und sprachlichen Informationen, indem es textuelle Beschreibungen der Bilddetails generiert."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Llama 3.2 1B instruct-Modell ist ein leichtgewichtiges mehrsprachiges Modell, das von Meta veröffentlicht wurde. Dieses Modell zielt darauf ab, die Effizienz zu steigern und bietet im Vergleich zu größeren Modellen signifikante Verbesserungen in Bezug auf Latenz und Kosten. Anwendungsbeispiele für dieses Modell sind Retrieval und Zusammenfassung."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Llama 3.2 3B instruct-Modell ist ein leichtgewichtiges mehrsprachiges Modell, das von Meta veröffentlicht wurde. Dieses Modell zielt darauf ab, die Effizienz zu steigern und bietet im Vergleich zu größeren Modellen signifikante Verbesserungen in Bezug auf Latenz und Kosten. Anwendungsbeispiele für dieses Modell sind Abfragen und Aufforderungsneuschreibungen sowie Schreibassistenz."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Meta's 90B Parameter instruct-Modell für Bildverarbeitung. Dieses Modell ist optimiert für visuelle Erkennung, Bildverarbeitung, Bildbeschreibung und die Beantwortung allgemeiner Fragen zu Bildern. Es kann visuelle Daten wie Diagramme und Grafiken verstehen und schließt die Lücke zwischen visuellen und sprachlichen Informationen, indem es textuelle Beschreibungen der Bilddetails generiert."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct ist die aktualisierte Version von Llama 3.1 70B aus dem Dezember. Dieses Modell wurde auf der Grundlage von Llama 3.1 70B (veröffentlicht im Juli 2024) verbessert und bietet erweiterte Funktionen für Toolaufrufe, mehrsprachige Textunterstützung sowie mathematische und Programmierfähigkeiten. Das Modell erreicht branchenführende Leistungen in den Bereichen Inferenz, Mathematik und Befehlsbefolgung und bietet eine ähnliche Leistung wie 3.1 405B, während es gleichzeitig signifikante Vorteile in Bezug auf Geschwindigkeit und Kosten bietet."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "Ein 24B-Parameter-Modell mit fortschrittlichen Fähigkeiten, die mit größeren Modellen vergleichbar sind."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Das Mixtral MoE 8x22B Instruct-Modell unterstützt durch seine große Anzahl an Parametern und Multi-Expert-Architektur die effiziente Verarbeitung komplexer Aufgaben."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Das Mixtral MoE 8x7B Instruct-Modell bietet durch seine Multi-Expert-Architektur effiziente Anweisungsverfolgung und -ausführung."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Das Mixtral MoE 8x7B Instruct-Modell (HF-Version) bietet die gleiche Leistung wie die offizielle Implementierung und eignet sich für verschiedene effiziente Anwendungsszenarien."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "Das MythoMax L2 13B-Modell kombiniert neuartige Kombinations-Technologien und ist besonders gut in Erzählungen und Rollenspielen."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "Das QwQ-Modell ist ein experimentelles Forschungsmodell, das vom Qwen-Team entwickelt wurde und sich auf die Verbesserung der KI-Inferenzfähigkeiten konzentriert."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "Die 72B-Version des Qwen-VL-Modells ist das neueste Ergebnis von Alibabas Iteration und repräsentiert fast ein Jahr an Innovation."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 ist eine Reihe von Sprachmodellen mit ausschließlich Decodern, die vom Alibaba Cloud Qwen-Team entwickelt wurde. Diese Modelle sind in verschiedenen Größen erhältlich, darunter 0.5B, 1.5B, 3B, 7B, 14B, 32B und 72B, mit Basis- und instruct-Varianten."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct ist die neueste Version der von Alibaba Cloud veröffentlichten Reihe von code-spezifischen großen Sprachmodellen. Dieses Modell basiert auf Qwen2.5 und wurde mit 55 Billionen Tokens trainiert, um die Fähigkeiten zur Codegenerierung, Inferenz und Fehlerbehebung erheblich zu verbessern. Es verbessert nicht nur die Codierungsfähigkeiten, sondern bewahrt auch die Vorteile in Mathematik und allgemeinen Fähigkeiten. Das Modell bietet eine umfassendere Grundlage für praktische Anwendungen wie Code-Agenten."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "Das StarCoder 15.5B-Modell unterstützt fortgeschrittene Programmieraufgaben und hat verbesserte mehrsprachige Fähigkeiten, die sich für komplexe Codegenerierung und -verständnis eignen."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "Das StarCoder 7B-Modell wurde für über 80 Programmiersprachen trainiert und bietet hervorragende Programmierausfüllfähigkeiten und Kontextverständnis."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Das Yi-Large-Modell bietet hervorragende mehrsprachige Verarbeitungsfähigkeiten und kann für verschiedene Sprachgenerierungs- und Verständnisaufgaben eingesetzt werden."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus ist das leistungsstärkste Modell von Anthropic zur Bearbeitung hochkomplexer Aufgaben. Es zeichnet sich durch hervorragende Leistung, Intelligenz, Flüssigkeit und Verständnis aus."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku ist das schnellste nächste Generation Modell von Anthropic. Im Vergleich zu Claude 3 Haiku hat Claude 3.5 Haiku in allen Fähigkeiten Fortschritte gemacht und übertrifft in vielen intellektuellen Benchmark-Tests das größte Modell der vorherigen Generation, Claude 3 Opus."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet bietet Fähigkeiten, die über Opus hinausgehen, und eine schnellere Geschwindigkeit als Sonnet, während es den gleichen Preis wie Sonnet beibehält. Sonnet ist besonders gut in Programmierung, Datenwissenschaft, visueller Verarbeitung und Agentenaufgaben."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet bietet eine ideale Balance zwischen Intelligenz und Geschwindigkeit für Unternehmensarbeitslasten. Es bietet maximalen Nutzen zu einem niedrigeren Preis, ist zuverlässig und für großflächige Bereitstellungen geeignet."
},
- "code-raccoon-v1": {
- "description": "Code-Raccoon ist ein intelligenter Softwareentwicklungsassistent, der auf dem großen Sprachmodell von SenseTime basiert und Softwareanforderungsanalyse, Architekturdesign, Codeerstellung, Softwaretests und andere Bereiche abdeckt, um die Bedürfnisse der Benutzer beim Programmieren und Lernen zu erfüllen. Code-Raccoon unterstützt über 90 gängige Programmiersprachen wie Python, Java, JavaScript, C++, Go, SQL sowie gängige IDEs wie VS Code und IntelliJ IDEA. In der praktischen Anwendung kann Code-Raccoon Entwicklern helfen, die Programmierproduktivität um über 50 % zu steigern."
- },
"codegeex-4": {
"description": "CodeGeeX-4 ist ein leistungsstarker AI-Programmierassistent, der intelligente Fragen und Codevervollständigung in verschiedenen Programmiersprachen unterstützt und die Entwicklungseffizienz steigert."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ ist ein leistungsstarkes großes Sprachmodell, das speziell für reale Unternehmensszenarien und komplexe Anwendungen entwickelt wurde."
},
+ "dall-e-2": {
+ "description": "Zweite Generation des DALL·E-Modells, unterstützt realistischere und genauere Bildgenerierung, mit einer Auflösung, die viermal so hoch ist wie die der ersten Generation."
+ },
+ "dall-e-3": {
+ "description": "Das neueste DALL·E-Modell, veröffentlicht im November 2023. Unterstützt realistischere und genauere Bildgenerierung mit verbesserter Detailgenauigkeit."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct bietet zuverlässige Anweisungsverarbeitungsfähigkeiten und unterstützt Anwendungen in verschiedenen Branchen."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 ist ein leistungsstarkes, kosteneffizientes hybrides Expertenmodell (MoE). Es wurde auf einem hochwertigen Korpus von 81 Billionen Tokens vortrainiert und durch überwachte Feinabstimmung (SFT) und verstärkendes Lernen (RL) weiter verbessert. Im Vergleich zu DeepSeek 67B bietet DeepSeek-V2 eine stärkere Leistung, spart 42,5 % der Trainingskosten, reduziert den KV-Cache um 93,3 % und erhöht die maximale Generierungsdurchsatzrate um das 5,76-fache. Dieses Modell unterstützt eine Kontextlänge von 128k und schneidet sowohl in Standard-Benchmark-Tests als auch in offenen Generierungsbewertungen hervorragend ab."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 ist ein durch verstärkendes Lernen (RL) gesteuertes Inferenzmodell, das die Probleme der Wiederholbarkeit und Lesbarkeit im Modell löst. Vor dem RL führte DeepSeek-R1 Kaltstartdaten ein, um die Inferenzleistung weiter zu optimieren. Es zeigt in mathematischen, programmierbezogenen und Inferenzaufgaben eine vergleichbare Leistung zu OpenAI-o1 und verbessert durch sorgfältig gestaltete Trainingsmethoden die Gesamteffizienz."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "Das DeepSeek-R1-Distill-Modell optimiert die Inferenzleistung durch verstärkendes Lernen und Kaltstartdaten. Das Open-Source-Modell setzt neue Maßstäbe für Multitasking."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B ist ein destilliertes Modell, das auf Llama-3.1-8B basiert. Dieses Modell wurde mit Beispielen, die von DeepSeek-R1 generiert wurden, feinabgestimmt und zeigt hervorragende Inferenzfähigkeiten. Es hat in mehreren Benchmark-Tests gut abgeschnitten, darunter eine Genauigkeit von 89,1 % in MATH-500, eine Bestehensquote von 50,4 % in AIME 2024 und eine Bewertung von 1205 in CodeForces, was starke mathematische und Programmierfähigkeiten für ein 8B-Modell demonstriert."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Das DeepSeek-R1-Distill-Modell optimiert die Inferenzleistung durch verstärkendes Lernen und Kaltstartdaten. Das Open-Source-Modell setzt neue Maßstäbe für Multitasking."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Das DeepSeek-R1-Distill-Modell optimiert die Inferenzleistung durch verstärkendes Lernen und Kaltstartdaten. Das Open-Source-Modell setzt neue Maßstäbe für Multitasking."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B ist ein Modell, das durch Wissensdestillation aus Qwen2.5-32B gewonnen wurde. Dieses Modell wurde mit 800.000 ausgewählten Beispielen, die von DeepSeek-R1 generiert wurden, feinabgestimmt und zeigt herausragende Leistungen in mehreren Bereichen wie Mathematik, Programmierung und Inferenz. Es hat in mehreren Benchmark-Tests, darunter AIME 2024, MATH-500 und GPQA Diamond, hervorragende Ergebnisse erzielt, wobei es in MATH-500 eine Genauigkeit von 94,3 % erreicht hat und damit starke mathematische Schlussfolgerungsfähigkeiten demonstriert."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B ist ein Modell, das durch Wissensdestillation aus Qwen2.5-Math-7B gewonnen wurde. Dieses Modell wurde mit 800.000 ausgewählten Beispielen, die von DeepSeek-R1 generiert wurden, feinabgestimmt und zeigt hervorragende Inferenzfähigkeiten. Es hat in mehreren Benchmark-Tests, darunter eine Genauigkeit von 92,8 % in MATH-500, eine Bestehensquote von 55,5 % in AIME 2024 und eine Bewertung von 1189 in CodeForces, was starke mathematische und Programmierfähigkeiten für ein 7B-Modell demonstriert."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 vereint die hervorragenden Merkmale früherer Versionen und verbessert die allgemeinen und kodierenden Fähigkeiten."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 ist ein hybrides Expertenmodell (MoE) mit 6710 Milliarden Parametern, das eine Multi-Head-Latent-Attention (MLA) und die DeepSeekMoE-Architektur verwendet, kombiniert mit einer Lastenausgleichsstrategie ohne Hilfskosten, um die Inferenz- und Trainingseffizienz zu optimieren. Durch das Pre-Training auf 14,8 Billionen hochwertigen Tokens und anschließendes überwachten Feintuning und verstärkendes Lernen übertrifft DeepSeek-V3 in der Leistung andere Open-Source-Modelle und nähert sich führenden Closed-Source-Modellen."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B ist ein fortschrittliches Modell, das für komplexe Dialoge trainiert wurde."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "Hochmodernes, effizientes LLM, das auf Schlussfolgern, Mathematik und Programmierung spezialisiert ist."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 ist ein hybrides Expertenmodell (MoE) für visuelle Sprache, das auf DeepSeekMoE-27B basiert und eine spärliche Aktivierung der MoE-Architektur verwendet, um außergewöhnliche Leistungen bei der Aktivierung von nur 4,5 Milliarden Parametern zu erzielen. Dieses Modell zeigt hervorragende Leistungen in mehreren Aufgaben, darunter visuelle Fragenbeantwortung, optische Zeichenerkennung, Dokument-/Tabellen-/Diagrammverständnis und visuelle Lokalisierung."
+ },
"deepseek-chat": {
"description": "Ein neues Open-Source-Modell, das allgemeine und Codefähigkeiten kombiniert. Es bewahrt nicht nur die allgemeinen Dialogfähigkeiten des ursprünglichen Chat-Modells und die leistungsstarken Codeverarbeitungsfähigkeiten des Coder-Modells, sondern stimmt auch besser mit menschlichen Präferenzen überein. Darüber hinaus hat DeepSeek-V2.5 in mehreren Bereichen wie Schreibaufgaben und Befolgung von Anweisungen erhebliche Verbesserungen erzielt."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 ist ein Open-Source-Mischexperten-Code-Modell, das in Codeaufgaben hervorragende Leistungen erbringt und mit GPT4-Turbo vergleichbar ist."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 ist ein durch verstärkendes Lernen (RL) gesteuertes Inferenzmodell, das die Probleme der Wiederholbarkeit und Lesbarkeit im Modell löst. Vor dem RL führte DeepSeek-R1 Kaltstartdaten ein, um die Inferenzleistung weiter zu optimieren. Es zeigt in mathematischen, programmierbezogenen und Inferenzaufgaben eine vergleichbare Leistung zu OpenAI-o1 und verbessert durch sorgfältig gestaltete Trainingsmethoden die Gesamteffizienz."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 – das größere und intelligentere Modell im DeepSeek-Paket – wurde in die Llama 70B-Architektur destilliert. Basierend auf Benchmark-Tests und menschlicher Bewertung ist dieses Modell intelligenter als das ursprüngliche Llama 70B, insbesondere bei Aufgaben, die mathematische und faktische Genauigkeit erfordern."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "Das DeepSeek-R1-Distill Modell wurde durch Wissensdistillationstechniken entwickelt, indem Proben, die von DeepSeek-R1 generiert wurden, auf Qwen, Llama und andere Open-Source-Modelle feinabgestimmt wurden."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "Das DeepSeek-R1-Distill Modell wurde durch Wissensdistillationstechniken entwickelt, indem Proben, die von DeepSeek-R1 generiert wurden, auf Qwen, Llama und andere Open-Source-Modelle feinabgestimmt wurden."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "Das DeepSeek-R1-Distill Modell wurde durch Wissensdistillationstechniken entwickelt, indem Proben, die von DeepSeek-R1 generiert wurden, auf Qwen, Llama und andere Open-Source-Modelle feinabgestimmt wurden."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "Das DeepSeek-R1-Distill Modell wurde durch Wissensdistillationstechniken entwickelt, indem Proben, die von DeepSeek-R1 generiert wurden, auf Qwen, Llama und andere Open-Source-Modelle feinabgestimmt wurden."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "Das DeepSeek-R1-Distill Modell wurde durch Wissensdistillationstechniken entwickelt, indem Proben, die von DeepSeek-R1 generiert wurden, auf Qwen, Llama und andere Open-Source-Modelle feinabgestimmt wurden."
+ },
+ "deepseek-reasoner": {
+ "description": "Das von DeepSeek entwickelte Inferenzmodell. Bevor das Modell die endgültige Antwort ausgibt, gibt es zunächst eine Denkprozesskette aus, um die Genauigkeit der endgültigen Antwort zu erhöhen."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 ist ein effizientes Mixture-of-Experts-Sprachmodell, das für wirtschaftliche Verarbeitungsanforderungen geeignet ist."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B ist das Design-Code-Modell von DeepSeek und bietet starke Fähigkeiten zur Codegenerierung."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 ist ein MoE-Modell, das von der Hangzhou DeepSeek Artificial Intelligence Technology Research Co., Ltd. entwickelt wurde. Es hat in mehreren Bewertungen herausragende Ergebnisse erzielt und belegt in den gängigen Rankings den ersten Platz unter den Open-Source-Modellen. Im Vergleich zum V2.5-Modell hat sich die Generierungsgeschwindigkeit um das Dreifache erhöht, was den Nutzern ein schnelleres und flüssigeres Nutzungserlebnis bietet."
+ },
"deepseek/deepseek-chat": {
"description": "Ein neues Open-Source-Modell, das allgemeine und Codefähigkeiten vereint. Es behält nicht nur die allgemeinen Dialogfähigkeiten des ursprünglichen Chat-Modells und die leistungsstarken Codeverarbeitungsfähigkeiten des Coder-Modells bei, sondern stimmt auch besser mit menschlichen Vorlieben überein. Darüber hinaus hat DeepSeek-V2.5 in vielen Bereichen wie Schreibaufgaben und Befehlsbefolgung erhebliche Verbesserungen erzielt."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 hat die Schlussfolgerungsfähigkeiten des Modells erheblich verbessert, selbst bei nur wenigen gekennzeichneten Daten. Bevor das Modell die endgültige Antwort ausgibt, gibt es zunächst eine Denkprozesskette aus, um die Genauigkeit der endgültigen Antwort zu erhöhen."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 hat die Schlussfolgerungsfähigkeiten des Modells erheblich verbessert, selbst bei nur wenigen gekennzeichneten Daten. Bevor das Modell die endgültige Antwort ausgibt, gibt es zunächst eine Denkprozesskette aus, um die Genauigkeit der endgültigen Antwort zu erhöhen."
+ },
"emohaa": {
"description": "Emohaa ist ein psychologisches Modell mit professionellen Beratungsfähigkeiten, das den Nutzern hilft, emotionale Probleme zu verstehen."
},
+ "ernie-3.5-128k": {
+ "description": "Das von Baidu entwickelte Flaggschiff-Modell für große Sprachmodelle deckt eine riesige Menge an chinesischen und englischen Korpora ab und bietet starke allgemeine Fähigkeiten, die die meisten Anforderungen an Dialogfragen, kreative Generierung und Plugin-Anwendungen erfüllen; es unterstützt die automatische Anbindung an das Baidu-Suchplugin, um die Aktualität der Antwortinformationen zu gewährleisten."
+ },
+ "ernie-3.5-8k": {
+ "description": "Das von Baidu entwickelte Flaggschiff-Modell für große Sprachmodelle deckt eine riesige Menge an chinesischen und englischen Korpora ab und bietet starke allgemeine Fähigkeiten, die die meisten Anforderungen an Dialogfragen, kreative Generierung und Plugin-Anwendungen erfüllen; es unterstützt die automatische Anbindung an das Baidu-Suchplugin, um die Aktualität der Antwortinformationen zu gewährleisten."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "Das von Baidu entwickelte Flaggschiff-Modell für große Sprachmodelle deckt eine riesige Menge an chinesischen und englischen Korpora ab und bietet starke allgemeine Fähigkeiten, die die meisten Anforderungen an Dialogfragen, kreative Generierung und Plugin-Anwendungen erfüllen; es unterstützt die automatische Anbindung an das Baidu-Suchplugin, um die Aktualität der Antwortinformationen zu gewährleisten."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "Das von Baidu entwickelte Flaggschiff-Modell für große Sprachmodelle hat im Vergleich zu ERNIE 3.5 eine umfassende Verbesserung der Modellfähigkeiten erreicht und ist weit verbreitet in komplexen Aufgabenbereichen anwendbar; es unterstützt die automatische Anbindung an das Baidu-Suchplugin, um die Aktualität der Antwortinformationen zu gewährleisten."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "Das von Baidu entwickelte Flaggschiff-Modell für große Sprachmodelle hat im Vergleich zu ERNIE 3.5 eine umfassende Verbesserung der Modellfähigkeiten erreicht und ist weit verbreitet in komplexen Aufgabenbereichen anwendbar; es unterstützt die automatische Anbindung an das Baidu-Suchplugin, um die Aktualität der Antwortinformationen zu gewährleisten."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "Das von Baidu entwickelte Flaggschiff-Modell für große Sprachmodelle zeigt hervorragende Gesamtergebnisse und ist weit verbreitet in komplexen Aufgabenbereichen anwendbar; es unterstützt die automatische Anbindung an das Baidu-Suchplugin, um die Aktualität der Antwortinformationen zu gewährleisten. Im Vergleich zu ERNIE 4.0 bietet es eine bessere Leistung."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "Das von Baidu entwickelte Flaggschiff-Modell für große Sprachmodelle zeigt hervorragende Gesamtergebnisse und ist weit verbreitet in komplexen Aufgabenbereichen anwendbar; es unterstützt die automatische Anbindung an das Baidu-Suchplugin, um die Aktualität der Antwortinformationen zu gewährleisten. Im Vergleich zu ERNIE 4.0 bietet es eine bessere Leistung."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "Das von Baidu entwickelte Flaggschiff-Modell für große Sprachmodelle zeigt hervorragende Gesamtergebnisse und ist weit verbreitet in komplexen Aufgabenbereichen anwendbar; es unterstützt die automatische Anbindung an das Baidu-Suchplugin, um die Aktualität der Antwortinformationen zu gewährleisten. Im Vergleich zu ERNIE 4.0 bietet es eine bessere Leistung."
+ },
+ "ernie-char-8k": {
+ "description": "Das von Baidu entwickelte große Sprachmodell für vertikale Szenarien eignet sich für Anwendungen wie NPCs in Spielen, Kundenservice-Dialoge und Rollenspiele, mit einem klareren und konsistenteren Charakterstil, einer stärkeren Befolgung von Anweisungen und besserer Inferenzleistung."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Das von Baidu entwickelte große Sprachmodell für vertikale Szenarien eignet sich für Anwendungen wie NPCs in Spielen, Kundenservice-Dialoge und Rollenspiele, mit einem klareren und konsistenteren Charakterstil, einer stärkeren Befolgung von Anweisungen und besserer Inferenzleistung."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite ist ein leichtgewichtiges großes Sprachmodell, das von Baidu entwickelt wurde und sowohl hervorragende Modellleistung als auch Inferenzleistung bietet, geeignet für die Verwendung mit AI-Beschleunigungskarten mit geringer Rechenleistung."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Das von Baidu entwickelte leichtgewichtige große Sprachmodell bietet sowohl hervorragende Modellleistung als auch Inferenzleistung, die besser ist als die von ERNIE Lite, und ist geeignet für die Verwendung mit AI-Beschleunigungskarten mit geringer Rechenleistung."
+ },
+ "ernie-novel-8k": {
+ "description": "Das von Baidu entwickelte allgemeine große Sprachmodell hat deutliche Vorteile in der Fähigkeit zur Fortsetzung von Romanen und kann auch in Szenarien wie Kurzdramen und Filmen eingesetzt werden."
+ },
+ "ernie-speed-128k": {
+ "description": "Das neueste hochleistungsfähige große Sprachmodell von Baidu, das 2024 veröffentlicht wurde, bietet hervorragende allgemeine Fähigkeiten und eignet sich gut als Basismodell für Feinabstimmungen, um spezifische Szenarien besser zu bewältigen, während es auch hervorragende Inferenzleistungen bietet."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "Das neueste hochleistungsfähige große Sprachmodell von Baidu, das 2024 veröffentlicht wurde, bietet hervorragende allgemeine Fähigkeiten und ist besser als ERNIE Speed, geeignet als Basismodell für Feinabstimmungen, um spezifische Szenarien besser zu bewältigen, während es auch hervorragende Inferenzleistungen bietet."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny ist ein hochleistungsfähiges großes Sprachmodell, dessen Bereitstellungs- und Feinabstimmungskosten die niedrigsten unter den Wenshin-Modellen sind."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Tuning) bietet stabile und anpassbare Leistung und ist die ideale Wahl für Lösungen komplexer Aufgaben."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro unterstützt bis zu 2 Millionen Tokens und ist die ideale Wahl für mittelgroße multimodale Modelle, die umfassende Unterstützung für komplexe Aufgaben bieten."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash bietet nächste Generation Funktionen und Verbesserungen, einschließlich außergewöhnlicher Geschwindigkeit, nativer Werkzeugnutzung, multimodaler Generierung und einem Kontextfenster von 1M Tokens."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash bietet nächste Generation Funktionen und Verbesserungen, einschließlich außergewöhnlicher Geschwindigkeit, nativer Werkzeugnutzung, multimodaler Generierung und einem Kontextfenster von 1M Tokens."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp ist Googles neuestes experimentelles multimodales KI-Modell mit Next-Gen-Funktionen, herausragender Geschwindigkeit, nativer Werkzeuganwendung und multimodaler Generierung."
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "Ein Gemini 2.0 Flash Modell, das auf Kosteneffizienz und niedrige Latenz optimiert wurde."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Exp ist Googles neuestes experimentelles multimodales KI-Modell mit der nächsten Generation von Funktionen, außergewöhnlicher Geschwindigkeit, nativer Tool-Nutzung und multimodaler Generierung."
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Exp ist Googles neuestes experimentelles multimodales KI-Modell mit der nächsten Generation von Funktionen, außergewöhnlicher Geschwindigkeit, nativen Toolaufrufen und multimodaler Generierung."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 ist Googles neuestes experimentelles multimodales KI-Modell, das über eine schnelle Verarbeitungskapazität verfügt und Texte, Bilder und Videoeingaben unterstützt, um eine effiziente Skalierung für verschiedene Aufgaben zu ermöglichen."
- },
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 ist Googles neuestes experimentelles multimodales KI-Modell, das über eine schnelle Verarbeitungskapazität verfügt und Texte, Bilder und Videoeingaben unterstützt, um eine effiziente Skalierung für verschiedene Aufgaben zu ermöglichen."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental ist Googles neuestes experimentelles multimodales KI-Modell, das im Vergleich zu früheren Versionen eine gewisse Qualitätsverbesserung aufweist, insbesondere in Bezug auf Weltwissen, Code und lange Kontexte."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 ist Googles neuestes experimentelles multimodales KI-Modell, das im Vergleich zu historischen Versionen eine gewisse Qualitätssteigerung aufweist."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus hat die Fähigkeit, Videoinhalte und mehrere Bilder zu verstehen und eignet sich für multimodale Aufgaben."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview verfügt über starke Fähigkeiten zur komplexen Schlussfolgerung und zeigt hervorragende Leistungen in den Bereichen logisches Denken, Mathematik und Programmierung."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash bietet nächste Generation Funktionen und Verbesserungen, einschließlich außergewöhnlicher Geschwindigkeit, nativer Werkzeugnutzung, multimodaler Generierung und einem Kontextfenster von 1M Tokens."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental ist Googles neuestes experimentelles multimodales KI-Modell, das im Vergleich zu früheren Versionen eine gewisse Qualitätsverbesserung aufweist, insbesondere in Bezug auf Weltwissen, Code und lange Kontexte."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash bietet optimierte multimodale Verarbeitungsfähigkeiten, die für verschiedene komplexe Aufgabenszenarien geeignet sind."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro kombiniert die neuesten Optimierungstechnologien und bietet eine effizientere Verarbeitung multimodaler Daten."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 ist ein effizientes Modell von Google, das eine Vielzahl von Anwendungsszenarien von kleinen Anwendungen bis hin zu komplexer Datenverarbeitung abdeckt."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 setzt das Designkonzept von Leichtbau und Effizienz fort."
},
"google/gemma-2-2b-it": {
"description": "Das leichtgewichtige Anweisungsoptimierungsmodell von Google."
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 ist ein effizientes Modell von Google, das eine Vielzahl von Anwendungsszenarien von kleinen Anwendungen bis hin zu komplexer Datenverarbeitung abdeckt."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 ist eine leichtgewichtige Open-Source-Textmodellreihe von Google."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo eignet sich für eine Vielzahl von Textgenerierungs- und Verständnisaufgaben. Derzeit verweist es auf gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, ein leistungsstarkes Textgenerierungsmodell, geeignet für komplexe Aufgaben."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo eignet sich für eine Vielzahl von Textgenerierungs- und Verständnisaufgaben. Derzeit verweist es auf gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o ist ein dynamisches Modell, das in Echtzeit aktualisiert wird, um die neueste Version zu gewährleisten. Es kombiniert starke Sprachverständnis- und Generierungsfähigkeiten und eignet sich für großangelegte Anwendungsbereiche, einschließlich Kundenservice, Bildung und technischen Support."
},
+ "gpt-4o-audio-preview": {
+ "description": "GPT-4o Audio-Modell, unterstützt Audioeingabe und -ausgabe."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini ist das neueste Modell von OpenAI, das nach GPT-4 Omni veröffentlicht wurde und sowohl Text- als auch Bildinput unterstützt. Als ihr fortschrittlichstes kleines Modell ist es viel günstiger als andere neueste Modelle und kostet über 60 % weniger als GPT-3.5 Turbo. Es behält die fortschrittliche Intelligenz bei und bietet gleichzeitig ein hervorragendes Preis-Leistungs-Verhältnis. GPT-4o mini erzielte 82 % im MMLU-Test und rangiert derzeit in den Chat-Präferenzen über GPT-4."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "Echtzeitversion von GPT-4o-mini, unterstützt Audio- und Texteingabe sowie -ausgabe in Echtzeit."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "Echtzeitversion von GPT-4o, unterstützt Audio- und Texteingabe sowie -ausgabe in Echtzeit."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "Echtzeitversion von GPT-4o, unterstützt Audio- und Texteingabe sowie -ausgabe in Echtzeit."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "Echtzeitversion von GPT-4o, unterstützt Audio- und Texteingabe sowie -ausgabe in Echtzeit."
+ },
"grok-2-1212": {
"description": "Dieses Modell hat Verbesserungen in Bezug auf Genauigkeit, Befolgung von Anweisungen und Mehrsprachigkeit erfahren."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "Das neueste MOE-Architektur-FunctionCall-Modell von Hunyuan, das mit hochwertigen FunctionCall-Daten trainiert wurde, hat ein Kontextfenster von 32K und führt in mehreren Bewertungsmetriken."
},
+ "hunyuan-large": {
+ "description": "Das Hunyuan-large Modell hat insgesamt etwa 389B Parameter, davon etwa 52B aktivierte Parameter, und ist das derzeit größte und leistungsstärkste Open-Source MoE-Modell mit Transformer-Architektur in der Branche."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Besonders gut geeignet für lange Textaufgaben wie Dokumentenzusammenfassungen und Dokumentenfragen, verfügt es auch über die Fähigkeit, allgemeine Textgenerierungsaufgaben zu bearbeiten. Es zeigt hervorragende Leistungen bei der Analyse und Generierung von langen Texten und kann effektiv mit komplexen und detaillierten Anforderungen an die Verarbeitung von langen Inhalten umgehen."
+ },
"hunyuan-lite": {
"description": "Aufgerüstet auf eine MOE-Struktur mit einem Kontextfenster von 256k, führt es in mehreren Bewertungssets in NLP, Code, Mathematik und Industrie zahlreiche Open-Source-Modelle an."
},
+ "hunyuan-lite-vision": {
+ "description": "Das neueste 7B multimodale Modell von Hunyuan, mit einem Kontextfenster von 32K, unterstützt multimodale Dialoge in Chinesisch und Englisch, Objekterkennung in Bildern, Dokumenten- und Tabellenverständnis sowie multimodale Mathematik und übertrifft in mehreren Dimensionen die Bewertungskennzahlen von 7B Wettbewerbsmodellen."
+ },
"hunyuan-pro": {
"description": "Ein MOE-32K-Modell für lange Texte mit einer Billion Parametern. Es erreicht in verschiedenen Benchmarks ein absolut führendes Niveau, hat komplexe Anweisungen und Schlussfolgerungen, verfügt über komplexe mathematische Fähigkeiten und unterstützt Funktionsaufrufe, mit Schwerpunkt auf Optimierung in den Bereichen mehrsprachige Übersetzung, Finanzrecht und Medizin."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Verwendet eine verbesserte Routing-Strategie und mildert gleichzeitig die Probleme der Lastenverteilung und Expertenkonvergenz. Bei langen Texten erreicht der Needle-in-a-Haystack-Indikator 99,9%. MOE-256K bricht in Länge und Effektivität weiter durch und erweitert die eingabefähige Länge erheblich."
},
+ "hunyuan-standard-vision": {
+ "description": "Das neueste multimodale Modell von Hunyuan, das mehrsprachige Antworten unterstützt und sowohl in Chinesisch als auch in Englisch ausgewogen ist."
+ },
"hunyuan-turbo": {
"description": "Die Vorschauversion des neuen großen Sprachmodells von Hunyuan verwendet eine neuartige hybride Expertenmodellstruktur (MoE) und bietet im Vergleich zu Hunyuan-Pro eine schnellere Inferenz und bessere Leistung."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Hunyuan-turbo Version vom 20. November 2024, eine feste Version, die zwischen hunyuan-turbo und hunyuan-turbo-latest liegt."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "Diese Version optimiert: Datenanweisungs-Skalierung, erhebliche Verbesserung der allgemeinen Generalisierungsfähigkeit des Modells; erhebliche Verbesserung der mathematischen, programmierbaren und logischen Denkfähigkeiten; Optimierung der Fähigkeiten im Textverständnis und der Wortverständnisfähigkeiten; Optimierung der Qualität der Inhaltserzeugung in der Texterstellung."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "Allgemeine Optimierung der Benutzererfahrung, einschließlich NLP-Verständnis, Texterstellung, Smalltalk, Wissensfragen, Übersetzung, Fachgebieten usw.; Verbesserung der Menschlichkeit, Optimierung der emotionalen Intelligenz des Modells; Verbesserung der Fähigkeit des Modells, bei unklaren Absichten aktiv Klarheit zu schaffen; Verbesserung der Bearbeitungsfähigkeit von Fragen zur Wort- und Satzanalyse; Verbesserung der Qualität und Interaktivität der Kreation; Verbesserung der Mehrfachinteraktionserfahrung."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "Das neue Flaggschiff-Modell der visuellen Sprache von Hunyuan, das eine brandneue Struktur des gemischten Expertenmodells (MoE) verwendet, bietet umfassende Verbesserungen in den Fähigkeiten zur grundlegenden Erkennung, Inhaltserstellung, Wissensfragen und Analyse sowie Schlussfolgerungen im Vergleich zum vorherigen Modell."
+ },
"hunyuan-vision": {
"description": "Das neueste multimodale Modell von Hunyuan unterstützt die Eingabe von Bildern und Text zur Generierung von Textinhalten."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "Unsere neueste Modellreihe mit herausragender Schlussfolgerungsleistung, die eine Kontextlänge von 1M unterstützt und über verbesserte Anweisungsbefolgung und Toolaufrufmöglichkeiten verfügt."
},
+ "internlm3-latest": {
+ "description": "Unsere neueste Modellreihe bietet herausragende Inferenzleistungen und führt die Open-Source-Modelle in ihrer Gewichtsklasse an. Standardmäßig verweist sie auf unser neuestes veröffentlichtes InternLM3-Modell."
+ },
+ "jina-deepsearch-v1": {
+ "description": "Die Tiefensuche kombiniert Websuche, Lesen und Schlussfolgern und ermöglicht umfassende Untersuchungen. Sie können es als einen Agenten betrachten, der Ihre Forschungsaufgaben übernimmt – er führt eine umfassende Suche durch und iteriert mehrfach, bevor er eine Antwort gibt. Dieser Prozess umfasst kontinuierliche Forschung, Schlussfolgerungen und die Lösung von Problemen aus verschiedenen Perspektiven. Dies unterscheidet sich grundlegend von den Standard-Großmodellen, die Antworten direkt aus vortrainierten Daten generieren, sowie von traditionellen RAG-Systemen, die auf einmaligen Oberflächensuchen basieren."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM ist ein experimentelles, aufgabenorientiertes Sprachmodell, das darauf trainiert wurde, den Prinzipien der Lernwissenschaft zu entsprechen und in Lehr- und Lernszenarien systematische Anweisungen zu befolgen, als Expertenmentor zu fungieren usw."
},
"lite": {
"description": "Spark Lite ist ein leichtgewichtiges großes Sprachmodell mit extrem niedriger Latenz und effizienter Verarbeitung, das vollständig kostenlos und offen ist und Echtzeitsuchfunktionen unterstützt. Seine schnelle Reaktionsfähigkeit macht es besonders geeignet für Inferenzanwendungen und Modellanpassungen auf Geräten mit geringer Rechenleistung und bietet den Nutzern ein hervorragendes Kosten-Nutzen-Verhältnis sowie ein intelligentes Erlebnis, insbesondere in den Bereichen Wissensabfragen, Inhaltserstellung und Suchszenarien."
},
- "llama-3.1-70b-instruct": {
- "description": "Das Llama 3.1 70B Instruct-Modell hat 70B Parameter und bietet herausragende Leistungen bei der Generierung großer Texte und Anweisungsaufgaben."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B bietet leistungsstarke KI-Schlussfolgerungsfähigkeiten, die für komplexe Anwendungen geeignet sind und eine hohe Rechenverarbeitung bei gleichzeitiger Effizienz und Genauigkeit unterstützen."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B ist ein leistungsstarkes Modell, das schnelle Textgenerierungsfähigkeiten bietet und sich hervorragend für Anwendungen eignet, die große Effizienz und Kosteneffektivität erfordern."
},
- "llama-3.1-8b-instruct": {
- "description": "Das Llama 3.1 8B Instruct-Modell hat 8B Parameter und unterstützt die effiziente Ausführung von bildbasierten Anweisungsaufgaben und bietet hochwertige Textgenerierungsfähigkeiten."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Das Llama 3.1 Sonar Huge Online-Modell hat 405B Parameter und unterstützt eine Kontextlänge von etwa 127.000 Markierungen, es wurde für komplexe Online-Chat-Anwendungen entwickelt."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Das Llama 3.1 Sonar Large Chat-Modell hat 70B Parameter und unterstützt eine Kontextlänge von etwa 127.000 Markierungen, es eignet sich für komplexe Offline-Chat-Aufgaben."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Das Llama 3.1 Sonar Large Online-Modell hat 70B Parameter und unterstützt eine Kontextlänge von etwa 127.000 Markierungen, es eignet sich für hochvolumige und vielfältige Chat-Aufgaben."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Das Llama 3.1 Sonar Small Chat-Modell hat 8B Parameter und wurde speziell für Offline-Chat entwickelt, es unterstützt eine Kontextlänge von etwa 127.000 Markierungen."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Das Llama 3.1 Sonar Small Online-Modell hat 8B Parameter und unterstützt eine Kontextlänge von etwa 127.000 Markierungen, es wurde speziell für Online-Chat entwickelt und kann verschiedene Textinteraktionen effizient verarbeiten."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 ist darauf ausgelegt, Aufgaben zu bearbeiten, die visuelle und textuelle Daten kombinieren. Es zeigt hervorragende Leistungen bei Aufgaben wie Bildbeschreibung und visuellen Fragen und Antworten und überbrückt die Kluft zwischen Sprachgenerierung und visueller Schlussfolgerung."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 ist das fortschrittlichste mehrsprachige Open-Source-Sprachmodell der Llama-Serie, das eine Leistung bietet, die mit einem 405B-Modell vergleichbar ist, und das zu extrem niedrigen Kosten. Es basiert auf der Transformer-Architektur und verbessert die Nützlichkeit und Sicherheit durch überwachte Feinabstimmung (SFT) und verstärkendes Lernen mit menschlichem Feedback (RLHF). Die auf Anweisungen optimierte Version ist speziell für mehrsprachige Dialoge optimiert und übertrifft in mehreren Branchenbenchmarks viele Open-Source- und geschlossene Chat-Modelle. Das Wissensdatum endet im Dezember 2023."
+ },
"llama-3.3-70b-versatile": {
"description": "Das Meta Llama 3.3 ist ein mehrsprachiges, großes Sprachmodell (LLM), das aus einem vortrainierten und anweisungsorientierten generativen Modell mit 70B (Text-Eingabe/Text-Ausgabe) besteht. Das anweisungsorientierte Modell von Llama 3.3 ist für mehrsprachige Dialoganwendungen optimiert und übertrifft viele verfügbare Open-Source- und Closed-Source-Chat-Modelle bei gängigen Branchenbenchmarks."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 ist darauf ausgelegt, Aufgaben zu bewältigen, die sowohl visuelle als auch Textdaten kombinieren. Es erzielt hervorragende Ergebnisse bei Aufgaben wie Bildbeschreibung und visueller Fragebeantwortung und überbrückt die Kluft zwischen Sprachgenerierung und visueller Schlussfolgerung."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 ist das fortschrittlichste mehrsprachige Open-Source-Sprachmodell der Llama-Serie, das zu extrem niedrigen Kosten eine Leistung bietet, die mit der eines 405B-Modells vergleichbar ist. Basierend auf der Transformer-Architektur und verbessert durch überwachte Feinabstimmung (SFT) und verstärkendes Lernen mit menschlichem Feedback (RLHF) für Nützlichkeit und Sicherheit. Die optimierte Version für Anweisungen ist speziell für mehrsprachige Dialoge optimiert und übertrifft in mehreren Branchenbenchmarks viele Open-Source- und geschlossene Chat-Modelle. Wissensstichtag ist der 31. Dezember 2023."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Das Meta Llama 3.3 mehrsprachige große Sprachmodell (LLM) ist ein vortrainiertes und anweisungsoptimiertes Generierungsmodell mit 70B (Textinput/Textoutput). Das anweisungsoptimierte reine Textmodell von Llama 3.3 wurde für mehrsprachige Dialoganwendungen optimiert und übertrifft viele verfügbare Open-Source- und geschlossene Chat-Modelle in gängigen Branchenbenchmarks."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 ist darauf ausgelegt, Aufgaben zu bewältigen, die sowohl visuelle als auch Textdaten kombinieren. Es erzielt hervorragende Ergebnisse bei Aufgaben wie Bildbeschreibung und visueller Fragebeantwortung und überbrückt die Kluft zwischen Sprachgenerierung und visueller Schlussfolgerung."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "Das 405B Llama 3.1 Turbo-Modell bietet eine enorme Kapazität zur Unterstützung von Kontexten für die Verarbeitung großer Datenmengen und zeigt herausragende Leistungen in groß angelegten KI-Anwendungen."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 ist das führende Modell von Meta, das bis zu 405B Parameter unterstützt und in komplexen Gesprächen, mehrsprachiger Übersetzung und Datenanalyse eingesetzt werden kann."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B bietet effiziente Dialogunterstützung in mehreren Sprachen."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 ist darauf ausgelegt, Aufgaben zu bearbeiten, die visuelle und textuelle Daten kombinieren. Es zeigt hervorragende Leistungen bei Aufgaben wie Bildbeschreibung und visuellem Fragen und Antworten und überbrückt die Kluft zwischen Sprachgenerierung und visueller Schlussfolgerung."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 ist das fortschrittlichste mehrsprachige Open-Source-Sprachmodell der Llama-Serie, das eine Leistung bietet, die mit einem 405B-Modell vergleichbar ist, und das zu extrem niedrigen Kosten. Es basiert auf der Transformer-Architektur und verbessert die Nützlichkeit und Sicherheit durch überwachte Feinabstimmung (SFT) und verstärkendes Lernen mit menschlichem Feedback (RLHF). Die auf Anweisungen optimierte Version ist speziell für mehrsprachige Dialoge optimiert und übertrifft in mehreren Branchenbenchmarks viele Open-Source- und geschlossene Chat-Modelle. Das Wissensdatum endet im Dezember 2023."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 ist das fortschrittlichste mehrsprachige Open-Source-Sprachmodell der Llama-Serie, das eine Leistung bietet, die mit einem 405B-Modell vergleichbar ist, und das zu extrem niedrigen Kosten. Es basiert auf der Transformer-Architektur und verbessert die Nützlichkeit und Sicherheit durch überwachte Feinabstimmung (SFT) und verstärkendes Lernen mit menschlichem Feedback (RLHF). Die auf Anweisungen optimierte Version ist speziell für mehrsprachige Dialoge optimiert und übertrifft in mehreren Branchenbenchmarks viele Open-Source- und geschlossene Chat-Modelle. Das Wissensdatum endet im Dezember 2023."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct ist das größte und leistungsstärkste Modell innerhalb des Llama 3.1 Instruct Modells. Es handelt sich um ein hochentwickeltes Modell für dialogbasierte Schlussfolgerungen und die Generierung synthetischer Daten, das auch als Grundlage für die professionelle kontinuierliche Vorab- und Feinabstimmung in bestimmten Bereichen verwendet werden kann. Die mehrsprachigen großen Sprachmodelle (LLMs) von Llama 3.1 sind eine Gruppe von vortrainierten, anweisungsoptimierten Generierungsmodellen, die in den Größen 8B, 70B und 405B (Text-Eingabe/Ausgabe) verfügbar sind. Die anweisungsoptimierten Textmodelle (8B, 70B, 405B) sind speziell für mehrsprachige Dialoganwendungen optimiert und haben in gängigen Branchenbenchmarks viele verfügbare Open-Source-Chat-Modelle übertroffen. Llama 3.1 ist für kommerzielle und Forschungszwecke in mehreren Sprachen konzipiert. Die anweisungsoptimierten Textmodelle eignen sich für assistentengleiche Chats, während die vortrainierten Modelle für verschiedene Aufgaben der natürlichen Sprachgenerierung angepasst werden können. Das Llama 3.1 Modell unterstützt auch die Nutzung seiner Ausgaben zur Verbesserung anderer Modelle, einschließlich der Generierung synthetischer Daten und der Verfeinerung. Llama 3.1 ist ein autoregressives Sprachmodell, das auf einer optimierten Transformer-Architektur basiert. Die angepasste Version verwendet überwachte Feinabstimmung (SFT) und verstärkendes Lernen mit menschlichem Feedback (RLHF), um den menschlichen Präferenzen für Hilfsbereitschaft und Sicherheit zu entsprechen."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 ist ein offenes großes Sprachmodell (LLM), das sich an Entwickler, Forscher und Unternehmen richtet und ihnen hilft, ihre Ideen für generative KI zu entwickeln, zu experimentieren und verantwortungsbewusst zu skalieren. Als Teil eines globalen Innovationssystems ist es besonders geeignet für Umgebungen mit begrenzter Rechenleistung und Ressourcen, für Edge-Geräte und schnellere Trainingszeiten."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "Fortgeschrittenes LLM, das die Generierung synthetischer Daten, Wissensverdichtung und Schlussfolgerungen unterstützt, geeignet für Chatbots, Programmierung und spezifische Aufgaben."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "Ermöglicht komplexe Gespräche mit hervorragendem Kontextverständnis, Schlussfolgerungsfähigkeiten und Textgenerierungsfähigkeiten."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "Fortschrittliches, hochmodernes Modell mit Sprachverständnis, hervorragenden Schlussfolgerungsfähigkeiten und Textgenerierungsfähigkeiten."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "Spitzenmäßiges visuelles Sprachmodell, das in der Lage ist, qualitativ hochwertige Schlussfolgerungen aus Bildern zu ziehen."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "Fortschrittliches, hochmodernes kleines Sprachmodell mit Sprachverständnis, hervorragenden Schlussfolgerungsfähigkeiten und Textgenerierungsfähigkeiten."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "Fortschrittliches, hochmodernes kleines Sprachmodell mit Sprachverständnis, hervorragenden Schlussfolgerungsfähigkeiten und Textgenerierungsfähigkeiten."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "Spitzenmäßiges visuelles Sprachmodell, das in der Lage ist, qualitativ hochwertige Schlussfolgerungen aus Bildern zu ziehen."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "Fortschrittliches LLM, das auf Schlussfolgern, Mathematik, Allgemeinwissen und Funktionsaufrufen spezialisiert ist."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 ist ein Sprachmodell von Microsoft AI, das in komplexen Dialogen, Mehrsprachigkeit, Inferenz und intelligenten Assistenten besonders gut abschneidet."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K ist ein Modell mit überragenden Fähigkeiten zur Verarbeitung von langen Kontexten, das für die Generierung von sehr langen Texten geeignet ist und die Anforderungen komplexer Generierungsaufgaben erfüllt. Es kann Inhalte mit bis zu 128.000 Tokens verarbeiten und eignet sich hervorragend für Anwendungen in der Forschung, Wissenschaft und der Erstellung großer Dokumente."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Das Kimi-Visionsmodell (einschließlich moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview usw.) kann Bildinhalte verstehen, einschließlich Bildtext, Bildfarbe und Objektformen."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K bietet die Fähigkeit zur Verarbeitung von mittellangen Kontexten und kann 32.768 Tokens verarbeiten, was es besonders geeignet für die Generierung verschiedener langer Dokumente und komplexer Dialoge macht, die in den Bereichen Inhaltserstellung, Berichtsgenerierung und Dialogsysteme eingesetzt werden."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Das Kimi-Visionsmodell (einschließlich moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview usw.) kann Bildinhalte verstehen, einschließlich Bildtext, Bildfarbe und Objektformen."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K ist für die Generierung von Kurztextaufgaben konzipiert und bietet eine effiziente Verarbeitungsleistung, die 8.192 Tokens verarbeiten kann. Es eignet sich hervorragend für kurze Dialoge, Notizen und schnelle Inhaltserstellung."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Das Kimi-Visionsmodell (einschließlich moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview usw.) kann Bildinhalte verstehen, einschließlich Bildtext, Bildfarbe und Objektformen."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B ist die aktualisierte Version von Nous Hermes 2 und enthält die neuesten intern entwickelten Datensätze."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B ist ein großes Sprachmodell, das von NVIDIA maßgeschneidert wurde, um die Hilfe von LLM-generierten Antworten auf Benutzeranfragen zu erhöhen."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B ist ein von NVIDIA maßgeschneidertes großes Sprachmodell, das darauf abzielt, die Hilfsfähigkeit der von LLM generierten Antworten auf Benutzeranfragen zu verbessern. Dieses Modell hat in Benchmark-Tests wie Arena Hard, AlpacaEval 2 LC und GPT-4-Turbo MT-Bench hervorragende Leistungen gezeigt und belegt bis zum 1. Oktober 2024 den ersten Platz in allen drei automatischen Ausrichtungsbenchmarks. Das Modell wurde mit RLHF (insbesondere REINFORCE), Llama-3.1-Nemotron-70B-Reward und HelpSteer2-Preference-Prompts auf dem Llama-3.1-70B-Instruct-Modell trainiert."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "Einzigartiges Sprachmodell, das unvergleichliche Genauigkeit und Effizienz bietet."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct ist ein von NVIDIA maßgeschneidertes großes Sprachmodell, das darauf abzielt, die Hilfsbereitschaft der von LLM generierten Antworten zu verbessern."
+ },
"o1": {
"description": "Konzentriert sich auf fortgeschrittene Inferenz und die Lösung komplexer Probleme, einschließlich mathematischer und wissenschaftlicher Aufgaben. Besonders geeignet für Anwendungen, die ein tiefes Verständnis des Kontexts und die Abwicklung von Arbeitsabläufen erfordern."
},
- "o1-2024-12-17": {
- "description": "o1 ist OpenAIs neues Inferenzmodell, das Text- und Bildinput unterstützt und Textausgaben erzeugt. Es eignet sich für komplexe Aufgaben, die umfassendes Allgemeinwissen erfordern. Dieses Modell hat einen Kontext von 200K und einen Wissensstand bis Oktober 2023."
- },
"o1-mini": {
"description": "o1-mini ist ein schnelles und kosteneffizientes Inferenzmodell, das für Programmier-, Mathematik- und Wissenschaftsanwendungen entwickelt wurde. Das Modell hat einen Kontext von 128K und einen Wissensstand bis Oktober 2023."
},
"o1-preview": {
"description": "o1 ist OpenAIs neues Inferenzmodell, das für komplexe Aufgaben geeignet ist, die umfangreiches Allgemeinwissen erfordern. Das Modell hat einen Kontext von 128K und einen Wissensstand bis Oktober 2023."
},
+ "o3-mini": {
+ "description": "o3-mini ist unser neuestes kompaktes Inferenzmodell, das bei den gleichen Kosten- und Verzögerungszielen wie o1-mini hohe Intelligenz bietet."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba ist ein auf die Codegenerierung spezialisiertes Mamba 2-Sprachmodell, das starke Unterstützung für fortschrittliche Code- und Schlussfolgerungsaufgaben bietet."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini ist das neueste Modell von OpenAI, das nach GPT-4 Omni veröffentlicht wurde und Text- und Bild-Eingaben unterstützt. Als ihr fortschrittlichstes kleines Modell ist es viel günstiger als andere neueste Modelle und über 60 % günstiger als GPT-3.5 Turbo. Es behält die fortschrittlichste Intelligenz bei und bietet gleichzeitig ein hervorragendes Preis-Leistungs-Verhältnis. GPT-4o mini erzielte 82 % im MMLU-Test und rangiert derzeit in den Chat-Präferenzen über GPT-4."
},
- "openai/o1": {
- "description": "o1 ist OpenAIs neues Inferenzmodell, das Text- und Bildinput unterstützt und Textausgaben erzeugt. Es eignet sich für komplexe Aufgaben, die umfassendes Allgemeinwissen erfordern. Dieses Modell hat einen Kontext von 200K und einen Wissensstand bis Oktober 2023."
- },
"openai/o1-mini": {
"description": "o1-mini ist ein schnelles und kosteneffizientes Inferenzmodell, das für Programmier-, Mathematik- und Wissenschaftsanwendungen entwickelt wurde. Das Modell hat einen Kontext von 128K und einen Wissensstand bis Oktober 2023."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K verfügt über eine außergewöhnliche Kontextverarbeitungsfähigkeit und kann bis zu 128K Kontextinformationen verarbeiten, was es besonders geeignet für die Analyse langer Texte und die Verarbeitung langfristiger logischer Zusammenhänge macht. Es bietet in komplexen Textkommunikationen flüssige und konsistente Logik sowie vielfältige Unterstützung für Zitate."
},
+ "qvq-72b-preview": {
+ "description": "Das QVQ-Modell ist ein experimentelles Forschungsmodell, das vom Qwen-Team entwickelt wurde und sich auf die Verbesserung der visuellen Schlussfolgerungsfähigkeiten konzentriert, insbesondere im Bereich der mathematischen Schlussfolgerungen."
+ },
"qwen-coder-plus-latest": {
"description": "Tongyi Qianwen Code-Modell."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Das Tongyi Qianwen Ultra-Scale Visuelle Sprachmodell. Im Vergleich zur verbesserten Version wurden die Fähigkeiten zur visuellen Schlussfolgerung und Befolgung von Anweisungen weiter gesteigert, was ein höheres Niveau an visueller Wahrnehmung und Kognition bietet."
},
+ "qwen-vl-ocr-latest": {
+ "description": "Tongyi Qianwen OCR ist ein spezialisiertes Modell zur Textextraktion, das sich auf die Textextraktionsfähigkeiten von Dokumenten, Tabellen, Prüfungsfragen und handschriftlichen Texten konzentriert. Es kann verschiedene Schriftarten erkennen und unterstützt derzeit folgende Sprachen: Chinesisch, Englisch, Französisch, Japanisch, Koreanisch, Deutsch, Russisch, Italienisch, Vietnamesisch und Arabisch."
+ },
"qwen-vl-plus-latest": {
"description": "Die verbesserte Version des Tongyi Qianwen, eines großangelegten visuellen Sprachmodells. Deutlich verbesserte Fähigkeiten zur Detailerkennung und Texterkennung, unterstützt Bildauflösungen von über einer Million Pixel und beliebige Seitenverhältnisse."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 ist eine neue Serie großer Sprachmodelle mit stärkeren Verständnis- und Generierungsfähigkeiten."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "LLM, das auf Chinesisch und Englisch ausgerichtet ist und sich auf Sprache, Programmierung, Mathematik, Schlussfolgern und andere Bereiche konzentriert."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "Fortgeschrittenes LLM, das die Codegenerierung, Schlussfolgerungen und Korrekturen unterstützt und gängige Programmiersprachen abdeckt."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "Leistungsstarkes, mittelgroßes Codierungsmodell, das 32K Kontextlängen unterstützt und in der mehrsprachigen Programmierung versiert ist."
+ },
"qwen2": {
"description": "Qwen2 ist das neue große Sprachmodell von Alibaba, das mit hervorragender Leistung eine Vielzahl von Anwendungsanforderungen unterstützt."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "Das 14B-Modell von Tongyi Qianwen 2.5 ist öffentlich zugänglich."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "Tongyi Qianwen 2.5 ist ein Open-Source-Modell mit einer Größe von 72B."
+ },
"qwen2.5-32b-instruct": {
"description": "Das 32B-Modell von Tongyi Qianwen 2.5 ist öffentlich zugänglich."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Das Qwen-Math-Modell verfügt über starke Fähigkeiten zur Lösung mathematischer Probleme."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "Verbesserte Befolgung von Anweisungen, Mathematik, Problemlösung und Programmierung, gesteigerte Erkennungsfähigkeiten für alle Arten von visuellen Elementen, Unterstützung für die präzise Lokalisierung visueller Elemente in verschiedenen Formaten, Verständnis von langen Videodateien (maximal 10 Minuten) und sekundengenauer Ereigniszeitpunktlokalisierung, Fähigkeit zur zeitlichen Einordnung und Geschwindigkeitsverständnis, Unterstützung für die Steuerung von OS- oder Mobile-Agenten basierend auf Analyse- und Lokalisierungsfähigkeiten, starke Fähigkeit zur Extraktion von Schlüsselinformationen und JSON-Format-Ausgabe. Diese Version ist die leistungsstärkste Version der 72B-Serie."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "Verbesserte Befolgung von Anweisungen, Mathematik, Problemlösung und Programmierung, gesteigerte Erkennungsfähigkeiten für alle Arten von visuellen Elementen, Unterstützung für die präzise Lokalisierung visueller Elemente in verschiedenen Formaten, Verständnis von langen Videodateien (maximal 10 Minuten) und sekundengenauer Ereigniszeitpunktlokalisierung, Fähigkeit zur zeitlichen Einordnung und Geschwindigkeitsverständnis, Unterstützung für die Steuerung von OS- oder Mobile-Agenten basierend auf Analyse- und Lokalisierungsfähigkeiten, starke Fähigkeit zur Extraktion von Schlüsselinformationen und JSON-Format-Ausgabe. Diese Version ist die leistungsstärkste Version der 72B-Serie."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 ist das neue, groß angelegte Sprachmodell der Alibaba-Gruppe, das hervorragende Leistungen zur Unterstützung vielfältiger Anwendungsbedürfnisse bietet."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro ist ein hochintelligentes LLM, das von Upstage entwickelt wurde und sich auf die Befolgung von Anweisungen mit einer einzigen GPU konzentriert, mit einem IFEval-Score von über 80. Derzeit unterstützt es Englisch, die offizielle Version ist für November 2024 geplant und wird die Sprachunterstützung und Kontextlänge erweitern."
},
+ "sonar": {
+ "description": "Ein leichtgewichtiges Suchprodukt, das auf kontextbezogener Suche basiert und schneller und günstiger ist als Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "Ein fortschrittliches Suchprodukt, das kontextbezogene Suche unterstützt und erweiterte Abfragen sowie Nachverfolgung ermöglicht."
+ },
+ "sonar-reasoning": {
+ "description": "Ein neues API-Produkt, das von DeepSeek-Inferenzmodellen unterstützt wird."
+ },
"step-1-128k": {
"description": "Bietet ein ausgewogenes Verhältnis zwischen Leistung und Kosten, geeignet für allgemeine Szenarien."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "Dieses Modell verfügt über starke Fähigkeiten zur Videoanalyse."
},
+ "step-1o-vision-32k": {
+ "description": "Dieses Modell verfügt über starke Fähigkeiten zur Bildverständnis. Im Vergleich zu den Modellen der Schritt-1v-Serie bietet es eine verbesserte visuelle Leistung."
+ },
"step-1v-32k": {
"description": "Unterstützt visuelle Eingaben und verbessert die multimodale Interaktionserfahrung."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Unterstützt groß angelegte Kontextinteraktionen und eignet sich für komplexe Dialogszenarien."
},
+ "step-2-mini": {
+ "description": "Ein ultraschnelles Großmodell, das auf der neuen, selbstentwickelten Attention-Architektur MFA basiert. Es erreicht mit extrem niedrigen Kosten ähnliche Ergebnisse wie Schritt 1 und bietet gleichzeitig eine höhere Durchsatzrate und schnellere Reaktionszeiten. Es kann allgemeine Aufgaben bearbeiten und hat besondere Fähigkeiten im Bereich der Codierung."
+ },
+ "taichu2_mm": {
+ "description": "Integriert Fähigkeiten zur Bildverstehung, Wissensübertragung und logischen Attribution und zeigt herausragende Leistungen im Bereich der Bild-Text-Fragen."
+ },
"taichu_llm": {
"description": "Das Zīdōng Taichu Sprachmodell verfügt über außergewöhnliche Sprachverständnisfähigkeiten sowie Fähigkeiten in Textgenerierung, Wissensabfrage, Programmierung, mathematischen Berechnungen, logischem Denken, Sentimentanalyse und Textzusammenfassung. Es kombiniert innovativ große Datenvortrainings mit reichhaltigem Wissen aus mehreren Quellen, verfeinert kontinuierlich die Algorithmen und absorbiert ständig neues Wissen aus umfangreichen Textdaten in Bezug auf Vokabular, Struktur, Grammatik und Semantik, um die Leistung des Modells kontinuierlich zu verbessern. Es bietet den Nutzern bequemere Informationen und Dienstleistungen sowie ein intelligenteres Erlebnis."
},
+ "text-embedding-3-large": {
+ "description": "Das leistungsstärkste Vektormodell, geeignet für englische und nicht-englische Aufgaben."
+ },
+ "text-embedding-3-small": {
+ "description": "Effizientes und kostengünstiges neues Embedding-Modell, geeignet für Wissensabruf, RAG-Anwendungen und andere Szenarien."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) bietet durch effiziente Strategien und Modellarchitekturen verbesserte Rechenfähigkeiten."
},
+ "tts-1": {
+ "description": "Das neueste Text-zu-Sprache-Modell, optimiert für Geschwindigkeit in Echtzeitszenarien."
+ },
+ "tts-1-hd": {
+ "description": "Das neueste Text-zu-Sprache-Modell, optimiert für Qualität."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) eignet sich für präzise Anweisungsaufgaben und bietet hervorragende Sprachverarbeitungsfähigkeiten."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet hebt den Branchenstandard an, übertrifft die Konkurrenzmodelle und Claude 3 Opus und zeigt in umfangreichen Bewertungen hervorragende Leistungen, während es die Geschwindigkeit und Kosten unserer mittelgroßen Modelle beibehält."
},
+ "whisper-1": {
+ "description": "Allgemeines Spracherkennungsmodell, unterstützt mehrsprachige Spracherkennung, Sprachübersetzung und Spracherkennung."
+ },
"wizardlm2": {
"description": "WizardLM 2 ist ein Sprachmodell von Microsoft AI, das in komplexen Dialogen, mehrsprachigen Anwendungen, Schlussfolgerungen und intelligenten Assistenten besonders gut abschneidet."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Modell für komplexe visuelle Aufgaben, das hohe Leistungsfähigkeit bei der Bildverarbeitung und -analyse bietet."
+ },
+ "yi-vision-v2": {
+ "description": "Ein Modell für komplexe visuelle Aufgaben, das leistungsstarke Verständnis- und Analysefähigkeiten auf der Grundlage mehrerer Bilder bietet."
}
}
diff --git a/locales/de-DE/providers.json b/locales/de-DE/providers.json
index e2e77447fc1ad..a2dcde1c2e424 100644
--- a/locales/de-DE/providers.json
+++ b/locales/de-DE/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure bietet eine Vielzahl fortschrittlicher KI-Modelle, darunter GPT-3.5 und die neueste GPT-4-Serie, die verschiedene Datentypen und komplexe Aufgaben unterstützen und sich auf sichere, zuverlässige und nachhaltige KI-Lösungen konzentrieren."
},
+ "azureai": {
+ "description": "Azure bietet eine Vielzahl fortschrittlicher KI-Modelle, darunter GPT-3.5 und die neueste GPT-4-Serie, die verschiedene Datentypen und komplexe Aufgaben unterstützen und sich auf sichere, zuverlässige und nachhaltige KI-Lösungen konzentrieren."
+ },
"baichuan": {
"description": "Baichuan Intelligent ist ein Unternehmen, das sich auf die Forschung und Entwicklung großer KI-Modelle spezialisiert hat. Ihre Modelle zeigen hervorragende Leistungen in chinesischen Aufgaben wie Wissensdatenbanken, Verarbeitung langer Texte und kreative Generierung und übertreffen die gängigen Modelle im Ausland. Baichuan Intelligent verfügt auch über branchenführende multimodale Fähigkeiten und hat in mehreren renommierten Bewertungen hervorragend abgeschnitten. Ihre Modelle umfassen Baichuan 4, Baichuan 3 Turbo und Baichuan 3 Turbo 128k, die jeweils für unterschiedliche Anwendungsszenarien optimiert sind und kosteneffiziente Lösungen bieten."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek ist ein Unternehmen, das sich auf die Forschung und Anwendung von KI-Technologien spezialisiert hat. Ihr neuestes Modell, DeepSeek-V2.5, kombiniert allgemeine Dialog- und Codeverarbeitungsfähigkeiten und hat signifikante Fortschritte in den Bereichen menschliche Präferenzanpassung, Schreibaufgaben und Befehlsbefolgung erzielt."
},
+ "doubao": {
+ "description": "Ein von ByteDance entwickeltes großes Modell. Durch die praktische Validierung in über 50 internen Geschäftsszenarien und die kontinuierliche Verfeinerung mit täglich Billionen von Tokens bietet es vielfältige Modalitäten und schafft mit hochwertigen Modellergebnissen ein reichhaltiges Geschäftserlebnis für Unternehmen."
+ },
"fireworksai": {
"description": "Fireworks AI ist ein führender Anbieter von fortschrittlichen Sprachmodellen, der sich auf Funktionsaufrufe und multimodale Verarbeitung spezialisiert hat. Ihr neuestes Modell, Firefunction V2, basiert auf Llama-3 und ist für Funktionsaufrufe, Dialoge und Befehlsbefolgung optimiert. Das visuelle Sprachmodell FireLLaVA-13B unterstützt gemischte Eingaben von Bildern und Text. Weitere bemerkenswerte Modelle sind die Llama-Serie und die Mixtral-Serie, die effiziente mehrsprachige Befehlsbefolgung und Generierungsunterstützung bieten."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "Eine Open-Source-Organisation, die sich der Forschung und Entwicklung von großen Modellen und Werkzeugketten widmet. Sie bietet allen KI-Entwicklern eine effiziente und benutzerfreundliche Open-Source-Plattform, die den Zugang zu den neuesten Technologien und Algorithmen für große Modelle ermöglicht."
},
+ "jina": {
+ "description": "Jina AI wurde 2020 gegründet und ist ein führendes Unternehmen im Bereich Such-KI. Unsere Suchplattform umfasst Vektormodelle, Re-Ranker und kleine Sprachmodelle, die Unternehmen dabei helfen, zuverlässige und qualitativ hochwertige generative KI- und multimodale Suchanwendungen zu entwickeln."
+ },
+ "lmstudio": {
+ "description": "LM Studio ist eine Desktop-Anwendung zum Entwickeln und Experimentieren mit LLMs auf Ihrem Computer."
+ },
"minimax": {
"description": "MiniMax ist ein im Jahr 2021 gegründetes Unternehmen für allgemeine künstliche Intelligenz, das sich der gemeinsamen Schaffung von Intelligenz mit den Nutzern widmet. MiniMax hat verschiedene multimodale allgemeine große Modelle entwickelt, darunter ein Textmodell mit Billionen von Parametern, ein Sprachmodell und ein Bildmodell. Außerdem wurden Anwendungen wie Conch AI eingeführt."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI ist eine Plattform, die eine Vielzahl von großen Sprachmodellen und API-Diensten für die KI-Bilderzeugung anbietet, die flexibel, zuverlässig und kosteneffektiv ist. Sie unterstützt die neuesten Open-Source-Modelle wie Llama3 und Mistral und bietet umfassende, benutzerfreundliche und automatisch skalierbare API-Lösungen für die Entwicklung generativer KI-Anwendungen, die für das schnelle Wachstum von KI-Startups geeignet sind."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ bietet Container für selbstgehostete, GPU-beschleunigte Inferenz-Mikrodienste, die die Bereitstellung von vortrainierten und benutzerdefinierten KI-Modellen in der Cloud, in Rechenzentren, auf RTX™ AI-PCs und Workstations unterstützen."
+ },
"ollama": {
"description": "Die von Ollama angebotenen Modelle decken ein breites Spektrum ab, darunter Code-Generierung, mathematische Berechnungen, mehrsprachige Verarbeitung und dialogbasierte Interaktionen, und unterstützen die vielfältigen Anforderungen an unternehmensgerechte und lokal angepasste Bereitstellungen."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "Das Institut für Automatisierung der Chinesischen Akademie der Wissenschaften und das Wuhan Institute of Artificial Intelligence haben ein neues Generation multimodales großes Modell eingeführt, das umfassende Frage-Antwort-Aufgaben unterstützt, darunter mehrstufige Fragen, Textgenerierung, Bildgenerierung, 3D-Verständnis und Signalverarbeitung, mit stärkeren kognitiven, verstehenden und kreativen Fähigkeiten, die ein neues interaktives Erlebnis bieten."
},
+ "tencentcloud": {
+ "description": "Die atomare Fähigkeit der Wissensmaschine (LLM Knowledge Engine Atomic Power) basiert auf der Entwicklung der Wissensmaschine und bietet eine umfassende Fähigkeit zur Wissensabfrage für Unternehmen und Entwickler. Sie können mit verschiedenen atomaren Fähigkeiten Ihren eigenen Modellservice erstellen und Dokumentenanalysen, -aufteilungen, Embeddings, mehrfache Umformulierungen und andere Dienste kombinieren, um maßgeschneiderte KI-Lösungen für Ihr Unternehmen zu entwickeln."
+ },
"togetherai": {
"description": "Together AI strebt an, durch innovative KI-Modelle führende Leistungen zu erzielen und bietet umfangreiche Anpassungsmöglichkeiten, einschließlich schneller Skalierungsunterstützung und intuitiver Bereitstellungsprozesse, um den unterschiedlichen Anforderungen von Unternehmen gerecht zu werden."
},
"upstage": {
"description": "Upstage konzentriert sich auf die Entwicklung von KI-Modellen für verschiedene geschäftliche Anforderungen, einschließlich Solar LLM und Dokumenten-KI, mit dem Ziel, künstliche allgemeine Intelligenz (AGI) zu erreichen. Es ermöglicht die Erstellung einfacher Dialogagenten über die Chat-API und unterstützt Funktionsaufrufe, Übersetzungen, Einbettungen und spezifische Anwendungsbereiche."
},
+ "vllm": {
+ "description": "vLLM ist eine schnelle und benutzerfreundliche Bibliothek für LLM-Inferenz und -Dienste."
+ },
+ "volcengine": {
+ "description": "Die von ByteDance eingeführte Entwicklungsplattform für große Modellservices bietet funktionsreiche, sichere und preislich wettbewerbsfähige Modellaufrufdienste. Sie bietet zudem End-to-End-Funktionen wie Moduldaten, Feinabstimmung, Inferenz und Bewertung, um die Entwicklung Ihrer KI-Anwendungen umfassend zu unterstützen."
+ },
"wenxin": {
"description": "Eine unternehmensweite, umfassende Plattform für die Entwicklung und den Service von großen Modellen und KI-nativen Anwendungen, die die vollständigsten und benutzerfreundlichsten Werkzeuge für die Entwicklung generativer KI-Modelle und den gesamten Anwendungsentwicklungsprozess bietet."
},
diff --git a/locales/de-DE/setting.json b/locales/de-DE/setting.json
index 9ce0ab6b16ff0..13a24aa532118 100644
--- a/locales/de-DE/setting.json
+++ b/locales/de-DE/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Maximale Token pro Antwort aktivieren"
},
+ "enableReasoningEffort": {
+ "title": "Aktivieren Sie die Anpassung der Schlussfolgerungsintensität"
+ },
"frequencyPenalty": {
- "desc": "Je höher der Wert, desto wahrscheinlicher ist es, dass sich wiederholende Wörter reduziert werden",
- "title": "Frequenzstrafe"
+ "desc": "Je höher der Wert, desto vielfältiger und abwechslungsreicher die Wortwahl; je niedriger der Wert, desto einfacher und schlichter die Wortwahl",
+ "title": "Wortvielfalt"
},
"maxTokens": {
"desc": "Maximale Anzahl von Tokens, die pro Interaktion verwendet werden",
@@ -212,19 +215,31 @@
"desc": "{{provider}} Modell",
"title": "Modell"
},
+ "params": {
+ "title": "Erweiterte Parameter"
+ },
"presencePenalty": {
- "desc": "Je höher der Wert, desto wahrscheinlicher ist es, dass sich das Gespräch auf neue Themen ausweitet",
- "title": "Themenfrische"
+ "desc": "Je höher der Wert, desto eher werden unterschiedliche Ausdrucksweisen bevorzugt, um Wiederholungen zu vermeiden; je niedriger der Wert, desto eher werden wiederholte Konzepte oder Erzählungen verwendet, was zu einer konsistenteren Ausdrucksweise führt",
+ "title": "Ausdrucksvielfalt"
+ },
+ "reasoningEffort": {
+ "desc": "Je höher der Wert, desto stärker die Schlussfolgerungsfähigkeit, aber dies kann die Antwortzeit und den Tokenverbrauch erhöhen.",
+ "options": {
+ "high": "Hoch",
+ "low": "Niedrig",
+ "medium": "Mittel"
+ },
+ "title": "Schlussfolgerungsintensität"
},
"temperature": {
- "desc": "Je höher der Wert, desto zufälliger die Antwort",
- "title": "Zufälligkeit",
- "titleWithValue": "Zufälligkeit {{value}}"
+ "desc": "Je höher der Wert, desto kreativer und einfallsreicher die Antworten; je niedriger der Wert, desto strenger die Antworten",
+ "title": "Kreativitätsgrad",
+ "warning": "Ein zu hoher Kreativitätsgrad kann zu unverständlichen Ausgaben führen"
},
- "title": "Modelleinstellungen",
+ "title": "Modell Einstellungen",
"topP": {
- "desc": "Ähnlich wie Zufälligkeit, aber nicht zusammen mit Zufälligkeit ändern",
- "title": "Top-P-Sampling"
+ "desc": "Wie viele Möglichkeiten in Betracht gezogen werden, je höher der Wert, desto mehr mögliche Antworten werden akzeptiert; je niedriger der Wert, desto eher wird die wahrscheinlichste Antwort gewählt. Es wird nicht empfohlen, dies zusammen mit dem Kreativitätsgrad zu ändern",
+ "title": "Offenheit des Denkens"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Allgemeine Einstellungen",
"experiment": "Experiment",
"llm": "Sprachmodell",
+ "provider": "KI-Dienstanbieter",
"sync": "Cloud-Synchronisierung",
"system-agent": "Systemassistent",
"tts": "Sprachdienste"
diff --git a/locales/en-US/changelog.json b/locales/en-US/changelog.json
index bba44698e4366..f9563ddf5ae1c 100644
--- a/locales/en-US/changelog.json
+++ b/locales/en-US/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "View all changelogs",
"description": "Stay updated on the new features and improvements of {{appName}}",
"pagination": {
- "older": "View Historical Changes",
- "prev": "Previous Page"
+ "next": "Next Page",
+ "older": "View Historical Changes"
},
"readDetails": "Read details",
"title": "Changelog",
diff --git a/locales/en-US/common.json b/locales/en-US/common.json
index 02d67d4766042..ca92f236b2efa 100644
--- a/locales/en-US/common.json
+++ b/locales/en-US/common.json
@@ -275,6 +275,7 @@
},
"temp": "Temporary",
"terms": "Terms of Service",
+ "update": "Update",
"updateAgent": "Update Assistant Information",
"upgradeVersion": {
"action": "Upgrade",
@@ -285,6 +286,7 @@
"anonymousNickName": "Anonymous User",
"billing": "Billing Management",
"cloud": "Launch {{name}}",
+ "community": "Community Edition",
"data": "Data Storage",
"defaultNickname": "Community User",
"discord": "Community Support",
@@ -294,8 +296,7 @@
"help": "Help Center",
"moveGuide": "The settings button has been moved here",
"plans": "Subscription Plans",
- "preview": "Preview",
- "profile": "Account Management",
+ "profile": "Account",
"setting": "Settings",
"usages": "Usage Statistics"
},
diff --git a/locales/en-US/components.json b/locales/en-US/components.json
index 9b7334f94b541..94f7b1d5c77f5 100644
--- a/locales/en-US/components.json
+++ b/locales/en-US/components.json
@@ -76,6 +76,7 @@
"custom": "Custom model, by default, supports both function call and visual recognition. Please verify the availability of the above capabilities based on actual situations.",
"file": "This model supports file upload for reading and recognition.",
"functionCall": "This model supports function call.",
+ "reasoning": "This model supports deep thinking",
"tokens": "This model supports up to {{tokens}} tokens in a single session.",
"vision": "This model supports visual recognition."
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "No enabled model. Please go to settings to enable.",
"provider": "Provider"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "Due to browser security restrictions, you need to configure cross-origin settings for Ollama to function properly.",
+ "linux": {
+ "env": "Add `Environment` under the [Service] section, and set the OLLAMA_ORIGINS environment variable:",
+ "reboot": "Reload systemd and restart Ollama",
+ "systemd": "Edit the ollama service using systemd:"
+ },
+ "macos": "Please open the 'Terminal' application, paste the following command, and press Enter to run it.",
+ "reboot": "Please restart the Ollama service after the execution is complete.",
+ "title": "Configure Ollama for Cross-Origin Access",
+ "windows": "On Windows, click 'Control Panel' and go to edit system environment variables. Create a new environment variable named 'OLLAMA_ORIGINS' for your user account, with the value set to *, and click 'OK/Apply' to save."
+ },
+ "install": {
+ "description": "Please ensure that you have started Ollama. If you haven't downloaded Ollama, please visit the official website to <1>download it.",
+ "docker": "If you prefer to use Docker, Ollama also provides an official Docker image, which you can pull using the following command:",
+ "linux": {
+ "command": "Install using the following command:",
+ "manual": "Alternatively, you can refer to the <1>Linux Manual Installation Guide for a manual installation."
+ },
+ "title": "Install and Start the Ollama Application Locally",
+ "windowsTab": "Windows (Preview)"
+ }
+ },
+ "Thinking": {
+ "thinking": "Deep in thought...",
+ "thought": "Deeply thought (took {{duration}} seconds)",
+ "thoughtWithDuration": "Deeply thought"
}
}
diff --git a/locales/en-US/discover.json b/locales/en-US/discover.json
index 84e71d0f6c367..3d2a702acad2a 100644
--- a/locales/en-US/discover.json
+++ b/locales/en-US/discover.json
@@ -126,6 +126,10 @@
"title": "Topic Freshness"
},
"range": "Range",
+ "reasoning_effort": {
+ "desc": "This setting controls the intensity of reasoning the model applies before generating a response. Low intensity prioritizes response speed and saves tokens, while high intensity provides more comprehensive reasoning but consumes more tokens and slows down response time. The default value is medium, balancing reasoning accuracy with response speed.",
+ "title": "Reasoning Intensity"
+ },
"temperature": {
"desc": "This setting affects the diversity of the model's responses. Lower values lead to more predictable and typical responses, while higher values encourage more diverse and less common responses. When set to 0, the model always gives the same response to a given input.",
"title": "Randomness"
diff --git a/locales/en-US/modelProvider.json b/locales/en-US/modelProvider.json
index 0cbd936f8e3a9..315607311264a 100644
--- a/locales/en-US/modelProvider.json
+++ b/locales/en-US/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API Key"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "The API version for Azure, following the YYYY-MM-DD format. Refer to the [latest version](https://learn.microsoft.com/en-us/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "Fetch List",
+ "title": "Azure API Version"
+ },
+ "endpoint": {
+ "desc": "Find the Azure AI model inference endpoint from the Azure AI project overview",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Azure AI Endpoint"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Find the API key from the Azure AI project overview",
+ "placeholder": "Azure Key",
+ "title": "Key"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "Enter AWS Access Key Id",
@@ -63,6 +81,46 @@
"title": "Cloudflare Account ID / API Address"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Please enter your API Key",
+ "title": "API Key"
+ },
+ "basicTitle": "Basic Information",
+ "configTitle": "Configuration Information",
+ "confirm": "Create",
+ "createSuccess": "Creation successful",
+ "description": {
+ "placeholder": "Provider description (optional)",
+ "title": "Provider Description"
+ },
+ "id": {
+ "desc": "Unique identifier for the service provider, which cannot be modified after creation",
+ "format": "Can only contain numbers, lowercase letters, hyphens (-), and underscores (_) ",
+ "placeholder": "Suggested all lowercase, e.g., openai, cannot be modified after creation",
+ "required": "Please enter the provider ID",
+ "title": "Provider ID"
+ },
+ "logo": {
+ "required": "Please upload a valid provider logo",
+ "title": "Provider Logo"
+ },
+ "name": {
+ "placeholder": "Please enter the display name of the provider",
+ "required": "Please enter the provider name",
+ "title": "Provider Name"
+ },
+ "proxyUrl": {
+ "required": "Please enter the proxy address",
+ "title": "Proxy URL"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Please select SDK type",
+ "title": "Request Format"
+ },
+ "title": "Create Custom AI Provider"
+ },
"github": {
"personalAccessToken": {
"desc": "Enter your GitHub PAT. Click [here](https://github.com/settings/tokens) to create one.",
@@ -77,6 +135,23 @@
"title": "HuggingFace Token"
}
},
+ "list": {
+ "title": {
+ "disabled": "Disabled",
+ "enabled": "Enabled"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Add Custom Provider",
+ "all": "All",
+ "list": {
+ "disabled": "Disabled",
+ "enabled": "Enabled"
+ },
+ "notFound": "No search results found",
+ "searchProviders": "Search Providers...",
+ "sort": "Custom Sort"
+ },
"ollama": {
"checker": {
"desc": "Test if the proxy address is correctly filled in",
@@ -94,33 +169,9 @@
"title": "Downloading model {{model}}"
},
"endpoint": {
- "desc": "Enter the Ollama interface proxy address, leave blank if not specified locally",
+ "desc": "Must include http(s)://; can be left blank if not specified locally.",
"title": "Interface proxy address"
},
- "setup": {
- "cors": {
- "description": "Due to browser security restrictions, you need to configure cross-origin settings for Ollama to function properly.",
- "linux": {
- "env": "Add `Environment` under [Service] section, and set the OLLAMA_ORIGINS environment variable:",
- "reboot": "Reload systemd and restart Ollama.",
- "systemd": "Invoke systemd to edit the ollama service:"
- },
- "macos": "Open the 'Terminal' application, paste the following command, and press Enter to run it.",
- "reboot": "Please restart the Ollama service after completion.",
- "title": "Configure Ollama for Cross-Origin Access",
- "windows": "On Windows, go to 'Control Panel' and edit system environment variables. Create a new environment variable named 'OLLAMA_ORIGINS' for your user account, set the value to '*', and click 'OK/Apply' to save."
- },
- "install": {
- "description": "Please make sure you have enabled Ollama. If you haven't downloaded Ollama yet, please visit the official website <1>to download.",
- "docker": "If you prefer using Docker, Ollama also provides an official Docker image. You can pull it using the following command:",
- "linux": {
- "command": "Install using the following command:",
- "manual": "Alternatively, you can refer to the <1>Linux Manual Installation Guide for manual installation."
- },
- "title": "Install and Start Ollama Locally",
- "windowsTab": "Windows (Preview)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Cancel Download",
@@ -131,25 +182,145 @@
"title": "Download specified Ollama model"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Enter the Access Key from the Baidu Qianfan platform",
- "placeholder": "Qianfan Access Key",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "Your key and proxy URL will be encrypted using <1>AES-GCM encryption algorithm",
+ "apiKey": {
+ "desc": "Please enter your {{name}} API Key",
+ "placeholder": "{{name}} API Key",
+ "title": "API Key"
+ },
+ "baseURL": {
+ "desc": "Must include http(s)://",
+ "invalid": "Please enter a valid URL",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "API Proxy URL"
+ },
+ "checker": {
+ "button": "Check",
+ "desc": "Test if the API Key and proxy URL are correctly filled",
+ "pass": "Check passed",
+ "title": "Connectivity Check"
+ },
+ "fetchOnClient": {
+ "desc": "Client request mode will initiate session requests directly from the browser, which can improve response speed",
+ "title": "Use Client Request Mode"
+ },
+ "helpDoc": "Configuration Guide",
+ "waitingForMore": "More models are currently <1>planned for integration, please stay tuned"
},
- "checker": {
- "desc": "Test if the Access Key / Secret Access is filled in correctly"
+ "createNew": {
+ "title": "Create Custom AI Model"
+ },
+ "item": {
+ "config": "Configure Model",
+ "customModelCards": {
+ "addNew": "Create and add {{id}} model",
+ "confirmDelete": "You are about to delete this custom model. Once deleted, it cannot be recovered. Please proceed with caution."
+ },
+ "delete": {
+ "confirm": "Are you sure you want to delete model {{displayName}}?",
+ "success": "Deletion successful",
+ "title": "Delete Model"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Field used for actual requests in Azure OpenAI",
+ "placeholder": "Please enter the model deployment name in Azure",
+ "title": "Model Deployment Name"
+ },
+ "deployName": {
+ "extra": "This field will be used as the model ID when sending requests",
+ "placeholder": "Please enter the actual deployment name or ID of the model",
+ "title": "Model Deployment Name"
+ },
+ "displayName": {
+ "placeholder": "Please enter the display name of the model, e.g., ChatGPT, GPT-4, etc.",
+ "title": "Model Display Name"
+ },
+ "files": {
+ "extra": "The current file upload implementation is just a hack solution, limited to self-experimentation. Please wait for complete file upload capabilities in future implementations.",
+ "title": "File Upload Support"
+ },
+ "functionCall": {
+ "extra": "This configuration will only enable the model's ability to use tools, allowing for the addition of tool-type plugins. However, whether the model can truly use the tools depends entirely on the model itself; please test for usability on your own.",
+ "title": "Support for Tool Usage"
+ },
+ "id": {
+ "extra": "This cannot be modified after creation and will be used as the model ID when calling AI",
+ "placeholder": "Please enter the model ID, e.g., gpt-4o or claude-3.5-sonnet",
+ "title": "Model ID"
+ },
+ "modalTitle": "Custom Model Configuration",
+ "reasoning": {
+ "extra": "This configuration will enable the model's deep thinking capabilities, and the specific effects depend entirely on the model itself. Please test whether this model has usable deep thinking abilities.",
+ "title": "Support Deep Thinking"
+ },
+ "tokens": {
+ "extra": "Set the maximum number of tokens supported by the model",
+ "title": "Maximum Context Window",
+ "unlimited": "Unlimited"
+ },
+ "vision": {
+ "extra": "This configuration will only enable image upload capabilities in the application. Whether recognition is supported depends entirely on the model itself. Please test the visual recognition capabilities of the model yourself.",
+ "title": "Support Vision"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/Image",
+ "inputCharts": "${{amount}}/M Characters",
+ "inputMinutes": "${{amount}}/Minutes",
+ "inputTokens": "Input ${{amount}}/M",
+ "outputTokens": "Output ${{amount}}/M"
+ },
+ "releasedAt": "Released at {{releasedAt}}"
},
- "secretKey": {
- "desc": "Enter the Secret Key from the Baidu Qianfan platform",
- "placeholder": "Qianfan Secret Key",
- "title": "Secret Key"
+ "list": {
+ "addNew": "Add Model",
+ "disabled": "Disabled",
+ "disabledActions": {
+ "showMore": "Show All"
+ },
+ "empty": {
+ "desc": "Please create a custom model or pull a model to get started.",
+ "title": "No available models"
+ },
+ "enabled": "Enabled",
+ "enabledActions": {
+ "disableAll": "Disable All",
+ "enableAll": "Enable All",
+ "sort": "Custom Model Sorting"
+ },
+ "enabledEmpty": "No enabled models available. Please enable your preferred models from the list below~",
+ "fetcher": {
+ "clear": "Clear fetched models",
+ "fetch": "Fetch models",
+ "fetching": "Fetching model list...",
+ "latestTime": "Last updated: {{time}}",
+ "noLatestTime": "Model list not yet fetched"
+ },
+ "resetAll": {
+ "conform": "Are you sure you want to reset all modifications to the current model? After resetting, the current model list will return to its default state.",
+ "success": "Reset successful",
+ "title": "Reset All Modifications"
+ },
+ "search": "Search Models...",
+ "searchResult": "{{count}} models found",
+ "title": "Model List",
+ "total": "{{count}} models available"
},
- "unlock": {
- "customRegion": "Custom Service Region",
- "description": "Input your Access Key / Secret Key to start the session. The application will not record your authentication configuration.",
- "title": "Use Custom Wenxin Yiyan Authentication Information"
- }
+ "searchNotFound": "No search results found"
+ },
+ "sortModal": {
+ "success": "Sort update successful",
+ "title": "Custom Order",
+ "update": "Update"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "You are about to delete this AI provider. Once deleted, it cannot be retrieved. Are you sure you want to delete?",
+ "deleteSuccess": "Deletion successful",
+ "tooltip": "Update provider basic configuration",
+ "updateSuccess": "Update successful"
},
"zeroone": {
"title": "01.AI Zero One Everything"
diff --git a/locales/en-US/models.json b/locales/en-US/models.json
index 0782114554e8a..4a96b12a1febc 100644
--- a/locales/en-US/models.json
+++ b/locales/en-US/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K emphasizes semantic safety and responsibility, designed specifically for applications with high content safety requirements, ensuring accuracy and robustness in user experience."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 builds a chain of thought using tree search and incorporates a reflection mechanism, trained with reinforcement learning, enabling the model to self-reflect and correct errors."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro is an advanced natural language processing model launched by 360, featuring exceptional text generation and understanding capabilities, particularly excelling in generation and creative tasks, capable of handling complex language transformations and role-playing tasks."
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1 uses tree search to build a chain of thought and introduces a reflection mechanism, utilizing reinforcement learning for training, enabling the model to possess self-reflection and error-correction capabilities."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra is the most powerful version in the Spark large model series, enhancing text content understanding and summarization capabilities while upgrading online search links. It is a comprehensive solution for improving office productivity and accurately responding to demands, leading the industry as an intelligent product."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "The leading model in the country, surpassing mainstream foreign models in Chinese tasks such as knowledge encyclopedias, long texts, and creative generation. It also possesses industry-leading multimodal capabilities, excelling in multiple authoritative evaluation benchmarks."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "The DeepSeek-R1 distillation model based on Qwen2.5-Math-1.5B optimizes inference performance through reinforcement learning and cold-start data, refreshing the benchmark for open-source models across multiple tasks."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "The DeepSeek-R1 distillation model based on Qwen2.5-14B optimizes inference performance through reinforcement learning and cold-start data, refreshing the benchmark for open-source models across multiple tasks."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "The DeepSeek-R1 series optimizes inference performance through reinforcement learning and cold-start data, refreshing the benchmark for open-source models across multiple tasks, surpassing the level of OpenAI-o1-mini."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "The DeepSeek-R1 distillation model based on Qwen2.5-Math-7B optimizes inference performance through reinforcement learning and cold-start data, refreshing the benchmark for open-source models across multiple tasks."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite provides extreme response speed and better cost-effectiveness, offering flexible options for various customer scenarios. It supports inference and fine-tuning with a 128k context window."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Baidu's self-developed flagship ultra-large-scale language model, which has achieved a comprehensive upgrade in model capabilities compared to ERNIE 3.5, widely applicable to complex task scenarios across various fields; supports automatic integration with Baidu search plugins to ensure the timeliness of Q&A information."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Baidu's self-developed flagship ultra-large-scale language model, demonstrating excellent overall performance, widely applicable to complex task scenarios across various fields; supports automatic integration with Baidu search plugins to ensure timely Q&A information. Compared to ERNIE 4.0, it performs even better."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "Baidu's self-developed flagship ultra-large-scale language model, demonstrating excellent overall performance, suitable for complex task scenarios across various fields; supports automatic integration with Baidu search plugins to ensure the timeliness of Q&A information. It offers better performance compared to ERNIE 4.0."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct is one of the latest large language models released by Alibaba Cloud. This 7B model shows significant improvements in coding and mathematics. It also provides multilingual support, covering over 29 languages, including Chinese and English. The model has made notable advancements in instruction following, understanding structured data, and generating structured outputs, especially JSON."
},
+ "MiniMax-Text-01": {
+ "description": "In the MiniMax-01 series of models, we have made bold innovations: for the first time, we have implemented a linear attention mechanism on a large scale, making the traditional Transformer architecture no longer the only option. This model has a parameter count of up to 456 billion, with a single activation of 45.9 billion. Its overall performance rivals that of top overseas models while efficiently handling the world's longest context of 4 million tokens, which is 32 times that of GPT-4o and 20 times that of Claude-3.5-Sonnet."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO is a highly flexible multi-model fusion designed to provide an exceptional creative experience."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 demonstrates exceptional performance across various visual language tasks, including document and chart understanding, scene text understanding, OCR, and solving scientific and mathematical problems."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 demonstrates exceptional performance across various visual language tasks, including document and chart understanding, scene text understanding, OCR, and solving scientific and mathematical problems."
- },
"Phi-3-medium-128k-instruct": {
"description": "The same Phi-3-medium model, but with a larger context size for RAG or few-shot prompting."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat is the open-source version of the GLM-4 series pre-trained models launched by Zhipu AI. This model excels in semantics, mathematics, reasoning, code, and knowledge. In addition to supporting multi-turn dialogues, GLM-4-9B-Chat also features advanced capabilities such as web browsing, code execution, custom tool invocation (Function Call), and long-text reasoning. The model supports 26 languages, including Chinese, English, Japanese, Korean, and German. In multiple benchmark tests, GLM-4-9B-Chat has demonstrated excellent performance, such as in AlignBench-v2, MT-Bench, MMLU, and C-Eval. The model supports a maximum context length of 128K, making it suitable for academic research and commercial applications."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 is a reinforcement learning (RL) driven inference model that addresses issues of repetitiveness and readability in models. Prior to RL, DeepSeek-R1 introduced cold start data to further optimize inference performance. It performs comparably to OpenAI-o1 in mathematical, coding, and reasoning tasks, and enhances overall effectiveness through carefully designed training methods."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 is a mixed expert (MoE) language model with 671 billion parameters, utilizing multi-head latent attention (MLA) and the DeepSeekMoE architecture, combined with a load balancing strategy without auxiliary loss to optimize inference and training efficiency. Pre-trained on 14.8 trillion high-quality tokens and fine-tuned with supervision and reinforcement learning, DeepSeek-V3 outperforms other open-source models and approaches leading closed-source models."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma is one of Google's lightweight, state-of-the-art open model series. It is a large language model with a decoder-only architecture, supporting English, and providing open weights, pre-trained variants, and instruction-tuned variants. The Gemma model is suitable for various text generation tasks, including question answering, summarization, and reasoning. This 9B model is trained on 80 trillion tokens. Its relatively small size allows it to be deployed in resource-constrained environments, such as laptops, desktops, or your own cloud infrastructure, making cutting-edge AI models more accessible and fostering innovation."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 is a family of multilingual large language models developed by Meta, including pre-trained and instruction-tuned variants with parameter sizes of 8B, 70B, and 405B. This 8B instruction-tuned model is optimized for multilingual dialogue scenarios and performs excellently in multiple industry benchmark tests. The model is trained using over 150 trillion tokens of public data and employs techniques such as supervised fine-tuning and human feedback reinforcement learning to enhance the model's usefulness and safety. Llama 3.1 supports text generation and code generation, with a knowledge cutoff date of December 2023."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview is an innovative natural language processing model capable of efficiently handling complex dialogue generation and context understanding tasks."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview is a research-oriented model developed by the Qwen team, focusing on visual reasoning capabilities, with unique advantages in understanding complex scenes and solving visually related mathematical problems."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview is Qwen's latest experimental research model, focusing on enhancing AI reasoning capabilities. By exploring complex mechanisms such as language mixing and recursive reasoning, its main advantages include strong analytical reasoning, mathematical, and programming abilities. However, it also faces challenges such as language switching issues, reasoning loops, safety considerations, and differences in other capabilities."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct is the latest version in Alibaba Cloud's series of code-specific large language models. This model significantly enhances code generation, reasoning, and repair capabilities based on Qwen2.5, trained on 55 trillion tokens. It not only improves coding abilities but also maintains advantages in mathematics and general capabilities, providing a more comprehensive foundation for practical applications such as code agents."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math focuses on problem-solving in the field of mathematics, providing expert solutions for challenging problems."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 is the latest series of the Qwen model, supporting 128k context. Compared to the current best open-source models, Qwen2-72B significantly surpasses leading models in natural language understanding, knowledge, coding, mathematics, and multilingual capabilities."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 is the latest series of the Qwen model, capable of outperforming optimal open-source models of similar size and even larger models. Qwen2 7B has achieved significant advantages in multiple evaluations, especially in coding and Chinese comprehension."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B is a powerful visual language model that supports multimodal processing of images and text, capable of accurately recognizing image content and generating relevant descriptions or answers."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct is a large language model with 14 billion parameters, delivering excellent performance, optimized for Chinese and multilingual scenarios, and supporting applications such as intelligent Q&A and content generation."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct is a large language model with 7 billion parameters, supporting function calls and seamless interaction with external systems, greatly enhancing flexibility and scalability. It is optimized for Chinese and multilingual scenarios, supporting applications such as intelligent Q&A and content generation."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct is a large-scale pre-trained programming instruction model with strong code understanding and generation capabilities, efficiently handling various programming tasks, particularly suited for intelligent code writing, automated script generation, and programming problem-solving."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct is a large language model specifically designed for code generation, code understanding, and efficient development scenarios, featuring an industry-leading 32 billion parameters to meet diverse programming needs."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "The TeleMM multimodal large model is a multimodal understanding model independently developed by China Telecom, capable of processing various modal inputs such as text and images, supporting functions like image understanding and chart analysis, providing users with cross-modal understanding services. The model can interact with users in a multimodal manner, accurately understand input content, answer questions, assist in creation, and efficiently provide multimodal information and inspiration support. It excels in fine-grained perception, logical reasoning, and other multimodal tasks."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large is the industry's largest open-source Transformer architecture MoE model, with a total of 389 billion parameters and 52 billion active parameters."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct is an instruction-tuned large language model in the Qwen2 series, with a parameter size of 72B. This model is based on the Transformer architecture and employs techniques such as the SwiGLU activation function, attention QKV bias, and group query attention. It can handle large-scale inputs. The model excels in language understanding, generation, multilingual capabilities, coding, mathematics, and reasoning across multiple benchmark tests, surpassing most open-source models and demonstrating competitive performance comparable to proprietary models in certain tasks."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct is one of the latest large language models released by Alibaba Cloud. This 72B model shows significant improvements in coding and mathematics. It also provides multilingual support, covering over 29 languages, including Chinese and English. The model has made notable advancements in instruction following, understanding structured data, and generating structured outputs, especially JSON."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Optimized for Chinese persona dialogue scenarios, providing smooth dialogue generation that aligns with Chinese expression habits."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Fireworks open-source function-calling model provides excellent instruction execution capabilities and customizable features."
+ "abab7-chat-preview": {
+ "description": "Significant improvements in capabilities such as long text, mathematics, and writing compared to the abab6.5 series models."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Fireworks' latest Firefunction-v2 is a high-performance function-calling model developed based on Llama-3, optimized for function calls, dialogues, and instruction following."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 is a state-of-the-art large language model optimized through reinforcement learning and cold-start data, excelling in reasoning, mathematics, and programming performance."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b is a visual language model that can accept both image and text inputs, trained on high-quality data, suitable for multimodal tasks."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "A powerful Mixture-of-Experts (MoE) language model provided by Deepseek, with a total parameter count of 671B, activating 37B parameters per token."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Llama 3 70B instruction model, optimized for multilingual dialogues and natural language understanding, outperforming most competitive models."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Llama 3 70B instruction model (HF version), aligned with official implementation results, suitable for high-quality instruction following tasks."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Llama 3 8B instruction model, optimized for dialogues and multilingual tasks, delivering outstanding and efficient performance."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Meta's 11B parameter instruction-tuned image reasoning model. This model is optimized for visual recognition, image reasoning, image description, and answering general questions about images. It understands visual data like charts and graphs, generating text descriptions of image details to bridge the gap between vision and language."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "The Llama 3.2 1B instruction model is a lightweight multilingual model introduced by Meta. This model aims to enhance efficiency, providing significant improvements in latency and cost compared to larger models. Sample use cases include retrieval and summarization."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "The Llama 3.2 3B instruction model is a lightweight multilingual model introduced by Meta. This model aims to enhance efficiency, providing significant improvements in latency and cost compared to larger models. Sample use cases include querying, prompt rewriting, and writing assistance."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Meta's 90B parameter instruction-tuned image reasoning model. This model is optimized for visual recognition, image reasoning, image description, and answering general questions about images. It understands visual data like charts and graphs, generating text descriptions of image details to bridge the gap between vision and language."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct is the December update of Llama 3.1 70B. This model builds upon Llama 3.1 70B (released in July 2024) with enhancements in tool invocation, multilingual text support, mathematics, and programming capabilities. It achieves industry-leading performance in reasoning, mathematics, and instruction following, providing similar performance to 3.1 405B while offering significant advantages in speed and cost."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "A 24B parameter model that possesses state-of-the-art capabilities comparable to larger models."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Mixtral MoE 8x22B instruction model, featuring large-scale parameters and a multi-expert architecture, fully supporting efficient processing of complex tasks."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Mixtral MoE 8x7B instruction model, with a multi-expert architecture providing efficient instruction following and execution."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Mixtral MoE 8x7B instruction model (HF version), performance consistent with official implementation, suitable for various efficient task scenarios."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "MythoMax L2 13B model, combining novel merging techniques, excels in narrative and role-playing."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "The QwQ model is an experimental research model developed by the Qwen team, focusing on enhancing AI reasoning capabilities."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "The 72B version of the Qwen-VL model is the latest iteration from Alibaba, representing nearly a year of innovation."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 is a series of decoder-only language models developed by the Alibaba Cloud Qwen team. These models come in different sizes including 0.5B, 1.5B, 3B, 7B, 14B, 32B, and 72B, available in both base and instruct variants."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct is the latest version in Alibaba Cloud's series of code-specific large language models. This model significantly enhances code generation, reasoning, and repair capabilities based on Qwen2.5, trained on 55 trillion tokens. It not only improves coding abilities but also maintains advantages in mathematics and general capabilities, providing a more comprehensive foundation for practical applications such as code agents."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "StarCoder 15.5B model supports advanced programming tasks, enhanced multilingual capabilities, suitable for complex code generation and understanding."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "StarCoder 7B model, trained on over 80 programming languages, boasts excellent code completion capabilities and contextual understanding."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Yi-Large model, featuring exceptional multilingual processing capabilities, suitable for various language generation and understanding tasks."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus is Anthropic's most powerful model for handling highly complex tasks. It excels in performance, intelligence, fluency, and comprehension."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku is Anthropic's fastest next-generation model. Compared to Claude 3 Haiku, Claude 3.5 Haiku shows improvements across various skills and surpasses the previous generation's largest model, Claude 3 Opus, in many intelligence benchmarks."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet offers capabilities that surpass Opus and faster speeds than Sonnet, while maintaining the same pricing as Sonnet. Sonnet excels particularly in programming, data science, visual processing, and agent tasks."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet provides an ideal balance of intelligence and speed for enterprise workloads. It offers maximum utility at a lower price, reliable and suitable for large-scale deployment."
},
- "code-raccoon-v1": {
- "description": "Code Raccoon is a software intelligent development assistant based on the SenseTime large language model, covering software requirement analysis, architecture design, code writing, software testing, and more, meeting various user needs for code writing and programming learning. Code Raccoon supports over 90 mainstream programming languages, including Python, Java, JavaScript, C++, Go, SQL, and popular IDEs like VS Code and IntelliJ IDEA. In practical applications, Code Raccoon can help developers improve programming efficiency by over 50%."
- },
"codegeex-4": {
"description": "CodeGeeX-4 is a powerful AI programming assistant that supports intelligent Q&A and code completion in various programming languages, enhancing development efficiency."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ is a high-performance large language model designed for real enterprise scenarios and complex applications."
},
+ "dall-e-2": {
+ "description": "The second generation DALL·E model, supporting more realistic and accurate image generation, with a resolution four times that of the first generation."
+ },
+ "dall-e-3": {
+ "description": "The latest DALL·E model, released in November 2023. It supports more realistic and accurate image generation with enhanced detail representation."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct provides highly reliable instruction processing capabilities, supporting applications across multiple industries."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 is a powerful and cost-effective mixture of experts (MoE) language model. It has been pre-trained on a high-quality corpus of 81 trillion tokens and further enhanced through supervised fine-tuning (SFT) and reinforcement learning (RL). Compared to DeepSeek 67B, DeepSeek-V2 offers stronger performance while saving 42.5% in training costs, reducing KV cache by 93.3%, and increasing maximum generation throughput by 5.76 times. The model supports a context length of 128k and performs excellently in standard benchmark tests and open-ended generation evaluations."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 is a reinforcement learning (RL) driven inference model that addresses issues of repetitiveness and readability within the model. Prior to RL, DeepSeek-R1 introduced cold start data to further optimize inference performance. It performs comparably to OpenAI-o1 in mathematical, coding, and reasoning tasks, and enhances overall effectiveness through meticulously designed training methods."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "The DeepSeek-R1 distillation model optimizes inference performance through reinforcement learning and cold-start data, refreshing the benchmark for open-source models across multiple tasks."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B is a distillation model developed based on Llama-3.1-8B. This model is fine-tuned using samples generated by DeepSeek-R1, demonstrating excellent reasoning capabilities. It has performed well in multiple benchmark tests, achieving an 89.1% accuracy rate on MATH-500, a 50.4% pass rate on AIME 2024, and a score of 1205 on CodeForces, showcasing strong mathematical and programming abilities as an 8B scale model."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "The DeepSeek-R1 distillation model optimizes inference performance through reinforcement learning and cold-start data, refreshing the benchmark for open-source models across multiple tasks."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "The DeepSeek-R1 distillation model optimizes inference performance through reinforcement learning and cold-start data, refreshing the benchmark for open-source models across multiple tasks."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B is a model obtained through knowledge distillation based on Qwen2.5-32B. This model is fine-tuned using 800,000 selected samples generated by DeepSeek-R1, demonstrating exceptional performance in mathematics, programming, and reasoning across multiple domains. It has achieved excellent results in various benchmark tests, including a 94.3% accuracy rate on MATH-500, showcasing strong mathematical reasoning capabilities."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B is a model obtained through knowledge distillation based on Qwen2.5-Math-7B. This model is fine-tuned using 800,000 selected samples generated by DeepSeek-R1, demonstrating excellent reasoning capabilities. It has performed outstandingly in multiple benchmark tests, achieving a 92.8% accuracy rate on MATH-500, a 55.5% pass rate on AIME 2024, and a score of 1189 on CodeForces, showcasing strong mathematical and programming abilities as a 7B scale model."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 combines the excellent features of previous versions, enhancing general and coding capabilities."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 is a mixture of experts (MoE) language model with 671 billion parameters, utilizing multi-head latent attention (MLA) and the DeepSeekMoE architecture, combined with a load balancing strategy that does not rely on auxiliary loss, optimizing inference and training efficiency. Pre-trained on 14.8 trillion high-quality tokens and fine-tuned with supervision and reinforcement learning, DeepSeek-V3 outperforms other open-source models and approaches leading closed-source models in performance."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B is an advanced model trained for highly complex conversations."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "A state-of-the-art efficient LLM skilled in reasoning, mathematics, and programming."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 is a mixture of experts (MoE) visual language model developed based on DeepSeekMoE-27B, employing a sparsely activated MoE architecture that achieves outstanding performance while activating only 4.5 billion parameters. This model excels in various tasks, including visual question answering, optical character recognition, document/table/chart understanding, and visual localization."
+ },
"deepseek-chat": {
"description": "A new open-source model that integrates general and coding capabilities, retaining the general conversational abilities of the original Chat model and the powerful code handling capabilities of the Coder model, while better aligning with human preferences. Additionally, DeepSeek-V2.5 has achieved significant improvements in writing tasks, instruction following, and more."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 is an open-source hybrid expert code model that performs excellently in coding tasks, comparable to GPT4-Turbo."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 is a reinforcement learning (RL) driven inference model that addresses issues of repetitiveness and readability within the model. Prior to RL, DeepSeek-R1 introduced cold start data to further optimize inference performance. It performs comparably to OpenAI-o1 in mathematical, coding, and reasoning tasks, and enhances overall effectiveness through meticulously designed training methods."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1—the larger and smarter model in the DeepSeek suite—has been distilled into the Llama 70B architecture. Based on benchmark tests and human evaluations, this model is smarter than the original Llama 70B, especially excelling in tasks requiring mathematical and factual accuracy."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "The DeepSeek-R1-Distill series models are fine-tuned versions of samples generated by DeepSeek-R1, using knowledge distillation techniques on open-source models like Qwen and Llama."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "The DeepSeek-R1-Distill series models are fine-tuned versions of samples generated by DeepSeek-R1, using knowledge distillation techniques on open-source models like Qwen and Llama."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "The DeepSeek-R1-Distill series models are fine-tuned versions of samples generated by DeepSeek-R1, using knowledge distillation techniques on open-source models like Qwen and Llama."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "The DeepSeek-R1-Distill series models are fine-tuned versions of samples generated by DeepSeek-R1, using knowledge distillation techniques on open-source models like Qwen and Llama."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "The DeepSeek-R1-Distill series models are fine-tuned versions of samples generated by DeepSeek-R1, using knowledge distillation techniques on open-source models like Qwen and Llama."
+ },
+ "deepseek-reasoner": {
+ "description": "The reasoning model launched by DeepSeek. Before outputting the final answer, the model first provides a chain of thought to enhance the accuracy of the final response."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 is an efficient Mixture-of-Experts language model, suitable for cost-effective processing needs."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B is the design code model of DeepSeek, providing powerful code generation capabilities."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 is a MoE model developed by Hangzhou DeepSeek Artificial Intelligence Technology Research Co., Ltd., achieving outstanding results in multiple evaluations and ranking first among open-source models on mainstream leaderboards. Compared to the V2.5 model, V3 has achieved a threefold increase in generation speed, providing users with a faster and smoother experience."
+ },
"deepseek/deepseek-chat": {
"description": "A new open-source model that integrates general and coding capabilities, retaining the general conversational abilities of the original Chat model and the powerful code handling capabilities of the Coder model, while better aligning with human preferences. Additionally, DeepSeek-V2.5 has achieved significant improvements in writing tasks, instruction following, and more."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 significantly enhances model reasoning capabilities with minimal labeled data. Before outputting the final answer, the model first provides a chain of thought to improve the accuracy of the final response."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 significantly enhances model reasoning capabilities with minimal labeled data. Before outputting the final answer, the model first provides a chain of thought to improve the accuracy of the final response."
+ },
"emohaa": {
"description": "Emohaa is a psychological model with professional counseling capabilities, helping users understand emotional issues."
},
+ "ernie-3.5-128k": {
+ "description": "Baidu's flagship large-scale language model, covering a vast amount of Chinese and English corpus, possesses strong general capabilities to meet the requirements of most dialogue Q&A, creative generation, and plugin application scenarios; it supports automatic integration with Baidu search plugins to ensure the timeliness of Q&A information."
+ },
+ "ernie-3.5-8k": {
+ "description": "Baidu's flagship large-scale language model, covering a vast amount of Chinese and English corpus, possesses strong general capabilities to meet the requirements of most dialogue Q&A, creative generation, and plugin application scenarios; it supports automatic integration with Baidu search plugins to ensure the timeliness of Q&A information."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "Baidu's flagship large-scale language model, covering a vast amount of Chinese and English corpus, possesses strong general capabilities to meet the requirements of most dialogue Q&A, creative generation, and plugin application scenarios; it supports automatic integration with Baidu search plugins to ensure the timeliness of Q&A information."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "Baidu's flagship ultra-large-scale language model, which has achieved a comprehensive upgrade in model capabilities compared to ERNIE 3.5, widely applicable to complex task scenarios across various fields; it supports automatic integration with Baidu search plugins to ensure the timeliness of Q&A information."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "Baidu's flagship ultra-large-scale language model, which has achieved a comprehensive upgrade in model capabilities compared to ERNIE 3.5, widely applicable to complex task scenarios across various fields; it supports automatic integration with Baidu search plugins to ensure the timeliness of Q&A information."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "Baidu's flagship ultra-large-scale language model, demonstrating outstanding overall performance, widely applicable to complex task scenarios across various fields; it supports automatic integration with Baidu search plugins to ensure the timeliness of Q&A information. It performs better than ERNIE 4.0 in terms of performance."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "Baidu's flagship ultra-large-scale language model, demonstrating outstanding overall performance, widely applicable to complex task scenarios across various fields; it supports automatic integration with Baidu search plugins to ensure the timeliness of Q&A information. It performs better than ERNIE 4.0 in terms of performance."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "Baidu's flagship ultra-large-scale language model, demonstrating outstanding overall performance, widely applicable to complex task scenarios across various fields; it supports automatic integration with Baidu search plugins to ensure the timeliness of Q&A information. It performs better than ERNIE 4.0 in terms of performance."
+ },
+ "ernie-char-8k": {
+ "description": "Baidu's vertical scene large language model, suitable for applications such as game NPCs, customer service dialogues, and role-playing conversations, with a more distinct and consistent character style, stronger instruction-following capabilities, and superior inference performance."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Baidu's vertical scene large language model, suitable for applications such as game NPCs, customer service dialogues, and role-playing conversations, with a more distinct and consistent character style, stronger instruction-following capabilities, and superior inference performance."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite is Baidu's lightweight large language model, balancing excellent model performance with inference efficiency, suitable for low-power AI acceleration card inference."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Baidu's lightweight large language model, balancing excellent model performance with inference efficiency, offering better performance than ERNIE Lite, suitable for low-power AI acceleration card inference."
+ },
+ "ernie-novel-8k": {
+ "description": "Baidu's general-purpose large language model, which has a significant advantage in novel continuation capabilities and can also be used in short plays, movies, and other scenarios."
+ },
+ "ernie-speed-128k": {
+ "description": "Baidu's latest self-developed high-performance large language model released in 2024, with excellent general capabilities, suitable as a base model for fine-tuning to better address specific scenario issues while also demonstrating excellent inference performance."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "Baidu's latest self-developed high-performance large language model released in 2024, with excellent general capabilities, offering better performance than ERNIE Speed, suitable as a base model for fine-tuning to better address specific scenario issues while also demonstrating excellent inference performance."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny is Baidu's ultra-high-performance large language model, with the lowest deployment and fine-tuning costs among the Wenxin series models."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Tuning) offers stable and tunable performance, making it an ideal choice for complex task solutions."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro supports up to 2 million tokens, making it an ideal choice for medium-sized multimodal models, providing multifaceted support for complex tasks."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash offers next-generation features and improvements, including exceptional speed, native tool usage, multimodal generation, and a 1M token context window."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash offers next-generation features and improvements, including exceptional speed, native tool usage, multimodal generation, and a 1M token context window."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp is Google's latest experimental multimodal AI model, featuring next-generation capabilities, exceptional speed, native tool integration, and multimodal generation."
},
- "gemini-2.0-flash-thinking-exp-1219": {
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "A Gemini 2.0 Flash model optimized for cost-effectiveness and low latency."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
"description": "Gemini 2.0 Flash Exp is Google's latest experimental multimodal AI model, featuring next-generation capabilities, exceptional speed, native tool invocation, and multimodal generation."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 is Google's latest experimental multimodal AI model, featuring rapid processing capabilities and supporting text, image, and video inputs, making it suitable for efficient scaling across various tasks."
+ "gemini-2.0-flash-thinking-exp-1219": {
+ "description": "Gemini 2.0 Flash Exp is Google's latest experimental multimodal AI model, featuring next-generation capabilities, exceptional speed, native tool invocation, and multimodal generation."
},
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 is Google's latest experimental multimodal AI model, featuring rapid processing capabilities and supporting text, image, and video inputs, making it efficient for a variety of tasks."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental is Google's latest experimental multimodal AI model, showing a quality improvement compared to previous versions, especially in world knowledge, coding, and long context."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 is Google's latest experimental multimodal AI model, offering significant quality improvements compared to previous versions."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus has the ability to understand video content and multiple images, suitable for multimodal tasks."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview possesses strong complex reasoning abilities, excelling in logical reasoning, mathematics, programming, and other fields."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash offers next-generation features and improvements, including exceptional speed, native tool usage, multimodal generation, and a 1M token context window."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental is Google's latest experimental multimodal AI model, showing a quality improvement compared to previous versions, especially in world knowledge, coding, and long context."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash offers optimized multimodal processing capabilities, suitable for various complex task scenarios."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro combines the latest optimization technologies to deliver more efficient multimodal data processing capabilities."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 is an efficient model launched by Google, covering a variety of application scenarios from small applications to complex data processing."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 continues the design philosophy of being lightweight and efficient."
},
"google/gemma-2-2b-it": {
"description": "Google's lightweight instruction-tuning model."
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 is an efficient model launched by Google, covering a variety of application scenarios from small applications to complex data processing."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 is Google's lightweight open-source text model series."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo is suitable for various text generation and understanding tasks. Currently points to gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, a high-capacity text generation model suitable for complex tasks."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo is suitable for various text generation and understanding tasks. Currently points to gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o is a dynamic model that updates in real-time to maintain the latest version. It combines powerful language understanding and generation capabilities, making it suitable for large-scale applications including customer service, education, and technical support."
},
+ "gpt-4o-audio-preview": {
+ "description": "GPT-4o Audio model, supporting audio input and output."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini is the latest model released by OpenAI after GPT-4 Omni, supporting both image and text input while outputting text. As their most advanced small model, it is significantly cheaper than other recent cutting-edge models, costing over 60% less than GPT-3.5 Turbo. It maintains state-of-the-art intelligence while offering remarkable cost-effectiveness. GPT-4o mini scored 82% on the MMLU test and currently ranks higher than GPT-4 in chat preferences."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "GPT-4o-mini real-time version, supporting real-time audio and text input and output."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "GPT-4o real-time version, supporting real-time audio and text input and output."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "GPT-4o real-time version, supporting real-time audio and text input and output."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "GPT-4o real-time version, supporting real-time audio and text input and output."
+ },
"grok-2-1212": {
"description": "This model has improved in accuracy, instruction adherence, and multilingual capabilities."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "The latest MOE architecture FunctionCall model from Hunyuan, trained on high-quality FunctionCall data, with a context window of 32K, leading in multiple dimensions of evaluation metrics."
},
+ "hunyuan-large": {
+ "description": "The Hunyuan-large model has a total parameter count of approximately 389B, with about 52B active parameters, making it the largest and most effective open-source MoE model in the industry based on the Transformer architecture."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Specializes in handling long text tasks such as document summarization and question answering, while also capable of general text generation tasks. It excels in analyzing and generating long texts, effectively addressing complex and detailed long-form content processing needs."
+ },
"hunyuan-lite": {
"description": "Upgraded to a MOE structure with a context window of 256k, leading many open-source models in various NLP, coding, mathematics, and industry benchmarks."
},
+ "hunyuan-lite-vision": {
+ "description": "The latest 7B multimodal model from Hunyuan, with a context window of 32K, supports multimodal dialogue in both Chinese and English scenarios, image object recognition, document table understanding, and multimodal mathematics, outperforming 7B competing models across multiple evaluation dimensions."
+ },
"hunyuan-pro": {
"description": "A trillion-parameter scale MOE-32K long text model. Achieves absolute leading levels across various benchmarks, capable of handling complex instructions and reasoning, with advanced mathematical abilities, supporting function calls, and optimized for applications in multilingual translation, finance, law, and healthcare."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Utilizes a superior routing strategy while alleviating issues of load balancing and expert convergence. For long texts, the needle-in-a-haystack metric reaches 99.9%. MOE-256K further breaks through in length and effectiveness, greatly expanding the input length capacity."
},
+ "hunyuan-standard-vision": {
+ "description": "The latest multimodal model from Hunyuan, supporting multilingual responses with balanced capabilities in both Chinese and English."
+ },
"hunyuan-turbo": {
"description": "The preview version of the next-generation Hunyuan large language model, featuring a brand-new mixed expert model (MoE) structure, which offers faster inference efficiency and stronger performance compared to Hunyuan Pro."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Hunyuan-turbo fixed version as of November 20, 2024, a version that lies between hunyuan-turbo and hunyuan-turbo-latest."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "This version optimizes: data instruction scaling, significantly enhancing the model's generalization capabilities; greatly improving mathematical, coding, and logical reasoning abilities; optimizing text understanding and word comprehension capabilities; enhancing the quality of content generation in text creation."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "General experience optimization, including NLP understanding, text creation, casual conversation, knowledge Q&A, translation, and domain-specific tasks; enhanced personification and emotional intelligence of the model; improved the model's ability to clarify when intentions are ambiguous; enhanced handling of word parsing-related questions; improved the quality and interactivity of creative outputs; enhanced multi-turn experience."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "The next-generation flagship visual language model from Hunyuan, utilizing a new mixed expert model (MoE) structure, with comprehensive improvements in basic recognition, content creation, knowledge Q&A, and analytical reasoning capabilities compared to the previous generation model."
+ },
"hunyuan-vision": {
"description": "The latest multimodal model from Hunyuan, supporting image + text input to generate textual content."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "Our latest model series, featuring exceptional reasoning performance, supporting a context length of 1M, and enhanced instruction following and tool invocation capabilities."
},
+ "internlm3-latest": {
+ "description": "Our latest model series boasts exceptional inference performance, leading the pack among open-source models of similar scale. It defaults to our most recently released InternLM3 series models."
+ },
+ "jina-deepsearch-v1": {
+ "description": "DeepSearch combines web search, reading, and reasoning for comprehensive investigations. You can think of it as an agent that takes on your research tasks—it conducts extensive searches and iterates multiple times before providing answers. This process involves ongoing research, reasoning, and problem-solving from various angles. This fundamentally differs from standard large models that generate answers directly from pre-trained data and traditional RAG systems that rely on one-time surface searches."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM is an experimental, task-specific language model trained to align with learning science principles, capable of following systematic instructions in teaching and learning scenarios, acting as an expert tutor, among other roles."
},
"lite": {
"description": "Spark Lite is a lightweight large language model with extremely low latency and efficient processing capabilities, completely free and open, supporting real-time online search functionality. Its quick response feature makes it excel in inference applications and model fine-tuning on low-power devices, providing users with excellent cost-effectiveness and intelligent experiences, particularly in knowledge Q&A, content generation, and search scenarios."
},
- "llama-3.1-70b-instruct": {
- "description": "Llama 3.1 70B Instruct model, featuring 70B parameters, delivers outstanding performance in large text generation and instruction tasks."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B provides enhanced AI reasoning capabilities, suitable for complex applications, supporting extensive computational processing while ensuring efficiency and accuracy."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B is a high-performance model that offers rapid text generation capabilities, making it ideal for applications requiring large-scale efficiency and cost-effectiveness."
},
- "llama-3.1-8b-instruct": {
- "description": "Llama 3.1 8B Instruct model, featuring 8B parameters, supports efficient execution of visual instruction tasks, providing high-quality text generation capabilities."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Llama 3.1 Sonar Huge Online model, featuring 405B parameters, supports a context length of approximately 127,000 tokens, designed for complex online chat applications."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Llama 3.1 Sonar Large Chat model, featuring 70B parameters, supports a context length of approximately 127,000 tokens, suitable for complex offline chat tasks."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Llama 3.1 Sonar Large Online model, featuring 70B parameters, supports a context length of approximately 127,000 tokens, suitable for high-capacity and diverse chat tasks."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Llama 3.1 Sonar Small Chat model, featuring 8B parameters, designed for offline chat, supports a context length of approximately 127,000 tokens."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Llama 3.1 Sonar Small Online model, featuring 8B parameters, supports a context length of approximately 127,000 tokens, designed for online chat, efficiently handling various text interactions."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 is designed to handle tasks that combine visual and textual data. It excels in tasks such as image description and visual question answering, bridging the gap between language generation and visual reasoning."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 is the most advanced multilingual open-source large language model in the Llama series, offering performance comparable to a 405B model at an extremely low cost. Based on the Transformer architecture, it enhances usability and safety through supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). Its instruction-tuned version is optimized for multilingual dialogue and outperforms many open-source and closed chat models on various industry benchmarks. Knowledge cutoff date is December 2023."
+ },
"llama-3.3-70b-versatile": {
"description": "Meta Llama 3.3 is a multilingual large language model (LLM) with 70 billion parameters (text input/text output), featuring pre-training and instruction-tuning. The instruction-tuned pure text model of Llama 3.3 is optimized for multilingual conversational use cases and outperforms many available open-source and closed chat models on common industry benchmarks."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 is designed for tasks involving both visual and textual data. It excels in tasks like image description and visual question answering, bridging the gap between language generation and visual reasoning."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 is the most advanced multilingual open-source large language model in the Llama series, offering performance comparable to 405B models at a very low cost. Based on the Transformer architecture, it enhances usability and safety through supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). Its instruction-tuned version is optimized for multilingual dialogue and outperforms many open-source and closed chat models on multiple industry benchmarks. Knowledge cutoff date is December 2023."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Meta Llama 3.3 is a multilingual large language model (LLM) that is a pre-trained and instruction-tuned generative model within the 70B (text input/text output) framework. The instruction-tuned pure text model is optimized for multilingual dialogue use cases and outperforms many available open-source and closed chat models on common industry benchmarks."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 is designed for tasks involving both visual and textual data. It excels in tasks like image description and visual question answering, bridging the gap between language generation and visual reasoning."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "The 405B Llama 3.1 Turbo model provides massive context support for big data processing, excelling in large-scale AI applications."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 is a leading model launched by Meta, supporting up to 405B parameters, applicable in complex conversations, multilingual translation, and data analysis."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B offers efficient conversational support in multiple languages."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 is designed to handle tasks that combine visual and textual data. It excels in tasks such as image description and visual question answering, bridging the gap between language generation and visual reasoning."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 is the most advanced multilingual open-source large language model in the Llama series, offering performance comparable to a 405B model at an extremely low cost. Based on the Transformer architecture, it enhances usability and safety through supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). Its instruction-tuned version is optimized for multilingual dialogue and outperforms many open-source and closed chat models on various industry benchmarks. Knowledge cutoff date is December 2023."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 is the most advanced multilingual open-source large language model in the Llama series, offering performance comparable to a 405B model at an extremely low cost. Based on the Transformer architecture, it enhances usability and safety through supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). Its instruction-tuned version is optimized for multilingual dialogue and outperforms many open-source and closed chat models on various industry benchmarks. Knowledge cutoff date is December 2023."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct is the largest and most powerful model in the Llama 3.1 Instruct series. It is a highly advanced conversational reasoning and synthetic data generation model, which can also serve as a foundation for specialized continuous pre-training or fine-tuning in specific domains. The multilingual large language models (LLMs) provided by Llama 3.1 are a set of pre-trained, instruction-tuned generative models, including sizes of 8B, 70B, and 405B (text input/output). The instruction-tuned text models (8B, 70B, 405B) are optimized for multilingual conversational use cases and have outperformed many available open-source chat models in common industry benchmarks. Llama 3.1 is designed for commercial and research purposes across multiple languages. The instruction-tuned text models are suitable for assistant-like chat, while the pre-trained models can adapt to various natural language generation tasks. The Llama 3.1 models also support improving other models using their outputs, including synthetic data generation and refinement. Llama 3.1 is an autoregressive language model built using an optimized transformer architecture. The tuned versions utilize supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 is an open large language model (LLM) aimed at developers, researchers, and enterprises, designed to help them build, experiment, and responsibly scale their generative AI ideas. As part of a foundational system for global community innovation, it is particularly suitable for those with limited computational power and resources, edge devices, and faster training times."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "An advanced LLM supporting synthetic data generation, knowledge distillation, and reasoning, suitable for chatbots, programming, and domain-specific tasks."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "Empowering complex conversations with exceptional context understanding, reasoning capabilities, and text generation abilities."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "An advanced cutting-edge model with language understanding, excellent reasoning capabilities, and text generation abilities."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "A state-of-the-art vision-language model adept at high-quality reasoning from images."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "A cutting-edge small language model with language understanding, excellent reasoning capabilities, and text generation abilities."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "A cutting-edge small language model with language understanding, excellent reasoning capabilities, and text generation abilities."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "A state-of-the-art vision-language model adept at high-quality reasoning from images."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "An advanced LLM skilled in reasoning, mathematics, common sense, and function calling."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 is a language model provided by Microsoft AI, excelling in complex dialogues, multilingual capabilities, reasoning, and intelligent assistant tasks."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K is a model with ultra-long context processing capabilities, suitable for generating extremely long texts, meeting the demands of complex generation tasks, capable of handling up to 128,000 tokens, making it ideal for research, academia, and large document generation."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "The Kimi visual model (including moonshot-v1-8k-vision-preview, moonshot-v1-32k-vision-preview, moonshot-v1-128k-vision-preview, etc.) can understand image content, including text in images, colors, and shapes of objects."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K offers medium-length context processing capabilities, able to handle 32,768 tokens, particularly suitable for generating various long documents and complex dialogues, applicable in content creation, report generation, and dialogue systems."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "The Kimi visual model (including moonshot-v1-8k-vision-preview, moonshot-v1-32k-vision-preview, moonshot-v1-128k-vision-preview, etc.) can understand image content, including text in images, colors, and shapes of objects."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K is designed for generating short text tasks, featuring efficient processing performance, capable of handling 8,192 tokens, making it ideal for brief dialogues, note-taking, and rapid content generation."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "The Kimi visual model (including moonshot-v1-8k-vision-preview, moonshot-v1-32k-vision-preview, moonshot-v1-128k-vision-preview, etc.) can understand image content, including text in images, colors, and shapes of objects."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B is an upgraded version of Nous Hermes 2, featuring the latest internally developed datasets."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B is a large language model customized by NVIDIA, designed to enhance the help provided by LLM-generated responses to user queries."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B is a large language model customized by NVIDIA, designed to enhance the helpfulness of LLM-generated responses to user queries. The model has excelled in benchmark tests such as Arena Hard, AlpacaEval 2 LC, and GPT-4-Turbo MT-Bench, ranking first in all three automatic alignment benchmarks as of October 1, 2024. The model is trained using RLHF (specifically REINFORCE), Llama-3.1-Nemotron-70B-Reward, and HelpSteer2-Preference prompts based on the Llama-3.1-70B-Instruct model."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "A unique language model offering unparalleled accuracy and efficiency."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct is a custom large language model by NVIDIA designed to enhance the helpfulness of LLM-generated responses."
+ },
"o1": {
"description": "Focused on advanced reasoning and solving complex problems, including mathematical and scientific tasks. It is particularly well-suited for applications that require deep contextual understanding and agent workflow."
},
- "o1-2024-12-17": {
- "description": "O1 is OpenAI's new reasoning model that supports both image and text input and outputs text, suitable for complex tasks requiring extensive general knowledge. This model has a context length of 200K and a knowledge cutoff date of October 2023."
- },
"o1-mini": {
"description": "o1-mini is a fast and cost-effective reasoning model designed for programming, mathematics, and scientific applications. This model features a 128K context and has a knowledge cutoff date of October 2023."
},
"o1-preview": {
"description": "o1 is OpenAI's new reasoning model, suitable for complex tasks that require extensive general knowledge. This model features a 128K context and has a knowledge cutoff date of October 2023."
},
+ "o3-mini": {
+ "description": "o3-mini is our latest small inference model that delivers high intelligence while maintaining the same cost and latency targets as o1-mini."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba is a language model focused on code generation, providing strong support for advanced coding and reasoning tasks."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini is the latest model released by OpenAI following GPT-4 Omni, supporting both text and image input while outputting text. As their most advanced small model, it is significantly cheaper than other recent cutting-edge models and over 60% cheaper than GPT-3.5 Turbo. It maintains state-of-the-art intelligence while offering remarkable cost-effectiveness. GPT-4o mini scored 82% on the MMLU test and currently ranks higher than GPT-4 in chat preferences."
},
- "openai/o1": {
- "description": "O1 is OpenAI's new reasoning model that supports both image and text input and outputs text, suitable for complex tasks requiring extensive general knowledge. This model has a context length of 200K and a knowledge cutoff date of October 2023."
- },
"openai/o1-mini": {
"description": "o1-mini is a fast and cost-effective reasoning model designed for programming, mathematics, and scientific applications. This model features a 128K context and has a knowledge cutoff date of October 2023."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K is equipped with an extra-large context processing capability, able to handle up to 128K of contextual information, making it particularly suitable for long-form content that requires comprehensive analysis and long-term logical connections, providing smooth and consistent logic and diverse citation support in complex text communication."
},
+ "qvq-72b-preview": {
+ "description": "The QVQ model is an experimental research model developed by the Qwen team, focusing on enhancing visual reasoning capabilities, particularly in the field of mathematical reasoning."
+ },
"qwen-coder-plus-latest": {
"description": "Tongyi Qianwen code model."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Tongyi Qianwen's ultra-large-scale visual language model. Compared to the enhanced version, it further improves visual reasoning and instruction-following abilities, providing a higher level of visual perception and cognition."
},
+ "qwen-vl-ocr-latest": {
+ "description": "The Tongyi Qianwen OCR is a proprietary model for text extraction, focusing on the ability to extract text from images of documents, tables, exam papers, and handwritten text. It can recognize multiple languages, currently supporting: Chinese, English, French, Japanese, Korean, German, Russian, Italian, Vietnamese, and Arabic."
+ },
"qwen-vl-plus-latest": {
"description": "Tongyi Qianwen's large-scale visual language model enhanced version. Significantly improves detail recognition and text recognition capabilities, supporting ultra-high pixel resolution and images of any aspect ratio."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 is a brand new series of large language models with enhanced understanding and generation capabilities."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "An LLM focused on both Chinese and English, targeting language, programming, mathematics, reasoning, and more."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "An advanced LLM supporting code generation, reasoning, and debugging, covering mainstream programming languages."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "A powerful medium-sized code model supporting 32K context length, proficient in multilingual programming."
+ },
"qwen2": {
"description": "Qwen2 is Alibaba's next-generation large-scale language model, supporting diverse application needs with excellent performance."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "The 14B model of Tongyi Qianwen 2.5 is open-sourced."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "The Tongyi Qianwen 2.5 model is open-sourced at a scale of 72B."
+ },
"qwen2.5-32b-instruct": {
"description": "The 32B model of Tongyi Qianwen 2.5 is open-sourced."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "The Qwen-Math model possesses strong capabilities for solving mathematical problems."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "This version enhances instruction following, mathematics, problem-solving, and coding capabilities, improving the ability to recognize various formats and accurately locate visual elements. It supports understanding long video files (up to 10 minutes) and pinpointing events in seconds, comprehending the sequence and speed of time, and based on parsing and locating capabilities, it supports controlling OS or Mobile agents. It has strong key information extraction and JSON output capabilities, and this version is the most powerful in the series at 72B."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "This version enhances instruction following, mathematics, problem-solving, and coding capabilities, improving the ability to recognize various formats and accurately locate visual elements. It supports understanding long video files (up to 10 minutes) and pinpointing events in seconds, comprehending the sequence and speed of time, and based on parsing and locating capabilities, it supports controlling OS or Mobile agents. It has strong key information extraction and JSON output capabilities, and this version is the most powerful in the series at 72B."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 is Alibaba's next-generation large-scale language model, supporting diverse application needs with outstanding performance."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro is a highly intelligent LLM launched by Upstage, focusing on single-GPU instruction-following capabilities, with an IFEval score above 80. Currently supports English, with a formal version planned for release in November 2024, which will expand language support and context length."
},
+ "sonar": {
+ "description": "A lightweight search product based on contextual search, faster and cheaper than Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "An advanced search product that supports contextual search, advanced queries, and follow-ups."
+ },
+ "sonar-reasoning": {
+ "description": "A new API product powered by the DeepSeek reasoning model."
+ },
"step-1-128k": {
"description": "Balances performance and cost, suitable for general scenarios."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "This model has powerful video understanding capabilities."
},
+ "step-1o-vision-32k": {
+ "description": "This model possesses powerful image understanding capabilities. Compared to the step-1v series models, it offers enhanced visual performance."
+ },
"step-1v-32k": {
"description": "Supports visual input, enhancing multimodal interaction experiences."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Supports large-scale context interactions, suitable for complex dialogue scenarios."
},
+ "step-2-mini": {
+ "description": "A high-speed large model based on the next-generation self-developed Attention architecture MFA, achieving results similar to step-1 at a very low cost, while maintaining higher throughput and faster response times. It is capable of handling general tasks and has specialized skills in coding."
+ },
+ "taichu2_mm": {
+ "description": "Integrating capabilities in image understanding, knowledge transfer, and logical attribution, it excels in the field of image-text question answering."
+ },
"taichu_llm": {
"description": "The ZD Taichu language model possesses strong language understanding capabilities and excels in text creation, knowledge Q&A, code programming, mathematical calculations, logical reasoning, sentiment analysis, and text summarization. It innovatively combines large-scale pre-training with rich knowledge from multiple sources, continuously refining algorithmic techniques and absorbing new knowledge in vocabulary, structure, grammar, and semantics from vast text data, resulting in an evolving model performance. It provides users with more convenient information and services, as well as a more intelligent experience."
},
+ "text-embedding-3-large": {
+ "description": "The most powerful vectorization model, suitable for both English and non-English tasks."
+ },
+ "text-embedding-3-small": {
+ "description": "An efficient and cost-effective next-generation embedding model, suitable for knowledge retrieval, RAG applications, and more."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) provides enhanced computational capabilities through efficient strategies and model architecture."
},
+ "tts-1": {
+ "description": "The latest text-to-speech model, optimized for speed in real-time scenarios."
+ },
+ "tts-1-hd": {
+ "description": "The latest text-to-speech model, optimized for quality."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) is suitable for refined instruction tasks, offering excellent language processing capabilities."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet raises the industry standard, outperforming competitor models and Claude 3 Opus, excelling in a wide range of evaluations while maintaining the speed and cost of our mid-tier models."
},
+ "whisper-1": {
+ "description": "A universal speech recognition model that supports multilingual speech recognition, speech translation, and language identification."
+ },
"wizardlm2": {
"description": "WizardLM 2 is a language model provided by Microsoft AI, excelling in complex dialogues, multilingual capabilities, reasoning, and intelligent assistant applications."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Model for complex visual tasks, providing high-performance image understanding and analysis capabilities."
+ },
+ "yi-vision-v2": {
+ "description": "A complex visual task model that provides high-performance understanding and analysis capabilities based on multiple images."
}
}
diff --git a/locales/en-US/providers.json b/locales/en-US/providers.json
index 3825edf49aa48..9fd2631972b6c 100644
--- a/locales/en-US/providers.json
+++ b/locales/en-US/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure offers a variety of advanced AI models, including GPT-3.5 and the latest GPT-4 series, supporting various data types and complex tasks, dedicated to secure, reliable, and sustainable AI solutions."
},
+ "azureai": {
+ "description": "Azure offers a variety of advanced AI models, including GPT-3.5 and the latest GPT-4 series, supporting multiple data types and complex tasks, dedicated to secure, reliable, and sustainable AI solutions."
+ },
"baichuan": {
"description": "Baichuan Intelligence is a company focused on the research and development of large AI models, with its models excelling in domestic knowledge encyclopedias, long text processing, and generative creation tasks in Chinese, surpassing mainstream foreign models. Baichuan Intelligence also possesses industry-leading multimodal capabilities, performing excellently in multiple authoritative evaluations. Its models include Baichuan 4, Baichuan 3 Turbo, and Baichuan 3 Turbo 128k, each optimized for different application scenarios, providing cost-effective solutions."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek is a company focused on AI technology research and application, with its latest model DeepSeek-V2.5 integrating general dialogue and code processing capabilities, achieving significant improvements in human preference alignment, writing tasks, and instruction following."
},
+ "doubao": {
+ "description": "A self-developed large model launched by ByteDance. Verified through practical applications in over 50 internal business scenarios, it continuously refines its capabilities with a daily usage of trillions of tokens, providing various modal abilities to create a rich business experience for enterprises with high-quality model performance."
+ },
"fireworksai": {
"description": "Fireworks AI is a leading provider of advanced language model services, focusing on functional calling and multimodal processing. Its latest model, Firefunction V2, is based on Llama-3, optimized for function calling, conversation, and instruction following. The visual language model FireLLaVA-13B supports mixed input of images and text. Other notable models include the Llama series and Mixtral series, providing efficient multilingual instruction following and generation support."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "An open-source organization dedicated to the research and development of large model toolchains. It provides an efficient and user-friendly open-source platform for all AI developers, making cutting-edge large models and algorithm technologies easily accessible."
},
+ "jina": {
+ "description": "Founded in 2020, Jina AI is a leading search AI company. Our search base platform includes vector models, rerankers, and small language models to help businesses build reliable and high-quality generative AI and multimodal search applications."
+ },
+ "lmstudio": {
+ "description": "LM Studio is a desktop application for developing and experimenting with LLMs on your computer."
+ },
"minimax": {
"description": "MiniMax is a general artificial intelligence technology company established in 2021, dedicated to co-creating intelligence with users. MiniMax has independently developed general large models of different modalities, including trillion-parameter MoE text models, voice models, and image models, and has launched applications such as Conch AI."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI is a platform providing a variety of large language models and AI image generation API services, flexible, reliable, and cost-effective. It supports the latest open-source models like Llama3 and Mistral, offering a comprehensive, user-friendly, and auto-scaling API solution for generative AI application development, suitable for the rapid growth of AI startups."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ provides containers for self-hosted GPU-accelerated inference microservices, supporting the deployment of pre-trained and custom AI models in the cloud, data centers, RTX™ AI personal computers, and workstations."
+ },
"ollama": {
"description": "Ollama provides models that cover a wide range of fields, including code generation, mathematical operations, multilingual processing, and conversational interaction, catering to diverse enterprise-level and localized deployment needs."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "The Institute of Automation, Chinese Academy of Sciences, and Wuhan Artificial Intelligence Research Institute have launched a new generation of multimodal large models, supporting comprehensive question-answering tasks such as multi-turn Q&A, text creation, image generation, 3D understanding, and signal analysis, with stronger cognitive, understanding, and creative abilities, providing a new interactive experience."
},
+ "tencentcloud": {
+ "description": "The Knowledge Engine Atomic Power, based on the Knowledge Engine, provides a comprehensive knowledge Q&A capability for enterprises and developers. It offers the ability to flexibly assemble and develop model applications. You can create your own model services using various atomic capabilities, integrating services such as document parsing, splitting, embedding, and multi-turn rewriting to customize AI solutions tailored to your business."
+ },
"togetherai": {
"description": "Together AI is dedicated to achieving leading performance through innovative AI models, offering extensive customization capabilities, including rapid scaling support and intuitive deployment processes to meet various enterprise needs."
},
"upstage": {
"description": "Upstage focuses on developing AI models for various business needs, including Solar LLM and document AI, aiming to achieve artificial general intelligence (AGI) for work. It allows for the creation of simple conversational agents through Chat API and supports functional calling, translation, embedding, and domain-specific applications."
},
+ "vllm": {
+ "description": "vLLM is a fast and easy-to-use library for LLM inference and serving."
+ },
+ "volcengine": {
+ "description": "A development platform for large model services launched by ByteDance, offering feature-rich, secure, and competitively priced model invocation services. It also provides end-to-end functionalities such as model data, fine-tuning, inference, and evaluation, ensuring comprehensive support for the development and implementation of your AI applications."
+ },
"wenxin": {
"description": "An enterprise-level one-stop platform for large model and AI-native application development and services, providing the most comprehensive and user-friendly toolchain for the entire process of generative artificial intelligence model development and application development."
},
diff --git a/locales/en-US/setting.json b/locales/en-US/setting.json
index 98540de3adee7..77930582729e5 100644
--- a/locales/en-US/setting.json
+++ b/locales/en-US/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Enable Max Tokens Limit"
},
+ "enableReasoningEffort": {
+ "title": "Enable Reasoning Effort Adjustment"
+ },
"frequencyPenalty": {
- "desc": "The higher the value, the more likely it is to reduce repeated words",
- "title": "Frequency Penalty"
+ "desc": "The higher the value, the more diverse and rich the vocabulary; the lower the value, the simpler and more straightforward the language.",
+ "title": "Vocabulary Richness"
},
"maxTokens": {
"desc": "The maximum number of tokens used for each interaction",
@@ -212,19 +215,31 @@
"desc": "{{provider}} model",
"title": "Model"
},
+ "params": {
+ "title": "Advanced Parameters"
+ },
"presencePenalty": {
- "desc": "The higher the value, the more likely it is to expand to new topics",
- "title": "Topic Freshness"
+ "desc": "The higher the value, the more inclined to use different expressions and avoid concept repetition; the lower the value, the more inclined to use repeated concepts or narratives, resulting in more consistent expression.",
+ "title": "Expression Divergence"
+ },
+ "reasoningEffort": {
+ "desc": "The higher the value, the stronger the reasoning ability, but it may increase response time and token consumption.",
+ "options": {
+ "high": "High",
+ "low": "Low",
+ "medium": "Medium"
+ },
+ "title": "Reasoning Effort"
},
"temperature": {
- "desc": "The higher the value, the more random the response",
- "title": "Randomness",
- "titleWithValue": "Randomness {{value}}"
+ "desc": "The higher the value, the more creative and imaginative the responses; the lower the value, the more rigorous the responses.",
+ "title": "Creativity Level",
+ "warning": "If the creativity level is set too high, the output may become garbled."
},
"title": "Model Settings",
"topP": {
- "desc": "Similar to randomness, but do not change together with randomness",
- "title": "Top P Sampling"
+ "desc": "How many possibilities to consider; a higher value accepts more potential answers, while a lower value tends to choose the most likely answer. It is not recommended to change this alongside the creativity level.",
+ "title": "Openness to Ideas"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Common Settings",
"experiment": "Experiment",
"llm": "Language Model",
+ "provider": "AI Service Provider",
"sync": "Cloud Sync",
"system-agent": "System Assistant",
"tts": "Text-to-Speech"
diff --git a/locales/es-ES/changelog.json b/locales/es-ES/changelog.json
index 2b35adebf5bb9..c289032f90389 100644
--- a/locales/es-ES/changelog.json
+++ b/locales/es-ES/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "Ver todos los registros de cambios",
"description": "Sigue las nuevas funciones y mejoras de {{appName}}",
"pagination": {
- "older": "Ver cambios anteriores",
- "prev": "Página anterior"
+ "next": "Siguiente página",
+ "older": "Ver cambios anteriores"
},
"readDetails": "Leer detalles",
"title": "Registro de cambios",
diff --git a/locales/es-ES/common.json b/locales/es-ES/common.json
index 8ce9f128abe32..05d56e0a1bc47 100644
--- a/locales/es-ES/common.json
+++ b/locales/es-ES/common.json
@@ -275,6 +275,7 @@
},
"temp": "Temporal",
"terms": "Términos de servicio",
+ "update": "Actualizar",
"updateAgent": "Actualizar información del asistente",
"upgradeVersion": {
"action": "Actualizar",
@@ -285,6 +286,7 @@
"anonymousNickName": "Usuario Anónimo",
"billing": "Gestión de facturación",
"cloud": "Prueba {{name}}",
+ "community": "Versión comunitaria",
"data": "Almacenamiento de datos",
"defaultNickname": "Usuario de la comunidad",
"discord": "Soporte de la comunidad",
@@ -294,7 +296,6 @@
"help": "Centro de ayuda",
"moveGuide": "El botón de configuración se ha movido aquí",
"plans": "Planes de suscripción",
- "preview": "Vista previa",
"profile": "Gestión de cuenta",
"setting": "Configuración de la aplicación",
"usages": "Estadísticas de uso"
diff --git a/locales/es-ES/components.json b/locales/es-ES/components.json
index ebcc3d74a924c..17088d988b58f 100644
--- a/locales/es-ES/components.json
+++ b/locales/es-ES/components.json
@@ -76,6 +76,7 @@
"custom": "Modelo personalizado: admite llamadas de función y reconocimiento visual. Verifique la disponibilidad de estas capacidades según sea necesario.",
"file": "Este modelo admite la carga y reconocimiento de archivos.",
"functionCall": "Este modelo admite llamadas de función.",
+ "reasoning": "Este modelo admite un pensamiento profundo",
"tokens": "Este modelo admite un máximo de {{tokens}} tokens por sesión.",
"vision": "Este modelo admite el reconocimiento visual."
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "No hay modelos habilitados. Vaya a la configuración para habilitarlos.",
"provider": "Proveedor"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "Debido a las restricciones de seguridad del navegador, necesitas configurar CORS para Ollama antes de poder usarlo correctamente.",
+ "linux": {
+ "env": "Agrega `Environment` en la sección [Service] y añade la variable de entorno OLLAMA_ORIGINS:",
+ "reboot": "Recarga systemd y reinicia Ollama",
+ "systemd": "Usa systemd para editar el servicio de ollama:"
+ },
+ "macos": "Abre la aplicación 'Terminal', pega el siguiente comando y presiona Enter para ejecutarlo",
+ "reboot": "Reinicia el servicio de Ollama después de completar la ejecución",
+ "title": "Configura Ollama para permitir el acceso CORS",
+ "windows": "En Windows, haz clic en 'Panel de control' y entra en la edición de variables de entorno del sistema. Crea una nueva variable de entorno llamada 'OLLAMA_ORIGINS' para tu cuenta de usuario, con el valor * y haz clic en 'OK/Aplicar' para guardar."
+ },
+ "install": {
+ "description": "Asegúrate de que has iniciado Ollama. Si no has descargado Ollama, visita el sitio web oficial <1>para descargar.",
+ "docker": "Si prefieres usar Docker, Ollama también ofrece una imagen oficial de Docker que puedes descargar con el siguiente comando:",
+ "linux": {
+ "command": "Instala con el siguiente comando:",
+ "manual": "O también puedes consultar la <1>guía de instalación manual de Linux para instalarlo tú mismo."
+ },
+ "title": "Instala y ejecuta la aplicación Ollama localmente",
+ "windowsTab": "Windows (versión preliminar)"
+ }
+ },
+ "Thinking": {
+ "thinking": "Pensando profundamente...",
+ "thought": "He pensado profundamente (durante {{duration}} segundos)",
+ "thoughtWithDuration": "He pensado profundamente"
}
}
diff --git a/locales/es-ES/discover.json b/locales/es-ES/discover.json
index 42386cb6e2230..1caad337e1e91 100644
--- a/locales/es-ES/discover.json
+++ b/locales/es-ES/discover.json
@@ -126,6 +126,10 @@
"title": "Novedad del tema"
},
"range": "Rango",
+ "reasoning_effort": {
+ "desc": "Esta configuración se utiliza para controlar la intensidad de razonamiento del modelo antes de generar una respuesta. Una baja intensidad prioriza la velocidad de respuesta y ahorra tokens, mientras que una alta intensidad proporciona un razonamiento más completo, pero consume más tokens y reduce la velocidad de respuesta. El valor predeterminado es medio, equilibrando la precisión del razonamiento con la velocidad de respuesta.",
+ "title": "Intensidad de razonamiento"
+ },
"temperature": {
"desc": "Esta configuración afecta la diversidad de las respuestas del modelo. Un valor más bajo resultará en respuestas más predecibles y típicas, mientras que un valor más alto alentará respuestas más diversas y menos comunes. Cuando el valor se establece en 0, el modelo siempre dará la misma respuesta para una entrada dada.",
"title": "Aleatoriedad"
diff --git a/locales/es-ES/modelProvider.json b/locales/es-ES/modelProvider.json
index fb6b3d73e8086..510b8f576378d 100644
--- a/locales/es-ES/modelProvider.json
+++ b/locales/es-ES/modelProvider.json
@@ -19,6 +19,24 @@
"title": "Clave API"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Versión de la API de Azure, siguiendo el formato AAAA-MM-DD, consulta la [última versión](https://learn.microsoft.com/es-es/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "Obtener lista",
+ "title": "Versión de la API de Azure"
+ },
+ "endpoint": {
+ "desc": "Encuentra el punto final de inferencia del modelo de Azure AI en la descripción general del proyecto de Azure AI",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Punto final de Azure AI"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Encuentra la clave API en la descripción general del proyecto de Azure AI",
+ "placeholder": "Clave de Azure",
+ "title": "Clave"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "Introduce tu AWS Access Key Id",
@@ -63,6 +81,46 @@
"title": "ID de cuenta de Cloudflare / dirección URL de API"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Por favor, introduce tu API Key",
+ "title": "API Key"
+ },
+ "basicTitle": "Información básica",
+ "configTitle": "Información de configuración",
+ "confirm": "Crear nuevo",
+ "createSuccess": "Creación exitosa",
+ "description": {
+ "placeholder": "Descripción del proveedor (opcional)",
+ "title": "Descripción del proveedor"
+ },
+ "id": {
+ "desc": "Identificador único del proveedor de servicios, no se puede modificar una vez creado",
+ "format": "Solo puede contener números, letras minúsculas, guiones (-) y guiones bajos (_) ",
+ "placeholder": "Se recomienda en minúsculas, por ejemplo openai, no se puede modificar después de crear",
+ "required": "Por favor, introduce el ID del proveedor",
+ "title": "ID del proveedor"
+ },
+ "logo": {
+ "required": "Por favor, sube un logo correcto del proveedor",
+ "title": "Logo del proveedor"
+ },
+ "name": {
+ "placeholder": "Por favor, introduce el nombre del proveedor",
+ "required": "Por favor, introduce el nombre del proveedor",
+ "title": "Nombre del proveedor"
+ },
+ "proxyUrl": {
+ "required": "Por favor, introduce la dirección del proxy",
+ "title": "Dirección del proxy"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Por favor, selecciona el tipo de SDK",
+ "title": "Formato de solicitud"
+ },
+ "title": "Crear proveedor de AI personalizado"
+ },
"github": {
"personalAccessToken": {
"desc": "Introduce tu PAT de Github, haz clic [aquí](https://github.com/settings/tokens) para crear uno",
@@ -77,6 +135,23 @@
"title": "Token de HuggingFace"
}
},
+ "list": {
+ "title": {
+ "disabled": "Proveedor no habilitado",
+ "enabled": "Proveedor habilitado"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Agregar proveedor personalizado",
+ "all": "Todo",
+ "list": {
+ "disabled": "No habilitado",
+ "enabled": "Habilitado"
+ },
+ "notFound": "No se encontraron resultados de búsqueda",
+ "searchProviders": "Buscar proveedores...",
+ "sort": "Orden personalizado"
+ },
"ollama": {
"checker": {
"desc": "Prueba si la dirección del proxy de la interfaz se ha introducido correctamente",
@@ -94,33 +169,9 @@
"title": "Descargando el modelo {{model}} "
},
"endpoint": {
- "desc": "Introduce la dirección del proxy de la interfaz de Ollama, déjalo en blanco si no se ha especificado localmente",
+ "desc": "Debe incluir http(s)://, se puede dejar vacío si no se especifica localmente",
"title": "Dirección del proxy de la interfaz"
},
- "setup": {
- "cors": {
- "description": "Debido a restricciones de seguridad del navegador, es necesario configurar Ollama para permitir el acceso entre dominios.",
- "linux": {
- "env": "En la sección [Service], agrega `Environment` y añade la variable de entorno OLLAMA_ORIGINS:",
- "reboot": "Recarga systemd y reinicia Ollama.",
- "systemd": "Edita el servicio ollama llamando a systemd:"
- },
- "macos": "Abre la aplicación 'Terminal', pega y ejecuta el siguiente comando, luego presiona Enter.",
- "reboot": "Reinicia el servicio de Ollama una vez completada la ejecución.",
- "title": "Configuración para permitir el acceso entre dominios en Ollama",
- "windows": "En Windows, ve a 'Panel de control', edita las variables de entorno del sistema. Crea una nueva variable de entorno llamada 'OLLAMA_ORIGINS' para tu cuenta de usuario, con el valor '*', y haz clic en 'OK/Aplicar' para guardar los cambios."
- },
- "install": {
- "description": "Por favor, asegúrate de que has activado Ollama. Si no has descargado Ollama, por favor visita el sitio web oficial para <1>descargarlo.",
- "docker": "Si prefieres usar Docker, Ollama también ofrece una imagen oficial en Docker. Puedes obtenerla con el siguiente comando:",
- "linux": {
- "command": "Instala con el siguiente comando:",
- "manual": "O también puedes consultar la <1>Guía de instalación manual en Linux para instalarlo por tu cuenta."
- },
- "title": "Instalación local y activación de la aplicación Ollama",
- "windowsTab": "Windows (Versión de vista previa)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Cancel Download",
@@ -131,25 +182,145 @@
"title": "Download specified Ollama model"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Introduce la Access Key de la plataforma Qianfan de Baidu",
- "placeholder": "Access Key de Qianfan",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "Tu clave y dirección del proxy se cifrarán utilizando el algoritmo de cifrado <1>AES-GCM",
+ "apiKey": {
+ "desc": "Por favor, introduce tu {{name}} API Key",
+ "placeholder": "{{name}} API Key",
+ "title": "API Key"
+ },
+ "baseURL": {
+ "desc": "Debe incluir http(s)://",
+ "invalid": "Por favor, introduce una URL válida",
+ "placeholder": "https://tu-direccion-proxy.com/v1",
+ "title": "Dirección del proxy API"
+ },
+ "checker": {
+ "button": "Verificar",
+ "desc": "Prueba si la API Key y la dirección del proxy están correctamente introducidas",
+ "pass": "Verificación exitosa",
+ "title": "Verificación de conectividad"
+ },
+ "fetchOnClient": {
+ "desc": "El modo de solicitud del cliente iniciará la solicitud de sesión directamente desde el navegador, lo que puede mejorar la velocidad de respuesta",
+ "title": "Usar modo de solicitud del cliente"
+ },
+ "helpDoc": "Guía de configuración",
+ "waitingForMore": "Más modelos están en <1>planificación de integración, por favor, espera"
},
- "checker": {
- "desc": "Verifica si la AccessKey / SecretAccess está correctamente ingresada"
+ "createNew": {
+ "title": "Crear modelo de AI personalizado"
},
- "secretKey": {
- "desc": "Introduce la Secret Key de la plataforma Qianfan de Baidu",
- "placeholder": "Secret Key de Qianfan",
- "title": "Secret Key"
+ "item": {
+ "config": "Configurar modelo",
+ "customModelCards": {
+ "addNew": "Crear y agregar modelo {{id}}",
+ "confirmDelete": "Estás a punto de eliminar este modelo personalizado, una vez eliminado no se puede recuperar, por favor actúa con precaución."
+ },
+ "delete": {
+ "confirm": "¿Confirmar eliminación del modelo {{displayName}}?",
+ "success": "Eliminación exitosa",
+ "title": "Eliminar modelo"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Campo solicitado en Azure OpenAI",
+ "placeholder": "Por favor, introduce el nombre de despliegue del modelo en Azure",
+ "title": "Nombre de despliegue del modelo"
+ },
+ "deployName": {
+ "extra": "Este campo se enviará como ID del modelo al hacer la solicitud",
+ "placeholder": "Introduce el nombre o ID real del modelo desplegado",
+ "title": "Nombre de despliegue del modelo"
+ },
+ "displayName": {
+ "placeholder": "Por favor, introduce el nombre de visualización del modelo, por ejemplo, ChatGPT, GPT-4, etc.",
+ "title": "Nombre de visualización del modelo"
+ },
+ "files": {
+ "extra": "La implementación actual de carga de archivos es solo una solución temporal, solo para prueba personal. La capacidad completa de carga de archivos estará disponible en futuras implementaciones.",
+ "title": "Soporte para carga de archivos"
+ },
+ "functionCall": {
+ "extra": "Esta configuración solo habilitará la capacidad del modelo para usar herramientas, lo que permite agregar complementos de tipo herramienta al modelo. Sin embargo, si realmente se admiten las herramientas depende completamente del modelo en sí, por favor pruebe su disponibilidad",
+ "title": "Soporte para el uso de herramientas"
+ },
+ "id": {
+ "extra": "No se puede modificar después de la creación, se utilizará como id del modelo al llamar a la IA",
+ "placeholder": "Introduce el id del modelo, por ejemplo gpt-4o o claude-3.5-sonnet",
+ "title": "ID del modelo"
+ },
+ "modalTitle": "Configuración del modelo personalizado",
+ "reasoning": {
+ "extra": "Esta configuración solo activará la capacidad de pensamiento profundo del modelo, el efecto específico depende completamente del modelo en sí, por favor, pruebe si este modelo tiene la capacidad de pensamiento profundo utilizable",
+ "title": "Soporte para pensamiento profundo"
+ },
+ "tokens": {
+ "extra": "Establecer el número máximo de tokens que el modelo puede soportar",
+ "title": "Máximo de ventana de contexto",
+ "unlimited": "Sin límite"
+ },
+ "vision": {
+ "extra": "Esta configuración solo habilitará la configuración de carga de imágenes en la aplicación, si se admite el reconocimiento depende completamente del modelo en sí, prueba la disponibilidad de la capacidad de reconocimiento visual de este modelo.",
+ "title": "Soporte para reconocimiento visual"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/imagen",
+ "inputCharts": "${{amount}}/M caracteres",
+ "inputMinutes": "${{amount}}/minuto",
+ "inputTokens": "Entrada ${{amount}}/M",
+ "outputTokens": "Salida ${{amount}}/M"
+ },
+ "releasedAt": "Publicado el {{releasedAt}}"
},
- "unlock": {
- "customRegion": "Región de servicio personalizada",
- "description": "Introduce tu AccessKey / SecretKey para comenzar la sesión. La aplicación no registrará tu configuración de autenticación",
- "title": "Usar información de autenticación personalizada de Wenxin Yiyan"
- }
+ "list": {
+ "addNew": "Agregar modelo",
+ "disabled": "No habilitado",
+ "disabledActions": {
+ "showMore": "Mostrar todo"
+ },
+ "empty": {
+ "desc": "Por favor, crea un modelo personalizado o importa un modelo para comenzar a usarlo.",
+ "title": "No hay modelos disponibles"
+ },
+ "enabled": "Habilitado",
+ "enabledActions": {
+ "disableAll": "Deshabilitar todo",
+ "enableAll": "Habilitar todo",
+ "sort": "Ordenar modelos personalizados"
+ },
+ "enabledEmpty": "No hay modelos habilitados, por favor habilita los modelos que te gusten de la lista a continuación~",
+ "fetcher": {
+ "clear": "Eliminar modelos obtenidos",
+ "fetch": "Obtener lista de modelos",
+ "fetching": "Obteniendo lista de modelos...",
+ "latestTime": "Última actualización: {{time}}",
+ "noLatestTime": "Lista aún no obtenida"
+ },
+ "resetAll": {
+ "conform": "¿Confirmar el restablecimiento de todas las modificaciones del modelo actual? Después del restablecimiento, la lista de modelos actuales volverá al estado predeterminado",
+ "success": "Restablecimiento exitoso",
+ "title": "Restablecer todas las modificaciones"
+ },
+ "search": "Buscar modelos...",
+ "searchResult": "Se encontraron {{count}} modelos",
+ "title": "Lista de modelos",
+ "total": "Un total de {{count}} modelos disponibles"
+ },
+ "searchNotFound": "No se encontraron resultados de búsqueda"
+ },
+ "sortModal": {
+ "success": "Orden actualizado con éxito",
+ "title": "Orden personalizado",
+ "update": "Actualizar"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "Estás a punto de eliminar este proveedor de AI, una vez eliminado no se puede recuperar, ¿confirmar eliminación?",
+ "deleteSuccess": "Eliminación exitosa",
+ "tooltip": "Actualizar configuración básica del proveedor",
+ "updateSuccess": "Actualización exitosa"
},
"zeroone": {
"title": "01.AI Cero Uno Todo"
diff --git a/locales/es-ES/models.json b/locales/es-ES/models.json
index 58d09b85a42c6..9f02be6e206eb 100644
--- a/locales/es-ES/models.json
+++ b/locales/es-ES/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K enfatiza la seguridad semántica y la responsabilidad, diseñado específicamente para aplicaciones que requieren altos estándares de seguridad de contenido, asegurando la precisión y robustez de la experiencia del usuario."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 utiliza la búsqueda en árbol para construir cadenas de pensamiento e introduce un mecanismo de reflexión, entrenado mediante aprendizaje por refuerzo, lo que le permite tener la capacidad de auto-reflexión y corrección de errores."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro es un modelo avanzado de procesamiento de lenguaje natural lanzado por la empresa 360, con una excelente capacidad de generación y comprensión de textos, destacándose especialmente en la generación y creación de contenido, capaz de manejar tareas complejas de conversión de lenguaje y representación de roles."
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1 utiliza búsqueda en árbol para construir cadenas de pensamiento e introduce un mecanismo de reflexión, entrenando el modelo con aprendizaje por refuerzo, lo que le confiere la capacidad de auto-reflexión y corrección de errores."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra es la versión más poderosa de la serie de modelos grandes de Xinghuo, mejorando la comprensión y capacidad de resumen de contenido textual al actualizar la conexión de búsqueda en línea. Es una solución integral para mejorar la productividad en la oficina y responder con precisión a las necesidades, siendo un producto inteligente líder en la industria."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "El modelo más potente del país, superando a los modelos principales extranjeros en tareas en chino como enciclopedias, textos largos y creación generativa. También cuenta con capacidades multimodales líderes en la industria, destacándose en múltiples evaluaciones de referencia."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "El modelo de destilación DeepSeek-R1 basado en Qwen2.5-Math-1.5B optimiza el rendimiento de inferencia mediante aprendizaje por refuerzo y datos de arranque en frío, actualizando el estándar de múltiples tareas en modelos de código abierto."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "El modelo de destilación DeepSeek-R1 basado en Qwen2.5-14B optimiza el rendimiento de inferencia mediante aprendizaje por refuerzo y datos de arranque en frío, actualizando el estándar de múltiples tareas en modelos de código abierto."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "La serie DeepSeek-R1 optimiza el rendimiento de inferencia mediante aprendizaje por refuerzo y datos de arranque en frío, actualizando el estándar de múltiples tareas en modelos de código abierto, superando el nivel de OpenAI-o1-mini."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "El modelo de destilación DeepSeek-R1 basado en Qwen2.5-Math-7B optimiza el rendimiento de inferencia mediante aprendizaje por refuerzo y datos de arranque en frío, actualizando el estándar de múltiples tareas en modelos de código abierto."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite presenta una velocidad de respuesta extrema y una mejor relación calidad-precio, ofreciendo opciones más flexibles para diferentes escenarios de clientes. Admite inferencia y ajuste fino con ventanas de contexto de 128k."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Modelo de lenguaje a gran escala ultra avanzado desarrollado por Baidu, que ha logrado una actualización completa de las capacidades del modelo en comparación con ERNIE 3.5, siendo ampliamente aplicable a escenarios de tareas complejas en diversos campos; soporta la integración automática con el plugin de búsqueda de Baidu, garantizando la actualidad de la información en las respuestas."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Modelo de lenguaje de gran escala de última generación desarrollado por Baidu, con un rendimiento sobresaliente en una variedad de escenarios de tareas complejas en diversos campos; soporta la integración automática con el plugin de búsqueda de Baidu, garantizando la actualidad de la información en las respuestas. En comparación con ERNIE 4.0, ofrece un rendimiento superior."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "Modelo de lenguaje a gran escala desarrollado por Baidu, con un rendimiento general excepcional, ampliamente aplicable a escenas complejas en diversos campos; soporta la conexión automática al complemento de búsqueda de Baidu, garantizando la actualidad de la información de las preguntas y respuestas. En comparación con ERNIE 4.0, tiene un rendimiento superior."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct es uno de los últimos modelos de lenguaje a gran escala lanzados por Alibaba Cloud. Este modelo de 7B ha mejorado significativamente en áreas como codificación y matemáticas. También ofrece soporte multilingüe, abarcando más de 29 idiomas, incluidos chino e inglés. El modelo ha mostrado mejoras significativas en el seguimiento de instrucciones, comprensión de datos estructurados y generación de salidas estructuradas (especialmente JSON)."
},
+ "MiniMax-Text-01": {
+ "description": "En la serie de modelos MiniMax-01, hemos realizado una innovación audaz: la implementación a gran escala del mecanismo de atención lineal, donde la arquitectura Transformer tradicional ya no es la única opción. Este modelo tiene una cantidad de parámetros de hasta 456 mil millones, con 45.9 mil millones por activación. El rendimiento general del modelo es comparable a los mejores modelos internacionales, y puede manejar de manera eficiente contextos de hasta 4 millones de tokens, que es 32 veces más que GPT-4o y 20 veces más que Claude-3.5-Sonnet."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO es una fusión de múltiples modelos altamente flexible, diseñada para ofrecer una experiencia creativa excepcional."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 ha demostrado un rendimiento sobresaliente en diversas tareas de lenguaje visual, incluidas la comprensión de documentos y gráficos, comprensión de texto en escenas, OCR, resolución de problemas científicos y matemáticos."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 ha demostrado un rendimiento sobresaliente en diversas tareas de lenguaje visual, incluidas la comprensión de documentos y gráficos, comprensión de texto en escenas, OCR, resolución de problemas científicos y matemáticos."
- },
"Phi-3-medium-128k-instruct": {
"description": "El mismo modelo Phi-3-medium, pero con un tamaño de contexto más grande para RAG o indicaciones de pocos disparos."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat es la versión de código abierto de la serie de modelos preentrenados GLM-4 lanzada por Zhipu AI. Este modelo destaca en semántica, matemáticas, razonamiento, código y conocimiento. Además de soportar diálogos de múltiples turnos, GLM-4-9B-Chat también cuenta con funciones avanzadas como navegación web, ejecución de código, llamadas a herramientas personalizadas (Function Call) y razonamiento de textos largos. El modelo admite 26 idiomas, incluidos chino, inglés, japonés, coreano y alemán. En múltiples pruebas de referencia, GLM-4-9B-Chat ha demostrado un rendimiento excepcional, como AlignBench-v2, MT-Bench, MMLU y C-Eval. Este modelo admite una longitud de contexto máxima de 128K, adecuado para investigación académica y aplicaciones comerciales."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 es un modelo de inferencia impulsado por aprendizaje por refuerzo (RL) que aborda problemas de repetitividad y legibilidad en el modelo. Antes del RL, DeepSeek-R1 introdujo datos de arranque en frío, optimizando aún más el rendimiento de inferencia. Se desempeña de manera comparable a OpenAI-o1 en tareas matemáticas, de código e inferencia, y mejora el rendimiento general a través de métodos de entrenamiento cuidadosamente diseñados."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 es un modelo de lenguaje de expertos mixtos (MoE) con 671 mil millones de parámetros, que utiliza atención potencial de múltiples cabezas (MLA) y la arquitectura DeepSeekMoE, combinando estrategias de balanceo de carga sin pérdidas auxiliares para optimizar la eficiencia de inferencia y entrenamiento. Preentrenado en 14.8 billones de tokens de alta calidad, y ajustado mediante supervisión y aprendizaje por refuerzo, DeepSeek-V3 supera a otros modelos de código abierto y se acerca a los modelos cerrados líderes."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma es una de las series de modelos abiertos más avanzados y ligeros desarrollados por Google. Es un modelo de lenguaje a gran escala solo de decodificación, que admite inglés y proporciona pesos abiertos, variantes preentrenadas y variantes de ajuste fino por instrucciones. El modelo Gemma es adecuado para diversas tareas de generación de texto, incluyendo preguntas y respuestas, resúmenes y razonamiento. Este modelo de 9B se ha entrenado con 80 billones de tokens. Su tamaño relativamente pequeño permite su implementación en entornos con recursos limitados, como computadoras portátiles, de escritorio o su propia infraestructura en la nube, lo que permite a más personas acceder a modelos de IA de vanguardia y fomentar la innovación."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 es parte de la familia de modelos de lenguaje a gran escala multilingües desarrollados por Meta, que incluye variantes preentrenadas y de ajuste fino por instrucciones con tamaños de parámetros de 8B, 70B y 405B. Este modelo de 8B ha sido optimizado para escenarios de diálogo multilingüe y ha destacado en múltiples pruebas de referencia de la industria. El entrenamiento del modelo utilizó más de 150 billones de tokens de datos públicos y empleó técnicas como ajuste fino supervisado y aprendizaje por refuerzo con retroalimentación humana para mejorar la utilidad y seguridad del modelo. Llama 3.1 admite generación de texto y generación de código, con una fecha límite de conocimiento hasta diciembre de 2023."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview es un modelo de procesamiento de lenguaje natural innovador, capaz de manejar de manera eficiente tareas complejas de generación de diálogos y comprensión del contexto."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview es un modelo de investigación desarrollado por el equipo de Qwen, enfocado en la capacidad de razonamiento visual, que tiene ventajas únicas en la comprensión de escenas complejas y en la resolución de problemas matemáticos relacionados con la visión."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview es el último modelo de investigación experimental de Qwen, enfocado en mejorar la capacidad de razonamiento de la IA. A través de la exploración de mecanismos complejos como la mezcla de lenguajes y el razonamiento recursivo, sus principales ventajas incluyen una poderosa capacidad de análisis de razonamiento, así como habilidades matemáticas y de programación. Sin embargo, también presenta problemas de cambio de idioma, ciclos de razonamiento, consideraciones de seguridad y diferencias en otras capacidades."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct es la última versión de la serie de modelos de lenguaje a gran escala específicos para código lanzada por Alibaba Cloud. Este modelo, basado en Qwen2.5, ha mejorado significativamente la generación, razonamiento y reparación de código a través de un entrenamiento con 55 billones de tokens. No solo ha mejorado la capacidad de codificación, sino que también ha mantenido ventajas en habilidades matemáticas y generales. El modelo proporciona una base más completa para aplicaciones prácticas como agentes de código."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math se centra en la resolución de problemas en el ámbito de las matemáticas, proporcionando respuestas profesionales a preguntas de alta dificultad."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 es la última serie del modelo Qwen, que admite un contexto de 128k. En comparación con los modelos de código abierto más óptimos actuales, Qwen2-72B supera significativamente a los modelos líderes actuales en comprensión del lenguaje natural, conocimiento, código, matemáticas y capacidades multilingües."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 es la última serie del modelo Qwen, capaz de superar a los modelos de código abierto de tamaño equivalente e incluso a modelos de mayor tamaño. Qwen2 7B ha logrado ventajas significativas en múltiples evaluaciones, especialmente en comprensión de código y chino."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B es un potente modelo de lenguaje visual que admite el procesamiento multimodal de imágenes y texto, capaz de identificar con precisión el contenido de las imágenes y generar descripciones o respuestas relacionadas."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct es un modelo de lenguaje grande de 14 mil millones de parámetros, con un rendimiento excelente, optimizado para escenarios en chino y multilingües, que admite aplicaciones de preguntas y respuestas inteligentes, generación de contenido, entre otros."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct es un modelo de lenguaje grande de 7 mil millones de parámetros, que admite llamadas a funciones e interacción sin problemas con sistemas externos, mejorando enormemente la flexibilidad y escalabilidad. Optimizado para escenarios en chino y multilingües, admite aplicaciones de preguntas y respuestas inteligentes, generación de contenido, entre otros."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct es un modelo de instrucciones de programación basado en un preentrenamiento a gran escala, con una potente capacidad de comprensión y generación de código, capaz de manejar eficientemente diversas tareas de programación, especialmente adecuado para la escritura inteligente de código, generación de scripts automatizados y resolución de problemas de programación."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct es un modelo de lenguaje grande diseñado específicamente para la generación de código, comprensión de código y escenarios de desarrollo eficiente, con una escala de 32B parámetros, líder en la industria, capaz de satisfacer diversas necesidades de programación."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "El modelo multimodal TeleMM ha sido desarrollado de manera independiente por China Telecom, siendo un modelo de comprensión multimodal que puede manejar entradas de múltiples modalidades como texto e imágenes, apoyando funciones como comprensión de imágenes y análisis de gráficos, proporcionando servicios de comprensión cruzada para los usuarios. El modelo puede interactuar con los usuarios de manera multimodal, entendiendo con precisión el contenido de entrada, respondiendo preguntas, asistiendo en la creación y proporcionando de manera eficiente información y apoyo inspirador multimodal. Ha mostrado un rendimiento excepcional en tareas multimodales como percepción de alta resolución y razonamiento lógico."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large es el modelo MoE de Transformer de código abierto más grande de la industria, con un total de 389 mil millones de parámetros y 52 mil millones de parámetros activados."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct es un modelo de lenguaje a gran escala de ajuste fino por instrucciones dentro de la serie Qwen2, con un tamaño de parámetros de 72B. Este modelo se basa en la arquitectura Transformer, utilizando funciones de activación SwiGLU, sesgos de atención QKV y atención de consulta agrupada, entre otras técnicas. Es capaz de manejar entradas a gran escala. Este modelo ha destacado en múltiples pruebas de referencia en comprensión del lenguaje, generación, capacidad multilingüe, codificación, matemáticas y razonamiento, superando a la mayoría de los modelos de código abierto y mostrando competitividad comparable a modelos propietarios en ciertas tareas."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct es uno de los últimos modelos de lenguaje a gran escala lanzados por Alibaba Cloud. Este modelo de 72B ha mejorado significativamente en áreas como codificación y matemáticas. También ofrece soporte multilingüe, abarcando más de 29 idiomas, incluidos chino e inglés. El modelo ha mostrado mejoras significativas en el seguimiento de instrucciones, comprensión de datos estructurados y generación de salidas estructuradas (especialmente JSON)."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Optimizado para escenarios de diálogo de personajes en chino, ofrece capacidades de generación de diálogos fluidos y acordes con las expresiones chinas."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Modelo de llamada de función de código abierto de Fireworks, que ofrece capacidades de ejecución de instrucciones sobresalientes y características personalizables."
+ "abab7-chat-preview": {
+ "description": "En comparación con la serie de modelos abab6.5, ha habido una mejora significativa en habilidades como el manejo de textos largos, matemáticas y escritura."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Firefunction-v2, lanzado por Fireworks, es un modelo de llamada de función de alto rendimiento, desarrollado sobre Llama-3 y optimizado para escenarios como llamadas de función, diálogos y seguimiento de instrucciones."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 es un modelo de lenguaje grande de última generación, optimizado mediante aprendizaje por refuerzo y datos de arranque en frío, con un rendimiento excepcional en razonamiento, matemáticas y programación."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b es un modelo de lenguaje visual que puede recibir entradas de imagen y texto simultáneamente, entrenado con datos de alta calidad, adecuado para tareas multimodales."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Modelo de lenguaje potente de Deepseek, basado en Mixture-of-Experts (MoE), con un total de 671B de parámetros, activando 37B de parámetros por cada token."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "El modelo de instrucciones Llama 3 70B está optimizado para diálogos multilingües y comprensión del lenguaje natural, superando el rendimiento de la mayoría de los modelos competidores."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "El modelo de instrucciones Llama 3 70B (versión HF) es consistente con los resultados de la implementación oficial, adecuado para tareas de seguimiento de instrucciones de alta calidad."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "El modelo de instrucciones Llama 3 8B está optimizado para diálogos y tareas multilingües, ofreciendo un rendimiento excepcional y eficiente."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Modelo de razonamiento de imágenes de 11B parámetros ajustado por Meta. Este modelo está optimizado para el reconocimiento visual, razonamiento de imágenes, descripción de imágenes y respuestas a preguntas generales sobre imágenes. Puede entender datos visuales, como gráficos y diagramas, y cerrar la brecha entre la visión y el lenguaje generando descripciones textuales de los detalles de las imágenes."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "El modelo de instrucciones Llama 3.2 1B es un modelo multilingüe ligero lanzado por Meta. Está diseñado para mejorar la eficiencia, ofreciendo mejoras significativas en latencia y costos en comparación con modelos más grandes. Ejemplos de uso de este modelo incluyen recuperación y resumen."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "El modelo de instrucciones Llama 3.2 3B es un modelo multilingüe ligero lanzado por Meta. Está diseñado para mejorar la eficiencia, ofreciendo mejoras significativas en latencia y costos en comparación con modelos más grandes. Ejemplos de uso de este modelo incluyen consultas, reescritura de indicaciones y asistencia en la escritura."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Modelo de razonamiento de imágenes de 90B parámetros ajustado por Meta. Este modelo está optimizado para el reconocimiento visual, razonamiento de imágenes, descripción de imágenes y respuestas a preguntas generales sobre imágenes. Puede entender datos visuales, como gráficos y diagramas, y cerrar la brecha entre la visión y el lenguaje generando descripciones textuales de los detalles de las imágenes."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct es la versión actualizada de diciembre de Llama 3.1 70B. Este modelo ha sido mejorado sobre la base de Llama 3.1 70B (lanzado en julio de 2024), mejorando la invocación de herramientas, el soporte de texto multilingüe, así como las capacidades matemáticas y de programación. El modelo alcanza niveles de liderazgo en la industria en razonamiento, matemáticas y cumplimiento de instrucciones, y puede ofrecer un rendimiento similar al de 3.1 405B, al tiempo que presenta ventajas significativas en velocidad y costo."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "Modelo de 24B parámetros, con capacidades de vanguardia comparables a modelos más grandes."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "El modelo de instrucciones Mixtral MoE 8x22B, con parámetros a gran escala y arquitectura de múltiples expertos, soporta de manera integral el procesamiento eficiente de tareas complejas."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "El modelo de instrucciones Mixtral MoE 8x7B, con una arquitectura de múltiples expertos, ofrece un seguimiento y ejecución de instrucciones eficientes."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "El modelo de instrucciones Mixtral MoE 8x7B (versión HF) tiene un rendimiento consistente con la implementación oficial, adecuado para una variedad de escenarios de tareas eficientes."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "El modelo MythoMax L2 13B combina técnicas de fusión innovadoras, destacándose en narración y juegos de rol."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "El modelo QwQ es un modelo de investigación experimental desarrollado por el equipo de Qwen, enfocado en mejorar la capacidad de razonamiento de la IA."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "La versión de 72B del modelo Qwen-VL es el resultado de la última iteración de Alibaba, representando casi un año de innovación."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 es una serie de modelos de lenguaje solo decodificadores desarrollados por el equipo Qwen de Alibaba Cloud. Estos modelos ofrecen diferentes tamaños, incluidos 0.5B, 1.5B, 3B, 7B, 14B, 32B y 72B, y tienen variantes base y de instrucciones."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct es la última versión de la serie de modelos de lenguaje a gran escala específicos para código lanzada por Alibaba Cloud. Este modelo, basado en Qwen2.5, ha mejorado significativamente la generación, razonamiento y reparación de código a través de un entrenamiento con 55 billones de tokens. No solo ha mejorado la capacidad de codificación, sino que también ha mantenido ventajas en habilidades matemáticas y generales. El modelo proporciona una base más completa para aplicaciones prácticas como agentes de código."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "El modelo StarCoder 15.5B soporta tareas de programación avanzadas, con capacidades multilingües mejoradas, adecuado para la generación y comprensión de código complejo."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "El modelo StarCoder 7B está entrenado en más de 80 lenguajes de programación, con una excelente capacidad de completado de código y comprensión del contexto."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "El modelo Yi-Large ofrece una capacidad de procesamiento multilingüe excepcional, adecuado para diversas tareas de generación y comprensión de lenguaje."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus es el modelo más potente de Anthropic para manejar tareas altamente complejas. Destaca en rendimiento, inteligencia, fluidez y comprensión."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku es el modelo de próxima generación más rápido de Anthropic. En comparación con Claude 3 Haiku, Claude 3.5 Haiku ha mejorado en todas las habilidades y ha superado al modelo más grande de la generación anterior, Claude 3 Opus, en muchas pruebas de inteligencia."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet ofrece capacidades que superan a Opus y una velocidad más rápida que Sonnet, manteniendo el mismo precio que Sonnet. Sonnet es especialmente hábil en programación, ciencia de datos, procesamiento visual y tareas de agente."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet proporciona un equilibrio ideal entre inteligencia y velocidad para cargas de trabajo empresariales. Ofrece la máxima utilidad a un costo más bajo, siendo fiable y adecuado para implementaciones a gran escala."
},
- "code-raccoon-v1": {
- "description": "Code Raccoon es un asistente de desarrollo inteligente basado en el modelo de lenguaje grande de SenseTime, que abarca análisis de requisitos de software, diseño de arquitectura, escritura de código, pruebas de software y más, satisfaciendo diversas necesidades de escritura de código y aprendizaje de programación. Code Raccoon admite más de 90 lenguajes de programación populares como Python, Java, JavaScript, C++, Go, SQL, y entornos de desarrollo integrados como VS Code, IntelliJ IDEA, entre otros. En la práctica, Code Raccoon puede ayudar a los desarrolladores a mejorar la eficiencia de programación en más del 50%."
- },
"codegeex-4": {
"description": "CodeGeeX-4 es un potente asistente de programación AI, que admite preguntas y respuestas inteligentes y autocompletado de código en varios lenguajes de programación, mejorando la eficiencia del desarrollo."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ es un modelo de lenguaje de gran tamaño de alto rendimiento, diseñado para escenarios empresariales reales y aplicaciones complejas."
},
+ "dall-e-2": {
+ "description": "El segundo modelo DALL·E, que admite generación de imágenes más realistas y precisas, con una resolución cuatro veces mayor que la de la primera generación."
+ },
+ "dall-e-3": {
+ "description": "El modelo DALL·E más reciente, lanzado en noviembre de 2023. Admite generación de imágenes más realistas y precisas, con una mayor capacidad de detalle."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct ofrece capacidades de procesamiento de instrucciones de alta fiabilidad, soportando aplicaciones en múltiples industrias."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 es un potente y rentable modelo de lenguaje de expertos mixtos (MoE). Se ha preentrenado en un corpus de alta calidad de 81 billones de tokens y ha mejorado aún más sus capacidades a través de ajuste fino supervisado (SFT) y aprendizaje por refuerzo (RL). En comparación con DeepSeek 67B, DeepSeek-V2 no solo ofrece un rendimiento superior, sino que también ahorra un 42.5% en costos de entrenamiento, reduce un 93.3% en caché KV y aumenta la máxima tasa de generación a 5.76 veces. Este modelo admite una longitud de contexto de 128k y ha destacado en pruebas de referencia estándar y evaluaciones de generación abierta."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 es un modelo de inferencia impulsado por aprendizaje reforzado (RL) que aborda los problemas de repetitividad y legibilidad en el modelo. Antes de RL, DeepSeek-R1 introdujo datos de arranque en frío, optimizando aún más el rendimiento de la inferencia. Su desempeño en tareas matemáticas, de código e inferencia es comparable al de OpenAI-o1, y ha mejorado su efectividad general a través de métodos de entrenamiento cuidadosamente diseñados."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "El modelo de destilación DeepSeek-R1 optimiza el rendimiento de inferencia mediante aprendizaje por refuerzo y datos de arranque en frío, actualizando el estándar de múltiples tareas en modelos de código abierto."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B es un modelo de destilación desarrollado a partir de Llama-3.1-8B. Este modelo se ajustó utilizando muestras generadas por DeepSeek-R1, mostrando una excelente capacidad de inferencia. Ha tenido un buen desempeño en múltiples pruebas de referencia, alcanzando una precisión del 89.1% en MATH-500, una tasa de aprobación del 50.4% en AIME 2024, y una puntuación de 1205 en CodeForces, demostrando una fuerte capacidad matemática y de programación como modelo de 8B."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "El modelo de destilación DeepSeek-R1 optimiza el rendimiento de inferencia mediante aprendizaje por refuerzo y datos de arranque en frío, actualizando el estándar de múltiples tareas en modelos de código abierto."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "El modelo de destilación DeepSeek-R1 optimiza el rendimiento de inferencia mediante aprendizaje por refuerzo y datos de arranque en frío, actualizando el estándar de múltiples tareas en modelos de código abierto."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B es un modelo obtenido mediante destilación de conocimiento basado en Qwen2.5-32B. Este modelo se ajustó utilizando 800,000 muestras seleccionadas generadas por DeepSeek-R1, mostrando un rendimiento excepcional en múltiples campos como matemáticas, programación e inferencia. Ha obtenido excelentes resultados en varias pruebas de referencia, alcanzando una precisión del 94.3% en MATH-500, demostrando una fuerte capacidad de razonamiento matemático."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B es un modelo obtenido mediante destilación de conocimiento basado en Qwen2.5-Math-7B. Este modelo se ajustó utilizando 800,000 muestras seleccionadas generadas por DeepSeek-R1, mostrando un rendimiento excepcional en múltiples campos como matemáticas, programación e inferencia. Ha obtenido excelentes resultados en varias pruebas de referencia, alcanzando una precisión del 92.8% en MATH-500, una tasa de aprobación del 55.5% en AIME 2024, y una puntuación de 1189 en CodeForces, demostrando una fuerte capacidad matemática y de programación como modelo de 7B."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 combina las excelentes características de versiones anteriores, mejorando la capacidad general y de codificación."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 es un modelo de lenguaje de expertos mixtos (MoE) con 6710 millones de parámetros, que utiliza atención latente de múltiples cabezas (MLA) y la arquitectura DeepSeekMoE, combinando una estrategia de balanceo de carga sin pérdidas auxiliares para optimizar la eficiencia de inferencia y entrenamiento. Al ser preentrenado en 14.8 billones de tokens de alta calidad y realizar ajustes supervisados y aprendizaje reforzado, DeepSeek-V3 supera en rendimiento a otros modelos de código abierto, acercándose a los modelos cerrados líderes."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B es un modelo avanzado entrenado para diálogos de alta complejidad."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "LLM eficiente de última generación, experto en razonamiento, matemáticas y programación."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 es un modelo de lenguaje visual de expertos mixtos (MoE) desarrollado sobre DeepSeekMoE-27B, que utiliza una arquitectura MoE de activación dispersa, logrando un rendimiento excepcional al activar solo 4.5B de parámetros. Este modelo destaca en múltiples tareas como preguntas visuales, reconocimiento óptico de caracteres, comprensión de documentos/tablas/gráficos y localización visual."
+ },
"deepseek-chat": {
"description": "Un nuevo modelo de código abierto que fusiona capacidades generales y de codificación, que no solo conserva la capacidad de diálogo general del modelo Chat original y la potente capacidad de procesamiento de código del modelo Coder, sino que también se alinea mejor con las preferencias humanas. Además, DeepSeek-V2.5 ha logrado mejoras significativas en tareas de escritura, seguimiento de instrucciones y más."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 es un modelo de código de expertos híbrido de código abierto, que destaca en tareas de codificación, comparable a GPT4-Turbo."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 es un modelo de inferencia impulsado por aprendizaje reforzado (RL) que aborda los problemas de repetitividad y legibilidad en el modelo. Antes de RL, DeepSeek-R1 introdujo datos de arranque en frío, optimizando aún más el rendimiento de la inferencia. Su desempeño en tareas matemáticas, de código e inferencia es comparable al de OpenAI-o1, y ha mejorado su efectividad general a través de métodos de entrenamiento cuidadosamente diseñados."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1, el modelo más grande e inteligente del conjunto DeepSeek, ha sido destilado en la arquitectura Llama 70B. Basado en pruebas de referencia y evaluaciones humanas, este modelo es más inteligente que el Llama 70B original, destacándose especialmente en tareas que requieren precisión matemática y factual."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "El modelo de la serie DeepSeek-R1-Distill se obtiene mediante la técnica de destilación de conocimiento, ajustando muestras generadas por DeepSeek-R1 a modelos de código abierto como Qwen y Llama."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "El modelo de la serie DeepSeek-R1-Distill se obtiene mediante la técnica de destilación de conocimiento, ajustando muestras generadas por DeepSeek-R1 a modelos de código abierto como Qwen y Llama."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "El modelo de la serie DeepSeek-R1-Distill se obtiene mediante la técnica de destilación de conocimiento, ajustando muestras generadas por DeepSeek-R1 a modelos de código abierto como Qwen y Llama."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "El modelo de la serie DeepSeek-R1-Distill se obtiene mediante la técnica de destilación de conocimiento, ajustando muestras generadas por DeepSeek-R1 a modelos de código abierto como Qwen y Llama."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "El modelo de la serie DeepSeek-R1-Distill se obtiene mediante la técnica de destilación de conocimiento, ajustando muestras generadas por DeepSeek-R1 a modelos de código abierto como Qwen y Llama."
+ },
+ "deepseek-reasoner": {
+ "description": "Modelo de inferencia lanzado por DeepSeek. Antes de proporcionar la respuesta final, el modelo genera primero una cadena de pensamiento para mejorar la precisión de la respuesta final."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 es un modelo de lenguaje Mixture-of-Experts eficiente, adecuado para necesidades de procesamiento económico."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B es el modelo de código de diseño de DeepSeek, que ofrece una potente capacidad de generación de código."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 es un modelo MoE desarrollado por Hangzhou DeepSeek Artificial Intelligence Technology Research Co., Ltd., que ha destacado en múltiples evaluaciones, ocupando el primer lugar en la lista de modelos de código abierto. En comparación con el modelo V2.5, la velocidad de generación se ha incrementado tres veces, brindando a los usuarios una experiencia de uso más rápida y fluida."
+ },
"deepseek/deepseek-chat": {
"description": "Un nuevo modelo de código abierto que fusiona capacidades generales y de codificación, no solo conserva la capacidad de diálogo general del modelo Chat original y la potente capacidad de procesamiento de código del modelo Coder, sino que también se alinea mejor con las preferencias humanas. Además, DeepSeek-V2.5 ha logrado mejoras significativas en tareas de escritura, seguimiento de instrucciones y más."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 mejora significativamente la capacidad de razonamiento del modelo con muy pocos datos etiquetados. Antes de proporcionar la respuesta final, el modelo genera una cadena de pensamiento para mejorar la precisión de la respuesta final."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 mejora significativamente la capacidad de razonamiento del modelo con muy pocos datos etiquetados. Antes de proporcionar la respuesta final, el modelo genera una cadena de pensamiento para mejorar la precisión de la respuesta final."
+ },
"emohaa": {
"description": "Emohaa es un modelo psicológico con capacidades de consulta profesional, ayudando a los usuarios a comprender problemas emocionales."
},
+ "ernie-3.5-128k": {
+ "description": "El modelo de lenguaje grande de bandera de Baidu, desarrollado internamente, cubre una vasta cantidad de corpus en chino e inglés, con potentes capacidades generales que satisfacen la mayoría de los requisitos de preguntas y respuestas en diálogos, generación creativa y aplicaciones de plugins; soporta la integración automática con el plugin de búsqueda de Baidu, garantizando la actualidad de la información de preguntas y respuestas."
+ },
+ "ernie-3.5-8k": {
+ "description": "El modelo de lenguaje grande de bandera de Baidu, desarrollado internamente, cubre una vasta cantidad de corpus en chino e inglés, con potentes capacidades generales que satisfacen la mayoría de los requisitos de preguntas y respuestas en diálogos, generación creativa y aplicaciones de plugins; soporta la integración automática con el plugin de búsqueda de Baidu, garantizando la actualidad de la información de preguntas y respuestas."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "El modelo de lenguaje grande de bandera de Baidu, desarrollado internamente, cubre una vasta cantidad de corpus en chino e inglés, con potentes capacidades generales que satisfacen la mayoría de los requisitos de preguntas y respuestas en diálogos, generación creativa y aplicaciones de plugins; soporta la integración automática con el plugin de búsqueda de Baidu, garantizando la actualidad de la información de preguntas y respuestas."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "El modelo de lenguaje grande de bandera de Baidu, desarrollado internamente, de ultra gran escala, ha logrado una actualización completa de capacidades en comparación con ERNIE 3.5, siendo ampliamente aplicable en escenarios de tareas complejas en diversos campos; soporta la integración automática con el plugin de búsqueda de Baidu, garantizando la actualidad de la información de preguntas y respuestas."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "El modelo de lenguaje grande de bandera de Baidu, desarrollado internamente, de ultra gran escala, ha logrado una actualización completa de capacidades en comparación con ERNIE 3.5, siendo ampliamente aplicable en escenarios de tareas complejas en diversos campos; soporta la integración automática con el plugin de búsqueda de Baidu, garantizando la actualidad de la información de preguntas y respuestas."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "El modelo de lenguaje grande de bandera de Baidu, desarrollado internamente, de ultra gran escala, muestra un rendimiento excepcional en general, siendo ampliamente aplicable en escenarios de tareas complejas en diversos campos; soporta la integración automática con el plugin de búsqueda de Baidu, garantizando la actualidad de la información de preguntas y respuestas. En comparación con ERNIE 4.0, presenta un rendimiento superior."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "El modelo de lenguaje grande de bandera de Baidu, desarrollado internamente, de ultra gran escala, muestra un rendimiento excepcional en general, siendo ampliamente aplicable en escenarios de tareas complejas en diversos campos; soporta la integración automática con el plugin de búsqueda de Baidu, garantizando la actualidad de la información de preguntas y respuestas. En comparación con ERNIE 4.0, presenta un rendimiento superior."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "El modelo de lenguaje grande de bandera de Baidu, desarrollado internamente, de ultra gran escala, muestra un rendimiento excepcional en general, siendo ampliamente aplicable en escenarios de tareas complejas en diversos campos; soporta la integración automática con el plugin de búsqueda de Baidu, garantizando la actualidad de la información de preguntas y respuestas. En comparación con ERNIE 4.0, presenta un rendimiento superior."
+ },
+ "ernie-char-8k": {
+ "description": "Modelo de lenguaje grande de escenario vertical desarrollado internamente por Baidu, adecuado para aplicaciones como NPC de juegos, diálogos de servicio al cliente y juegos de rol de diálogos, con un estilo de personaje más distintivo y consistente, y una mayor capacidad de seguimiento de instrucciones y rendimiento de inferencia."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Modelo de lenguaje grande de escenario vertical desarrollado internamente por Baidu, adecuado para aplicaciones como NPC de juegos, diálogos de servicio al cliente y juegos de rol de diálogos, con un estilo de personaje más distintivo y consistente, y una mayor capacidad de seguimiento de instrucciones y rendimiento de inferencia."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite es un modelo de lenguaje grande ligero desarrollado internamente por Baidu, que combina un excelente rendimiento del modelo con una buena capacidad de inferencia, adecuado para su uso en tarjetas de aceleración de IA de bajo consumo."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Modelo de lenguaje grande ligero desarrollado internamente por Baidu, que combina un excelente rendimiento del modelo con una buena capacidad de inferencia, con un rendimiento superior al de ERNIE Lite, adecuado para su uso en tarjetas de aceleración de IA de bajo consumo."
+ },
+ "ernie-novel-8k": {
+ "description": "Modelo de lenguaje grande general desarrollado internamente por Baidu, con ventajas notables en la capacidad de continuar novelas, también aplicable en escenarios de cortometrajes y películas."
+ },
+ "ernie-speed-128k": {
+ "description": "El modelo de lenguaje grande de alto rendimiento desarrollado internamente por Baidu, lanzado en 2024, tiene capacidades generales excepcionales, adecuado como modelo base para ajustes finos, manejando mejor problemas específicos de escenarios, y con un excelente rendimiento de inferencia."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "El modelo de lenguaje grande de alto rendimiento desarrollado internamente por Baidu, lanzado en 2024, tiene capacidades generales excepcionales, con un rendimiento superior al de ERNIE Speed, adecuado como modelo base para ajustes finos, manejando mejor problemas específicos de escenarios, y con un excelente rendimiento de inferencia."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny es un modelo de lenguaje grande de alto rendimiento desarrollado internamente por Baidu, con los costos de implementación y ajuste más bajos entre los modelos de la serie Wenxin."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Ajuste) ofrece un rendimiento estable y ajustable, siendo una opción ideal para soluciones de tareas complejas."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro admite hasta 2 millones de tokens, siendo una opción ideal para modelos multimodales de tamaño medio, adecuados para un soporte multifacético en tareas complejas."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash ofrece funciones y mejoras de próxima generación, incluyendo velocidad excepcional, uso de herramientas nativas, generación multimodal y una ventana de contexto de 1M tokens."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash ofrece funciones y mejoras de próxima generación, incluyendo velocidad excepcional, uso de herramientas nativas, generación multimodal y una ventana de contexto de 1M tokens."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp es el último modelo experimental de inteligencia artificial multimodal de Google, que cuenta con características de próxima generación, velocidad superior, llamadas nativas a herramientas y generación multimodal."
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "Un modelo Gemini 2.0 Flash optimizado para objetivos de costo-efectividad y baja latencia."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Exp es el último modelo experimental de IA multimodal de Google, con características de próxima generación, velocidad excepcional, llamadas nativas a herramientas y generación multimodal."
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Exp es el último modelo de IA multimodal experimental de Google, que cuenta con características de próxima generación, velocidad excepcional, llamadas a herramientas nativas y generación multimodal."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 es el modelo de IA multimodal experimental más reciente de Google, que cuenta con una capacidad de procesamiento rápido y admite entradas de texto, imagen y video, adecuado para una amplia gama de tareas de manera eficiente."
- },
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 es el último modelo experimental de IA multimodal de Google, que cuenta con una capacidad de procesamiento rápido y admite entradas de texto, imágenes y videos, adecuado para una amplia gama de tareas de manera eficiente."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental es el último modelo de IA multimodal experimental de Google, con mejoras de calidad en comparación con versiones anteriores, especialmente en conocimiento del mundo, código y contextos largos."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 es el último modelo de IA multimodal experimental de Google, con mejoras de calidad en comparación con versiones anteriores."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus tiene la capacidad de entender contenido de video y múltiples imágenes, adecuado para tareas multimodales."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview posee una poderosa capacidad de razonamiento complejo, destacándose en áreas como razonamiento lógico, matemáticas y programación."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash ofrece funciones y mejoras de próxima generación, incluyendo velocidad excepcional, uso de herramientas nativas, generación multimodal y una ventana de contexto de 1M tokens."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental es el último modelo de IA multimodal experimental de Google, con mejoras de calidad en comparación con versiones anteriores, especialmente en conocimiento del mundo, código y contextos largos."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash ofrece capacidades de procesamiento multimodal optimizadas, adecuadas para una variedad de escenarios de tareas complejas."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro combina las últimas tecnologías de optimización, proporcionando una capacidad de procesamiento de datos multimodal más eficiente."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 es un modelo eficiente lanzado por Google, que abarca una variedad de escenarios de aplicación desde aplicaciones pequeñas hasta procesamiento de datos complejos."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 continúa con el concepto de diseño ligero y eficiente."
},
"google/gemma-2-2b-it": {
"description": "Modelo de ajuste de instrucciones ligero de Google."
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 es un modelo eficiente lanzado por Google, que abarca una variedad de escenarios de aplicación desde aplicaciones pequeñas hasta procesamiento de datos complejos."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 es una serie de modelos de texto de código abierto y ligeros de Google."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo, adecuado para diversas tareas de generación y comprensión de texto, actualmente apunta a gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, un modelo de generación de texto de alta capacidad, adecuado para tareas complejas."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo, adecuado para diversas tareas de generación y comprensión de texto, actualmente apunta a gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o es un modelo dinámico que se actualiza en tiempo real para mantener la versión más reciente. Combina una poderosa comprensión del lenguaje con habilidades de generación, adecuada para escenarios de aplicación a gran escala, incluidos servicio al cliente, educación y soporte técnico."
},
+ "gpt-4o-audio-preview": {
+ "description": "Modelo de audio GPT-4o, que admite entrada y salida de audio."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini es el último modelo lanzado por OpenAI después de GPT-4 Omni, que admite entradas de texto e imagen y genera texto como salida. Como su modelo más avanzado de menor tamaño, es mucho más económico que otros modelos de vanguardia recientes y es más de un 60% más barato que GPT-3.5 Turbo. Mantiene una inteligencia de vanguardia mientras ofrece una relación calidad-precio significativa. GPT-4o mini obtuvo un puntaje del 82% en la prueba MMLU y actualmente se clasifica por encima de GPT-4 en preferencias de chat."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "Versión en tiempo real de GPT-4o-mini, que admite entrada y salida de audio y texto en tiempo real."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "Versión en tiempo real de GPT-4o, que admite entrada y salida de audio y texto en tiempo real."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "Versión en tiempo real de GPT-4o, que admite entrada y salida de audio y texto en tiempo real."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "Versión en tiempo real de GPT-4o, que admite entrada y salida de audio y texto en tiempo real."
+ },
"grok-2-1212": {
"description": "Este modelo ha mejorado en precisión, cumplimiento de instrucciones y capacidades multilingües."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "El último modelo FunctionCall de Hunyuan con arquitectura MOE, entrenado con datos de FunctionCall de alta calidad, con una ventana de contexto de 32K, liderando en múltiples dimensiones de métricas de evaluación."
},
+ "hunyuan-large": {
+ "description": "El modelo Hunyuan-large tiene un total de aproximadamente 389B de parámetros, con aproximadamente 52B de parámetros activados, siendo el modelo MoE de código abierto con la mayor escala de parámetros y el mejor rendimiento en la arquitectura Transformer en la industria actual."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Especializado en tareas de texto largo como resúmenes de documentos y preguntas y respuestas de documentos, también tiene la capacidad de manejar tareas generales de generación de texto. Destaca en el análisis y generación de textos largos, pudiendo abordar eficazmente las necesidades de procesamiento de contenido largo y complejo."
+ },
"hunyuan-lite": {
"description": "Actualizado a una estructura MOE, con una ventana de contexto de 256k, lidera en múltiples conjuntos de evaluación en NLP, código, matemáticas, industria y más, superando a muchos modelos de código abierto."
},
+ "hunyuan-lite-vision": {
+ "description": "El modelo multimodal más reciente de 7B de Hunyuan, con una ventana de contexto de 32K, soporta diálogos multimodales en chino e inglés, reconocimiento de objetos en imágenes, comprensión de documentos y tablas, matemáticas multimodales, entre otros, superando a modelos competidores de 7B en múltiples dimensiones de evaluación."
+ },
"hunyuan-pro": {
"description": "Modelo de texto largo MOE-32K con un tamaño de parámetros de billones. Alcanzando niveles de liderazgo absoluto en varios benchmarks, con capacidades complejas de instrucciones y razonamiento, habilidades matemáticas complejas, soporte para llamadas a funciones, optimizado para aplicaciones en traducción multilingüe, finanzas, derecho y medicina."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Adopta una estrategia de enrutamiento mejorada, al tiempo que mitiga problemas de equilibrio de carga y convergencia de expertos. En el caso de textos largos, el índice de precisión alcanza el 99.9%. MOE-256K rompe barreras en longitud y efectividad, ampliando enormemente la longitud de entrada permitida."
},
+ "hunyuan-standard-vision": {
+ "description": "El modelo multimodal más reciente de Hunyuan, que soporta respuestas en múltiples idiomas, con capacidades equilibradas en chino e inglés."
+ },
"hunyuan-turbo": {
"description": "Versión preliminar de la nueva generación del modelo de lenguaje de Hunyuan, que utiliza una nueva estructura de modelo de expertos mixtos (MoE), con una eficiencia de inferencia más rápida y un rendimiento más fuerte en comparación con Hunyuan-Pro."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Versión fija de hunyuan-turbo del 20 de noviembre de 2024, una versión intermedia entre hunyuan-turbo y hunyuan-turbo-latest."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "Optimización de esta versión: escalado de instrucciones de datos, mejora significativa de la capacidad de generalización del modelo; mejora significativa de las capacidades de matemáticas, código y razonamiento lógico; optimización de la comprensión de texto y de palabras relacionadas; optimización de la calidad de generación de contenido en la creación de texto."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "Optimización de la experiencia general, incluyendo comprensión de NLP, creación de texto, conversación casual, preguntas y respuestas de conocimiento, traducción, entre otros; mejora de la humanización, optimización de la inteligencia emocional del modelo; mejora de la capacidad del modelo para aclarar proactivamente en caso de ambigüedad en la intención; mejora de la capacidad de manejo de problemas de análisis de palabras; mejora de la calidad y la interactividad de la creación; mejora de la experiencia en múltiples turnos."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "El nuevo modelo insignia de lenguaje visual de Hunyuan de nueva generación, que utiliza una nueva estructura de modelo de expertos mixtos (MoE), mejorando de manera integral las capacidades de reconocimiento básico, creación de contenido, preguntas y respuestas de conocimiento, y análisis y razonamiento en comparación con la generación anterior de modelos."
+ },
"hunyuan-vision": {
"description": "El último modelo multimodal de Hunyuan, que admite la entrada de imágenes y texto para generar contenido textual."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "Nuestra última serie de modelos, con un rendimiento de inferencia excepcional, que admite una longitud de contexto de 1M y una mayor capacidad de seguimiento de instrucciones y llamadas a herramientas."
},
+ "internlm3-latest": {
+ "description": "Nuestra última serie de modelos, con un rendimiento de inferencia excepcional, lidera el mercado de modelos de código abierto de tamaño similar. Apunta por defecto a nuestra serie de modelos InternLM3 más reciente."
+ },
+ "jina-deepsearch-v1": {
+ "description": "La búsqueda profunda combina la búsqueda en la web, la lectura y el razonamiento para realizar investigaciones exhaustivas. Puedes considerarlo como un agente que acepta tus tareas de investigación: realiza una búsqueda amplia y pasa por múltiples iteraciones antes de proporcionar una respuesta. Este proceso implica una investigación continua, razonamiento y resolución de problemas desde diferentes ángulos. Esto es fundamentalmente diferente de los grandes modelos estándar que generan respuestas directamente a partir de datos preentrenados y de los sistemas RAG tradicionales que dependen de búsquedas superficiales únicas."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM es un modelo de lenguaje experimental y específico para tareas, entrenado para cumplir con los principios de la ciencia del aprendizaje, capaz de seguir instrucciones sistemáticas en escenarios de enseñanza y aprendizaje, actuando como un tutor experto, entre otros."
},
"lite": {
"description": "Spark Lite es un modelo de lenguaje grande y ligero, con una latencia extremadamente baja y una capacidad de procesamiento eficiente, completamente gratuito y de código abierto, que admite funciones de búsqueda en línea en tiempo real. Su característica de respuesta rápida lo hace destacar en aplicaciones de inferencia y ajuste de modelos en dispositivos de baja potencia, brindando a los usuarios una excelente relación costo-beneficio y experiencia inteligente, especialmente en escenarios de preguntas y respuestas, generación de contenido y búsqueda."
},
- "llama-3.1-70b-instruct": {
- "description": "El modelo Llama 3.1 70B Instruct, con 70B de parámetros, puede ofrecer un rendimiento excepcional en tareas de generación de texto y de instrucciones a gran escala."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B ofrece una capacidad de razonamiento AI más potente, adecuada para aplicaciones complejas, soportando un procesamiento computacional extenso y garantizando eficiencia y precisión."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B es un modelo de alto rendimiento que ofrece una rápida capacidad de generación de texto, ideal para aplicaciones que requieren eficiencia a gran escala y rentabilidad."
},
- "llama-3.1-8b-instruct": {
- "description": "El modelo Llama 3.1 8B Instruct, con 8B de parámetros, soporta la ejecución eficiente de tareas de instrucciones visuales, ofreciendo una excelente capacidad de generación de texto."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "El modelo Llama 3.1 Sonar Huge Online, con 405B de parámetros, soporta una longitud de contexto de aproximadamente 127,000 tokens, diseñado para aplicaciones de chat en línea complejas."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "El modelo Llama 3.1 Sonar Large Chat, con 70B de parámetros, soporta una longitud de contexto de aproximadamente 127,000 tokens, adecuado para tareas de chat fuera de línea complejas."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "El modelo Llama 3.1 Sonar Large Online, con 70B de parámetros, soporta una longitud de contexto de aproximadamente 127,000 tokens, adecuado para tareas de chat de alta capacidad y diversidad."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "El modelo Llama 3.1 Sonar Small Chat, con 8B de parámetros, está diseñado para chat fuera de línea, soportando una longitud de contexto de aproximadamente 127,000 tokens."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "El modelo Llama 3.1 Sonar Small Online, con 8B de parámetros, soporta una longitud de contexto de aproximadamente 127,000 tokens, diseñado para chat en línea, capaz de manejar eficientemente diversas interacciones textuales."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 está diseñado para manejar tareas que combinan datos visuales y textuales. Destaca en tareas como la descripción de imágenes y preguntas visuales, cruzando la brecha entre la generación de lenguaje y el razonamiento visual."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 es el modelo de lenguaje de código abierto multilingüe más avanzado de la serie Llama, que ofrece un rendimiento comparable al modelo de 405B a un costo extremadamente bajo. Basado en la estructura Transformer, y mejorado en utilidad y seguridad a través de ajuste fino supervisado (SFT) y aprendizaje por refuerzo con retroalimentación humana (RLHF). Su versión ajustada para instrucciones está optimizada para diálogos multilingües, superando a muchos modelos de chat de código abierto y cerrado en múltiples benchmarks de la industria. La fecha límite de conocimiento es diciembre de 2023."
+ },
"llama-3.3-70b-versatile": {
"description": "El modelo de lenguaje multilingüe Meta Llama 3.3 (LLM) es un modelo generativo preentrenado y ajustado para instrucciones de 70B (entrada/salida de texto). El modelo de texto puro ajustado para instrucciones de Llama 3.3 está optimizado para casos de uso de conversación multilingüe y supera a muchos modelos de chat de código abierto y cerrado en benchmarks industriales comunes."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 está diseñado para manejar tareas que combinan datos visuales y textuales. Se destaca en tareas como descripción de imágenes y preguntas visuales, cruzando la brecha entre la generación de lenguaje y el razonamiento visual."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 es el modelo de lenguaje de código abierto multilingüe más avanzado de la serie Llama, que ofrece un rendimiento comparable al modelo de 405B a un costo muy bajo. Basado en la estructura Transformer, y mejorado en utilidad y seguridad a través de ajuste fino supervisado (SFT) y aprendizaje por refuerzo con retroalimentación humana (RLHF). Su versión ajustada por instrucciones está optimizada para diálogos multilingües, superando a muchos modelos de chat de código abierto y cerrado en múltiples benchmarks de la industria. La fecha de corte de conocimiento es diciembre de 2023."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "El modelo de lenguaje grande multilingüe Meta Llama 3.3 (LLM) es un modelo generativo preentrenado y ajustado por instrucciones de 70B (entrada de texto/salida de texto). El modelo de texto puro ajustado por instrucciones de Llama 3.3 está optimizado para casos de uso de diálogo multilingüe y supera a muchos modelos de chat de código abierto y cerrados en benchmarks de la industria."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 está diseñado para manejar tareas que combinan datos visuales y textuales. Se destaca en tareas como descripción de imágenes y preguntas visuales, cruzando la brecha entre la generación de lenguaje y el razonamiento visual."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "El modelo Llama 3.1 Turbo de 405B proporciona un soporte de contexto de gran capacidad para el procesamiento de grandes datos, destacándose en aplicaciones de inteligencia artificial a gran escala."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 es el modelo líder lanzado por Meta, que soporta hasta 405B de parámetros, aplicable en diálogos complejos, traducción multilingüe y análisis de datos."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B proporciona soporte de conversación eficiente en múltiples idiomas."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 está diseñado para manejar tareas que combinan datos visuales y textuales. Destaca en tareas como la descripción de imágenes y preguntas visuales, superando la brecha entre la generación de lenguaje y el razonamiento visual."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 es el modelo de lenguaje de código abierto multilingüe más avanzado de la serie Llama, que ofrece un rendimiento comparable al modelo de 405B a un costo extremadamente bajo. Basado en la estructura Transformer, y mejorado en utilidad y seguridad a través de ajuste fino supervisado (SFT) y aprendizaje por refuerzo con retroalimentación humana (RLHF). Su versión ajustada para instrucciones está optimizada para diálogos multilingües, superando a muchos modelos de chat de código abierto y cerrado en múltiples benchmarks de la industria. La fecha límite de conocimiento es diciembre de 2023."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 es el modelo de lenguaje de código abierto multilingüe más avanzado de la serie Llama, que ofrece un rendimiento comparable al modelo de 405B a un costo extremadamente bajo. Basado en la estructura Transformer, y mejorado en utilidad y seguridad a través de ajuste fino supervisado (SFT) y aprendizaje por refuerzo con retroalimentación humana (RLHF). Su versión ajustada para instrucciones está optimizada para diálogos multilingües, superando a muchos modelos de chat de código abierto y cerrado en múltiples benchmarks de la industria. La fecha límite de conocimiento es diciembre de 2023."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct es el modelo más grande y potente de la serie Llama 3.1 Instruct, un modelo de generación de datos de diálogo y razonamiento altamente avanzado, que también puede servir como base para un preentrenamiento o ajuste fino especializado en dominios específicos. Los modelos de lenguaje de gran tamaño (LLMs) multilingües que ofrece Llama 3.1 son un conjunto de modelos generativos preentrenados y ajustados por instrucciones, que incluyen tamaños de 8B, 70B y 405B (entrada/salida de texto). Los modelos de texto ajustados por instrucciones de Llama 3.1 (8B, 70B, 405B) están optimizados para casos de uso de diálogo multilingüe y superan a muchos modelos de chat de código abierto disponibles en pruebas de referencia de la industria. Llama 3.1 está diseñado para usos comerciales y de investigación en múltiples idiomas. Los modelos de texto ajustados por instrucciones son adecuados para chats similares a asistentes, mientras que los modelos preentrenados pueden adaptarse a diversas tareas de generación de lenguaje natural. El modelo Llama 3.1 también admite el uso de su salida para mejorar otros modelos, incluida la generación de datos sintéticos y el refinamiento. Llama 3.1 es un modelo de lenguaje autorregresivo que utiliza una arquitectura de transformador optimizada. Las versiones ajustadas utilizan ajuste fino supervisado (SFT) y aprendizaje por refuerzo con retroalimentación humana (RLHF) para alinearse con las preferencias humanas de ayuda y seguridad."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 es un modelo de lenguaje de gran tamaño (LLM) abierto dirigido a desarrolladores, investigadores y empresas, diseñado para ayudarles a construir, experimentar y escalar de manera responsable sus ideas de IA generativa. Como parte de un sistema base para la innovación de la comunidad global, es ideal para dispositivos de borde con recursos y capacidades computacionales limitadas, así como para tiempos de entrenamiento más rápidos."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "LLM avanzado, que soporta generación de datos sintéticos, destilación de conocimiento y razonamiento, adecuado para chatbots, programación y tareas de dominio específico."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "Potencia diálogos complejos, con excelente comprensión del contexto, capacidad de razonamiento y generación de texto."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "Modelo de última generación avanzado, con comprensión del lenguaje, excelente capacidad de razonamiento y generación de texto."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "Modelo de visión-lenguaje de vanguardia, experto en razonamiento de alta calidad a partir de imágenes."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "Modelo de lenguaje pequeño de última generación, con comprensión del lenguaje, excelente capacidad de razonamiento y generación de texto."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "Modelo de lenguaje pequeño de última generación, con comprensión del lenguaje, excelente capacidad de razonamiento y generación de texto."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "Modelo de visión-lenguaje de vanguardia, experto en razonamiento de alta calidad a partir de imágenes."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "Modelo LLM avanzado, experto en razonamiento, matemáticas, sentido común y llamadas a funciones."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 es un modelo de lenguaje proporcionado por Microsoft AI, que destaca en diálogos complejos, multilingüismo, razonamiento y asistentes inteligentes."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K es un modelo con capacidad de procesamiento de contexto ultra largo, adecuado para generar textos extensos, satisfaciendo las demandas de tareas de generación complejas, capaz de manejar hasta 128,000 tokens, ideal para aplicaciones en investigación, académicas y generación de documentos grandes."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "El modelo visual Kimi (incluyendo moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, etc.) puede entender el contenido de las imágenes, incluyendo texto en imágenes, colores de imágenes y formas de objetos."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K ofrece capacidad de procesamiento de contexto de longitud media, capaz de manejar 32,768 tokens, especialmente adecuado para generar diversos documentos largos y diálogos complejos, aplicable en creación de contenido, generación de informes y sistemas de diálogo."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "El modelo visual Kimi (incluyendo moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, etc.) puede entender el contenido de las imágenes, incluyendo texto en imágenes, colores de imágenes y formas de objetos."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K está diseñado para tareas de generación de texto corto, con un rendimiento de procesamiento eficiente, capaz de manejar 8,192 tokens, ideal para diálogos breves, toma de notas y generación rápida de contenido."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "El modelo visual Kimi (incluyendo moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, etc.) puede entender el contenido de las imágenes, incluyendo texto en imágenes, colores de imágenes y formas de objetos."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B es una versión mejorada de Nous Hermes 2, que incluye los conjuntos de datos más recientes desarrollados internamente."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B es un modelo de lenguaje de gran tamaño personalizado por NVIDIA, diseñado para aumentar el grado de ayuda que las respuestas generadas por LLM ofrecen a las consultas de los usuarios."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B es un modelo de lenguaje a gran escala personalizado por NVIDIA, diseñado para mejorar la utilidad de las respuestas generadas por LLM a las consultas de los usuarios. Este modelo ha destacado en pruebas de referencia como Arena Hard, AlpacaEval 2 LC y GPT-4-Turbo MT-Bench, ocupando el primer lugar en los tres benchmarks de alineación automática hasta el 1 de octubre de 2024. El modelo se entrena utilizando RLHF (especialmente REINFORCE), Llama-3.1-Nemotron-70B-Reward y HelpSteer2-Preference sobre la base del modelo Llama-3.1-70B-Instruct."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "Modelo de lenguaje único, que ofrece una precisión y eficiencia inigualables."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct es un modelo de lenguaje grande personalizado por NVIDIA, diseñado para mejorar la utilidad de las respuestas generadas por LLM."
+ },
"o1": {
"description": "Se centra en el razonamiento avanzado y la resolución de problemas complejos, incluidas tareas matemáticas y científicas. Es muy adecuado para aplicaciones que requieren una comprensión profunda del contexto y flujos de trabajo de agentes."
},
- "o1-2024-12-17": {
- "description": "o1 es el nuevo modelo de inferencia de OpenAI, que admite entradas de texto e imagen y produce texto como salida, adecuado para tareas complejas que requieren un amplio conocimiento general. Este modelo tiene un contexto de 200K y una fecha límite de conocimiento de octubre de 2023."
- },
"o1-mini": {
"description": "o1-mini es un modelo de inferencia rápido y rentable diseñado para aplicaciones de programación, matemáticas y ciencias. Este modelo tiene un contexto de 128K y una fecha de corte de conocimiento en octubre de 2023."
},
"o1-preview": {
"description": "o1 es el nuevo modelo de inferencia de OpenAI, adecuado para tareas complejas que requieren un amplio conocimiento general. Este modelo tiene un contexto de 128K y una fecha de corte de conocimiento en octubre de 2023."
},
+ "o3-mini": {
+ "description": "o3-mini es nuestro último modelo de inferencia de tamaño pequeño, que ofrece alta inteligencia con los mismos objetivos de costo y latencia que o1-mini."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba es un modelo de lenguaje Mamba 2 enfocado en la generación de código, que proporciona un fuerte apoyo para tareas avanzadas de codificación y razonamiento."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini es el modelo más reciente de OpenAI, lanzado después de GPT-4 Omni, que admite entradas de texto e imagen y genera texto como salida. Como su modelo más avanzado de tamaño pequeño, es mucho más económico que otros modelos de vanguardia recientes y más de un 60% más barato que GPT-3.5 Turbo. Mantiene una inteligencia de vanguardia mientras ofrece una relación calidad-precio notable. GPT-4o mini obtuvo un puntaje del 82% en la prueba MMLU y actualmente se clasifica por encima de GPT-4 en preferencias de chat."
},
- "openai/o1": {
- "description": "o1 es el nuevo modelo de inferencia de OpenAI, que admite entradas de texto e imagen y produce texto como salida, adecuado para tareas complejas que requieren un amplio conocimiento general. Este modelo tiene un contexto de 200K y una fecha límite de conocimiento de octubre de 2023."
- },
"openai/o1-mini": {
"description": "o1-mini es un modelo de inferencia rápido y rentable diseñado para aplicaciones de programación, matemáticas y ciencias. Este modelo tiene un contexto de 128K y una fecha de corte de conocimiento en octubre de 2023."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K está equipado con una capacidad de procesamiento de contexto extragrande, capaz de manejar hasta 128K de información contextual, especialmente adecuado para el análisis completo y el manejo de relaciones lógicas a largo plazo en contenido extenso, proporcionando una lógica fluida y coherente y un soporte diverso de citas en comunicaciones de texto complejas."
},
+ "qvq-72b-preview": {
+ "description": "El modelo QVQ es un modelo de investigación experimental desarrollado por el equipo de Qwen, enfocado en mejorar la capacidad de razonamiento visual, especialmente en el ámbito del razonamiento matemático."
+ },
"qwen-coder-plus-latest": {
"description": "Modelo de código Qwen de Tongyi."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Modelo de lenguaje visual a ultra gran escala Tongyi Qianwen. En comparación con la versión mejorada, mejora aún más la capacidad de razonamiento visual y de seguimiento de instrucciones, ofreciendo un nivel más alto de percepción y cognición visual."
},
+ "qwen-vl-ocr-latest": {
+ "description": "Qwen OCR es un modelo especializado en extracción de texto, enfocado en la capacidad de extraer texto de imágenes de documentos, tablas, exámenes, escritura a mano, entre otros. Puede reconocer múltiples idiomas, actualmente soporta: chino, inglés, francés, japonés, coreano, alemán, ruso, italiano, vietnamita y árabe."
+ },
"qwen-vl-plus-latest": {
"description": "Versión mejorada del modelo de lenguaje visual a gran escala Tongyi Qianwen. Mejora significativamente la capacidad de reconocimiento de detalles y de texto, soportando imágenes con resolución de más de un millón de píxeles y proporciones de ancho y alto arbitrarias."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 es una nueva serie de modelos de lenguaje de gran tamaño, con una mayor capacidad de comprensión y generación."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "LLM orientado a chino e inglés, enfocado en áreas como lenguaje, programación, matemáticas y razonamiento."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "LLM avanzado, que soporta generación de código, razonamiento y corrección, abarcando lenguajes de programación populares."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "Poderoso modelo de código de tamaño mediano, que soporta longitudes de contexto de 32K, experto en programación multilingüe."
+ },
"qwen2": {
"description": "Qwen2 es el nuevo modelo de lenguaje a gran escala de Alibaba, que ofrece un rendimiento excepcional para satisfacer diversas necesidades de aplicación."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "El modelo de 14B de Tongyi Qwen 2.5, de código abierto."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "El modelo de 72B de Qwen2.5 es de código abierto."
+ },
"qwen2.5-32b-instruct": {
"description": "El modelo de 32B de Tongyi Qwen 2.5, de código abierto."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "El modelo Qwen-Math tiene una poderosa capacidad para resolver problemas matemáticos."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "Mejora general en seguimiento de instrucciones, matemáticas, resolución de problemas y código, con capacidades de reconocimiento de objetos mejoradas, soporta formatos diversos para localizar elementos visuales con precisión, y puede entender archivos de video largos (hasta 10 minutos) y localizar eventos en segundos, comprendiendo la secuencia y velocidad del tiempo, soportando el control de agentes en OS o móviles, con fuerte capacidad de extracción de información clave y salida en formato Json. Esta versión es la de 72B, la más potente de la serie."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "Mejora general en seguimiento de instrucciones, matemáticas, resolución de problemas y código, con capacidades de reconocimiento de objetos mejoradas, soporta formatos diversos para localizar elementos visuales con precisión, y puede entender archivos de video largos (hasta 10 minutos) y localizar eventos en segundos, comprendiendo la secuencia y velocidad del tiempo, soportando el control de agentes en OS o móviles, con fuerte capacidad de extracción de información clave y salida en formato Json. Esta versión es la de 72B, la más potente de la serie."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 es la nueva generación de modelos de lenguaje a gran escala de Alibaba, que ofrece un rendimiento excepcional para satisfacer diversas necesidades de aplicación."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro es un LLM de alta inteligencia lanzado por Upstage, enfocado en la capacidad de seguimiento de instrucciones en un solo GPU, con una puntuación IFEval superior a 80. Actualmente soporta inglés, y se planea lanzar la versión oficial en noviembre de 2024, ampliando el soporte de idiomas y la longitud del contexto."
},
+ "sonar": {
+ "description": "Producto de búsqueda ligero basado en contexto de búsqueda, más rápido y económico que Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "Producto de búsqueda avanzada que soporta contexto de búsqueda, consultas avanzadas y seguimiento."
+ },
+ "sonar-reasoning": {
+ "description": "Nuevo producto API respaldado por el modelo de razonamiento de DeepSeek."
+ },
"step-1-128k": {
"description": "Equilibrio entre rendimiento y costo, adecuado para escenarios generales."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "Este modelo tiene una potente capacidad de comprensión de video."
},
+ "step-1o-vision-32k": {
+ "description": "Este modelo posee una poderosa capacidad de comprensión de imágenes. En comparación con la serie de modelos step-1v, ofrece un rendimiento visual superior."
+ },
"step-1v-32k": {
"description": "Soporta entradas visuales, mejorando la experiencia de interacción multimodal."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Soporta interacciones de contexto a gran escala, adecuado para escenarios de diálogo complejos."
},
+ "step-2-mini": {
+ "description": "Un modelo de gran velocidad basado en la nueva arquitectura de atención autogestionada MFA, que logra efectos similares a los de step1 a un costo muy bajo, manteniendo al mismo tiempo un mayor rendimiento y tiempos de respuesta más rápidos. Capaz de manejar tareas generales, con habilidades destacadas en programación."
+ },
+ "taichu2_mm": {
+ "description": "Integra capacidades de comprensión de imágenes, transferencia de conocimiento, atribución lógica, destacándose en el campo de preguntas y respuestas basadas en texto e imagen."
+ },
"taichu_llm": {
"description": "El modelo de lenguaje Taichu de Zīdōng tiene una poderosa capacidad de comprensión del lenguaje, así como habilidades en creación de textos, preguntas y respuestas, programación de código, cálculos matemáticos, razonamiento lógico, análisis de sentimientos y resúmenes de texto. Combina de manera innovadora el preentrenamiento con grandes datos y un conocimiento rico de múltiples fuentes, perfeccionando continuamente la tecnología algorítmica y absorbiendo nuevos conocimientos en vocabulario, estructura, gramática y semántica de grandes volúmenes de datos textuales, logrando una evolución constante del modelo. Proporciona a los usuarios información y servicios más convenientes, así como una experiencia más inteligente."
},
+ "text-embedding-3-large": {
+ "description": "El modelo de vectorización más potente, adecuado para tareas en inglés y no inglés."
+ },
+ "text-embedding-3-small": {
+ "description": "Un modelo de Embedding de nueva generación, eficiente y económico, adecuado para la recuperación de conocimiento, aplicaciones RAG y más."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) proporciona una capacidad de cálculo mejorada a través de estrategias y arquitecturas de modelos eficientes."
},
+ "tts-1": {
+ "description": "El modelo más reciente de texto a voz, optimizado para velocidad en escenarios en tiempo real."
+ },
+ "tts-1-hd": {
+ "description": "El modelo más reciente de texto a voz, optimizado para calidad."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) es adecuado para tareas de instrucciones detalladas, ofreciendo una excelente capacidad de procesamiento de lenguaje."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet eleva el estándar de la industria, superando a modelos competidores y a Claude 3 Opus, destacándose en evaluaciones amplias, mientras mantiene la velocidad y costo de nuestros modelos de nivel medio."
},
+ "whisper-1": {
+ "description": "Modelo de reconocimiento de voz general, que admite reconocimiento de voz multilingüe, traducción de voz y reconocimiento de idiomas."
+ },
"wizardlm2": {
"description": "WizardLM 2 es un modelo de lenguaje proporcionado por Microsoft AI, que destaca en diálogos complejos, multilingües, razonamiento y asistentes inteligentes."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Modelo para tareas visuales complejas, que ofrece un alto rendimiento en comprensión y análisis de imágenes."
+ },
+ "yi-vision-v2": {
+ "description": "Modelo para tareas visuales complejas, que ofrece capacidades de comprensión y análisis de alto rendimiento basadas en múltiples imágenes."
}
}
diff --git a/locales/es-ES/providers.json b/locales/es-ES/providers.json
index 31602edcc04fd..7bcbd8cb2795c 100644
--- a/locales/es-ES/providers.json
+++ b/locales/es-ES/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure ofrece una variedad de modelos de IA avanzados, incluidos GPT-3.5 y la última serie GPT-4, que admiten múltiples tipos de datos y tareas complejas, comprometidos con soluciones de IA seguras, confiables y sostenibles."
},
+ "azureai": {
+ "description": "Azure ofrece una variedad de modelos de IA avanzados, incluidos GPT-3.5 y la última serie GPT-4, que admiten múltiples tipos de datos y tareas complejas, comprometidos con soluciones de IA seguras, confiables y sostenibles."
+ },
"baichuan": {
"description": "Baichuan Intelligent es una empresa centrada en el desarrollo de modelos de gran tamaño de inteligencia artificial, cuyos modelos han demostrado un rendimiento excepcional en tareas en chino como enciclopedias de conocimiento, procesamiento de textos largos y creación de contenido, superando a los modelos principales extranjeros. Baichuan Intelligent también posee capacidades multimodales líderes en la industria, destacándose en múltiples evaluaciones de autoridad. Sus modelos incluyen Baichuan 4, Baichuan 3 Turbo y Baichuan 3 Turbo 128k, optimizados para diferentes escenarios de aplicación, ofreciendo soluciones de alta relación calidad-precio."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek es una empresa centrada en la investigación y aplicación de tecnologías de inteligencia artificial, cuyo modelo más reciente, DeepSeek-V2.5, combina capacidades de diálogo general y procesamiento de código, logrando mejoras significativas en alineación con preferencias humanas, tareas de escritura y seguimiento de instrucciones."
},
+ "doubao": {
+ "description": "Un modelo grande desarrollado internamente por ByteDance. Validado a través de más de 50 escenarios de negocio internos, con un uso diario de tokens en billones que se perfecciona continuamente, ofrece múltiples capacidades modales y crea experiencias comerciales ricas para las empresas con un rendimiento de modelo de alta calidad."
+ },
"fireworksai": {
"description": "Fireworks AI es un proveedor líder de servicios de modelos de lenguaje avanzados, enfocado en la llamada de funciones y el procesamiento multimodal. Su modelo más reciente, Firefunction V2, basado en Llama-3, está optimizado para llamadas de funciones, diálogos y seguimiento de instrucciones. El modelo de lenguaje visual FireLLaVA-13B admite entradas mixtas de imágenes y texto. Otros modelos notables incluyen la serie Llama y la serie Mixtral, que ofrecen un soporte eficiente para el seguimiento y generación de instrucciones multilingües."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "Organización de código abierto dedicada a la investigación y desarrollo de herramientas para modelos grandes. Proporciona a todos los desarrolladores de IA una plataforma de código abierto eficiente y fácil de usar, permitiendo el acceso a las tecnologías y algoritmos más avanzados."
},
+ "jina": {
+ "description": "Jina AI, fundada en 2020, es una empresa líder en búsqueda de IA. Nuestra plataforma de búsqueda base incluye modelos vectoriales, reordenadores y pequeños modelos de lenguaje, que ayudan a las empresas a construir aplicaciones de búsqueda generativa y multimodal confiables y de alta calidad."
+ },
+ "lmstudio": {
+ "description": "LM Studio es una aplicación de escritorio para desarrollar y experimentar con LLMs en su computadora."
+ },
"minimax": {
"description": "MiniMax es una empresa de tecnología de inteligencia artificial general fundada en 2021, dedicada a co-crear inteligencia con los usuarios. MiniMax ha desarrollado de forma independiente modelos de gran tamaño de diferentes modalidades, que incluyen un modelo de texto MoE de un billón de parámetros, un modelo de voz y un modelo de imagen. También ha lanzado aplicaciones como Conch AI."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI es una plataforma que ofrece servicios API para múltiples modelos de lenguaje de gran tamaño y generación de imágenes de IA, siendo flexible, confiable y rentable. Soporta los últimos modelos de código abierto como Llama3 y Mistral, proporcionando soluciones API completas, amigables para el usuario y autoescalables para el desarrollo de aplicaciones de IA, adecuadas para el rápido crecimiento de startups de IA."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ proporciona contenedores que se pueden utilizar para microservicios de inferencia acelerados por GPU autohospedados, admitiendo el despliegue de modelos de IA preentrenados y personalizados en la nube, centros de datos, PC RTX™ AI y estaciones de trabajo."
+ },
"ollama": {
"description": "Los modelos ofrecidos por Ollama abarcan ampliamente áreas como la generación de código, cálculos matemáticos, procesamiento multilingüe e interacciones conversacionales, apoyando diversas necesidades de implementación empresarial y local."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "El Instituto de Automatización de la Academia de Ciencias de China y el Instituto de Investigación de Inteligencia Artificial de Wuhan han lanzado una nueva generación de modelos de gran tamaño multimodal, que apoyan tareas de preguntas y respuestas de múltiples rondas, creación de texto, generación de imágenes, comprensión 3D, análisis de señales y más, con capacidades de cognición, comprensión y creación más fuertes, ofreciendo una nueva experiencia de interacción."
},
+ "tencentcloud": {
+ "description": "La capacidad atómica del motor de conocimiento (LLM Knowledge Engine Atomic Power) se basa en el desarrollo del motor de conocimiento y ofrece una capacidad completa de preguntas y respuestas, dirigida a empresas y desarrolladores, proporcionando la capacidad de construir y desarrollar aplicaciones de modelos de manera flexible. Puede ensamblar su propio servicio de modelo utilizando varias capacidades atómicas, invocando servicios de análisis de documentos, división, embedding, reescritura en múltiples turnos, entre otros, para personalizar un negocio de IA exclusivo para su empresa."
+ },
"togetherai": {
"description": "Together AI se dedica a lograr un rendimiento líder a través de modelos de IA innovadores, ofreciendo amplias capacidades de personalización, incluyendo soporte para escalado rápido y procesos de implementación intuitivos, satisfaciendo diversas necesidades empresariales."
},
"upstage": {
"description": "Upstage se centra en desarrollar modelos de IA para diversas necesidades comerciales, incluidos Solar LLM y Document AI, con el objetivo de lograr una inteligencia general artificial (AGI) que trabaje para las personas. Crea agentes de diálogo simples a través de la API de Chat y admite llamadas de funciones, traducción, incrustaciones y aplicaciones de dominio específico."
},
+ "vllm": {
+ "description": "vLLM es una biblioteca rápida y fácil de usar para la inferencia y el servicio de LLM."
+ },
+ "volcengine": {
+ "description": "Plataforma de desarrollo de servicios de modelos grandes lanzada por ByteDance, que ofrece servicios de invocación de modelos ricos en funciones, seguros y competitivos en precio, al tiempo que proporciona datos de modelos, ajuste fino, inferencia, evaluación y otras funciones de extremo a extremo, garantizando de manera integral el desarrollo y la implementación de sus aplicaciones de IA."
+ },
"wenxin": {
"description": "Plataforma de desarrollo y servicios de modelos grandes y aplicaciones nativas de IA de nivel empresarial, que ofrece la cadena de herramientas más completa y fácil de usar para el desarrollo de modelos de inteligencia artificial generativa y el desarrollo de aplicaciones en todo el proceso."
},
diff --git a/locales/es-ES/setting.json b/locales/es-ES/setting.json
index 5cfa5452bb99c..79c315c7d333a 100644
--- a/locales/es-ES/setting.json
+++ b/locales/es-ES/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Activar límite de tokens por respuesta"
},
+ "enableReasoningEffort": {
+ "title": "Activar ajuste de intensidad de razonamiento"
+ },
"frequencyPenalty": {
- "desc": "Cuanto mayor sea el valor, más probable es que se reduzcan las repeticiones de palabras",
- "title": "Penalización de frecuencia"
+ "desc": "Cuanto mayor sea el valor, más rica y variada será la elección de palabras; cuanto menor sea el valor, más simples y directas serán las palabras.",
+ "title": "Riqueza del vocabulario"
},
"maxTokens": {
"desc": "Número máximo de tokens utilizados en una interacción",
@@ -212,19 +215,31 @@
"desc": "{{provider}} modelo",
"title": "Modelo"
},
+ "params": {
+ "title": "Parámetros avanzados"
+ },
"presencePenalty": {
- "desc": "Cuanto mayor sea el valor, más probable es que se amplíe a nuevos temas",
- "title": "Penalización de novedad del tema"
+ "desc": "Cuanto mayor sea el valor, más se inclinará hacia diferentes formas de expresión, evitando la repetición de conceptos; cuanto menor sea el valor, más se inclinará hacia el uso de conceptos o narrativas repetidas, expresando mayor consistencia.",
+ "title": "Diversidad de expresión"
+ },
+ "reasoningEffort": {
+ "desc": "Cuanto mayor sea el valor, más fuerte será la capacidad de razonamiento, pero puede aumentar el tiempo de respuesta y el consumo de tokens.",
+ "options": {
+ "high": "Alto",
+ "low": "Bajo",
+ "medium": "Medio"
+ },
+ "title": "Intensidad de razonamiento"
},
"temperature": {
- "desc": "Cuanto mayor sea el valor, más aleatoria será la respuesta",
- "title": "Temperatura",
- "titleWithValue": "Temperatura {{value}}"
+ "desc": "Cuanto mayor sea el valor, más creativas e imaginativas serán las respuestas; cuanto menor sea el valor, más rigurosas serán las respuestas",
+ "title": "Nivel de creatividad",
+ "warning": "Un valor de creatividad demasiado alto puede generar salidas confusas"
},
"title": "Configuración del modelo",
"topP": {
- "desc": "Similar a la temperatura, pero no se debe cambiar junto con la temperatura",
- "title": "Muestreo de núcleo"
+ "desc": "Cuántas posibilidades se consideran, cuanto mayor sea el valor, más respuestas posibles se aceptan; cuanto menor sea el valor, se tiende a elegir la respuesta más probable. No se recomienda cambiarlo junto con el nivel de creatividad",
+ "title": "Apertura mental"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Configuración común",
"experiment": "Experimento",
"llm": "Modelo de lenguaje",
+ "provider": "Proveedor de servicios de IA",
"sync": "Sincronización en la nube",
"system-agent": "Asistente del sistema",
"tts": "Servicio de voz"
diff --git a/locales/fa-IR/changelog.json b/locales/fa-IR/changelog.json
index bad13c844875c..e70752a47d27d 100644
--- a/locales/fa-IR/changelog.json
+++ b/locales/fa-IR/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "تمام تغییرات را مشاهده کنید",
"description": "بهروزرسانیهای جدید و بهبودهای {{appName}} را دنبال کنید",
"pagination": {
- "older": "مشاهده تغییرات قبلی",
- "prev": "صفحه قبلی"
+ "next": "صفحه بعدی",
+ "older": "مشاهده تغییرات قبلی"
},
"readDetails": "جزئیات را بخوانید",
"title": "تغییرات",
diff --git a/locales/fa-IR/common.json b/locales/fa-IR/common.json
index c31a94b510ca9..b9a4cf0b1552c 100644
--- a/locales/fa-IR/common.json
+++ b/locales/fa-IR/common.json
@@ -275,6 +275,7 @@
},
"temp": "موقت",
"terms": "شرایط خدمات",
+ "update": "بهروزرسانی",
"updateAgent": "بهروزرسانی اطلاعات دستیار",
"upgradeVersion": {
"action": "ارتقاء",
@@ -285,6 +286,7 @@
"anonymousNickName": "کاربر ناشناس",
"billing": "مدیریت صورتحساب",
"cloud": "تجربه {{name}}",
+ "community": "نسخه جامعه",
"data": "ذخیرهسازی داده",
"defaultNickname": "کاربر نسخه جامعه",
"discord": "پشتیبانی جامعه",
@@ -294,7 +296,6 @@
"help": "مرکز راهنما",
"moveGuide": "دکمه تنظیمات به اینجا منتقل شد",
"plans": "طرحهای اشتراک",
- "preview": "نسخه پیشنمایش",
"profile": "مدیریت حساب",
"setting": "تنظیمات برنامه",
"usages": "آمار استفاده"
diff --git a/locales/fa-IR/components.json b/locales/fa-IR/components.json
index 5f363085fdaee..a1cb4af04315b 100644
--- a/locales/fa-IR/components.json
+++ b/locales/fa-IR/components.json
@@ -76,6 +76,7 @@
"custom": "مدل سفارشی، تنظیمات پیشفرض از فراخوانی توابع و تشخیص بصری پشتیبانی میکند، لطفاً قابلیتهای فوق را بر اساس شرایط واقعی بررسی کنید",
"file": "این مدل از بارگذاری و شناسایی فایلها پشتیبانی میکند",
"functionCall": "این مدل از فراخوانی توابع (Function Call) پشتیبانی میکند",
+ "reasoning": "این مدل از تفکر عمیق پشتیبانی میکند",
"tokens": "این مدل در هر جلسه حداکثر از {{tokens}} توکن پشتیبانی میکند",
"vision": "این مدل از تشخیص بصری پشتیبانی میکند"
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "هیچ مدلی فعال نیست، لطفاً به تنظیمات بروید و آن را فعال کنید",
"provider": "ارائهدهنده"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "به دلیل محدودیتهای امنیتی مرورگر، شما باید تنظیمات跨域 برای Ollama را انجام دهید تا بتوانید به درستی از آن استفاده کنید.",
+ "linux": {
+ "env": "در بخش [Service]، `Environment` را اضافه کنید و متغیر محیطی OLLAMA_ORIGINS را اضافه کنید:",
+ "reboot": "systemd را بارگذاری مجدد کرده و Ollama را راهاندازی مجدد کنید",
+ "systemd": "برای ویرایش سرویس ollama از systemd استفاده کنید:"
+ },
+ "macos": "لطفاً برنامه «ترمینال» را باز کرده و دستورات زیر را کپی کرده و با فشار دادن Enter اجرا کنید",
+ "reboot": "لطفاً پس از اتمام اجرا، سرویس Ollama را راهاندازی مجدد کنید",
+ "title": "تنظیمات Ollama برای اجازه دسترسی跨域",
+ "windows": "در ویندوز، بر روی «کنترل پنل» کلیک کنید و به ویرایش متغیرهای محیطی سیستم بروید. برای حساب کاربری خود یک متغیر محیطی جدید به نام «OLLAMA_ORIGINS» با مقدار * ایجاد کنید و بر روی «OK/اعمال» کلیک کنید تا ذخیره شود."
+ },
+ "install": {
+ "description": "لطفاً اطمینان حاصل کنید که Ollama را فعال کردهاید. اگر Ollama را دانلود نکردهاید، لطفاً به وبسایت رسمی <1>دانلود بروید.",
+ "docker": "اگر تمایل دارید از Docker استفاده کنید، Ollama همچنین تصویر Docker رسمی را ارائه میدهد که میتوانید با استفاده از دستور زیر آن را بارگیری کنید:",
+ "linux": {
+ "command": "با استفاده از دستور زیر نصب کنید:",
+ "manual": "یا میتوانید به <1>راهنمای نصب دستی لینوکس مراجعه کنید و خودتان نصب کنید."
+ },
+ "title": "نصب و راهاندازی محلی برنامه Ollama",
+ "windowsTab": "ویندوز (نسخه پیشنمایش)"
+ }
+ },
+ "Thinking": {
+ "thinking": "در حال تفکر عمیق...",
+ "thought": "بهطور عمیق فکر شده است (مدت زمان {{duration}} ثانیه)",
+ "thoughtWithDuration": "بهطور عمیق فکر شده است"
}
}
diff --git a/locales/fa-IR/discover.json b/locales/fa-IR/discover.json
index c44d426d0b42e..486075494a6a7 100644
--- a/locales/fa-IR/discover.json
+++ b/locales/fa-IR/discover.json
@@ -126,6 +126,10 @@
"title": "تازگی موضوع"
},
"range": "محدوده",
+ "reasoning_effort": {
+ "desc": "این تنظیم برای کنترل شدت استدلال مدل قبل از تولید پاسخ استفاده میشود. شدت پایین به سرعت پاسخدهی اولویت میدهد و توکن را صرفهجویی میکند، در حالی که شدت بالا استدلال کاملتری ارائه میدهد اما توکن بیشتری مصرف کرده و سرعت پاسخدهی را کاهش میدهد. مقدار پیشفرض متوسط است که تعادل بین دقت استدلال و سرعت پاسخدهی را برقرار میکند.",
+ "title": "شدت استدلال"
+ },
"temperature": {
"desc": "این تنظیمات بر تنوع پاسخهای مدل تأثیر میگذارد. مقادیر پایینتر منجر به پاسخهای قابل پیشبینیتر و معمولیتر میشود، در حالی که مقادیر بالاتر تنوع و پاسخهای غیرمعمولتر را تشویق میکند. وقتی مقدار به 0 تنظیم شود، مدل همیشه برای ورودی داده شده یک پاسخ یکسان ارائه میدهد.",
"title": "تصادفی بودن"
diff --git a/locales/fa-IR/modelProvider.json b/locales/fa-IR/modelProvider.json
index 5dc0afc74777f..bc96e9d3fb902 100644
--- a/locales/fa-IR/modelProvider.json
+++ b/locales/fa-IR/modelProvider.json
@@ -19,6 +19,24 @@
"title": "کلید API"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "نسخه API آژور، با فرمت YYYY-MM-DD، برای مشاهده [آخرین نسخه](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "دریافت لیست",
+ "title": "نسخه API آژور"
+ },
+ "endpoint": {
+ "desc": "نقطه پایانی استنتاج مدل آژور AI را از نمای کلی پروژه آژور AI پیدا کنید",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "نقطه پایانی آژور AI"
+ },
+ "title": "آژور OpenAI",
+ "token": {
+ "desc": "کلید API را از نمای کلی پروژه آژور AI پیدا کنید",
+ "placeholder": "کلید آژور",
+ "title": "کلید"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "AWS Access Key Id را وارد کنید",
@@ -63,6 +81,46 @@
"title": "شناسه حساب Cloudflare / آدرس API"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "لطفاً کلید API خود را وارد کنید",
+ "title": "کلید API"
+ },
+ "basicTitle": "اطلاعات پایه",
+ "configTitle": "اطلاعات پیکربندی",
+ "confirm": "ایجاد جدید",
+ "createSuccess": "ایجاد با موفقیت انجام شد",
+ "description": {
+ "placeholder": "توضیحات ارائهدهنده (اختیاری)",
+ "title": "توضیحات ارائهدهنده"
+ },
+ "id": {
+ "desc": "به عنوان شناسه منحصر به فرد ارائهدهنده خدمات، پس از ایجاد قابل ویرایش نخواهد بود",
+ "format": "فقط میتواند شامل اعداد، حروف کوچک، خط تیره (-) و زیرخط (_) باشد",
+ "placeholder": "توصیه میشود تماماً با حروف کوچک باشد، مانند openai، پس از ایجاد قابل ویرایش نخواهد بود",
+ "required": "لطفاً شناسه ارائهدهنده را وارد کنید",
+ "title": "شناسه ارائهدهنده"
+ },
+ "logo": {
+ "required": "لطفاً لوگوی صحیح ارائهدهنده را بارگذاری کنید",
+ "title": "لوگوی ارائهدهنده"
+ },
+ "name": {
+ "placeholder": "لطفاً نام نمایشی ارائهدهنده را وارد کنید",
+ "required": "لطفاً نام ارائهدهنده را وارد کنید",
+ "title": "نام ارائهدهنده"
+ },
+ "proxyUrl": {
+ "required": "لطفاً آدرس پروکسی را وارد کنید",
+ "title": "آدرس پروکسی"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "لطفاً نوع SDK را انتخاب کنید",
+ "title": "فرمت درخواست"
+ },
+ "title": "ایجاد ارائهدهنده AI سفارشی"
+ },
"github": {
"personalAccessToken": {
"desc": "توکن دسترسی شخصی Github خود را وارد کنید، برای ایجاد [اینجا](https://github.com/settings/tokens) کلیک کنید",
@@ -77,6 +135,23 @@
"title": "توکن HuggingFace"
}
},
+ "list": {
+ "title": {
+ "disabled": "سرویسدهنده غیرفعال",
+ "enabled": "سرویسدهنده فعال"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "اضافه کردن ارائهدهنده سفارشی",
+ "all": "همه",
+ "list": {
+ "disabled": "غیرفعال",
+ "enabled": "فعال"
+ },
+ "notFound": "نتیجهای برای جستجو پیدا نشد",
+ "searchProviders": "جستجوی ارائهدهندگان...",
+ "sort": "مرتبسازی سفارشی"
+ },
"ollama": {
"checker": {
"desc": "آزمایش کنید که آیا آدرس پروکسی به درستی وارد شده است",
@@ -94,33 +169,9 @@
"title": "در حال دانلود مدل {{model}} "
},
"endpoint": {
- "desc": "آدرس پروکسی رابط Ollama را وارد کنید، اگر به صورت محلی تنظیم نشده است، میتوانید خالی بگذارید",
+ "desc": "باید شامل http(s):// باشد، اگر محلی به طور اضافی مشخص نشده باشد میتوان خالی گذاشت",
"title": "آدرس سرویس Ollama"
},
- "setup": {
- "cors": {
- "description": "به دلیل محدودیتهای امنیتی مرورگر، شما باید پیکربندیهای Cross-Origin را برای Ollama انجام دهید تا بتوانید به درستی از آن استفاده کنید.",
- "linux": {
- "env": "در بخش [Service]، `Environment` را اضافه کنید و متغیر محیطی OLLAMA_ORIGINS را اضافه کنید:",
- "reboot": "systemd را مجدداً بارگذاری کرده و Ollama را راهاندازی مجدد کنید",
- "systemd": "با استفاده از systemd سرویس ollama را ویرایش کنید:"
- },
- "macos": "لطفاً برنامه «ترمینال» را باز کنید و دستور زیر را کپی و اجرا کنید",
- "reboot": "پس از اتمام، سرویس Ollama را مجدداً راهاندازی کنید",
- "title": "پیکربندی Ollama برای دسترسی Cross-Origin",
- "windows": "در ویندوز، به «کنترل پنل» بروید و متغیرهای محیطی سیستم را ویرایش کنید. برای حساب کاربری خود یک متغیر محیطی جدید به نام «OLLAMA_ORIGINS» با مقدار * ایجاد کنید و روی «OK/اعمال» کلیک کنید تا ذخیره شود"
- },
- "install": {
- "description": "لطفاً مطمئن شوید که Ollama را فعال کردهاید. اگر Ollama را دانلود نکردهاید، لطفاً به وبسایت رسمی مراجعه کرده و <1>دانلود کنید",
- "docker": "اگر ترجیح میدهید از Docker استفاده کنید، Ollama یک تصویر رسمی Docker نیز ارائه میدهد که میتوانید با دستور زیر آن را دریافت کنید:",
- "linux": {
- "command": "با دستور زیر نصب کنید:",
- "manual": "یا میتوانید به <1>راهنمای نصب دستی لینوکس مراجعه کنید"
- },
- "title": "Ollama را به صورت محلی نصب و فعال کنید",
- "windowsTab": "ویندوز (نسخه پیشنمایش)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "لغو دانلود",
@@ -131,25 +182,145 @@
"title": "دانلود مدل مشخص شده Ollama"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "کلید دسترسی Qianfan پلتفرم بایدو را وارد کنید",
- "placeholder": "کلید دسترسی Qianfan",
- "title": "کلید دسترسی"
+ "providerModels": {
+ "config": {
+ "aesGcm": "کلید شما و آدرس پروکسی و غیره با استفاده از <1>AES-GCM رمزگذاری خواهد شد",
+ "apiKey": {
+ "desc": "لطفاً کلید API {{name}} خود را وارد کنید",
+ "placeholder": "{{name}} کلید API",
+ "title": "کلید API"
+ },
+ "baseURL": {
+ "desc": "باید شامل http(s):// باشد",
+ "invalid": "لطفاً یک URL معتبر وارد کنید",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "آدرس پروکسی API"
+ },
+ "checker": {
+ "button": "بررسی",
+ "desc": "آزمون کلید API و آدرس پروکسی برای صحت",
+ "pass": "بررسی موفقیتآمیز",
+ "title": "بررسی اتصال"
+ },
+ "fetchOnClient": {
+ "desc": "مدل درخواست کلاینت به طور مستقیم از مرورگر درخواست جلسه را آغاز میکند و میتواند سرعت پاسخ را افزایش دهد",
+ "title": "استفاده از مدل درخواست کلاینت"
+ },
+ "helpDoc": "راهنمای پیکربندی",
+ "waitingForMore": "مدلهای بیشتری در حال <1>برنامهریزی برای اتصال هستند، لطفاً منتظر بمانید"
},
- "checker": {
- "desc": "تست کنید که آیا AccessKey / SecretAccess به درستی وارد شده است"
+ "createNew": {
+ "title": "ایجاد مدل AI سفارشی"
},
- "secretKey": {
- "desc": "کلید مخفی Qianfan پلتفرم بایدو را وارد کنید",
- "placeholder": "کلید مخفی Qianfan",
- "title": "کلید مخفی"
+ "item": {
+ "config": "پیکربندی مدل",
+ "customModelCards": {
+ "addNew": "ایجاد و افزودن مدل {{id}}",
+ "confirmDelete": "در حال حذف این مدل سفارشی هستید، پس از حذف قابل بازیابی نخواهد بود، لطفاً با احتیاط عمل کنید."
+ },
+ "delete": {
+ "confirm": "آیا مطمئن هستید که میخواهید مدل {{displayName}} را حذف کنید؟",
+ "success": "حذف با موفقیت انجام شد",
+ "title": "حذف مدل"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "فیلدی که در Azure OpenAI درخواست واقعی میشود",
+ "placeholder": "لطفاً نام استقرار مدل در Azure را وارد کنید",
+ "title": "نام استقرار مدل"
+ },
+ "deployName": {
+ "extra": "این فیلد به عنوان شناسه مدل هنگام ارسال درخواست استفاده میشود",
+ "placeholder": "لطفاً نام یا شناسه واقعی مدل را وارد کنید",
+ "title": "نام مدل برای استقرار"
+ },
+ "displayName": {
+ "placeholder": "لطفاً نام نمایشی مدل را وارد کنید، مانند ChatGPT، GPT-4 و غیره",
+ "title": "نام نمایشی مدل"
+ },
+ "files": {
+ "extra": "پیادهسازی بارگذاری فایل فعلی تنها یک راهحل Hack است و فقط برای آزمایش شخصی محدود است. لطفاً منتظر پیادهسازی کامل قابلیت بارگذاری فایل باشید",
+ "title": "پشتیبانی از بارگذاری فایل"
+ },
+ "functionCall": {
+ "extra": "این پیکربندی تنها قابلیت استفاده از ابزارها را برای مدل فعال میکند و به این ترتیب میتوان افزونههای نوع ابزار را به مدل اضافه کرد. اما اینکه آیا واقعاً از ابزارها استفاده میشود به خود مدل بستگی دارد، لطفاً قابلیت استفاده را خودتان آزمایش کنید",
+ "title": "پشتیبانی از استفاده از ابزار"
+ },
+ "id": {
+ "extra": "پس از ایجاد قابل ویرایش نیست و در هنگام فراخوانی AI به عنوان شناسه مدل استفاده خواهد شد",
+ "placeholder": "لطفاً شناسه مدل را وارد کنید، مانند gpt-4o یا claude-3.5-sonnet",
+ "title": "شناسه مدل"
+ },
+ "modalTitle": "پیکربندی مدل سفارشی",
+ "reasoning": {
+ "extra": "این تنظیم فقط قابلیت تفکر عمیق مدل را فعال میکند و تأثیر دقیق آن کاملاً به خود مدل بستگی دارد، لطفاً خودتان آزمایش کنید که آیا این مدل قابلیت تفکر عمیق قابل استفاده را دارد یا خیر",
+ "title": "پشتیبانی از تفکر عمیق"
+ },
+ "tokens": {
+ "extra": "حداکثر تعداد توکنهای پشتیبانی شده توسط مدل را تنظیم کنید",
+ "title": "حداکثر پنجره زمینه",
+ "unlimited": "بدون محدودیت"
+ },
+ "vision": {
+ "extra": "این پیکربندی تنها قابلیت بارگذاری تصویر در برنامه را فعال میکند، اینکه آیا شناسایی پشتیبانی میشود به خود مدل بستگی دارد، لطفاً قابلیت استفاده از شناسایی بصری این مدل را آزمایش کنید",
+ "title": "پشتیبانی از شناسایی بصری"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/تصویر",
+ "inputCharts": "${{amount}}/M کاراکتر",
+ "inputMinutes": "${{amount}}/دقیقه",
+ "inputTokens": "ورودی ${{amount}}/M",
+ "outputTokens": "خروجی ${{amount}}/M"
+ },
+ "releasedAt": "منتشر شده در {{releasedAt}}"
},
- "unlock": {
- "customRegion": "منطقه خدمات سفارشی",
- "description": "کلید دسترسی / کلید مخفی خود را وارد کنید تا مکالمه را شروع کنید. برنامه تنظیمات احراز هویت شما را ذخیره نخواهد کرد",
- "title": "استفاده از اطلاعات احراز هویت سفارشی Wenxin Yiyan"
- }
+ "list": {
+ "addNew": "مدل جدید اضافه کنید",
+ "disabled": "غیرفعال",
+ "disabledActions": {
+ "showMore": "نمایش همه"
+ },
+ "empty": {
+ "desc": "لطفاً یک مدل سفارشی ایجاد کنید یا پس از بارگذاری مدلها، شروع به استفاده کنید",
+ "title": "مدل قابل استفادهای وجود ندارد"
+ },
+ "enabled": "فعال",
+ "enabledActions": {
+ "disableAll": "غیرفعال کردن همه",
+ "enableAll": "فعال کردن همه",
+ "sort": "مرتبسازی مدلهای سفارشی"
+ },
+ "enabledEmpty": "مدل فعال وجود ندارد، لطفاً از لیست زیر مدل مورد نظر خود را فعال کنید~",
+ "fetcher": {
+ "clear": "پاک کردن مدلهای دریافت شده",
+ "fetch": "دریافت لیست مدلها",
+ "fetching": "در حال دریافت لیست مدلها...",
+ "latestTime": "آخرین زمان بهروزرسانی: {{time}}",
+ "noLatestTime": "لیست هنوز دریافت نشده است"
+ },
+ "resetAll": {
+ "conform": "آیا مطمئن هستید که میخواهید تمام تغییرات مدل فعلی را بازنشانی کنید؟ پس از بازنشانی، لیست مدلهای فعلی به حالت پیشفرض باز خواهد گشت",
+ "success": "بازنشانی با موفقیت انجام شد",
+ "title": "بازنشانی تمام تغییرات"
+ },
+ "search": "جستجوی مدل...",
+ "searchResult": "{{count}} مدل پیدا شد",
+ "title": "لیست مدلها",
+ "total": "در مجموع {{count}} مدل در دسترس است"
+ },
+ "searchNotFound": "نتیجهای برای جستجو پیدا نشد"
+ },
+ "sortModal": {
+ "success": "بهروزرسانی مرتبسازی با موفقیت انجام شد",
+ "title": "مرتبسازی سفارشی",
+ "update": "بهروزرسانی"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "در حال حذف این ارائهدهنده AI هستید، پس از حذف قابل بازیابی نخواهد بود، آیا مطمئن هستید که میخواهید حذف کنید؟",
+ "deleteSuccess": "حذف با موفقیت انجام شد",
+ "tooltip": "بهروزرسانی پیکربندی پایه ارائهدهنده",
+ "updateSuccess": "بهروزرسانی با موفقیت انجام شد"
},
"zeroone": {
"title": "01.AI صفر و یک همه چیز"
diff --git a/locales/fa-IR/models.json b/locales/fa-IR/models.json
index 1e7d0b05c82ca..14b0c299400b7 100644
--- a/locales/fa-IR/models.json
+++ b/locales/fa-IR/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K بر امنیت معنایی و مسئولیتپذیری تأکید دارد و بهطور ویژه برای سناریوهایی طراحی شده است که نیاز بالایی به امنیت محتوا دارند، تا دقت و پایداری تجربه کاربری را تضمین کند."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 از جستجوی درخت برای ساخت زنجیرههای تفکر استفاده میکند و مکانیزم بازتاب را معرفی کرده است و با استفاده از یادگیری تقویتی آموزش دیده است، این مدل توانایی خودبازتابی و اصلاح خطا را دارد."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro مدل پیشرفته پردازش زبان طبیعی است که توسط شرکت 360 ارائه شده است. این مدل دارای تواناییهای برجستهای در تولید و درک متن است و به ویژه در زمینه تولید و خلاقیت عملکرد فوقالعادهای دارد. همچنین قادر به انجام وظایف پیچیده تبدیل زبان و ایفای نقش میباشد."
},
+ "360zhinao2-o1": {
+ "description": "مدل 360zhinao2-o1 با استفاده از جستجوی درختی زنجیره تفکر را ایجاد کرده و مکانیزم بازتاب را معرفی کرده است و با استفاده از یادگیری تقویتی آموزش دیده است، این مدل توانایی خودبازتابی و اصلاح خطا را دارد."
+ },
"4.0Ultra": {
"description": "Spark Ultra قدرتمندترین نسخه از سری مدلهای بزرگ Spark است که با ارتقاء مسیر جستجوی متصل به شبکه، توانایی درک و خلاصهسازی محتوای متنی را بهبود میبخشد. این یک راهحل جامع برای افزایش بهرهوری در محیط کار و پاسخگویی دقیق به نیازها است و به عنوان یک محصول هوشمند پیشرو در صنعت شناخته میشود."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "توانایی مدل در کشور اول است و در وظایف چینی مانند دانشنامه، متنهای طولانی و تولید خلاقانه از مدلهای اصلی خارجی پیشی میگیرد. همچنین دارای قابلیتهای چندرسانهای پیشرفته در صنعت است و در چندین معیار ارزیابی معتبر عملکرد عالی دارد."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "مدل تقطیر DeepSeek-R1 مبتنی بر Qwen2.5-Math-1.5B است که با استفاده از یادگیری تقویتی و دادههای شروع سرد عملکرد استدلال را بهینهسازی کرده و مدلهای متنباز را به روز کرده است."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "مدل تقطیر DeepSeek-R1 مبتنی بر Qwen2.5-14B است که با استفاده از یادگیری تقویتی و دادههای شروع سرد عملکرد استدلال را بهینهسازی کرده و مدلهای متنباز را به روز کرده است."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "سری DeepSeek-R1 با استفاده از یادگیری تقویتی و دادههای شروع سرد عملکرد استدلال را بهینهسازی کرده و مدلهای متنباز را به روز کرده و از سطح OpenAI-o1-mini فراتر رفته است."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "مدل تقطیر DeepSeek-R1 مبتنی بر Qwen2.5-Math-7B است که با استفاده از یادگیری تقویتی و دادههای شروع سرد عملکرد استدلال را بهینهسازی کرده و مدلهای متنباز را به روز کرده است."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite دارای سرعت پاسخدهی بسیار بالا و قیمت مناسبتر است و برای سناریوهای مختلف مشتریان گزینههای منعطفتری ارائه میدهد. این مدل از استنتاج و تنظیم 128k پنجره متنی پشتیبانی میکند."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "مدل زبان بزرگ مقیاس پرچمدار توسعهیافته توسط بایدو، در مقایسه با ERNIE 3.5 ارتقاء کامل تواناییهای مدل را به ارمغان آورده و برای وظایف پیچیده در حوزههای مختلف مناسب است؛ از افزونه جستجوی بایدو پشتیبانی میکند تا اطلاعات پرسش و پاسخ بهروز بماند."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "مدل زبان بزرگ و پرچمدار خودساخته بایدو که عملکردی عالی دارد و به طور گستردهای در سناریوهای پیچیده در زمینههای مختلف کاربرد دارد؛ از اتصال خودکار به افزونه جستجوی بایدو پشتیبانی میکند و به روز بودن اطلاعات پرسش و پاسخ را تضمین میکند. نسبت به ERNIE 4.0 در عملکرد برتر است."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "مدل زبان بزرگ و پیشرفتهای که توسط بایدو توسعه یافته است، با عملکرد برجسته در زمینههای مختلف و مناسب برای وظایف پیچیده؛ از افزونه جستجوی بایدو بهطور خودکار پشتیبانی میکند تا اطلاعات بهروز را در پاسخها تضمین کند. در مقایسه با ERNIE 4.0، عملکرد بهتری دارد."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct یکی از جدیدترین سری مدلهای زبانی بزرگ منتشر شده توسط Alibaba Cloud است. این مدل 7B در زمینههای کدنویسی و ریاضی دارای تواناییهای بهبود یافته قابل توجهی است. این مدل همچنین از پشتیبانی چند زبانه برخوردار است و بیش از 29 زبان از جمله چینی و انگلیسی را پوشش میدهد. این مدل در پیروی از دستورات، درک دادههای ساختاری و تولید خروجیهای ساختاری (به ویژه JSON) به طور قابل توجهی بهبود یافته است."
},
+ "MiniMax-Text-01": {
+ "description": "در سری مدلهای MiniMax-01، ما نوآوریهای جسورانهای انجام دادهایم: برای اولین بار مکانیزم توجه خطی را به طور وسیع پیادهسازی کردهایم و معماری سنتی Transformer دیگر تنها گزینه نیست. این مدل دارای 456 میلیارد پارامتر است که در یک بار فعالسازی 45.9 میلیارد است. عملکرد کلی این مدل با بهترین مدلهای خارجی برابری میکند و در عین حال میتواند به طور مؤثر به متنهای طولانی جهانی با 4 میلیون توکن رسیدگی کند، که 32 برابر GPT-4o و 20 برابر Claude-3.5-Sonnet است."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO یک ترکیب چندمدلی بسیار انعطافپذیر است که برای ارائه تجربهای خلاقانه و برجسته طراحی شده است."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 در وظایف مختلف زبان تصویری عملکرد برجستهای از خود نشان داده است، از جمله درک اسناد و نمودارها، درک متن صحنه، OCR، حل مسائل علمی و ریاضی و غیره."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 در وظایف مختلف زبان تصویری عملکرد برجستهای از خود نشان داده است، از جمله درک اسناد و نمودارها، درک متن صحنه، OCR، حل مسائل علمی و ریاضی و غیره."
- },
"Phi-3-medium-128k-instruct": {
"description": "همان مدل Phi-3-medium، اما با اندازه بزرگتر زمینه، مناسب برای RAG یا تعداد کمی از دستورات."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat نسخه متن باز از مدلهای پیشآموزش شده سری GLM-4 است که توسط AI Zhizhu ارائه شده است. این مدل در زمینههای معنایی، ریاضی، استدلال، کد و دانش عملکرد عالی دارد. علاوه بر پشتیبانی از گفتگوی چند دور، GLM-4-9B-Chat همچنین دارای قابلیتهای پیشرفتهای مانند مرور وب، اجرای کد، فراخوانی ابزارهای سفارشی (Function Call) و استدلال متن طولانی است. این مدل از 26 زبان پشتیبانی میکند، از جمله چینی، انگلیسی، ژاپنی، کرهای و آلمانی. در چندین آزمون معیار، GLM-4-9B-Chat عملکرد عالی نشان داده است، مانند AlignBench-v2، MT-Bench، MMLU و C-Eval. این مدل از حداکثر طول زمینه 128K پشتیبانی میکند و برای تحقیقات علمی و کاربردهای تجاری مناسب است."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 یک مدل استنتاجی مبتنی بر یادگیری تقویتی (RL) است که مشکلات تکرار و خوانایی را در مدل حل میکند. قبل از RL، DeepSeek-R1 دادههای شروع سرد را معرفی کرده و عملکرد استنتاج را بهینهسازی کرده است. این مدل در وظایف ریاضی، کد و استنتاج با OpenAI-o1 عملکرد مشابهی دارد و از طریق روشهای آموزشی به دقت طراحی شده، عملکرد کلی را بهبود میبخشد."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 یک مدل زبان با 671 میلیارد پارامتر است که از معماری متخصصان ترکیبی (MoE) و توجه چندسر (MLA) استفاده میکند و با استراتژی تعادل بار بدون ضرر کمکی بهینهسازی کارایی استنتاج و آموزش را انجام میدهد. این مدل با پیشآموزش بر روی 14.8 تریلیون توکن با کیفیت بالا و انجام تنظیم دقیق نظارتی و یادگیری تقویتی، در عملکرد از سایر مدلهای متنباز پیشی میگیرد و به مدلهای بسته پیشرو نزدیک میشود."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma یکی از مدلهای پیشرفته و سبک وزن متن باز است که توسط Google توسعه یافته است. این یک مدل زبانی بزرگ با تنها دیکودر است که از زبان انگلیسی پشتیبانی میکند و وزنهای باز، واریانتهای پیشآموزش شده و واریانتهای تنظیم دقیق دستوری را ارائه میدهد. مدل Gemma برای انواع وظایف تولید متن، از جمله پرسش و پاسخ، خلاصهسازی و استدلال مناسب است. این مدل 9B از طریق 8 تریلیون توکن آموزش دیده است. اندازه نسبتاً کوچک آن امکان استقرار در محیطهای با منابع محدود، مانند لپتاپ، دسکتاپ یا زیرساخت ابری خود را فراهم میکند و به این ترتیب دسترسی به مدلهای پیشرفته AI را برای افراد بیشتری فراهم میکند و نوآوری را تسهیل میکند."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 یکی از خانوادههای مدلهای زبانی بزرگ چند زبانه است که توسط Meta توسعه یافته و شامل واریانتهای پیشآموزش شده و تنظیم دقیق دستوری با اندازههای پارامتر 8B، 70B و 405B است. این مدل 8B به طور خاص برای سناریوهای گفتگوی چند زبانه بهینهسازی شده و در چندین آزمون معیار صنعتی عملکرد عالی دارد. آموزش مدل با استفاده از بیش از 15 تریلیون توکن دادههای عمومی انجام شده و از تکنیکهای تنظیم دقیق نظارتی و یادگیری تقویتی با بازخورد انسانی برای افزایش مفید بودن و ایمنی مدل استفاده شده است. Llama 3.1 از تولید متن و تولید کد پشتیبانی میکند و تاریخ قطع دانش آن دسامبر 2023 است."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview یک مدل پردازش زبان طبیعی نوآورانه است که قادر به پردازش کارآمد مکالمات پیچیده و درک زمینه است."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview یک مدل تحقیقاتی است که توسط تیم Qwen توسعه یافته و بر روی تواناییهای استنتاج بصری تمرکز دارد و در درک صحنههای پیچیده و حل مسائل ریاضی مرتبط با بصری دارای مزیتهای منحصر به فردی است."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview جدیدترین مدل تحقیقاتی تجربی Qwen است که بر بهبود توانایی استدلال AI تمرکز دارد. با کاوش در مکانیزمهای پیچیدهای مانند ترکیب زبان و استدلال بازگشتی، مزایای اصلی شامل توانایی تحلیل استدلال قوی، توانایی ریاضی و برنامهنویسی است. در عین حال، مشکلاتی مانند تغییر زبان، حلقههای استدلال، ملاحظات ایمنی و تفاوتهای دیگر در تواناییها وجود دارد."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct جدیدترین نسخه از سری مدلهای زبانی بزرگ خاص کد است که توسط Alibaba Cloud منتشر شده است. این مدل بر اساس Qwen2.5 و با آموزش 5.5 تریلیون توکن، توانایی تولید کد، استدلال و اصلاح را به طور قابل توجهی افزایش داده است. این مدل نه تنها توانایی کدنویسی را تقویت کرده بلکه مزایای ریاضی و عمومی را نیز حفظ کرده است. این مدل پایهای جامعتر برای کاربردهای عملی مانند عاملهای کد فراهم میکند."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math بر حل مسائل در حوزه ریاضیات تمرکز دارد و پاسخهای حرفهای برای مسائل پیچیده ارائه میدهد."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 جدیدترین سری مدلهای Qwen است که از 128k زمینه پشتیبانی میکند. در مقایسه با بهترین مدلهای متنباز فعلی، Qwen2-72B در درک زبان طبیعی، دانش، کد، ریاضی و چندزبانگی به طور قابل توجهی از مدلهای پیشرو فعلی فراتر رفته است."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 جدیدترین سری مدلهای Qwen است که میتواند از مدلهای متنباز با مقیاس مشابه و حتی بزرگتر فراتر رود. Qwen2 7B در چندین ارزیابی برتری قابل توجهی به دست آورده است، به ویژه در درک کد و زبان چینی."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B یک مدل زبان بصری قدرتمند است که از پردازش چندرسانهای تصویر و متن پشتیبانی میکند و میتواند محتوای تصویر را به دقت شناسایی کرده و توصیف یا پاسخهای مرتبط تولید کند."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct یک مدل زبان بزرگ با 140 میلیارد پارامتر است که عملکرد عالی دارد و بهینهسازی شده برای سناریوهای چینی و چند زبانه، از کاربردهایی مانند پرسش و پاسخ هوشمند و تولید محتوا پشتیبانی میکند."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct یک مدل زبان بزرگ با 70 میلیارد پارامتر است که از تماسهای تابع و تعامل بینقص با سیستمهای خارجی پشتیبانی میکند و به طور قابل توجهی انعطافپذیری و مقیاسپذیری را افزایش میدهد. این مدل بهینهسازی شده برای سناریوهای چینی و چند زبانه، از کاربردهایی مانند پرسش و پاسخ هوشمند و تولید محتوا پشتیبانی میکند."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct یک مدل دستور برنامهنویسی مبتنی بر پیشآموزش وسیع است که دارای تواناییهای قوی در درک و تولید کد است و میتواند به طور مؤثر به انواع وظایف برنامهنویسی رسیدگی کند، به ویژه برای نوشتن کد هوشمند، تولید اسکریپتهای خودکار و پاسخ به مسائل برنامهنویسی مناسب است."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct یک مدل زبان بزرگ است که به طور خاص برای تولید کد، درک کد و سناریوهای توسعه کارآمد طراحی شده است و از مقیاس 32B پارامتر پیشرفته در صنعت بهره میبرد و میتواند نیازهای متنوع برنامهنویسی را برآورده کند."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "مدل بزرگ چندرسانهای TeleMM یک مدل بزرگ درک چندرسانهای است که توسط China Telecom به طور مستقل توسعه یافته و قادر به پردازش ورودیهای چندرسانهای از جمله متن و تصویر است و از قابلیتهایی مانند درک تصویر و تحلیل نمودار پشتیبانی میکند و خدمات درک چندرسانهای را به کاربران ارائه میدهد. این مدل قادر به تعامل چندرسانهای با کاربران است و محتوا را به دقت درک کرده و به سوالات پاسخ میدهد، به خلاقیت کمک میکند و به طور کارآمد اطلاعات و الهام چندرسانهای را ارائه میدهد. این مدل در وظایف چندرسانهای مانند درک دقیق، استدلال منطقی و غیره عملکرد خوبی دارد."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large بزرگترین مدل MoE با ساختار Transformer متن باز در صنعت است که دارای ۳۸۹۰ میلیارد پارامتر کل و ۵۲۰ میلیارد پارامتر فعال است."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct یک مدل زبانی بزرگ با تنظیم دقیق دستوری در سری Qwen2 است که اندازه پارامتر آن 72B است. این مدل بر اساس معماری Transformer ساخته شده و از تکنیکهای SwiGLU،偏置 QKV توجه و توجه گروهی استفاده میکند. این مدل قادر به پردازش ورودیهای بزرگ مقیاس است. این مدل در درک زبان، تولید، توانایی چند زبانه، کدنویسی، ریاضی و استدلال در چندین آزمون معیار عملکرد عالی دارد و از اکثر مدلهای متن باز پیشی گرفته و در برخی وظایف رقابت قابل توجهی با مدلهای اختصاصی نشان میدهد."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct یکی از جدیدترین سری مدلهای زبانی بزرگ منتشر شده توسط Alibaba Cloud است. این مدل 72B در زمینههای کدنویسی و ریاضی دارای تواناییهای بهبود یافته قابل توجهی است. این مدل همچنین از پشتیبانی چند زبانه برخوردار است و بیش از 29 زبان از جمله چینی و انگلیسی را پوشش میدهد. این مدل در پیروی از دستورات، درک دادههای ساختاری و تولید خروجیهای ساختاری (به ویژه JSON) به طور قابل توجهی بهبود یافته است."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "بهینهسازی شده برای سناریوهای مکالمه با شخصیتهای چینی، ارائه توانایی تولید مکالمات روان و مطابق با عادات بیانی چینی."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "مدل فراخوانی تابع Fireworks منبعباز، ارائهدهنده قابلیت اجرای دستورالعملهای برجسته و ویژگیهای باز و قابل سفارشیسازی."
+ "abab7-chat-preview": {
+ "description": "نسبت به مدلهای سری abab6.5، در تواناییهای مربوط به متنهای طولانی، ریاضیات و نوشتن به طور قابل توجهی بهبود یافته است."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "مدل Firefunction-v2 که به تازگی توسط شرکت Fireworks معرفی شده است، یک مدل فراخوانی تابع با عملکرد عالی است که بر اساس Llama-3 توسعه یافته و با بهینهسازیهای گسترده، بهویژه برای فراخوانی توابع، مکالمات و پیروی از دستورات مناسب است."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 یک مدل زبان بزرگ پیشرفته است که با یادگیری تقویتی و بهینهسازی دادههای راهاندازی سرد، عملکرد استدلال، ریاضیات و برنامهنویسی فوقالعادهای دارد."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b یک مدل زبان بصری است که میتواند بهطور همزمان ورودیهای تصویر و متن را دریافت کند. این مدل با دادههای باکیفیت آموزش دیده و برای وظایف چندوجهی مناسب است."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "مدل زبان قدرتمند Mixture-of-Experts (MoE) ارائه شده توسط Deepseek، با مجموع پارامترها به میزان 671B و فعالسازی 37B پارامتر برای هر نشانه."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "مدل Llama 3 70B دستورالعمل، بهطور ویژه برای مکالمات چندزبانه و درک زبان طبیعی بهینهسازی شده است و عملکردی بهتر از اکثر مدلهای رقیب دارد."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "مدل Llama 3 70B دستورالعمل (نسخه HF)، با نتایج پیادهسازی رسمی سازگار است و برای وظایف پیروی از دستورالعمل با کیفیت بالا مناسب است."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "مدل Llama 3 8B دستورالعمل، بهینهسازی شده برای مکالمه و وظایف چندزبانه، با عملکرد برجسته و کارآمد."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "مدل استنتاج تصویر با ۱۱ میلیارد پارامتر از Meta که برای دستورالعملها تنظیم شده است. این مدل برای تشخیص بصری، استنتاج تصویر، توصیف تصویر و پاسخ به سوالات عمومی درباره تصاویر بهینهسازی شده است. این مدل قادر به درک دادههای بصری مانند نمودارها و گرافها است و با تولید توضیحات متنی از جزئیات تصاویر، فاصله بین دیداری و زبانی را پر میکند."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "مدل Llama 3.2 1B دستورالعمل یک مدل چندزبانه سبک است که توسط Meta ارائه شده است. این مدل با هدف بهبود کارایی طراحی شده و در مقایسه با مدلهای بزرگتر، بهبودهای قابل توجهی در تأخیر و هزینه ارائه میدهد. موارد استفاده نمونه از این مدل شامل بازیابی و خلاصهسازی است."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "مدل Llama 3.2 3B دستورالعمل یک مدل چندزبانه سبک است که توسط Meta ارائه شده است. این مدل با هدف بهبود کارایی طراحی شده و در مقایسه با مدلهای بزرگتر، بهبودهای قابل توجهی در تأخیر و هزینه ارائه میدهد. نمونههای کاربردی این مدل شامل بازنویسی پرسشها و دستورات و همچنین کمک به نوشتن است."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "مدل استنتاج تصویر با 90 میلیارد پارامتر از Meta که برای دستورالعملها تنظیم شده است. این مدل برای تشخیص بصری، استنتاج تصویر، توصیف تصویر و پاسخ به سوالات عمومی در مورد تصاویر بهینهسازی شده است. این مدل قادر است دادههای بصری مانند نمودارها و گرافها را درک کند و با تولید توضیحات متنی از جزئیات تصویر، فاصله بین دیداری و زبانی را پر کند."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "مدل Llama 3.3 70B Instruct نسخه بهروزرسانی شده Llama 3.1 70B در دسامبر است. این مدل بر اساس Llama 3.1 70B (منتشر شده در ژوئیه 2024) بهبود یافته و قابلیتهای فراخوانی ابزار، پشتیبانی از متن چند زبانه، ریاضیات و برنامهنویسی را تقویت کرده است. این مدل در استدلال، ریاضیات و پیروی از دستورات به سطح پیشرفتهای در صنعت رسیده و میتواند عملکردی مشابه با 3.1 405B ارائه دهد، در حالی که از نظر سرعت و هزینه مزایای قابل توجهی دارد."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "مدل 24B با پارامترهایی که قابلیتهای پیشرفتهای مشابه مدلهای بزرگتر را داراست."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "مدل Mixtral MoE 8x22B دستوری، با پارامترهای بزرگ و معماری چندین متخصص، پشتیبانی کامل از پردازش کارآمد وظایف پیچیده."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "مدل Mixtral MoE 8x7B، معماری چندین متخصص برای پیروی و اجرای دستورات بهصورت کارآمد ارائه میدهد."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "مدل Mixtral MoE 8x7B دستورالعمل (نسخه HF)، با عملکردی مشابه با پیادهسازی رسمی، مناسب برای انواع سناریوهای وظایف کارآمد."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "مدل MythoMax L2 13B، با استفاده از تکنیکهای ترکیبی نوآورانه، در روایت داستان و نقشآفرینی مهارت دارد."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "مدل QwQ یک مدل تحقیقاتی تجربی است که توسط تیم Qwen توسعه یافته و بر تقویت توانایی استدلال AI تمرکز دارد."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "نسخه 72B مدل Qwen-VL نتیجه جدیدترین بهروزرسانیهای علیبابا است که نمایانگر نوآوریهای نزدیک به یک سال اخیر است."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 مجموعهای از مدلهای زبانی است که تنها شامل رمزگشاها میباشد و توسط تیم Qwen علیبابا کلود توسعه یافته است. این مدلها در اندازههای مختلف از جمله 0.5B، 1.5B، 3B، 7B، 14B، 32B و 72B ارائه میشوند و دارای دو نوع پایه (base) و دستوری (instruct) هستند."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct جدیدترین نسخه از سری مدلهای زبانی بزرگ خاص کد است که توسط Alibaba Cloud منتشر شده است. این مدل بر اساس Qwen2.5 و با آموزش 5.5 تریلیون توکن، توانایی تولید کد، استدلال و اصلاح را به طور قابل توجهی افزایش داده است. این مدل نه تنها توانایی کدنویسی را تقویت کرده بلکه مزایای ریاضی و عمومی را نیز حفظ کرده است. این مدل پایهای جامعتر برای کاربردهای عملی مانند عاملهای کد فراهم میکند."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "مدل StarCoder 15.5B، پشتیبانی از وظایف پیشرفته برنامهنویسی، با قابلیتهای چندزبانه تقویت شده، مناسب برای تولید و درک کدهای پیچیده."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "مدل StarCoder 7B، آموزشدیده بر روی بیش از ۸۰ زبان برنامهنویسی مختلف، با توانایی برجسته در تکمیل کد و درک زمینه."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "مدل Yi-Large، با توانایی برجسته در پردازش چندزبانه، مناسب برای انواع وظایف تولید و درک زبان."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus قدرتمندترین مدل Anthropic برای انجام وظایف بسیار پیچیده است. این مدل در عملکرد، هوش، روانی و درک عالی عمل میکند."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku سریعترین مدل نسل بعدی Anthropic است. در مقایسه با Claude 3 Haiku، Claude 3.5 Haiku در تمام مهارتها بهبود یافته و در بسیاری از آزمونهای هوش از بزرگترین مدل نسل قبلی، Claude 3 Opus پیشی گرفته است."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet تواناییهایی فراتر از Opus ارائه میدهد و سرعتی سریعتر از Sonnet دارد، در حالی که قیمت آن با Sonnet یکسان است. Sonnet بهویژه در برنامهنویسی، علم داده، پردازش بصری و وظایف نمایندگی مهارت دارد."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet تعادلی ایدهآل بین هوش و سرعت برای بارهای کاری سازمانی فراهم میکند. این محصول با قیمتی پایینتر حداکثر بهرهوری را ارائه میدهد، قابل اعتماد است و برای استقرار در مقیاس بزرگ مناسب میباشد."
},
- "code-raccoon-v1": {
- "description": "کد راکون یک دستیار هوشمند توسعه نرمافزار است که بر اساس مدل زبان بزرگ سنتنگ طراحی شده و شامل تحلیل نیازمندیهای نرمافزار، طراحی معماری، نوشتن کد و تست نرمافزار است و نیازهای مختلف کاربران در نوشتن کد و یادگیری برنامهنویسی را برآورده میکند. کد راکون از بیش از 90 زبان برنامهنویسی اصلی مانند Python، Java، JavaScript، C++، Go، SQL و IDEهای اصلی مانند VS Code و IntelliJ IDEA پشتیبانی میکند. در کاربردهای واقعی، کد راکون میتواند به توسعهدهندگان کمک کند تا کارایی برنامهنویسی خود را بیش از 50% افزایش دهند."
- },
"codegeex-4": {
"description": "CodeGeeX-4 یک دستیار برنامهنویسی قدرتمند مبتنی بر هوش مصنوعی است که از پرسش و پاسخ هوشمند و تکمیل کد در زبانهای برنامهنویسی مختلف پشتیبانی میکند و بهرهوری توسعه را افزایش میدهد."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ یک مدل زبان بزرگ با عملکرد بالا است که برای سناریوهای واقعی کسبوکار و کاربردهای پیچیده طراحی شده است."
},
+ "dall-e-2": {
+ "description": "مدل نسل دوم DALL·E، پشتیبانی از تولید تصاویر واقعیتر و دقیقتر، با وضوح 4 برابر نسل اول."
+ },
+ "dall-e-3": {
+ "description": "جدیدترین مدل DALL·E، منتشر شده در نوامبر 2023. پشتیبانی از تولید تصاویر واقعیتر و دقیقتر، با جزئیات بیشتر."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct قابلیت پردازش دستورات با قابلیت اطمینان بالا را فراهم میکند و از کاربردهای چندین صنعت پشتیبانی میکند."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 یک مدل زبانی ترکیبی (MoE) قوی و اقتصادی است. این مدل بر روی 8.1 تریلیون توکن از یک مجموعه داده با کیفیت بالا پیشآموزش دیده و از طریق تنظیم دقیق نظارتی (SFT) و یادگیری تقویتی (RL) تواناییهای مدل را بیشتر افزایش داده است. در مقایسه با DeepSeek 67B، DeepSeek-V2 در حالی که 42.5% از هزینههای آموزشی را صرفهجویی کرده و 93.3% از کش KV را کاهش داده، عملکرد بهتری دارد و حداکثر توان تولید را به 5.76 برابر افزایش داده است. این مدل از طول زمینه 128k پشتیبانی میکند و در آزمونهای معیار استاندارد و ارزیابی تولید باز عملکرد عالی دارد."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 یک مدل استنتاجی مبتنی بر یادگیری تقویتی (RL) است که به مشکلات تکرار و خوانایی در مدل پرداخته است. قبل از RL، DeepSeek-R1 دادههای شروع سرد را معرفی کرد و عملکرد استنتاج را بهینهتر کرد. این مدل در وظایف ریاضی، کدنویسی و استنتاج با OpenAI-o1 عملکرد مشابهی دارد و با استفاده از روشهای آموزشی به دقت طراحی شده، کیفیت کلی را بهبود بخشیده است."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "مدل تقطیر DeepSeek-R1 که با استفاده از یادگیری تقویتی و دادههای شروع سرد عملکرد استدلال را بهینهسازی کرده و مدلهای متنباز را به روز کرده است."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B مدلی است که بر اساس Llama-3.1-8B توسعه یافته است. این مدل با استفاده از نمونههای تولید شده توسط DeepSeek-R1 برای تنظیم دقیق، توانایی استدلال عالی را نشان میدهد. در چندین آزمون معیار عملکرد خوبی داشته است، به طوری که در MATH-500 به دقت 89.1% و در AIME 2024 به نرخ قبولی 50.4% دست یافته و در CodeForces امتیاز 1205 را کسب کرده است و به عنوان مدلی با مقیاس 8B تواناییهای ریاضی و برنامهنویسی قوی را نشان میدهد."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "مدل تقطیر DeepSeek-R1 که با استفاده از یادگیری تقویتی و دادههای شروع سرد عملکرد استدلال را بهینهسازی کرده و مدلهای متنباز را به روز کرده است."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "مدل تقطیر DeepSeek-R1 که با استفاده از یادگیری تقویتی و دادههای شروع سرد عملکرد استدلال را بهینهسازی کرده و مدلهای متنباز را به روز کرده است."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B مدلی است که از تقطیر دانش بر اساس Qwen2.5-32B به دست آمده است. این مدل با استفاده از 800000 نمونه منتخب تولید شده توسط DeepSeek-R1 برای تنظیم دقیق، در زمینههای مختلفی از جمله ریاضیات، برنامهنویسی و استدلال عملکرد برجستهای را نشان میدهد. در چندین آزمون معیار از جمله AIME 2024، MATH-500 و GPQA Diamond نتایج عالی کسب کرده است، به طوری که در MATH-500 به دقت 94.3% دست یافته و توانایی استدلال ریاضی قوی را نشان میدهد."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B مدلی است که از تقطیر دانش بر اساس Qwen2.5-Math-7B به دست آمده است. این مدل با استفاده از 800000 نمونه منتخب تولید شده توسط DeepSeek-R1 برای تنظیم دقیق، توانایی استدلال عالی را نشان میدهد. در چندین آزمون معیار عملکرد برجستهای داشته است، به طوری که در MATH-500 به دقت 92.8% و در AIME 2024 به نرخ قبولی 55.5% دست یافته و در CodeForces امتیاز 1189 را کسب کرده است و به عنوان مدلی با مقیاس 7B تواناییهای ریاضی و برنامهنویسی قوی را نشان میدهد."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 ویژگیهای برجسته نسخههای قبلی را گرد هم آورده و تواناییهای عمومی و کدنویسی را تقویت کرده است."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 یک مدل زبانی ترکیبی از متخصصان (MoE) با 671 میلیارد پارامتر است که از توجه چندسر (MLA) و معماری DeepSeekMoE استفاده میکند و با ترکیب استراتژی تعادل بار بدون ضرر کمکی، کارایی استنتاج و آموزش را بهینه میکند. با پیشآموزش بر روی 14.8 تریلیون توکن با کیفیت بالا و انجام تنظیم دقیق نظارتی و یادگیری تقویتی، DeepSeek-V3 در عملکرد از سایر مدلهای متنباز پیشی میگیرد و به مدلهای بسته پیشرو نزدیک میشود."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek LLM Chat (67B) یک مدل نوآورانه هوش مصنوعی است که توانایی درک عمیق زبان و تعامل را فراهم میکند."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "مدل LLM پیشرفته و کارآمد که در استدلال، ریاضیات و برنامهنویسی مهارت دارد."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 یک مدل زبانی بصری مبتنی بر DeepSeekMoE-27B است که از معماری MoE با فعالسازی پراکنده استفاده میکند و در حالی که تنها 4.5 میلیارد پارامتر فعال است، عملکرد فوقالعادهای را ارائه میدهد. این مدل در چندین وظیفه از جمله پرسش و پاسخ بصری، شناسایی کاراکتر نوری، درک اسناد/جدولها/نمودارها و مکانیابی بصری عملکرد عالی دارد."
+ },
"deepseek-chat": {
"description": "مدل متنباز جدیدی که تواناییهای عمومی و کدنویسی را ترکیب میکند. این مدل نه تنها توانایی گفتگوی عمومی مدل Chat و توانایی قدرتمند پردازش کد مدل Coder را حفظ کرده است، بلکه به ترجیحات انسانی نیز بهتر همسو شده است. علاوه بر این، DeepSeek-V2.5 در وظایف نوشتاری، پیروی از دستورات و سایر جنبهها نیز بهبودهای قابل توجهی داشته است."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 یک مدل کد نویسی ترکیبی و متنباز است که در وظایف کدنویسی عملکرد بسیار خوبی دارد و با GPT4-Turbo قابل مقایسه است."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 یک مدل استنتاجی مبتنی بر یادگیری تقویتی (RL) است که به مشکلات تکرار و خوانایی در مدل پرداخته است. قبل از RL، DeepSeek-R1 دادههای شروع سرد را معرفی کرد و عملکرد استنتاج را بهینهتر کرد. این مدل در وظایف ریاضی، کدنویسی و استنتاج با OpenAI-o1 عملکرد مشابهی دارد و با استفاده از روشهای آموزشی به دقت طراحی شده، کیفیت کلی را بهبود بخشیده است."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 - مدل بزرگتر و هوشمندتر در مجموعه DeepSeek - به معماری Llama 70B تقطیر شده است. بر اساس آزمونهای معیار و ارزیابیهای انسانی، این مدل از Llama 70B اصلی هوشمندتر است، به ویژه در وظایفی که نیاز به دقت ریاضی و واقعی دارند."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "مدلهای سری DeepSeek-R1-Distill از طریق تکنیک تقطیر دانش، نمونههای تولید شده توسط DeepSeek-R1 را برای تنظیم دقیق بر روی مدلهای متنباز مانند Qwen و Llama به کار میبرند."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "مدلهای سری DeepSeek-R1-Distill از طریق تکنیک تقطیر دانش، نمونههای تولید شده توسط DeepSeek-R1 را برای تنظیم دقیق بر روی مدلهای متنباز مانند Qwen و Llama به کار میبرند."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "مدلهای سری DeepSeek-R1-Distill از طریق تکنیک تقطیر دانش، نمونههای تولید شده توسط DeepSeek-R1 را برای تنظیم دقیق بر روی مدلهای متنباز مانند Qwen و Llama به کار میبرند."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "مدلهای سری DeepSeek-R1-Distill از طریق تکنیک تقطیر دانش، نمونههای تولید شده توسط DeepSeek-R1 را برای تنظیم دقیق بر روی مدلهای متنباز مانند Qwen و Llama به کار میبرند."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "مدلهای سری DeepSeek-R1-Distill از طریق تکنیک تقطیر دانش، نمونههای تولید شده توسط DeepSeek-R1 را برای تنظیم دقیق بر روی مدلهای متنباز مانند Qwen و Llama به کار میبرند."
+ },
+ "deepseek-reasoner": {
+ "description": "مدل استدلالی ارائه شده توسط DeepSeek. قبل از ارائه پاسخ نهایی، مدل ابتدا یک زنجیره تفکر را تولید میکند تا دقت پاسخ نهایی را افزایش دهد."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 یک مدل زبانی Mixture-of-Experts کارآمد است که برای پردازش نیازهای اقتصادی و کارآمد مناسب میباشد."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B مدل طراحی کد DeepSeek است که تواناییهای قدرتمندی در تولید کد ارائه میدهد."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 مدل MoE توسعه یافته توسط شرکت تحقیقاتی فناوری هوش مصنوعی DeepSeek در هانگژو است که در چندین ارزیابی عملکرد برجستهای دارد و در لیستهای اصلی در صدر مدلهای متنباز قرار دارد. V3 نسبت به مدل V2.5 سرعت تولید را 3 برابر افزایش داده و تجربه کاربری سریعتر و روانتری را برای کاربران فراهم میکند."
+ },
"deepseek/deepseek-chat": {
"description": "مدل متنباز جدیدی که تواناییهای عمومی و کدنویسی را ترکیب میکند. این مدل نه تنها توانایی گفتگوی عمومی مدل Chat و قدرت پردازش کد مدل Coder را حفظ کرده است، بلکه به ترجیحات انسانی نیز بهتر همسو شده است. علاوه بر این، DeepSeek-V2.5 در وظایف نوشتاری، پیروی از دستورات و سایر جنبهها نیز بهبودهای قابل توجهی داشته است."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 با وجود دادههای برچسبگذاری شده بسیار کم، توانایی استدلال مدل را به طرز چشمگیری افزایش میدهد. قبل از ارائه پاسخ نهایی، مدل ابتدا یک زنجیره تفکر را تولید میکند تا دقت پاسخ نهایی را افزایش دهد."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 با وجود دادههای برچسبگذاری شده بسیار کم، توانایی استدلال مدل را به طرز چشمگیری افزایش میدهد. قبل از ارائه پاسخ نهایی، مدل ابتدا یک زنجیره تفکر را تولید میکند تا دقت پاسخ نهایی را افزایش دهد."
+ },
"emohaa": {
"description": "Emohaa یک مدل روانشناختی است که دارای توانایی مشاوره حرفهای بوده و به کاربران در درک مسائل احساسی کمک میکند."
},
+ "ernie-3.5-128k": {
+ "description": "مدل زبان بزرگ پرچمدار خود توسعه یافته توسط بایدو، که شامل حجم وسیعی از متون چینی و انگلیسی است و دارای تواناییهای عمومی قوی است که میتواند نیازهای اکثر موارد پرسش و پاسخ، تولید خلاقانه و کاربردهای افزونه را برآورده کند؛ از اتصال خودکار به افزونه جستجوی بایدو پشتیبانی میکند تا اطلاعات پرسش و پاسخ به روز باشد."
+ },
+ "ernie-3.5-8k": {
+ "description": "مدل زبان بزرگ پرچمدار خود توسعه یافته توسط بایدو، که شامل حجم وسیعی از متون چینی و انگلیسی است و دارای تواناییهای عمومی قوی است که میتواند نیازهای اکثر موارد پرسش و پاسخ، تولید خلاقانه و کاربردهای افزونه را برآورده کند؛ از اتصال خودکار به افزونه جستجوی بایدو پشتیبانی میکند تا اطلاعات پرسش و پاسخ به روز باشد."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "مدل زبان بزرگ پرچمدار خود توسعه یافته توسط بایدو، که شامل حجم وسیعی از متون چینی و انگلیسی است و دارای تواناییهای عمومی قوی است که میتواند نیازهای اکثر موارد پرسش و پاسخ، تولید خلاقانه و کاربردهای افزونه را برآورده کند؛ از اتصال خودکار به افزونه جستجوی بایدو پشتیبانی میکند تا اطلاعات پرسش و پاسخ به روز باشد."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "مدل زبان بزرگ فوقالعاده پرچمدار خود توسعه یافته توسط بایدو، که نسبت به ERNIE 3.5 بهروزرسانیهای جامعتری در تواناییهای مدل داشته و به طور گستردهای در زمینههای مختلف برای وظایف پیچیده کاربرد دارد؛ از اتصال خودکار به افزونه جستجوی بایدو پشتیبانی میکند تا اطلاعات پرسش و پاسخ به روز باشد."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "مدل زبان بزرگ فوقالعاده پرچمدار خود توسعه یافته توسط بایدو، که نسبت به ERNIE 3.5 بهروزرسانیهای جامعتری در تواناییهای مدل داشته و به طور گستردهای در زمینههای مختلف برای وظایف پیچیده کاربرد دارد؛ از اتصال خودکار به افزونه جستجوی بایدو پشتیبانی میکند تا اطلاعات پرسش و پاسخ به روز باشد."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "مدل زبان بزرگ فوقالعاده پرچمدار خود توسعه یافته توسط بایدو، که عملکرد کلی آن بسیار خوب است و به طور گستردهای در زمینههای مختلف برای وظایف پیچیده کاربرد دارد؛ از اتصال خودکار به افزونه جستجوی بایدو پشتیبانی میکند تا اطلاعات پرسش و پاسخ به روز باشد. نسبت به ERNIE 4.0 در عملکرد بهتر است."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "مدل زبان بزرگ فوقالعاده پرچمدار خود توسعه یافته توسط بایدو، که عملکرد کلی آن بسیار خوب است و به طور گستردهای در زمینههای مختلف برای وظایف پیچیده کاربرد دارد؛ از اتصال خودکار به افزونه جستجوی بایدو پشتیبانی میکند تا اطلاعات پرسش و پاسخ به روز باشد. نسبت به ERNIE 4.0 در عملکرد بهتر است."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "مدل زبان بزرگ فوقالعاده پرچمدار خود توسعه یافته توسط بایدو، که عملکرد کلی آن بسیار خوب است و به طور گستردهای در زمینههای مختلف برای وظایف پیچیده کاربرد دارد؛ از اتصال خودکار به افزونه جستجوی بایدو پشتیبانی میکند تا اطلاعات پرسش و پاسخ به روز باشد. نسبت به ERNIE 4.0 در عملکرد بهتر است."
+ },
+ "ernie-char-8k": {
+ "description": "مدل زبان بزرگ با کاربرد خاص که توسط بایدو توسعه یافته است و برای کاربردهایی مانند NPCهای بازی، مکالمات خدمات مشتری، و نقشآفرینی در مکالمات مناسب است، سبک شخصیت آن واضحتر و یکدستتر است و توانایی پیروی از دستورات و عملکرد استدلال بهتری دارد."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "مدل زبان بزرگ با کاربرد خاص که توسط بایدو توسعه یافته است و برای کاربردهایی مانند NPCهای بازی، مکالمات خدمات مشتری، و نقشآفرینی در مکالمات مناسب است، سبک شخصیت آن واضحتر و یکدستتر است و توانایی پیروی از دستورات و عملکرد استدلال بهتری دارد."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite مدل زبان بزرگ سبک خود توسعه یافته توسط بایدو است که تعادل خوبی بین عملکرد مدل و عملکرد استدلال دارد و برای استفاده در کارتهای تسریع AI با توان محاسباتی پایین مناسب است."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "مدل زبان بزرگ سبک خود توسعه یافته توسط بایدو که تعادل خوبی بین عملکرد مدل و عملکرد استدلال دارد و عملکرد بهتری نسبت به ERNIE Lite دارد و برای استفاده در کارتهای تسریع AI با توان محاسباتی پایین مناسب است."
+ },
+ "ernie-novel-8k": {
+ "description": "مدل زبان بزرگ عمومی خود توسعه یافته توسط بایدو که در توانایی ادامه نوشتن رمان مزیت قابل توجهی دارد و همچنین میتواند در صحنههای کوتاهنمایش و فیلمها استفاده شود."
+ },
+ "ernie-speed-128k": {
+ "description": "مدل زبان بزرگ با عملکرد بالا که به تازگی در سال 2024 توسط بایدو منتشر شده است، دارای تواناییهای عمومی عالی است و برای تنظیم دقیق به عنوان مدل پایه مناسب است و میتواند به خوبی مسائل خاص را مدیریت کند و همچنین دارای عملکرد استدلال بسیار خوبی است."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "مدل زبان بزرگ با عملکرد بالا که به تازگی در سال 2024 توسط بایدو منتشر شده است، دارای تواناییهای عمومی عالی است و عملکرد بهتری نسبت به ERNIE Speed دارد و برای تنظیم دقیق به عنوان مدل پایه مناسب است و میتواند به خوبی مسائل خاص را مدیریت کند و همچنین دارای عملکرد استدلال بسیار خوبی است."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny مدل زبان بزرگ با عملکرد فوقالعاده بالا است که هزینههای استقرار و تنظیم آن در بین مدلهای سری Wenxin کمترین است."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (تنظیم) عملکردی پایدار و قابل تنظیم ارائه میدهد و انتخابی ایدهآل برای راهحلهای وظایف پیچیده است."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro از حداکثر ۲ میلیون توکن پشتیبانی میکند و انتخابی ایدهآل برای مدلهای چندوجهی متوسط است که برای پشتیبانی از وظایف پیچیده مناسب میباشد."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash ویژگیها و بهبودهای نسل بعدی را ارائه میدهد، از جمله سرعت عالی، استفاده از ابزارهای بومی، تولید چندرسانهای و پنجره متن 1M توکن."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash ویژگیها و بهبودهای نسل بعدی را ارائه میدهد، از جمله سرعت عالی، استفاده از ابزارهای بومی، تولید چندرسانهای و پنجره متن 1M توکن."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp یک مدل هوش مصنوعی چندمدلی آزمایشی از گوگل است که دارای ویژگیهای نسل بعدی، سرعت فوقالعاده، فراخوانی ابزارهای بومی و تولید چندمدلی است."
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "مدل Gemini 2.0 Flash که برای بهینهسازی هزینه و تأخیر کم طراحی شده است."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Exp جدیدترین مدل AI چندرسانهای آزمایشی گوگل است که دارای ویژگیهای نسل بعدی، سرعت فوقالعاده، فراخوانی ابزار بومی و تولید چندرسانهای است."
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Exp جدیدترین مدل هوش مصنوعی چندرسانهای آزمایشی Google است که ویژگیهای نسل بعدی، سرعت فوقالعاده، فراخوانی ابزارهای بومی و تولید چندرسانهای را داراست."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 جدیدترین مدل AI چند وجهی تجربی Google است که دارای قابلیت پردازش سریع میباشد و از ورودیهای متنی، تصویری و ویدیویی پشتیبانی میکند و برای گسترش مؤثر در چندین وظیفه مناسب است."
- },
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 جدیدترین مدل آزمایشی چندرسانهای AI گوگل است که دارای قابلیت پردازش سریع بوده و از ورودیهای متنی، تصویری و ویدیویی پشتیبانی میکند و برای گسترش کارآمد در انواع وظایف مناسب است."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental جدیدترین مدل AI چندرسانهای آزمایشی گوگل است که نسبت به نسخههای قبلی خود بهبود کیفیت قابل توجهی داشته است، به ویژه در زمینه دانش جهانی، کد و متنهای طولانی."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 جدیدترین مدل AI چندرسانهای آزمایشی Google است که نسبت به نسخههای تاریخی خود دارای ارتقاء کیفیت قابل توجهی میباشد."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus توانایی درک محتوای ویدئویی و تصاویر متعدد را دارد و برای وظایف چندرسانهای مناسب است."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview دارای تواناییهای پیچیده استدلال است و در زمینههای استدلال منطقی، ریاضیات، برنامهنویسی و غیره عملکرد عالی دارد."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash ویژگیها و بهبودهای نسل بعدی را ارائه میدهد، از جمله سرعت عالی، استفاده از ابزارهای بومی، تولید چندرسانهای و پنجره متن 1M توکن."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental جدیدترین مدل AI چندرسانهای آزمایشی گوگل است که نسبت به نسخههای قبلی خود بهبود کیفیت قابل توجهی داشته است، به ویژه در زمینه دانش جهانی، کد و متنهای طولانی."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash قابلیت پردازش چندوجهی بهینهشده را ارائه میدهد و برای انواع سناریوهای پیچیده مناسب است."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro با ترکیب آخرین فناوریهای بهینهسازی، توانایی پردازش دادههای چندحالته را با کارایی بالاتر ارائه میدهد."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 مدل کارآمدی است که توسط Google ارائه شده و شامل طیف وسیعی از کاربردها از برنامههای کوچک تا پردازش دادههای پیچیده است."
+ },
"google/gemma-2-27b-it": {
"description": "جمما ۲ ادامهدهندهی ایده طراحی سبک و کارآمد است."
},
"google/gemma-2-2b-it": {
"description": "مدل بهینهسازی دستورات سبک گوگل"
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 مدل کارآمدی است که توسط Google ارائه شده و شامل طیف وسیعی از کاربردها از برنامههای کوچک تا پردازش دادههای پیچیده است."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 یک سری مدلهای متنی سبک و متنباز از Google است."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 توربو، مناسب برای انواع وظایف تولید و درک متن، در حال حاضر به gpt-3.5-turbo-0125 اشاره میکند"
},
+ "gpt-3.5-turbo-16k": {
+ "description": "مدل تولید متن با ظرفیت بالا GPT 3.5 Turbo 16k، مناسب برای وظایف پیچیده."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 توربو، مناسب برای انواع وظایف تولید و درک متن، در حال حاضر به gpt-3.5-turbo-0125 اشاره میکند"
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o یک مدل پویا است که به طور مداوم بهروز رسانی میشود تا نسخه فعلی و جدیدی را حفظ کند. این مدل قدرت فهم و تولید زبان را ترکیب کرده و مناسب برای کاربردهای مقیاس بزرگ مانند خدمات مشتری، آموزش و پشتیبانی فنی است."
},
+ "gpt-4o-audio-preview": {
+ "description": "مدل صوتی GPT-4o، پشتیبانی از ورودی و خروجی صوتی."
+ },
"gpt-4o-mini": {
"description": "یک راهحل هوش مصنوعی مقرونبهصرفه که برای انواع وظایف متنی و تصویری مناسب است."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "نسخه زنده GPT-4o-mini، پشتیبانی از ورودی و خروجی صوتی و متنی به صورت زنده."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "نسخه زنده GPT-4o، پشتیبانی از ورودی و خروجی صوتی و متنی به صورت زنده."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "نسخه زنده GPT-4o، پشتیبانی از ورودی و خروجی صوتی و متنی به صورت زنده."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "نسخه زنده GPT-4o، پشتیبانی از ورودی و خروجی صوتی و متنی به صورت زنده."
+ },
"grok-2-1212": {
"description": "این مدل در دقت، پیروی از دستورات و توانایی چند زبانه بهبود یافته است."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "مدل FunctionCall با معماری MOE جدید Hunyuan، آموزشدیده با دادههای باکیفیت FunctionCall، با پنجره متنی تا 32K و پیشرو در چندین شاخص ارزیابی."
},
+ "hunyuan-large": {
+ "description": "مدل Hunyuan-large دارای مجموع پارامترها حدود 389B و پارامترهای فعال حدود 52B است، که بزرگترین و بهترین مدل MoE با ساختار Transformer در صنعت به شمار میرود."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "متخصص در پردازش وظایف متنی طولانی مانند خلاصهسازی اسناد و پرسش و پاسخ اسنادی، همچنین توانایی پردازش وظایف تولید متن عمومی را دارد. در تحلیل و تولید متنهای طولانی عملکرد فوقالعادهای دارد و میتواند بهطور مؤثر به نیازهای پیچیده و دقیق پردازش محتوای طولانی پاسخ دهد."
+ },
"hunyuan-lite": {
"description": "به ساختار MOE ارتقا یافته است، پنجره متنی 256k دارد و در چندین مجموعه ارزیابی در زمینههای NLP، کد، ریاضیات و صنایع از بسیاری از مدلهای متنباز پیشی گرفته است."
},
+ "hunyuan-lite-vision": {
+ "description": "مدل چندرسانهای 7B جدید Hunyuan، با پنجره زمینه 32K، از گفتگوی چندرسانهای در صحنههای چینی و انگلیسی، شناسایی اشیاء در تصاویر، درک جداول اسناد و ریاضیات چندرسانهای پشتیبانی میکند و در چندین بعد، معیارهای ارزیابی را نسبت به مدلهای رقیب 7B بهبود میبخشد."
+ },
"hunyuan-pro": {
"description": "مدل MOE-32K با مقیاس پارامتر تریلیونها. در انواع بنچمارکها به سطح پیشرو مطلق دست یافته است، توانایی پردازش دستورالعملها و استدلالهای پیچیده، دارای قابلیتهای ریاضی پیچیده، پشتیبانی از functioncall، و بهطور ویژه در حوزههای ترجمه چندزبانه، مالی، حقوقی و پزشکی بهینهسازی شده است."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "با استفاده از استراتژی مسیریابی بهینهتر، در عین حال مشکلات توازن بار و همگرایی کارشناسان را کاهش داده است. در زمینه متون طولانی، شاخص «یافتن سوزن در انبار کاه» به ۹۹.۹٪ رسیده است. MOE-256K در طول و عملکرد پیشرفت بیشتری داشته و به طور قابل توجهی طول ورودی قابل قبول را گسترش داده است."
},
+ "hunyuan-standard-vision": {
+ "description": "مدل چندرسانهای جدید Hunyuan، از پاسخگویی به چند زبان پشتیبانی میکند و تواناییهای چینی و انگلیسی را بهطور متوازن ارائه میدهد."
+ },
"hunyuan-turbo": {
"description": "نسخه پیشنمایش مدل زبان بزرگ نسل جدید HunYuan که از ساختار مدل متخصص ترکیبی (MoE) جدید استفاده میکند. در مقایسه با hunyuan-pro، کارایی استنتاج سریعتر و عملکرد بهتری دارد."
},
+ "hunyuan-turbo-20241120": {
+ "description": "نسخه ثابت hunyuan-turbo 20 نوامبر 2024، نسخهای بین hunyuan-turbo و hunyuan-turbo-latest."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "بهینهسازیهای این نسخه: مقیاسدهی دستورات داده، بهطور قابل توجهی توانایی تعمیم عمومی مدل را افزایش میدهد؛ بهطور قابل توجهی تواناییهای ریاضی، کدنویسی و استدلال منطقی را بهبود میبخشد؛ بهینهسازی تواناییهای درک متن و کلمات مرتبط با آن؛ بهینهسازی کیفیت تولید محتوای خلق متن."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "بهینهسازی تجربه عمومی، شامل درک NLP، خلق متن، گپزنی، پرسش و پاسخ دانش، ترجمه و حوزههای مختلف؛ افزایش انساننمایی، بهینهسازی هوش عاطفی مدل؛ افزایش توانایی مدل در روشنسازی فعال زمانی که نیت مبهم است؛ افزایش توانایی پردازش مسائل مربوط به تجزیه و تحلیل کلمات؛ افزایش کیفیت و قابلیت تعامل در خلق محتوا؛ بهبود تجربه چند دور."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "مدل بزرگ زبان بصری نسل جدید Hunyuan، با استفاده از ساختار جدید مدلهای متخصص ترکیبی (MoE)، در تواناییهای مربوط به درک تصویر و متن، خلق محتوا، پرسش و پاسخ دانش و تحلیل استدلال نسبت به مدلهای نسل قبلی بهطور جامع بهبود یافته است."
+ },
"hunyuan-vision": {
"description": "جدیدترین مدل چندوجهی هونیوان، پشتیبانی از ورودی تصویر + متن برای تولید محتوای متنی."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "جدیدترین سری مدلهای ما با عملکرد استدلال عالی، از طول متن ۱M پشتیبانی میکند و تواناییهای قویتری در پیروی از دستورات و فراخوانی ابزارها دارد."
},
+ "internlm3-latest": {
+ "description": "سری جدیدترین مدلهای ما با عملکرد استدلال برجسته، پیشتاز مدلهای متنباز در همان سطح هستند. به طور پیشفرض به جدیدترین مدلهای سری InternLM3 ما اشاره دارد."
+ },
+ "jina-deepsearch-v1": {
+ "description": "جستجوی عمیق ترکیبی از جستجوی اینترنتی، خواندن و استدلال است که میتواند تحقیقات جامع را انجام دهد. میتوانید آن را به عنوان یک نماینده در نظر بگیرید که وظایف تحقیق شما را میپذیرد - این نماینده جستجوی گستردهای انجام میدهد و پس از چندین بار تکرار، پاسخ را ارائه میدهد. این فرآیند شامل تحقیق مداوم، استدلال و حل مسئله از زوایای مختلف است. این با مدلهای بزرگ استاندارد که مستقیماً از دادههای پیشآموزش شده پاسخ تولید میکنند و سیستمهای RAG سنتی که به جستجوی سطحی یکباره وابستهاند، تفاوت اساسی دارد."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM یک مدل زبانی تجربی و خاص برای وظایف است که برای مطابقت با اصول علم یادگیری آموزش دیده است و میتواند در سناریوهای آموزشی و یادگیری از دستورات سیستم پیروی کند و به عنوان مربی متخصص عمل کند."
},
"lite": {
"description": "Spark Lite یک مدل زبان بزرگ سبک است که دارای تأخیر بسیار کم و توانایی پردازش کارآمد میباشد. بهطور کامل رایگان و باز است و از قابلیت جستجوی آنلاین در زمان واقعی پشتیبانی میکند. ویژگی پاسخدهی سریع آن باعث میشود که در کاربردهای استنتاجی و تنظیم مدل در دستگاههای با توان محاسباتی پایین عملکرد برجستهای داشته باشد و تجربهای هوشمند و مقرونبهصرفه برای کاربران فراهم کند. بهویژه در زمینههای پرسش و پاسخ دانش، تولید محتوا و جستجو عملکرد خوبی دارد."
},
- "llama-3.1-70b-instruct": {
- "description": "مدل Llama 3.1 70B Instruct با 70 میلیارد پارامتر، قادر به ارائه عملکرد برجسته در تولید متنهای بزرگ و انجام وظایف دستوری است."
- },
"llama-3.1-70b-versatile": {
"description": "لاما 3.1 70B توانایی استدلال هوش مصنوعی قویتری را ارائه میدهد، مناسب برای برنامههای پیچیده، پشتیبانی از پردازشهای محاسباتی فراوان و تضمین کارایی و دقت بالا."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B یک مدل با کارایی بالا است که توانایی تولید سریع متن را فراهم میکند و برای کاربردهایی که به بهرهوری و صرفهجویی در هزینه در مقیاس بزرگ نیاز دارند، بسیار مناسب است."
},
- "llama-3.1-8b-instruct": {
- "description": "مدل Llama 3.1 8B Instruct با 8 میلیارد پارامتر، قابلیت اجرای کارهای مبتنی بر دستورات تصویری را بهطور مؤثر دارد و توانایی تولید متن با کیفیت بالا را ارائه میدهد."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "مدل Llama 3.1 Sonar Huge Online با 405 میلیارد پارامتر، پشتیبانی از طول زمینه حدود 127,000 نشانه، طراحی شده برای برنامههای چت آنلاین پیچیده."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "مدل Llama 3.1 Sonar Large Chat با 70 میلیارد پارامتر، پشتیبانی از طول زمینه حدود 127,000 نشانه، مناسب برای وظایف پیچیده چت آفلاین."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "مدل Llama 3.1 Sonar Large Online با 70 میلیارد پارامتر، پشتیبانی از طول زمینه حدود 127,000 نشانه، مناسب برای وظایف چت با حجم بالا و متنوع."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "مدل Llama 3.1 Sonar Small Chat با 8 میلیارد پارامتر، بهطور خاص برای چت آفلاین طراحی شده و از طول زمینهای حدود 127,000 نشانه پشتیبانی میکند."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "مدل Llama 3.1 Sonar Small Online با 8 میلیارد پارامتر، پشتیبانی از طول زمینهای حدود 127,000 نشانه، بهطور ویژه برای چت آنلاین طراحی شده و میتواند بهطور کارآمد انواع تعاملات متنی را پردازش کند."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "لاما 3.2 برای انجام وظایفی که ترکیبی از دادههای بصری و متنی هستند طراحی شده است. این مدل در وظایفی مانند توصیف تصاویر و پرسش و پاسخ بصری عملکرد بسیار خوبی دارد و فاصله بین تولید زبان و استدلال بصری را پر میکند."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 پیشرفتهترین مدل زبان چندزبانه و متنباز در سری Llama است که تجربهای با هزینه بسیار پایین مشابه عملکرد مدل 405B را ارائه میدهد. این مدل بر اساس ساختار Transformer طراحی شده و از طریق تنظیم دقیق نظارتی (SFT) و یادگیری تقویتی با بازخورد انسانی (RLHF) بهبود کارایی و ایمنی یافته است. نسخه بهینهسازی شده آن برای مکالمات چندزبانه طراحی شده و در چندین معیار صنعتی از بسیاری از مدلهای چت متنباز و بسته بهتر عمل میکند. تاریخ قطع دانش آن دسامبر 2023 است."
+ },
"llama-3.3-70b-versatile": {
"description": "مدل زبان بزرگ چند زبانه Meta Llama 3.3 (LLM) یک مدل تولیدی پیشآموزش دیده و تنظیمشده در 70B (ورودی متن/خروجی متن) است. مدل متن خالص Llama 3.3 برای کاربردهای گفتگوی چند زبانه بهینهسازی شده و در معیارهای صنعتی معمول در مقایسه با بسیاری از مدلهای چت متنباز و بسته عملکرد بهتری دارد."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 برای انجام وظایفی که ترکیبی از دادههای بصری و متنی هستند طراحی شده است. این مدل در وظایفی مانند توصیف تصویر و پرسش و پاسخ بصری عملکرد بسیار خوبی دارد و فاصله بین تولید زبان و استدلال بصری را پر میکند."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 پیشرفتهترین مدل زبان بزرگ چند زبانه متن باز از سری Llama، با هزینه بسیار کم، تجربهای مشابه با عملکرد مدل 405B. بر پایه ساختار Transformer و با بهبود کارایی و ایمنی از طریق تنظیم دقیق نظارتی (SFT) و یادگیری تقویتی با بازخورد انسانی (RLHF). نسخه بهینهسازی شده برای دستورالعملها به طور خاص برای مکالمات چند زبانه بهینهسازی شده و در چندین معیار صنعتی بهتر از بسیاری از مدلهای چت متن باز و بسته عمل میکند. تاریخ قطع دانش تا دسامبر 2023."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "مدل بزرگ زبان چند زبانه Meta Llama 3.3 (LLM) یک مدل تولیدی پیشآموزش و تنظیم دستوری در 70B (ورودی متن/خروجی متن) است. مدل تنظیم دستوری Llama 3.3 به طور خاص برای موارد استفاده مکالمه چند زبانه بهینهسازی شده و در معیارهای صنعتی رایج از بسیاری از مدلهای چت متنباز و بسته موجود بهتر عمل میکند."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 برای انجام وظایفی که ترکیبی از دادههای بصری و متنی هستند طراحی شده است. این مدل در وظایفی مانند توصیف تصویر و پرسش و پاسخ بصری عملکرد بسیار خوبی دارد و فاصله بین تولید زبان و استدلال بصری را پر میکند."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "مدل Llama 3.1 Turbo با ظرفیت 405B، پشتیبانی از زمینههای بسیار بزرگ برای پردازش دادههای عظیم را فراهم میکند و در کاربردهای هوش مصنوعی در مقیاس بسیار بزرگ عملکرد برجستهای دارد."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 مدل پیشرو ارائه شده توسط Meta است که از حداکثر 405B پارامتر پشتیبانی میکند و میتواند در زمینههای گفتگوهای پیچیده، ترجمه چند زبانه و تحلیل دادهها استفاده شود."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B پشتیبانی کارآمد از مکالمات چندزبانه را ارائه میدهد."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 برای انجام وظایفی طراحی شده است که دادههای بصری و متنی را با هم ترکیب میکند. این مدل در وظایفی مانند توصیف تصویر و پرسش و پاسخ بصری عملکرد بسیار خوبی دارد و فاصله بین تولید زبان و استدلال بصری را پر میکند."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 پیشرفتهترین مدل زبان چندزبانه و متنباز در سری Llama است که تجربهای با هزینه بسیار پایین مشابه عملکرد مدل 405B را ارائه میدهد. این مدل بر اساس ساختار Transformer طراحی شده و از طریق تنظیم دقیق نظارتی (SFT) و یادگیری تقویتی با بازخورد انسانی (RLHF) بهبود کارایی و ایمنی یافته است. نسخه بهینهسازی شده آن برای مکالمات چندزبانه طراحی شده و در چندین معیار صنعتی از بسیاری از مدلهای چت متنباز و بسته بهتر عمل میکند. تاریخ قطع دانش آن دسامبر 2023 است."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 پیشرفتهترین مدل زبان چندزبانه و متنباز در سری Llama است که تجربهای با هزینه بسیار پایین مشابه عملکرد مدل 405B را ارائه میدهد. این مدل بر اساس ساختار Transformer طراحی شده و از طریق تنظیم دقیق نظارتی (SFT) و یادگیری تقویتی با بازخورد انسانی (RLHF) بهبود کارایی و ایمنی یافته است. نسخه بهینهسازی شده آن برای مکالمات چندزبانه طراحی شده و در چندین معیار صنعتی از بسیاری از مدلهای چت متنباز و بسته بهتر عمل میکند. تاریخ قطع دانش آن دسامبر 2023 است."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct بزرگترین و قدرتمندترین مدل در میان مدلهای Llama 3.1 Instruct است. این یک مدل بسیار پیشرفته برای استدلال مکالمهای و تولید دادههای مصنوعی است و همچنین میتواند به عنوان پایهای برای پیشتمرین یا تنظیم دقیق مداوم در حوزههای خاص استفاده شود. Llama 3.1 مجموعهای از مدلهای زبان بزرگ چندزبانه (LLMs) است که از پیش آموزش دیده و برای دستورالعملها تنظیم شدهاند و شامل اندازههای 8B، 70B و 405B (ورودی/خروجی متنی) میباشد. مدلهای متنی تنظیمشده بر اساس دستورالعملهای Llama 3.1 (8B، 70B، 405B) بهطور خاص برای موارد استفاده مکالمه چندزبانه بهینهسازی شدهاند و در بسیاری از معیارهای استاندارد صنعتی از مدلهای چت منبع باز موجود پیشی گرفتهاند. Llama 3.1 برای استفادههای تجاری و تحقیقاتی در زبانهای مختلف طراحی شده است. مدلهای متنی تنظیمشده بر اساس دستورالعملها برای چتهای مشابه دستیار مناسب هستند، در حالی که مدلهای پیشآموزشدیده میتوانند برای انواع وظایف تولید زبان طبیعی سازگار شوند. مدلهای Llama 3.1 همچنین از خروجیهای خود برای بهبود سایر مدلها، از جمله تولید دادههای مصنوعی و پالایش، پشتیبانی میکنند. Llama 3.1 یک مدل زبان خودبازگشتی است که از معماری بهینهشده ترانسفورمر استفاده میکند. نسخههای تنظیمشده از تنظیم دقیق نظارتشده (SFT) و یادگیری تقویتی با بازخورد انسانی (RLHF) برای تطابق با ترجیحات انسانی در مورد کمکرسانی و ایمنی استفاده میکنند."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 یک مدل زبان بزرگ باز (LLM) است که برای توسعهدهندگان، پژوهشگران و شرکتها طراحی شده است تا به آنها در ساخت، آزمایش و گسترش مسئولانه ایدههای هوش مصنوعی مولد کمک کند. به عنوان بخشی از سیستم پایه نوآوری جامعه جهانی، این مدل برای دستگاههای با توان محاسباتی و منابع محدود، دستگاههای لبه و زمانهای آموزش سریعتر بسیار مناسب است."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "مدل LLM پیشرفته که از تولید دادههای ترکیبی، تقطیر دانش و استدلال پشتیبانی میکند و برای رباتهای چت، برنامهنویسی و وظایف خاص مناسب است."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "توانمندسازی گفتگوهای پیچیده با درک زمینهای عالی، توانایی استدلال و قابلیت تولید متن."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "مدل پیشرفته و پیشرفته که دارای درک زبان، توانایی استدلال عالی و قابلیت تولید متن است."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "مدل بینایی-زبان پیشرفته که در استدلال با کیفیت بالا از تصاویر مهارت دارد."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "مدل زبان کوچک پیشرفته و پیشرفته که دارای درک زبان، توانایی استدلال عالی و قابلیت تولید متن است."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "مدل زبان کوچک پیشرفته و پیشرفته که دارای درک زبان، توانایی استدلال عالی و قابلیت تولید متن است."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "مدل بینایی-زبان پیشرفته که در استدلال با کیفیت بالا از تصاویر مهارت دارد."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "مدل LLM پیشرفته که در استدلال، ریاضیات، دانش عمومی و فراخوانی توابع مهارت دارد."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 یک مدل زبانی است که توسط AI مایکروسافت ارائه شده و در زمینههای گفتگوی پیچیده، چند زبانه، استدلال و دستیار هوشمند به ویژه عملکرد خوبی دارد."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K یک مدل با قابلیت پردازش متن طولانی است که برای تولید متون بسیار طولانی مناسب است. این مدل میتواند تا 128,000 توکن را پردازش کند و برای کاربردهایی مانند پژوهش، علمی و تولید اسناد بزرگ بسیار مناسب است."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "مدل بصری Kimi (شامل moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview و غیره) قادر به درک محتوای تصویر است، از جمله متن تصویر، رنگ تصویر و شکل اشیاء."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K توانایی پردازش متن با طول متوسط را فراهم میکند و قادر به پردازش 32,768 توکن است. این مدل بهویژه برای تولید اسناد طولانی و مکالمات پیچیده مناسب است و در زمینههایی مانند تولید محتوا، ایجاد گزارش و سیستمهای مکالمه کاربرد دارد."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "مدل بصری Kimi (شامل moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview و غیره) قادر به درک محتوای تصویر است، از جمله متن تصویر، رنگ تصویر و شکل اشیاء."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K بهطور ویژه برای تولید متنهای کوتاه طراحی شده است و دارای عملکرد پردازشی کارآمدی است که میتواند ۸,۱۹۲ توکن را پردازش کند. این مدل برای مکالمات کوتاه، یادداشتبرداری سریع و تولید محتوای سریع بسیار مناسب است."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "مدل بصری Kimi (شامل moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview و غیره) قادر به درک محتوای تصویر است، از جمله متن تصویر، رنگ تصویر و شکل اشیاء."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "هرمس ۲ پرو لاما ۳ ۸B نسخه ارتقاء یافته Nous Hermes 2 است که شامل جدیدترین مجموعه دادههای توسعهیافته داخلی میباشد."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B یک مدل زبان بزرگ است که توسط NVIDIA سفارشیسازی شده و هدف آن افزایش مفید بودن پاسخهای تولید شده توسط LLM به پرسشهای کاربران است."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B یک مدل زبانی بزرگ سفارشی شده توسط NVIDIA است که به منظور افزایش کمک به پاسخهای تولید شده توسط LLM برای پرسشهای کاربران طراحی شده است. این مدل در آزمونهای معیار مانند Arena Hard، AlpacaEval 2 LC و GPT-4-Turbo MT-Bench عملکرد عالی داشته و تا تاریخ 1 اکتبر 2024 در تمامی سه آزمون خودکار همراستایی در رتبه اول قرار دارد. این مدل با استفاده از RLHF (به ویژه REINFORCE)، Llama-3.1-Nemotron-70B-Reward و HelpSteer2-Preference در مدل Llama-3.1-70B-Instruct آموزش دیده است."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "مدل زبان منحصر به فرد که دقت و کارایی بینظیری را ارائه میدهد."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B یک مدل زبان بزرگ سفارشی از NVIDIA است که به منظور افزایش کمکپذیری پاسخهای تولید شده توسط LLM طراحی شده است."
+ },
"o1": {
"description": "متمرکز بر استدلال پیشرفته و حل مسائل پیچیده، از جمله وظایف ریاضی و علمی. بسیار مناسب برای برنامههایی که به درک عمیق زمینه و مدیریت جریانهای کاری نیاز دارند."
},
- "o1-2024-12-17": {
- "description": "o1 مدل استدلال جدید OpenAI است که ورودیهای متنی و تصویری را پشتیبانی میکند و خروجی متنی تولید میکند و برای وظایف پیچیدهای که به دانش عمومی گسترده نیاز دارند، مناسب است. این مدل دارای 200K زمینه و تاریخ قطع دانش در اکتبر 2023 است."
- },
"o1-mini": {
"description": "کوچکتر و سریعتر از o1-preview، با ۸۰٪ هزینه کمتر، و عملکرد خوب در تولید کد و عملیات با زمینههای کوچک."
},
"o1-preview": {
"description": "تمرکز بر استدلال پیشرفته و حل مسائل پیچیده، از جمله وظایف ریاضی و علمی. بسیار مناسب برای برنامههایی که نیاز به درک عمیق از زمینه و جریان کاری خودمختار دارند."
},
+ "o3-mini": {
+ "description": "o3-mini جدیدترین مدل استنتاج کوچک ماست که هوش بالایی را با هزینه و هدف تأخیر مشابه o1-mini ارائه میدهد."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba یک مدل زبان Mamba 2 است که بر تولید کد تمرکز دارد و پشتیبانی قدرتمندی برای وظایف پیشرفته کدنویسی و استدلال ارائه میدهد."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini جدیدترین مدل OpenAI است که پس از GPT-4 Omni عرضه شده و از ورودیهای تصویری و متنی پشتیبانی میکند و خروجی متنی ارائه میدهد. به عنوان پیشرفتهترین مدل کوچک آنها، این مدل بسیار ارزانتر از سایر مدلهای پیشرفته اخیر است و بیش از ۶۰٪ ارزانتر از GPT-3.5 Turbo میباشد. این مدل هوشمندی پیشرفته را حفظ کرده و در عین حال از نظر اقتصادی بسیار مقرون به صرفه است. GPT-4o mini در آزمون MMLU امتیاز ۸۲٪ را کسب کرده و در حال حاضر در ترجیحات چت بالاتر از GPT-4 رتبهبندی شده است."
},
- "openai/o1": {
- "description": "o1 مدل استدلال جدید OpenAI است که ورودیهای متنی و تصویری را پشتیبانی میکند و خروجی متنی تولید میکند و برای وظایف پیچیدهای که به دانش عمومی گسترده نیاز دارند، مناسب است. این مدل دارای 200K زمینه و تاریخ قطع دانش در اکتبر 2023 است."
- },
"openai/o1-mini": {
"description": "o1-mini یک مدل استنتاج سریع و مقرونبهصرفه است که برای برنامهنویسی، ریاضیات و کاربردهای علمی طراحی شده است. این مدل دارای ۱۲۸ هزار بایت زمینه و تاریخ قطع دانش تا اکتبر ۲۰۲۳ میباشد."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K با قابلیت پردازش متن بسیار بزرگ، قادر به پردازش تا 128K اطلاعات متنی است. این ویژگی بهویژه برای تحلیل کامل و پردازش ارتباطات منطقی طولانیمدت در محتوای متنی طولانی مناسب است و میتواند در ارتباطات متنی پیچیده، پشتیبانی از منطق روان و یکپارچه و ارجاعات متنوع را فراهم کند."
},
+ "qvq-72b-preview": {
+ "description": "مدل QVQ یک مدل تحقیقاتی تجربی است که توسط تیم Qwen توسعه یافته و بر بهبود توانایی استدلال بصری، بهویژه در زمینه استدلال ریاضی تمرکز دارد."
+ },
"qwen-coder-plus-latest": {
"description": "مدل کد Qwen با قابلیتهای جامع."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "مدل زبان بصری فوقالعاده بزرگ Qwen-VL. در مقایسه با نسخه تقویتشده، توانایی استدلال بصری و پیروی از دستورات را دوباره بهبود میبخشد و سطح بالاتری از ادراک و شناخت بصری را ارائه میدهد."
},
+ "qwen-vl-ocr-latest": {
+ "description": "مدل OCR Qwen برای استخراج متن، بر روی توانایی استخراج متن از انواع تصاویر مانند اسناد، جداول، سوالات و متنهای دستنویس تمرکز دارد. این مدل قادر به شناسایی انواع مختلف متون است و زبانهای پشتیبانی شده شامل: چینی، انگلیسی، فرانسوی، ژاپنی، کرهای، آلمانی، روسی، ایتالیایی، ویتنامی و عربی میباشد."
+ },
"qwen-vl-plus-latest": {
"description": "نسخه تقویتشده مدل زبان تصویری بزرگ تونگی چیانون. بهبود قابل توجه در توانایی تشخیص جزئیات و شناسایی متن، پشتیبانی از وضوح بیش از یک میلیون پیکسل و تصاویر با هر نسبت طول به عرض."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 یک سری جدید از مدلهای زبان بزرگ است که دارای تواناییهای درک و تولید قویتری میباشد."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "مدل LLM برای زبانهای چینی و انگلیسی که در زمینههای زبان، برنامهنویسی، ریاضیات و استدلال تخصص دارد."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "مدل LLM پیشرفته که از تولید کد، استدلال و اصلاح پشتیبانی میکند و شامل زبانهای برنامهنویسی اصلی است."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "مدل کد قدرتمند و متوسط که از طول زمینه 32K پشتیبانی میکند و در برنامهنویسی چند زبانه مهارت دارد."
+ },
"qwen2": {
"description": "Qwen2 مدل زبان بزرگ نسل جدید علیبابا است که با عملکرد عالی از نیازهای متنوع کاربردی پشتیبانی میکند."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "مدل 14B مقیاس Qwen 2.5 که به صورت منبع باز ارائه شده است."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "مدل 72B مقیاس Qwen2.5 که به صورت متنباز ارائه شده است."
+ },
"qwen2.5-32b-instruct": {
"description": "مدل 32B مقیاس Qwen 2.5 که به صورت منبع باز ارائه شده است."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "مدل Qwen-Math دارای توانایی قوی در حل مسائل ریاضی است."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "پیروی از دستورات، ریاضیات، حل مسائل، بهبود کلی کد، بهبود توانایی شناسایی همه چیز، پشتیبانی از فرمتهای مختلف برای شناسایی دقیق عناصر بصری، پشتیبانی از درک فایلهای ویدیویی طولانی (حداکثر 10 دقیقه) و شناسایی لحظات رویداد در سطح ثانیه، توانایی درک زمان و سرعت، بر اساس توانایی تجزیه و تحلیل و شناسایی، پشتیبانی از کنترل عاملهای OS یا Mobile، توانایی استخراج اطلاعات کلیدی و خروجی به فرمت Json قوی، این نسخه 72B است و قویترین نسخه در این سری است."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "پیروی از دستورات، ریاضیات، حل مسائل، بهبود کلی کد، بهبود توانایی شناسایی همه چیز، پشتیبانی از فرمتهای مختلف برای شناسایی دقیق عناصر بصری، پشتیبانی از درک فایلهای ویدیویی طولانی (حداکثر 10 دقیقه) و شناسایی لحظات رویداد در سطح ثانیه، توانایی درک زمان و سرعت، بر اساس توانایی تجزیه و تحلیل و شناسایی، پشتیبانی از کنترل عاملهای OS یا Mobile، توانایی استخراج اطلاعات کلیدی و خروجی به فرمت Json قوی، این نسخه 72B است و قویترین نسخه در این سری است."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 نسل جدید مدل زبانی مقیاس بزرگ Alibaba است که با عملکرد عالی از نیازهای متنوع کاربردی پشتیبانی میکند."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro یک مدل هوش مصنوعی پیشرفته از Upstage است که بر توانایی پیروی از دستورات با استفاده از یک GPU تمرکز دارد و امتیاز IFEval بالای 80 را کسب کرده است. در حال حاضر از زبان انگلیسی پشتیبانی میکند و نسخه رسمی آن برای نوامبر 2024 برنامهریزی شده است که پشتیبانی از زبانهای بیشتر و طول زمینه را گسترش خواهد داد."
},
+ "sonar": {
+ "description": "محصول جستجوی سبک بر اساس زمینه جستجو که سریعتر و ارزانتر از Sonar Pro است."
+ },
+ "sonar-pro": {
+ "description": "محصول جستجوی پیشرفته که از جستجوی زمینه پشتیبانی میکند و قابلیتهای پیشرفتهای برای پرسش و پیگیری دارد."
+ },
+ "sonar-reasoning": {
+ "description": "محصول جدید API که توسط مدل استدلال DeepSeek پشتیبانی میشود."
+ },
"step-1-128k": {
"description": "تعادل بین عملکرد و هزینه، مناسب برای سناریوهای عمومی."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "این مدل دارای تواناییهای قوی در درک ویدیو است."
},
+ "step-1o-vision-32k": {
+ "description": "این مدل دارای تواناییهای قوی در درک تصویر است. در مقایسه با مدلهای سری step-1v، عملکرد بصری بهتری دارد."
+ },
"step-1v-32k": {
"description": "پشتیبانی از ورودی بصری، تقویت تجربه تعامل چندحالته."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "پشتیبانی از تعاملات متنی گسترده، مناسب برای سناریوهای مکالمه پیچیده."
},
+ "step-2-mini": {
+ "description": "مدل بزرگ فوقالعاده سریع مبتنی بر معماری توجه MFA که بهطور خودجوش توسعه یافته است، با هزینه بسیار کم به نتایجی مشابه با مرحله ۱ دست مییابد و در عین حال توانایی پردازش بالاتر و زمان پاسخ سریعتری را حفظ میکند. این مدل قادر به انجام وظایف عمومی است و در تواناییهای کدنویسی تخصص دارد."
+ },
+ "taichu2_mm": {
+ "description": "ترکیبی از درک تصویر، انتقال دانش، استدلال منطقی و غیره، در زمینه پرسش و پاسخ تصویری و متنی عملکرد برجستهای دارد."
+ },
"taichu_llm": {
"description": "Taichu 2.0 بر اساس حجم زیادی از دادههای با کیفیت بالا آموزش دیده است و دارای تواناییهای قویتری در درک متن، تولید محتوا، پرسش و پاسخ در مکالمه و غیره میباشد."
},
+ "text-embedding-3-large": {
+ "description": "قدرتمندترین مدل وکتور سازی، مناسب برای وظایف انگلیسی و غیرانگلیسی."
+ },
+ "text-embedding-3-small": {
+ "description": "مدل جدید و کارآمد Embedding، مناسب برای جستجوی دانش، کاربردهای RAG و سایر سناریوها."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) با استفاده از استراتژیها و معماری مدل کارآمد، توان محاسباتی بهبودیافتهای را ارائه میدهد."
},
+ "tts-1": {
+ "description": "جدیدترین مدل تبدیل متن به گفتار، بهینهسازی شده برای سرعت در سناریوهای زنده."
+ },
+ "tts-1-hd": {
+ "description": "جدیدترین مدل تبدیل متن به گفتار، بهینهسازی شده برای کیفیت."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) مناسب برای وظایف دقیق دستوری، ارائهدهنده تواناییهای برجسته در پردازش زبان."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet استانداردهای صنعتی را ارتقا داده و عملکردی فراتر از مدلهای رقیب و Claude 3 Opus دارد و در ارزیابیهای گستردهای عملکرد عالی از خود نشان میدهد، در حالی که سرعت و هزینه مدلهای سطح متوسط ما را نیز داراست."
},
+ "whisper-1": {
+ "description": "مدل شناسایی گفتار عمومی، پشتیبانی از شناسایی گفتار چند زبانه، ترجمه گفتار و شناسایی زبان."
+ },
"wizardlm2": {
"description": "WizardLM 2 یک مدل زبانی ارائه شده توسط هوش مصنوعی مایکروسافت است که در مکالمات پیچیده، چندزبانه، استدلال و دستیارهای هوشمند عملکرد برجستهای دارد."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "مدل وظایف پیچیده بینایی، ارائه دهنده قابلیتهای درک و تحلیل تصویر با عملکرد بالا."
+ },
+ "yi-vision-v2": {
+ "description": "مدلهای پیچیده بصری که قابلیتهای درک و تحلیل با عملکرد بالا را بر اساس چندین تصویر ارائه میدهند."
}
}
diff --git a/locales/fa-IR/providers.json b/locales/fa-IR/providers.json
index 908fb0bbac5b1..67d6c771107d7 100644
--- a/locales/fa-IR/providers.json
+++ b/locales/fa-IR/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure انواع مدلهای پیشرفته AI را ارائه میدهد، از جمله GPT-3.5 و جدیدترین سری GPT-4، که از انواع دادهها و وظایف پیچیده پشتیبانی میکند و به ارائه راهحلهای AI ایمن، قابل اعتماد و پایدار متعهد است."
},
+ "azureai": {
+ "description": "Azure مجموعهای از مدلهای پیشرفته AI را ارائه میدهد، از جمله GPT-3.5 و جدیدترین سری GPT-4، که از انواع مختلف دادهها و وظایف پیچیده پشتیبانی میکند و به دنبال راهحلهای AI ایمن، قابل اعتماد و پایدار است."
+ },
"baichuan": {
"description": "بایچوان هوش مصنوعی یک شرکت متمرکز بر توسعه مدلهای بزرگ هوش مصنوعی است. مدلهای این شرکت در وظایف چینی مانند دانشنامه، پردازش متون طولانی و تولید محتوا عملکرد برجستهای دارند و از مدلهای اصلی خارجی پیشی گرفتهاند. بایچوان هوش مصنوعی همچنین دارای تواناییهای چندوجهی پیشرو در صنعت است و در چندین ارزیابی معتبر عملکرد عالی داشته است. مدلهای آن شامل Baichuan 4، Baichuan 3 Turbo و Baichuan 3 Turbo 128k هستند که برای سناریوهای مختلف بهینهسازی شدهاند و راهحلهای مقرونبهصرفهای ارائه میدهند."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek یک شرکت متمرکز بر تحقیق و کاربرد فناوری هوش مصنوعی است. مدل جدید آن، DeepSeek-V2.5، تواناییهای مکالمه عمومی و پردازش کد را ترکیب کرده و در زمینههایی مانند همترازی با ترجیحات انسانی، وظایف نوشتاری و پیروی از دستورات بهبود قابل توجهی داشته است."
},
+ "doubao": {
+ "description": "مدل بزرگ خودساخته شده توسط بایتدANCE. با تأیید در بیش از 50 سناریوی تجاری داخلی بایتدANCE، با استفاده روزانه از تریلیونها توکن، به طور مداوم بهبود یافته و تواناییهای چندگانهای را ارائه میدهد تا تجربههای تجاری غنی را با کیفیت مدل بالا برای شرکتها ایجاد کند."
+ },
"fireworksai": {
"description": "Fireworks AI یک ارائهدهنده پیشرو در خدمات مدلهای زبان پیشرفته است که بر فراخوانی توابع و پردازش چندوجهی تمرکز دارد. جدیدترین مدل آن، Firefunction V2، بر اساس Llama-3 ساخته شده و برای فراخوانی توابع، مکالمه و پیروی از دستورات بهینهسازی شده است. مدل زبان تصویری FireLLaVA-13B از ورودیهای ترکیبی تصویر و متن پشتیبانی میکند. سایر مدلهای قابل توجه شامل سری Llama و سری Mixtral هستند که پشتیبانی کارآمدی از پیروی دستورات چندزبانه و تولید ارائه میدهند."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "سازمان متن باز متعهد به تحقیق و توسعه ابزارهای مدلهای بزرگ. ارائه یک پلتفرم متن باز کارآمد و آسان برای تمام توسعهدهندگان هوش مصنوعی، تا جدیدترین مدلها و تکنیکهای الگوریتمی در دسترس باشد."
},
+ "jina": {
+ "description": "Jina AI در سال 2020 تأسیس شد و یک شرکت پیشرو در زمینه AI جستجو است. پلتفرم پایه جستجوی ما شامل مدلهای برداری، بازچینشگرها و مدلهای زبانی کوچک است که به کسبوکارها کمک میکند تا برنامههای جستجوی تولیدی و چندرسانهای قابل اعتماد و با کیفیت بالا بسازند."
+ },
+ "lmstudio": {
+ "description": "LM Studio یک برنامه دسکتاپ برای توسعه و آزمایش LLM ها بر روی رایانه شما است."
+ },
"minimax": {
"description": "MiniMax یک شرکت فناوری هوش مصنوعی عمومی است که در سال 2021 تأسیس شد و به همکاری با کاربران برای ایجاد هوش مصنوعی متعهد است. MiniMax بهطور مستقل مدلهای بزرگ عمومی چندگانهای را توسعه داده است، از جمله مدل متنی MoE با تریلیونها پارامتر، مدل صوتی و مدل تصویری. همچنین برنامههایی مانند حلزون AI را معرفی کرده است."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI یک پلتفرم ارائهدهنده خدمات API برای مدلهای بزرگ زبانی و تولید تصاویر هوش مصنوعی است که انعطافپذیر، قابلاعتماد و مقرونبهصرفه میباشد. این پلتفرم از جدیدترین مدلهای متنباز مانند Llama3 و Mistral پشتیبانی میکند و راهحلهای API جامع، کاربرپسند و خودکار برای توسعه برنامههای هوش مصنوعی مولد ارائه میدهد که مناسب رشد سریع استارتاپهای هوش مصنوعی است."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ کانتینرهایی را ارائه میدهد که میتوانند برای استنتاج میکروسرویسهای GPU تسریع شده خود میزبان استفاده شوند و از استقرار مدلهای AI پیشآموزشدیده و سفارشی در ابر، مراکز داده، رایانههای شخصی RTX™ AI و ایستگاههای کاری پشتیبانی میکند."
+ },
"ollama": {
"description": "مدلهای ارائهشده توسط Ollama طیف گستردهای از تولید کد، محاسبات ریاضی، پردازش چندزبانه و تعاملات گفتگویی را پوشش میدهند و از نیازهای متنوع استقرار در سطح سازمانی و محلی پشتیبانی میکنند."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "موسسه اتوماسیون آکادمی علوم چین و موسسه هوش مصنوعی ووهان نسل جدیدی از مدلهای چندوجهی را معرفی کردهاند که از پرسش و پاسخ چندمرحلهای، تولید متن، تولید تصویر، درک سهبعدی، تحلیل سیگنال و سایر وظایف جامع پرسش و پاسخ پشتیبانی میکند. این مدل دارای تواناییهای شناختی، درک و خلاقیت قویتری است و تجربه تعاملی جدیدی را به ارمغان میآورد."
},
+ "tencentcloud": {
+ "description": "قدرت اتمی موتور دانش (LLM Knowledge Engine Atomic Power) بر اساس موتور دانش توسعه یافته و قابلیت کامل پرسش و پاسخ را برای شرکتها و توسعهدهندگان ارائه میدهد. شما میتوانید با استفاده از چندین قدرت اتمی، خدمات مدل اختصاصی خود را بسازید و از خدماتی مانند تجزیه و تحلیل اسناد، تقسیم، جاسازی، بازنویسی چند دور و غیره برای سفارشیسازی کسب و کار هوش مصنوعی اختصاصی خود استفاده کنید."
+ },
"togetherai": {
"description": "Together AI متعهد به دستیابی به عملکرد پیشرو از طریق مدلهای نوآورانه هوش مصنوعی است و قابلیتهای سفارشیسازی گستردهای را ارائه میدهد، از جمله پشتیبانی از مقیاسپذیری سریع و فرآیندهای استقرار شهودی، که نیازهای مختلف شرکتها را برآورده میکند."
},
"upstage": {
"description": "Upstage بر توسعه مدلهای هوش مصنوعی برای نیازهای مختلف تجاری تمرکز دارد، از جمله Solar LLM و هوش مصنوعی اسناد، که هدف آن دستیابی به هوش عمومی مصنوعی (AGI) برای کار است. با استفاده از Chat API، میتوانید نمایندگان مکالمه ساده ایجاد کنید و از قابلیتهای فراخوانی عملکرد، ترجمه، تعبیه و کاربردهای خاص حوزه پشتیبانی کنید."
},
+ "vllm": {
+ "description": "vLLM یک کتابخانه سریع و آسان برای استفاده است که برای استنتاج و خدمات LLM طراحی شده است."
+ },
+ "volcengine": {
+ "description": "پلتفرم توسعه خدمات مدلهای بزرگ که توسط بایتدANCE راهاندازی شده است، خدمات فراوان، ایمن و با قیمت رقابتی برای فراخوانی مدلها را ارائه میدهد. همچنین امکاناتی از جمله دادههای مدل، تنظیم دقیق، استنتاج و ارزیابی را به صورت end-to-end فراهم میکند و به طور جامع از توسعه و پیادهسازی برنامههای هوش مصنوعی شما حمایت میکند."
+ },
"wenxin": {
"description": "پلتفرم جامع توسعه و خدمات مدلهای بزرگ و برنامههای بومی هوش مصنوعی در سطح سازمانی، ارائهدهنده کاملترین و کاربرپسندترین زنجیره ابزارهای توسعه مدلهای هوش مصنوعی مولد و توسعه برنامهها"
},
diff --git a/locales/fa-IR/setting.json b/locales/fa-IR/setting.json
index 6e232668d53bb..295320b2a8ae0 100644
--- a/locales/fa-IR/setting.json
+++ b/locales/fa-IR/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "فعالسازی محدودیت پاسخ"
},
+ "enableReasoningEffort": {
+ "title": "فعالسازی تنظیم شدت استدلال"
+ },
"frequencyPenalty": {
- "desc": "هرچه مقدار بیشتر باشد، احتمال کاهش تکرار کلمات بیشتر است",
- "title": "مجازات تکرار"
+ "desc": "هر چه مقدار بزرگتر باشد، واژگان متنوعتر و غنیتری استفاده میشود؛ هر چه مقدار کوچکتر باشد، واژگان سادهتر و عادیتر خواهند بود.",
+ "title": "تنوع واژگان"
},
"maxTokens": {
"desc": "حداکثر تعداد توکنهای استفادهشده در هر تعامل",
@@ -212,19 +215,31 @@
"desc": "مدل {{provider}}",
"title": "مدل"
},
+ "params": {
+ "title": "پارامترهای پیشرفته"
+ },
"presencePenalty": {
- "desc": "هرچه مقدار بیشتر باشد، احتمال گسترش به موضوعات جدید بیشتر است",
- "title": "تازگی موضوع"
+ "desc": "هر چه مقدار بزرگتر باشد، تمایل به استفاده از عبارات مختلف بیشتر میشود و از تکرار مفاهیم جلوگیری میکند؛ هر چه مقدار کوچکتر باشد، تمایل به استفاده از مفاهیم یا روایتهای تکراری بیشتر میشود و بیان یکدستتری خواهد داشت.",
+ "title": "گستردگی بیان"
+ },
+ "reasoningEffort": {
+ "desc": "هرچه مقدار بیشتر باشد، توانایی استدلال قویتر است، اما ممکن است زمان پاسخ و مصرف توکن را افزایش دهد",
+ "options": {
+ "high": "بالا",
+ "low": "پایین",
+ "medium": "متوسط"
+ },
+ "title": "شدت استدلال"
},
"temperature": {
- "desc": "هرچه مقدار بیشتر باشد، پاسخها تصادفیتر خواهند بود",
- "title": "تصادفی بودن",
- "titleWithValue": "تصادفی بودن {{value}}"
+ "desc": "هر چه عدد بزرگتر باشد، پاسخها خلاقانهتر و تخیلیتر خواهند بود؛ هر چه عدد کوچکتر باشد، پاسخها دقیقتر خواهند بود",
+ "title": "فعالیت خلاقانه",
+ "warning": "اگر عدد فعالیت خلاقانه بیش از حد بزرگ باشد، خروجی ممکن است دچار اختلال شود"
},
"title": "تنظیمات مدل",
"topP": {
- "desc": "مشابه تصادفی بودن است، اما نباید همزمان با آن تغییر کند",
- "title": "نمونهگیری هستهای"
+ "desc": "چند احتمال را در نظر میگیرد، هر چه عدد بزرگتر باشد، پاسخهای بیشتری را میپذیرد؛ هر چه عدد کوچکتر باشد، تمایل به انتخاب پاسخهای محتملتر دارد. تغییر همزمان با فعالیت خلاقانه توصیه نمیشود",
+ "title": "باز بودن ذهن"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "تنظیمات عمومی",
"experiment": "آزمایش",
"llm": "مدل زبان",
+ "provider": "ارائه دهنده خدمات هوش مصنوعی",
"sync": "همگامسازی ابری",
"system-agent": "دستیار سیستم",
"tts": "خدمات صوتی"
diff --git a/locales/fr-FR/changelog.json b/locales/fr-FR/changelog.json
index 2f9f68c59c736..ecff38c56de44 100644
--- a/locales/fr-FR/changelog.json
+++ b/locales/fr-FR/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "Voir tous les journaux de mise à jour",
"description": "Suivez en continu les nouvelles fonctionnalités et améliorations de {{appName}}",
"pagination": {
- "older": "Voir les modifications antérieures",
- "prev": "Page précédente"
+ "next": "Page suivante",
+ "older": "Voir les modifications antérieures"
},
"readDetails": "Lire les détails",
"title": "Journal des mises à jour",
diff --git a/locales/fr-FR/common.json b/locales/fr-FR/common.json
index ca37d5b2920ee..d4122f5f67c98 100644
--- a/locales/fr-FR/common.json
+++ b/locales/fr-FR/common.json
@@ -275,6 +275,7 @@
},
"temp": "Temporaire",
"terms": "Conditions de service",
+ "update": "Mise à jour",
"updateAgent": "Mettre à jour les informations de l'agent",
"upgradeVersion": {
"action": "Mettre à jour",
@@ -285,6 +286,7 @@
"anonymousNickName": "Utilisateur anonyme",
"billing": "Gestion de la facturation",
"cloud": "Découvrir {{name}}",
+ "community": "Version communautaire",
"data": "Stockage des données",
"defaultNickname": "Utilisateur de la version communautaire",
"discord": "Support de la communauté",
@@ -294,7 +296,6 @@
"help": "Centre d'aide",
"moveGuide": "Le bouton de configuration a été déplacé ici",
"plans": "Forfaits d'abonnement",
- "preview": "Aperçu",
"profile": "Gestion du compte",
"setting": "Paramètres de l'application",
"usages": "Statistiques d'utilisation"
diff --git a/locales/fr-FR/components.json b/locales/fr-FR/components.json
index 668dae2116ecd..eb2347cec6ebf 100644
--- a/locales/fr-FR/components.json
+++ b/locales/fr-FR/components.json
@@ -76,6 +76,7 @@
"custom": "Modèle personnalisé par défaut prenant en charge à la fois les appels de fonction et la reconnaissance visuelle. Veuillez vérifier la disponibilité de ces capacités en fonction de vos besoins réels.",
"file": "Ce modèle prend en charge la lecture et la reconnaissance de fichiers téléchargés.",
"functionCall": "Ce modèle prend en charge les appels de fonction.",
+ "reasoning": "Ce modèle prend en charge une réflexion approfondie",
"tokens": "Ce modèle prend en charge jusqu'à {{tokens}} jetons par session.",
"vision": "Ce modèle prend en charge la reconnaissance visuelle."
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "Aucun modèle activé. Veuillez vous rendre dans les paramètres pour l'activer.",
"provider": "Fournisseur"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "En raison des restrictions de sécurité des navigateurs, vous devez configurer les paramètres CORS pour utiliser Ollama correctement.",
+ "linux": {
+ "env": "Ajoutez `Environment` sous la section [Service], et ajoutez la variable d'environnement OLLAMA_ORIGINS :",
+ "reboot": "Rechargez systemd et redémarrez Ollama",
+ "systemd": "Appelez systemd pour éditer le service ollama :"
+ },
+ "macos": "Veuillez ouvrir l'application « Terminal », collez la commande suivante et appuyez sur Entrée pour l'exécuter",
+ "reboot": "Veuillez redémarrer le service Ollama après l'exécution",
+ "title": "Configurer Ollama pour autoriser l'accès CORS",
+ "windows": "Sous Windows, cliquez sur « Panneau de configuration », puis accédez à l'édition des variables d'environnement système. Créez une nouvelle variable d'environnement nommée « OLLAMA_ORIGINS » pour votre compte utilisateur, avec la valeur * , puis cliquez sur « OK/Appliquer » pour enregistrer"
+ },
+ "install": {
+ "description": "Veuillez vous assurer que vous avez démarré Ollama. Si vous n'avez pas téléchargé Ollama, veuillez vous rendre sur le site officiel <1>pour le télécharger",
+ "docker": "Si vous préférez utiliser Docker, Ollama propose également une image Docker officielle que vous pouvez tirer avec la commande suivante :",
+ "linux": {
+ "command": "Installez avec la commande suivante :",
+ "manual": "Ou, vous pouvez également consulter le <1>guide d'installation manuelle pour Linux pour l'installer vous-même"
+ },
+ "title": "Installer et démarrer l'application Ollama localement",
+ "windowsTab": "Windows (version préliminaire)"
+ }
+ },
+ "Thinking": {
+ "thinking": "En pleine réflexion...",
+ "thought": "Pensée approfondie (durée : {{duration}} secondes)",
+ "thoughtWithDuration": "Pensée approfondie"
}
}
diff --git a/locales/fr-FR/discover.json b/locales/fr-FR/discover.json
index 508b8d010bb06..5fb3a7588ef88 100644
--- a/locales/fr-FR/discover.json
+++ b/locales/fr-FR/discover.json
@@ -126,6 +126,10 @@
"title": "Fraîcheur des sujets"
},
"range": "Plage",
+ "reasoning_effort": {
+ "desc": "Ce paramètre contrôle l'intensité de raisonnement du modèle avant de générer une réponse. Une faible intensité privilégie la rapidité de réponse et économise des tokens, tandis qu'une forte intensité offre un raisonnement plus complet, mais consomme plus de tokens et ralentit la réponse. La valeur par défaut est moyenne, équilibrant précision du raisonnement et rapidité de réponse.",
+ "title": "Intensité de raisonnement"
+ },
"temperature": {
"desc": "Ce paramètre influence la diversité des réponses du modèle. Des valeurs plus basses entraînent des réponses plus prévisibles et typiques, tandis que des valeurs plus élevées encouragent des réponses plus variées et moins courantes. Lorsque la valeur est fixée à 0, le modèle donne toujours la même réponse pour une entrée donnée.",
"title": "Aléatoire"
diff --git a/locales/fr-FR/modelProvider.json b/locales/fr-FR/modelProvider.json
index 4628b03f64c19..40e9dcc47fa1f 100644
--- a/locales/fr-FR/modelProvider.json
+++ b/locales/fr-FR/modelProvider.json
@@ -19,6 +19,24 @@
"title": "Clé API"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Version de l'API Azure, au format YYYY-MM-DD. Consultez la [dernière version](https://learn.microsoft.com/fr-fr/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "Obtenir la liste",
+ "title": "Version de l'API Azure"
+ },
+ "endpoint": {
+ "desc": "Trouvez le point de terminaison d'inférence du modèle Azure AI dans l'aperçu du projet Azure AI",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Point de terminaison Azure AI"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Trouvez la clé API dans l'aperçu du projet Azure AI",
+ "placeholder": "Clé Azure",
+ "title": "Clé"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "Saisissez l'ID de clé d'accès AWS",
@@ -63,6 +81,46 @@
"title": "ID de compte Cloudflare / adresse API"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Veuillez entrer votre clé API",
+ "title": "Clé API"
+ },
+ "basicTitle": "Informations de base",
+ "configTitle": "Informations de configuration",
+ "confirm": "Créer",
+ "createSuccess": "Création réussie",
+ "description": {
+ "placeholder": "Description du fournisseur (facultatif)",
+ "title": "Description du fournisseur"
+ },
+ "id": {
+ "desc": "Identifiant unique du fournisseur de services, qui ne peut pas être modifié après sa création",
+ "format": "Ne peut contenir que des chiffres, des lettres minuscules, des tirets (-) et des underscores (_) ",
+ "placeholder": "Utilisez uniquement des lettres minuscules, par exemple openai, non modifiable après création",
+ "required": "Veuillez entrer l'ID du fournisseur",
+ "title": "ID du fournisseur"
+ },
+ "logo": {
+ "required": "Veuillez télécharger un logo valide pour le fournisseur",
+ "title": "Logo du fournisseur"
+ },
+ "name": {
+ "placeholder": "Veuillez entrer le nom d'affichage du fournisseur",
+ "required": "Veuillez entrer le nom du fournisseur",
+ "title": "Nom du fournisseur"
+ },
+ "proxyUrl": {
+ "required": "Veuillez remplir l'adresse du proxy",
+ "title": "Adresse du proxy"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Veuillez sélectionner le type de SDK",
+ "title": "Format de requête"
+ },
+ "title": "Créer un fournisseur AI personnalisé"
+ },
"github": {
"personalAccessToken": {
"desc": "Entrez votre PAT GitHub, cliquez [ici](https://github.com/settings/tokens) pour en créer un.",
@@ -77,6 +135,23 @@
"title": "Jeton HuggingFace"
}
},
+ "list": {
+ "title": {
+ "disabled": "Fournisseur non activé",
+ "enabled": "Fournisseur activé"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Ajouter un fournisseur personnalisé",
+ "all": "Tout",
+ "list": {
+ "disabled": "Non activé",
+ "enabled": "Activé"
+ },
+ "notFound": "Aucun résultat trouvé",
+ "searchProviders": "Rechercher des fournisseurs...",
+ "sort": "Tri personnalisé"
+ },
"ollama": {
"checker": {
"desc": "Vérifiez si l'adresse du proxy est correctement saisie",
@@ -94,33 +169,9 @@
"title": "Téléchargement du modèle {{model}} en cours"
},
"endpoint": {
- "desc": "Saisissez l'adresse du proxy Ollama, laissez vide si non spécifié localement",
+ "desc": "Doit inclure http(s)://, peut rester vide si non spécifié localement",
"title": "Adresse du proxy"
},
- "setup": {
- "cors": {
- "description": "En raison des restrictions de sécurité du navigateur, vous devez configurer CORS pour Ollama afin de l'utiliser correctement.",
- "linux": {
- "env": "Ajoutez `Environment` sous la section [Service], en ajoutant la variable d'environnement OLLAMA_ORIGINS :",
- "reboot": "Rechargez systemd et redémarrez Ollama",
- "systemd": "Appelez systemd pour éditer le service ollama :"
- },
- "macos": "Veuillez ouvrir l'application « Terminal » et coller la commande suivante, puis appuyez sur Entrée pour l'exécuter.",
- "reboot": "Veuillez redémarrer le service Ollama après l'exécution.",
- "title": "Configurer Ollama pour autoriser l'accès CORS",
- "windows": "Sous Windows, cliquez sur « Panneau de configuration », puis accédez à l'édition des variables d'environnement système. Créez une nouvelle variable d'environnement nommée « OLLAMA_ORIGINS » pour votre compte utilisateur, avec la valeur *, puis cliquez sur « OK/Appliquer » pour enregistrer."
- },
- "install": {
- "description": "Veuillez vous assurer que vous avez démarré Ollama. Si vous n'avez pas téléchargé Ollama, veuillez vous rendre sur le site officiel <1>pour le télécharger.",
- "docker": "Si vous préférez utiliser Docker, Ollama propose également une image Docker officielle que vous pouvez tirer avec la commande suivante :",
- "linux": {
- "command": "Installez avec la commande suivante :",
- "manual": "Ou, vous pouvez également consulter le <1>guide d'installation manuelle pour Linux pour l'installer vous-même."
- },
- "title": "Installer et démarrer l'application Ollama localement",
- "windowsTab": "Windows (version préliminaire)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Annuler le téléchargement",
@@ -131,25 +182,145 @@
"title": "Télécharger le modèle Ollama spécifié"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Entrez la clé d'accès de la plateforme Qianfan de Baidu",
- "placeholder": "Clé d'accès Qianfan",
- "title": "Clé d'accès"
+ "providerModels": {
+ "config": {
+ "aesGcm": "Votre clé et votre adresse de proxy seront chiffrées à l'aide de l'algorithme de chiffrement <1>AES-GCM",
+ "apiKey": {
+ "desc": "Veuillez entrer votre {{name}} clé API",
+ "placeholder": "{{name}} clé API",
+ "title": "Clé API"
+ },
+ "baseURL": {
+ "desc": "Doit inclure http(s)://",
+ "invalid": "Veuillez entrer une URL valide",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "Adresse du proxy API"
+ },
+ "checker": {
+ "button": "Vérifier",
+ "desc": "Tester si la clé API et l'adresse de proxy sont correctement renseignées",
+ "pass": "Vérification réussie",
+ "title": "Vérification de connectivité"
+ },
+ "fetchOnClient": {
+ "desc": "Le mode de requête client lancera directement la requête de session depuis le navigateur, ce qui peut améliorer la vitesse de réponse",
+ "title": "Utiliser le mode de requête client"
+ },
+ "helpDoc": "Guide de configuration",
+ "waitingForMore": "D'autres modèles sont en <1>planification d'intégration, restez à l'écoute"
},
- "checker": {
- "desc": "Vérifiez si la clé d'accès / SecretAccess est correctement remplie"
+ "createNew": {
+ "title": "Créer un modèle AI personnalisé"
+ },
+ "item": {
+ "config": "Configurer le modèle",
+ "customModelCards": {
+ "addNew": "Créer et ajouter le modèle {{id}}",
+ "confirmDelete": "Vous allez supprimer ce modèle personnalisé, une fois supprimé, il ne pourra pas être récupéré, veuillez agir avec prudence."
+ },
+ "delete": {
+ "confirm": "Confirmer la suppression du modèle {{displayName}} ?",
+ "success": "Suppression réussie",
+ "title": "Supprimer le modèle"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Champ utilisé pour la demande réelle dans Azure OpenAI",
+ "placeholder": "Veuillez entrer le nom de déploiement du modèle dans Azure",
+ "title": "Nom de déploiement du modèle"
+ },
+ "deployName": {
+ "extra": "Ce champ sera utilisé comme ID de modèle lors de l'envoi de la demande",
+ "placeholder": "Veuillez entrer le nom ou l'ID de déploiement réel du modèle",
+ "title": "Nom de déploiement du modèle"
+ },
+ "displayName": {
+ "placeholder": "Veuillez entrer le nom d'affichage du modèle, par exemple ChatGPT, GPT-4, etc.",
+ "title": "Nom d'affichage du modèle"
+ },
+ "files": {
+ "extra": "La mise en œuvre actuelle du téléchargement de fichiers n'est qu'une solution de contournement, à essayer à vos risques et périls. Veuillez attendre la mise en œuvre complète des capacités de téléchargement de fichiers.",
+ "title": "Téléchargement de fichiers pris en charge"
+ },
+ "functionCall": {
+ "extra": "Cette configuration activera uniquement la capacité du modèle à utiliser des outils, permettant ainsi d'ajouter des plugins de type outil au modèle. Cependant, la prise en charge de l'utilisation réelle des outils dépend entièrement du modèle lui-même, veuillez tester la disponibilité par vous-même.",
+ "title": "Support de l'utilisation des outils"
+ },
+ "id": {
+ "extra": "Une fois créé, il ne peut pas être modifié et sera utilisé comme identifiant du modèle lors de l'appel à l'IA",
+ "placeholder": "Veuillez entrer l'identifiant du modèle, par exemple gpt-4o ou claude-3.5-sonnet",
+ "title": "ID du modèle"
+ },
+ "modalTitle": "Configuration du modèle personnalisé",
+ "reasoning": {
+ "extra": "Cette configuration activera uniquement la capacité de réflexion approfondie du modèle. Les résultats dépendent entièrement du modèle lui-même, veuillez tester si ce modèle possède une capacité de réflexion approfondie utilisable.",
+ "title": "Support de la réflexion approfondie"
+ },
+ "tokens": {
+ "extra": "Définir le nombre maximal de tokens pris en charge par le modèle",
+ "title": "Fenêtre de contexte maximale",
+ "unlimited": "Illimité"
+ },
+ "vision": {
+ "extra": "Cette configuration n'activera que la configuration de téléchargement d'images dans l'application, la prise en charge de la reconnaissance dépend entièrement du modèle lui-même, veuillez tester la disponibilité des capacités de reconnaissance visuelle de ce modèle.",
+ "title": "Reconnaissance visuelle prise en charge"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/image",
+ "inputCharts": "${{amount}}/M caractères",
+ "inputMinutes": "${{amount}}/minutes",
+ "inputTokens": "Entrée ${{amount}}/M",
+ "outputTokens": "Sortie ${{amount}}/M"
+ },
+ "releasedAt": "Publié le {{releasedAt}}"
},
- "secretKey": {
- "desc": "Entrez la clé secrète de la plateforme Qianfan de Baidu",
- "placeholder": "Clé secrète Qianfan",
- "title": "Clé secrète"
+ "list": {
+ "addNew": "Ajouter un modèle",
+ "disabled": "Non activé",
+ "disabledActions": {
+ "showMore": "Afficher tout"
+ },
+ "empty": {
+ "desc": "Veuillez créer un modèle personnalisé ou importer un modèle pour commencer à l'utiliser.",
+ "title": "Aucun modèle disponible"
+ },
+ "enabled": "Activé",
+ "enabledActions": {
+ "disableAll": "Désactiver tout",
+ "enableAll": "Activer tout",
+ "sort": "Trier les modèles personnalisés"
+ },
+ "enabledEmpty": "Aucun modèle activé pour le moment, veuillez activer vos modèles préférés dans la liste ci-dessous~",
+ "fetcher": {
+ "clear": "Effacer les modèles récupérés",
+ "fetch": "Récupérer la liste des modèles",
+ "fetching": "Récupération de la liste des modèles en cours...",
+ "latestTime": "Dernière mise à jour : {{time}}",
+ "noLatestTime": "Aucune liste récupérée pour le moment"
+ },
+ "resetAll": {
+ "conform": "Êtes-vous sûr de vouloir réinitialiser toutes les modifications du modèle actuel ? Après la réinitialisation, la liste des modèles actuels reviendra à l'état par défaut",
+ "success": "Réinitialisation réussie",
+ "title": "Réinitialiser toutes les modifications"
+ },
+ "search": "Rechercher des modèles...",
+ "searchResult": "Trouvé {{count}} modèle(s)",
+ "title": "Liste des modèles",
+ "total": "Un total de {{count}} modèles disponibles"
},
- "unlock": {
- "customRegion": "Région de service personnalisée",
- "description": "Entrez votre AccessKey / SecretKey pour commencer la session. L'application ne conservera pas votre configuration d'authentification",
- "title": "Utiliser les informations d'authentification personnalisées de Wenxin Yiyan"
- }
+ "searchNotFound": "Aucun résultat trouvé"
+ },
+ "sortModal": {
+ "success": "Mise à jour du tri réussie",
+ "title": "Tri personnalisé",
+ "update": "Mettre à jour"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "Vous allez supprimer ce fournisseur AI, une fois supprimé, il ne pourra pas être récupéré, confirmez-vous la suppression ?",
+ "deleteSuccess": "Suppression réussie",
+ "tooltip": "Mettre à jour la configuration de base du fournisseur",
+ "updateSuccess": "Mise à jour réussie"
},
"zeroone": {
"title": "01.AI Zéro Un Tout"
diff --git a/locales/fr-FR/models.json b/locales/fr-FR/models.json
index fcd38911fda1b..c07c00ca5ea41 100644
--- a/locales/fr-FR/models.json
+++ b/locales/fr-FR/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K met l'accent sur la sécurité sémantique et l'orientation vers la responsabilité, conçu pour des scénarios d'application exigeant une sécurité de contenu élevée, garantissant l'exactitude et la robustesse de l'expérience utilisateur."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 utilise une recherche arborescente pour construire des chaînes de pensée et introduit un mécanisme de réflexion, entraîné par apprentissage par renforcement, permettant au modèle d'avoir des capacités d'auto-réflexion et de correction."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro est un modèle avancé de traitement du langage naturel lancé par la société 360, offrant d'excellentes capacités de génération et de compréhension de texte, en particulier dans le domaine de la création et de la génération."
},
+ "360zhinao2-o1": {
+ "description": "Le modèle 360zhinao2-o1 utilise une recherche arborescente pour construire une chaîne de pensée et introduit un mécanisme de réflexion, formé par apprentissage par renforcement, permettant au modèle d'avoir la capacité de réflexion et de correction autonome."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra est la version la plus puissante de la série de grands modèles Xinghuo, améliorant la compréhension et la capacité de résumé du contenu textuel tout en mettant à jour le lien de recherche en ligne. C'est une solution complète pour améliorer la productivité au bureau et répondre avec précision aux besoins, représentant un produit intelligent de premier plan dans l'industrie."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "Le modèle le plus performant en Chine, surpassant les modèles dominants étrangers dans les tâches en chinois telles que les encyclopédies, les longs textes et la création. Il possède également des capacités multimodales de pointe, avec d'excellentes performances dans plusieurs évaluations de référence."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Le modèle distillé DeepSeek-R1 basé sur Qwen2.5-Math-1.5B optimise les performances d'inférence grâce à l'apprentissage par renforcement et aux données de démarrage à froid, rafraîchissant les références multi-tâches des modèles open source."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Le modèle distillé DeepSeek-R1 basé sur Qwen2.5-14B optimise les performances d'inférence grâce à l'apprentissage par renforcement et aux données de démarrage à froid, rafraîchissant les références multi-tâches des modèles open source."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "La série DeepSeek-R1 optimise les performances d'inférence grâce à l'apprentissage par renforcement et aux données de démarrage à froid, rafraîchissant les références multi-tâches des modèles open source, dépassant le niveau d'OpenAI-o1-mini."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "Le modèle distillé DeepSeek-R1 basé sur Qwen2.5-Math-7B optimise les performances d'inférence grâce à l'apprentissage par renforcement et aux données de démarrage à froid, rafraîchissant les références multi-tâches des modèles open source."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite présente une rapidité de réponse exceptionnelle et un excellent rapport qualité-prix, offrant des choix plus flexibles pour différents scénarios clients. Prend en charge le raisonnement et le réglage fin avec une fenêtre de contexte de 128k."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Modèle de langage ultra-large de premier plan développé par Baidu, ayant réalisé une mise à niveau complète des capacités par rapport à ERNIE 3.5, largement applicable à des scénarios de tâches complexes dans divers domaines ; prend en charge l'intégration automatique avec le plugin de recherche Baidu, garantissant l'actualité des informations de réponse."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Modèle de langage ultra-large de premier plan développé par Baidu, offrant d'excellentes performances globales, largement applicable à des scénarios de tâches complexes dans divers domaines ; prend en charge l'intégration automatique avec le plugin de recherche Baidu, garantissant la pertinence des informations de réponse. Par rapport à ERNIE 4.0, il offre de meilleures performances."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "Modèle linguistique ultramoderne et de grande taille auto-développé par Baidu, avec d'excellentes performances générales, largement applicable à divers scénarios de tâches complexes ; prend en charge la connexion automatique aux plugins de recherche Baidu pour assurer la pertinence des informations de réponse. Par rapport à ERNIE 4.0, il affiche de meilleures performances."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct est l'un des derniers modèles de langage à grande échelle publiés par Alibaba Cloud. Ce modèle 7B présente des capacités considérablement améliorées dans des domaines tels que le codage et les mathématiques. Le modèle offre également un support multilingue, couvrant plus de 29 langues, y compris le chinois et l'anglais. Il a montré des améliorations significatives dans le suivi des instructions, la compréhension des données structurées et la génération de sorties structurées (en particulier JSON)."
},
+ "MiniMax-Text-01": {
+ "description": "Dans la série de modèles MiniMax-01, nous avons réalisé une innovation audacieuse : la première mise en œuvre à grande échelle d'un mécanisme d'attention linéaire, rendant l'architecture Transformer traditionnelle non plus le seul choix. Ce modèle possède un nombre de paramètres atteignant 456 milliards, avec 45,9 milliards d'activations par instance. Les performances globales du modèle rivalisent avec celles des meilleurs modèles étrangers, tout en étant capable de traiter efficacement un contexte mondial de 4 millions de tokens, soit 32 fois celui de GPT-4o et 20 fois celui de Claude-3.5-Sonnet."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO est une fusion de modèles hautement flexible, visant à offrir une expérience créative exceptionnelle."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 a démontré des performances exceptionnelles sur diverses tâches de langage visuel, y compris la compréhension de documents et de graphiques, la compréhension de texte de scène, l'OCR, ainsi que la résolution de problèmes scientifiques et mathématiques."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 a démontré des performances exceptionnelles sur diverses tâches de langage visuel, y compris la compréhension de documents et de graphiques, la compréhension de texte de scène, l'OCR, ainsi que la résolution de problèmes scientifiques et mathématiques."
- },
"Phi-3-medium-128k-instruct": {
"description": "Même modèle Phi-3-medium, mais avec une taille de contexte plus grande pour RAG ou un prompt à quelques exemples."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat est la version open source de la série de modèles pré-entraînés GLM-4 lancée par Zhipu AI. Ce modèle excelle dans plusieurs domaines tels que la sémantique, les mathématiques, le raisonnement, le code et les connaissances. En plus de prendre en charge des dialogues multi-tours, GLM-4-9B-Chat dispose également de fonctionnalités avancées telles que la navigation sur le web, l'exécution de code, l'appel d'outils personnalisés (Function Call) et le raisonnement sur de longs textes. Le modèle prend en charge 26 langues, y compris le chinois, l'anglais, le japonais, le coréen et l'allemand. Dans plusieurs tests de référence, GLM-4-9B-Chat a montré d'excellentes performances, comme AlignBench-v2, MT-Bench, MMLU et C-Eval. Ce modèle prend en charge une longueur de contexte maximale de 128K, adapté à la recherche académique et aux applications commerciales."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 est un modèle d'inférence piloté par l'apprentissage par renforcement (RL), qui résout les problèmes de répétition et de lisibilité dans le modèle. Avant le RL, DeepSeek-R1 a introduit des données de démarrage à froid, optimisant encore les performances d'inférence. Il se compare à OpenAI-o1 dans les tâches mathématiques, de code et d'inférence, et améliore l'ensemble des performances grâce à des méthodes d'entraînement soigneusement conçues."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 est un modèle de langage à experts mixtes (MoE) avec 671 milliards de paramètres, utilisant une attention potentielle multi-tête (MLA) et une architecture DeepSeekMoE, combinant une stratégie d'équilibrage de charge sans perte auxiliaire pour optimiser l'efficacité d'inférence et d'entraînement. Pré-entraîné sur 14,8 billions de tokens de haute qualité, et affiné par supervision et apprentissage par renforcement, DeepSeek-V3 surpasse d'autres modèles open source et se rapproche des modèles fermés de premier plan."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma est l'une des séries de modèles open source légers et avancés développés par Google. C'est un modèle de langage à grande échelle uniquement décodeur, prenant en charge l'anglais, offrant des poids ouverts, des variantes pré-entraînées et des variantes d'ajustement d'instructions. Le modèle Gemma est adapté à diverses tâches de génération de texte, y compris les questions-réponses, les résumés et le raisonnement. Ce modèle 9B a été formé avec 80 trillions de tokens. Sa taille relativement petite permet de le déployer dans des environnements à ressources limitées, tels que des ordinateurs portables, des ordinateurs de bureau ou votre propre infrastructure cloud, rendant ainsi les modèles d'IA de pointe plus accessibles et favorisant l'innovation."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 est une famille de modèles de langage à grande échelle multilingues développée par Meta, comprenant des variantes pré-entraînées et d'ajustement d'instructions de tailles de paramètres de 8B, 70B et 405B. Ce modèle d'ajustement d'instructions 8B est optimisé pour des scénarios de dialogue multilingue, montrant d'excellentes performances dans plusieurs tests de référence de l'industrie. L'entraînement du modèle a utilisé plus de 150 trillions de tokens de données publiques, et des techniques telles que l'ajustement supervisé et l'apprentissage par renforcement basé sur les retours humains ont été appliquées pour améliorer l'utilité et la sécurité du modèle. Llama 3.1 prend en charge la génération de texte et de code, avec une date limite de connaissances fixée à décembre 2023."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview est un modèle de traitement du langage naturel innovant, capable de gérer efficacement des tâches complexes de génération de dialogues et de compréhension contextuelle."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview est un modèle de recherche développé par l'équipe Qwen, axé sur les capacités de raisonnement visuel, qui possède des avantages uniques dans la compréhension de scènes complexes et la résolution de problèmes mathématiques liés à la vision."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview est le dernier modèle de recherche expérimental de Qwen, axé sur l'amélioration des capacités de raisonnement de l'IA. En explorant des mécanismes complexes tels que le mélange de langues et le raisonnement récursif, ses principaux avantages incluent de puissantes capacités d'analyse de raisonnement, ainsi que des compétences en mathématiques et en programmation. Cependant, il existe également des problèmes de changement de langue, des cycles de raisonnement, des considérations de sécurité et des différences dans d'autres capacités."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct est la dernière version de la série de modèles de langage à grande échelle spécifique au code publiée par Alibaba Cloud. Ce modèle, basé sur Qwen2.5, a été formé avec 55 trillions de tokens, améliorant considérablement les capacités de génération, de raisonnement et de correction de code. Il renforce non seulement les capacités de codage, mais maintient également des avantages en mathématiques et en compétences générales. Le modèle fournit une base plus complète pour des applications pratiques telles que les agents de code."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math se concentre sur la résolution de problèmes dans le domaine des mathématiques, fournissant des réponses professionnelles pour des questions de haute difficulté."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 est la dernière série du modèle Qwen, prenant en charge un contexte de 128k. Comparé aux meilleurs modèles open source actuels, Qwen2-72B surpasse de manière significative les modèles leaders dans des domaines tels que la compréhension du langage naturel, les connaissances, le code, les mathématiques et le multilinguisme."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 est la dernière série du modèle Qwen, capable de surpasser les meilleurs modèles open source de taille équivalente, voire de plus grande taille. Qwen2 7B a obtenu des résultats significatifs dans plusieurs évaluations, en particulier en ce qui concerne la compréhension du code et du chinois."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B est un puissant modèle de langage visuel, prenant en charge le traitement multimodal d'images et de textes, capable de reconnaître avec précision le contenu des images et de générer des descriptions ou des réponses pertinentes."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct est un grand modèle de langage de 14 milliards de paramètres, offrant d'excellentes performances, optimisé pour les scénarios en chinois et multilingues, prenant en charge des applications telles que les questions-réponses intelligentes et la génération de contenu."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct est un grand modèle de langage de 7 milliards de paramètres, prenant en charge les appels de fonction et l'interaction transparente avec des systèmes externes, améliorant considérablement la flexibilité et l'évolutivité. Optimisé pour les scénarios en chinois et multilingues, il prend en charge des applications telles que les questions-réponses intelligentes et la génération de contenu."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct est un modèle d'instructions de programmation basé sur un pré-entraînement à grande échelle, doté d'une puissante capacité de compréhension et de génération de code, capable de traiter efficacement diverses tâches de programmation, particulièrement adapté à la rédaction de code intelligent, à la génération de scripts automatisés et à la résolution de problèmes de programmation."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct est un grand modèle de langage conçu pour la génération de code, la compréhension de code et les scénarios de développement efficaces, avec une échelle de 32 milliards de paramètres, répondant à des besoins de programmation variés."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "Le grand modèle multimodal TeleMM est un modèle de compréhension multimodale développé de manière autonome par China Telecom, capable de traiter des entrées multimodales telles que du texte et des images, prenant en charge des fonctionnalités telles que la compréhension d'images et l'analyse de graphiques, fournissant des services de compréhension intermodale aux utilisateurs. Le modèle peut interagir avec les utilisateurs de manière multimodale, comprendre avec précision le contenu d'entrée, répondre à des questions, assister à la création, et fournir efficacement des informations et un soutien d'inspiration multimodale. Il excelle dans des tâches multimodales telles que la perception fine et le raisonnement logique."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large est le plus grand modèle MoE à architecture Transformer open source de l'industrie, avec un total de 389 milliards de paramètres et 52 milliards de paramètres activés."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct est un modèle de langage à grande échelle de la série Qwen2, avec une taille de paramètre de 72B. Ce modèle est basé sur l'architecture Transformer, utilisant des fonctions d'activation SwiGLU, des biais d'attention QKV et des techniques d'attention par groupe. Il est capable de traiter de grandes entrées. Ce modèle excelle dans la compréhension du langage, la génération, les capacités multilingues, le codage, les mathématiques et le raisonnement dans plusieurs tests de référence, surpassant la plupart des modèles open source et montrant une compétitivité comparable à celle des modèles propriétaires dans certaines tâches."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct est l'un des derniers modèles de langage à grande échelle publiés par Alibaba Cloud. Ce modèle 72B présente des capacités considérablement améliorées dans des domaines tels que le codage et les mathématiques. Le modèle offre également un support multilingue, couvrant plus de 29 langues, y compris le chinois et l'anglais. Il a montré des améliorations significatives dans le suivi des instructions, la compréhension des données structurées et la génération de sorties structurées (en particulier JSON)."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Optimisé pour des scénarios de dialogue en chinois, offrant une capacité de génération de dialogues fluide et conforme aux habitudes d'expression en chinois."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Le modèle d'appel de fonction open source de Fireworks offre d'excellentes capacités d'exécution d'instructions et des caractéristiques personnalisables."
+ "abab7-chat-preview": {
+ "description": "Par rapport à la série de modèles abab6.5, il y a eu une amélioration significative des capacités en matière de textes longs, de mathématiques et d'écriture."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Firefunction-v2, récemment lancé par Fireworks, est un modèle d'appel de fonction performant, développé sur la base de Llama-3 et optimisé pour des scénarios tels que les appels de fonction, les dialogues et le suivi d'instructions."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 est un modèle de langage de grande taille à la pointe de la technologie, optimisé par apprentissage renforcé et données de démarrage à froid, offrant d'excellentes performances en raisonnement, mathématiques et programmation."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b est un modèle de langage visuel capable de recevoir simultanément des entrées d'images et de texte, entraîné sur des données de haute qualité, adapté aux tâches multimodales."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Modèle de langage puissant de Deepseek basé sur un mélange d'experts (MoE), avec un total de 671B de paramètres, activant 37B de paramètres par jeton."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Le modèle d'instructions Llama 3 70B est optimisé pour les dialogues multilingues et la compréhension du langage naturel, surpassant la plupart des modèles concurrents."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Le modèle d'instructions Llama 3 70B (version HF) est conforme aux résultats de l'implémentation officielle, adapté aux tâches de suivi d'instructions de haute qualité."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Le modèle d'instructions Llama 3 8B est optimisé pour les dialogues et les tâches multilingues, offrant des performances exceptionnelles et efficaces."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Modèle d'inférence d'image ajusté par instructions de Meta avec 11B paramètres. Ce modèle est optimisé pour la reconnaissance visuelle, l'inférence d'image, la description d'image et pour répondre à des questions générales sur l'image. Il est capable de comprendre des données visuelles, comme des graphiques et des diagrammes, et de combler le fossé entre la vision et le langage en générant des descriptions textuelles des détails de l'image."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Le modèle d'instructions Llama 3.2 1B est un modèle multilingue léger lancé par Meta. Ce modèle vise à améliorer l'efficacité, offrant des améliorations significatives en matière de latence et de coût par rapport aux modèles plus grands. Les cas d'utilisation incluent la recherche et le résumé."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Le modèle d'instructions Llama 3.2 3B est un modèle multilingue léger lancé par Meta. Ce modèle vise à améliorer l'efficacité, offrant des améliorations significatives en matière de latence et de coût par rapport aux modèles plus grands. Les cas d'utilisation incluent les requêtes, la réécriture de prompts et l'assistance à l'écriture."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Modèle d'inférence d'image ajusté par instructions de Meta avec 90B paramètres. Ce modèle est optimisé pour la reconnaissance visuelle, l'inférence d'image, la description d'image et pour répondre à des questions générales sur l'image. Il est capable de comprendre des données visuelles, comme des graphiques et des diagrammes, et de combler le fossé entre la vision et le langage en générant des descriptions textuelles des détails de l'image."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct est la version mise à jour de Llama 3.1 70B de décembre. Ce modèle a été amélioré par rapport à Llama 3.1 70B (publié en juillet 2024), renforçant les appels d'outils, le support multilingue, ainsi que les capacités en mathématiques et en programmation. Ce modèle atteint des niveaux de performance de pointe dans le raisonnement, les mathématiques et le respect des instructions, tout en offrant des performances similaires à celles du 3.1 405B, avec des avantages significatifs en termes de vitesse et de coût."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "Modèle de 24B paramètres, doté de capacités de pointe comparables à celles de modèles plus grands."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Le modèle d'instructions Mixtral MoE 8x22B, avec des paramètres à grande échelle et une architecture multi-experts, prend en charge efficacement le traitement de tâches complexes."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Le modèle d'instructions Mixtral MoE 8x7B, avec une architecture multi-experts, offre un suivi et une exécution d'instructions efficaces."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Le modèle d'instructions Mixtral MoE 8x7B (version HF) offre des performances conformes à l'implémentation officielle, adapté à divers scénarios de tâches efficaces."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "Le modèle MythoMax L2 13B, combinant des techniques de fusion novatrices, excelle dans la narration et le jeu de rôle."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "Le modèle QwQ est un modèle de recherche expérimental développé par l'équipe Qwen, axé sur l'amélioration des capacités de raisonnement de l'IA."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "La version 72B du modèle Qwen-VL est le fruit de la dernière itération d'Alibaba, représentant près d'un an d'innovation."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 est une série de modèles de langage à décodage uniquement développée par l'équipe Qwen d'Alibaba Cloud. Ces modèles sont offerts en différentes tailles, y compris 0.5B, 1.5B, 3B, 7B, 14B, 32B et 72B, avec des variantes de base (base) et d'instruction (instruct)."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct est la dernière version de la série de modèles de langage à grande échelle spécifique au code publiée par Alibaba Cloud. Ce modèle, basé sur Qwen2.5, a été formé avec 55 trillions de tokens, améliorant considérablement les capacités de génération, de raisonnement et de correction de code. Il renforce non seulement les capacités de codage, mais maintient également des avantages en mathématiques et en compétences générales. Le modèle fournit une base plus complète pour des applications pratiques telles que les agents de code."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "Le modèle StarCoder 15.5B prend en charge des tâches de programmation avancées, avec des capacités multilingues améliorées, adapté à la génération et à la compréhension de code complexes."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "Le modèle StarCoder 7B est entraîné sur plus de 80 langages de programmation, offrant d'excellentes capacités de complétion de code et de compréhension contextuelle."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Le modèle Yi-Large offre d'excellentes capacités de traitement multilingue, adapté à diverses tâches de génération et de compréhension de langage."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus est le modèle le plus puissant d'Anthropic pour traiter des tâches hautement complexes. Il excelle en termes de performance, d'intelligence, de fluidité et de compréhension."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku est le modèle de nouvelle génération le plus rapide d'Anthropic. Par rapport à Claude 3 Haiku, Claude 3.5 Haiku présente des améliorations dans toutes les compétences et surpasse le plus grand modèle de la génération précédente, Claude 3 Opus, dans de nombreux tests de référence intellectuels."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet offre des capacités supérieures à celles d'Opus et une vitesse plus rapide que Sonnet, tout en maintenant le même prix que Sonnet. Sonnet excelle particulièrement dans la programmation, la science des données, le traitement visuel et les tâches d'agent."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet offre un équilibre idéal entre intelligence et vitesse pour les charges de travail d'entreprise. Il fournit une utilité maximale à un coût inférieur, fiable et adapté à un déploiement à grande échelle."
},
- "code-raccoon-v1": {
- "description": "Code Raccoon est un assistant de développement intelligent basé sur le grand modèle de langage de SenseTime, couvrant l'analyse des besoins logiciels, la conception d'architecture, la rédaction de code, les tests logiciels, etc., répondant aux divers besoins des utilisateurs en matière de rédaction de code et d'apprentissage de la programmation. Code Raccoon prend en charge plus de 90 langages de programmation populaires tels que Python, Java, JavaScript, C++, Go, SQL, ainsi que des IDE populaires comme VS Code et IntelliJ IDEA. Dans les applications pratiques, Code Raccoon peut aider les développeurs à améliorer leur efficacité de programmation de plus de 50 %."
- },
"codegeex-4": {
"description": "CodeGeeX-4 est un puissant assistant de programmation AI, prenant en charge des questions intelligentes et l'achèvement de code dans divers langages de programmation, améliorant l'efficacité du développement."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ est un modèle de langage de grande taille à haute performance, conçu pour des scénarios d'entreprise réels et des applications complexes."
},
+ "dall-e-2": {
+ "description": "Le deuxième modèle DALL·E, prenant en charge la génération d'images plus réalistes et précises, avec une résolution quatre fois supérieure à celle de la première génération."
+ },
+ "dall-e-3": {
+ "description": "Le dernier modèle DALL·E, publié en novembre 2023. Prend en charge la génération d'images plus réalistes et précises, avec une meilleure expressivité des détails."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct offre des capacités de traitement d'instructions hautement fiables, prenant en charge des applications dans divers secteurs."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 est un puissant modèle de langage hybride (MoE) économique. Il a été pré-entraîné sur un corpus de haute qualité de 81 trillions de tokens et a été amélioré par un ajustement supervisé (SFT) et un apprentissage par renforcement (RL). Comparé à DeepSeek 67B, DeepSeek-V2 offre de meilleures performances tout en économisant 42,5 % des coûts d'entraînement, réduisant de 93,3 % le cache KV et augmentant le débit de génération maximal de 5,76 fois. Ce modèle prend en charge une longueur de contexte de 128k et excelle dans les tests de référence standard et les évaluations de génération ouverte."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 est un modèle d'inférence alimenté par l'apprentissage par renforcement (RL), qui résout les problèmes de répétitivité et de lisibilité dans le modèle. Avant le RL, DeepSeek-R1 a introduit des données de démarrage à froid, optimisant ainsi les performances d'inférence. Il se compare à OpenAI-o1 en matière de tâches mathématiques, de code et d'inférence, et améliore l'efficacité globale grâce à des méthodes d'entraînement soigneusement conçues."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "Le modèle distillé DeepSeek-R1 optimise les performances d'inférence grâce à l'apprentissage par renforcement et aux données de démarrage à froid, rafraîchissant les références multi-tâches des modèles open source."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B est un modèle distillé basé sur Llama-3.1-8B. Ce modèle a été affiné avec des échantillons générés par DeepSeek-R1, montrant d'excellentes capacités d'inférence. Il a bien performé dans plusieurs tests de référence, atteignant 89,1 % de précision dans MATH-500, 50,4 % de taux de réussite dans AIME 2024, et un score de 1205 sur CodeForces, démontrant de fortes capacités en mathématiques et en programmation pour un modèle de 8B."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Le modèle distillé DeepSeek-R1 optimise les performances d'inférence grâce à l'apprentissage par renforcement et aux données de démarrage à froid, rafraîchissant les références multi-tâches des modèles open source."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Le modèle distillé DeepSeek-R1 optimise les performances d'inférence grâce à l'apprentissage par renforcement et aux données de démarrage à froid, rafraîchissant les références multi-tâches des modèles open source."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B est un modèle obtenu par distillation de Qwen2.5-32B. Ce modèle a été affiné avec 800 000 échantillons sélectionnés générés par DeepSeek-R1, montrant des performances exceptionnelles dans plusieurs domaines tels que les mathématiques, la programmation et le raisonnement. Il a obtenu d'excellents résultats dans plusieurs tests de référence, atteignant 94,3 % de précision dans MATH-500, démontrant une forte capacité de raisonnement mathématique."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B est un modèle obtenu par distillation de Qwen2.5-Math-7B. Ce modèle a été affiné avec 800 000 échantillons sélectionnés générés par DeepSeek-R1, montrant d'excellentes capacités d'inférence. Il a obtenu d'excellents résultats dans plusieurs tests de référence, atteignant 92,8 % de précision dans MATH-500, 55,5 % de taux de réussite dans AIME 2024, et un score de 1189 sur CodeForces, démontrant de fortes capacités en mathématiques et en programmation pour un modèle de 7B."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 intègre les excellentes caractéristiques des versions précédentes, renforçant les capacités générales et de codage."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 est un modèle de langage à experts mixtes (MoE) avec 6710 milliards de paramètres, utilisant une attention potentielle multi-tête (MLA) et l'architecture DeepSeekMoE, combinée à une stratégie d'équilibrage de charge sans perte auxiliaire, optimisant ainsi l'efficacité d'inférence et d'entraînement. En pré-entraînant sur 14,8 billions de tokens de haute qualité, suivi d'un ajustement supervisé et d'apprentissage par renforcement, DeepSeek-V3 surpasse les autres modèles open source en termes de performance, se rapprochant des modèles fermés de premier plan."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B est un modèle avancé formé pour des dialogues de haute complexité."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "LLM avancé et efficace, spécialisé dans le raisonnement, les mathématiques et la programmation."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 est un modèle de langage visuel à experts mixtes (MoE) développé sur la base de DeepSeekMoE-27B, utilisant une architecture MoE à activation sparse, réalisant des performances exceptionnelles tout en n'activant que 4,5 milliards de paramètres. Ce modèle excelle dans plusieurs tâches telles que la question-réponse visuelle, la reconnaissance optique de caractères, la compréhension de documents/tableaux/graphes et le positionnement visuel."
+ },
"deepseek-chat": {
"description": "Un nouveau modèle open source qui fusionne des capacités générales et de code, conservant non seulement la capacité de dialogue général du modèle Chat d'origine et la puissante capacité de traitement de code du modèle Coder, mais s'alignant également mieux sur les préférences humaines. De plus, DeepSeek-V2.5 a réalisé des améliorations significatives dans plusieurs domaines tels que les tâches d'écriture et le suivi des instructions."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 est un modèle de code open source de type expert mixte, performant dans les tâches de code, rivalisant avec GPT4-Turbo."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 est un modèle d'inférence alimenté par l'apprentissage par renforcement (RL), qui résout les problèmes de répétitivité et de lisibilité dans le modèle. Avant le RL, DeepSeek-R1 a introduit des données de démarrage à froid, optimisant ainsi les performances d'inférence. Il se compare à OpenAI-o1 en matière de tâches mathématiques, de code et d'inférence, et améliore l'efficacité globale grâce à des méthodes d'entraînement soigneusement conçues."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 — le modèle plus grand et plus intelligent de la suite DeepSeek — a été distillé dans l'architecture Llama 70B. Basé sur des tests de référence et des évaluations humaines, ce modèle est plus intelligent que le Llama 70B d'origine, en particulier dans les tâches nécessitant précision mathématique et factuelle."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "Le modèle de la série DeepSeek-R1-Distill est obtenu par la technique de distillation des connaissances, en ajustant les échantillons générés par DeepSeek-R1 sur des modèles open source tels que Qwen et Llama."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "Le modèle de la série DeepSeek-R1-Distill est obtenu par la technique de distillation des connaissances, en ajustant les échantillons générés par DeepSeek-R1 sur des modèles open source tels que Qwen et Llama."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "Le modèle de la série DeepSeek-R1-Distill est obtenu par la technique de distillation des connaissances, en ajustant les échantillons générés par DeepSeek-R1 sur des modèles open source tels que Qwen et Llama."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "Le modèle de la série DeepSeek-R1-Distill est obtenu par la technique de distillation des connaissances, en ajustant les échantillons générés par DeepSeek-R1 sur des modèles open source tels que Qwen et Llama."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "Le modèle de la série DeepSeek-R1-Distill est obtenu par la technique de distillation des connaissances, en ajustant les échantillons générés par DeepSeek-R1 sur des modèles open source tels que Qwen et Llama."
+ },
+ "deepseek-reasoner": {
+ "description": "Modèle d'inférence proposé par DeepSeek. Avant de fournir la réponse finale, le modèle génère d'abord une chaîne de pensée pour améliorer l'exactitude de la réponse finale."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 est un modèle de langage Mixture-of-Experts efficace, adapté aux besoins de traitement économique."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B est le modèle de code de conception de DeepSeek, offrant de puissantes capacités de génération de code."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 est un modèle MoE développé par la société Hangzhou DeepSeek AI Technology Research Co., Ltd., avec des performances exceptionnelles dans plusieurs évaluations, se classant au premier rang des modèles open source dans les classements principaux. Par rapport au modèle V2.5, la vitesse de génération a été multipliée par 3, offrant aux utilisateurs une expérience d'utilisation plus rapide et fluide."
+ },
"deepseek/deepseek-chat": {
"description": "Un nouveau modèle open source fusionnant des capacités générales et de codage, qui non seulement conserve les capacités de dialogue général du modèle Chat d'origine et la puissante capacité de traitement de code du modèle Coder, mais s'aligne également mieux sur les préférences humaines. De plus, DeepSeek-V2.5 a également réalisé des améliorations significatives dans plusieurs domaines tels que les tâches d'écriture et le suivi d'instructions."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 améliore considérablement les capacités de raisonnement du modèle avec très peu de données annotées. Avant de fournir la réponse finale, le modèle génère d'abord une chaîne de pensée pour améliorer l'exactitude de la réponse finale."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 améliore considérablement les capacités de raisonnement du modèle avec très peu de données annotées. Avant de fournir la réponse finale, le modèle génère d'abord une chaîne de pensée pour améliorer l'exactitude de la réponse finale."
+ },
"emohaa": {
"description": "Emohaa est un modèle psychologique, doté de compétences de conseil professionnel, aidant les utilisateurs à comprendre les problèmes émotionnels."
},
+ "ernie-3.5-128k": {
+ "description": "Le modèle de langage de grande taille phare développé par Baidu, couvrant une vaste quantité de corpus en chinois et en anglais, avec de puissantes capacités générales, capable de répondre à la plupart des exigences en matière de questions-réponses, de génération créative et d'applications de plugins ; supporte l'intégration automatique avec le plugin de recherche Baidu, garantissant la pertinence des informations de réponse."
+ },
+ "ernie-3.5-8k": {
+ "description": "Le modèle de langage de grande taille phare développé par Baidu, couvrant une vaste quantité de corpus en chinois et en anglais, avec de puissantes capacités générales, capable de répondre à la plupart des exigences en matière de questions-réponses, de génération créative et d'applications de plugins ; supporte l'intégration automatique avec le plugin de recherche Baidu, garantissant la pertinence des informations de réponse."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "Le modèle de langage de grande taille phare développé par Baidu, couvrant une vaste quantité de corpus en chinois et en anglais, avec de puissantes capacités générales, capable de répondre à la plupart des exigences en matière de questions-réponses, de génération créative et d'applications de plugins ; supporte l'intégration automatique avec le plugin de recherche Baidu, garantissant la pertinence des informations de réponse."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "Le modèle de langage de très grande taille phare développé par Baidu, par rapport à ERNIE 3.5, a réalisé une mise à niveau complète des capacités du modèle, largement applicable à des scénarios de tâches complexes dans divers domaines ; supporte l'intégration automatique avec le plugin de recherche Baidu, garantissant la pertinence des informations de réponse."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "Le modèle de langage de très grande taille phare développé par Baidu, par rapport à ERNIE 3.5, a réalisé une mise à niveau complète des capacités du modèle, largement applicable à des scénarios de tâches complexes dans divers domaines ; supporte l'intégration automatique avec le plugin de recherche Baidu, garantissant la pertinence des informations de réponse."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "Le modèle de langage de très grande taille phare développé par Baidu, avec d'excellentes performances globales, largement applicable à des scénarios de tâches complexes dans divers domaines ; supporte l'intégration automatique avec le plugin de recherche Baidu, garantissant la pertinence des informations de réponse. Par rapport à ERNIE 4.0, il offre de meilleures performances."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "Le modèle de langage de très grande taille phare développé par Baidu, avec d'excellentes performances globales, largement applicable à des scénarios de tâches complexes dans divers domaines ; supporte l'intégration automatique avec le plugin de recherche Baidu, garantissant la pertinence des informations de réponse. Par rapport à ERNIE 4.0, il offre de meilleures performances."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "Le modèle de langage de très grande taille phare développé par Baidu, avec d'excellentes performances globales, largement applicable à des scénarios de tâches complexes dans divers domaines ; supporte l'intégration automatique avec le plugin de recherche Baidu, garantissant la pertinence des informations de réponse. Par rapport à ERNIE 4.0, il offre de meilleures performances."
+ },
+ "ernie-char-8k": {
+ "description": "Le modèle de langage pour des scénarios verticaux développé par Baidu, adapté aux dialogues de NPC de jeux, aux dialogues de service client, aux jeux de rôle, avec un style de personnage plus distinct et cohérent, une meilleure capacité de suivi des instructions et des performances d'inférence supérieures."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Le modèle de langage pour des scénarios verticaux développé par Baidu, adapté aux dialogues de NPC de jeux, aux dialogues de service client, aux jeux de rôle, avec un style de personnage plus distinct et cohérent, une meilleure capacité de suivi des instructions et des performances d'inférence supérieures."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite est un modèle de langage léger développé par Baidu, alliant d'excellentes performances du modèle et performances d'inférence, adapté à une utilisation sur des cartes d'accélération AI à faible puissance."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Un modèle de langage léger développé par Baidu, alliant d'excellentes performances du modèle et performances d'inférence, avec des résultats supérieurs à ceux d'ERNIE Lite, adapté à une utilisation sur des cartes d'accélération AI à faible puissance."
+ },
+ "ernie-novel-8k": {
+ "description": "Le modèle de langage général développé par Baidu, avec un avantage évident dans la capacité de continuation de romans, également applicable à des scénarios de courtes pièces, de films, etc."
+ },
+ "ernie-speed-128k": {
+ "description": "Le modèle de langage haute performance développé par Baidu, publié en 2024, avec d'excellentes capacités générales, adapté comme modèle de base pour un affinage, permettant de mieux traiter des problèmes spécifiques, tout en offrant d'excellentes performances d'inférence."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "Le modèle de langage haute performance développé par Baidu, publié en 2024, avec d'excellentes capacités générales, offrant de meilleures performances que l'ERNIE Speed, adapté comme modèle de base pour un affinage, permettant de mieux traiter des problèmes spécifiques, tout en offrant d'excellentes performances d'inférence."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny est un modèle de langage à très haute performance développé par Baidu, avec les coûts de déploiement et d'affinage les plus bas parmi les modèles de la série Wenxin."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Ajustement) offre des performances stables et ajustables, ce qui en fait un choix idéal pour des solutions de tâches complexes."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro prend en charge jusqu'à 2 millions de tokens, ce qui en fait un choix idéal pour un modèle multimodal de taille moyenne, adapté à un soutien polyvalent pour des tâches complexes."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash propose des fonctionnalités et des améliorations de nouvelle génération, y compris une vitesse exceptionnelle, l'utilisation d'outils natifs, la génération multimodale et une fenêtre de contexte de 1M tokens."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash propose des fonctionnalités et des améliorations de nouvelle génération, y compris une vitesse exceptionnelle, l'utilisation d'outils natifs, la génération multimodale et une fenêtre de contexte de 1M tokens."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp est le dernier modèle d'IA multimodal expérimental de Google, avec des caractéristiques de nouvelle génération, une vitesse exceptionnelle, des appels d'outils natifs et une génération multimodale."
},
- "gemini-2.0-flash-thinking-exp-1219": {
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "Un modèle Gemini 2.0 Flash optimisé pour des objectifs de rentabilité et de faible latence."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
"description": "Gemini 2.0 Flash Exp est le dernier modèle d'IA multimodal expérimental de Google, doté de caractéristiques de nouvelle génération, d'une vitesse exceptionnelle, d'appels d'outils natifs et de génération multimodale."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 est le dernier modèle d'IA multimodale expérimental de Google, offrant une capacité de traitement rapide, prenant en charge les entrées de texte, d'image et de vidéo, et permettant une extension efficace pour une variété de tâches."
+ "gemini-2.0-flash-thinking-exp-1219": {
+ "description": "Gemini 2.0 Flash Exp est le dernier modèle d'IA multimodal expérimental de Google, doté de caractéristiques de nouvelle génération, d'une vitesse exceptionnelle, d'appels d'outils natifs et de génération multimodale."
},
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 est le dernier modèle d'IA multimodal expérimental de Google, capable de traiter rapidement des entrées textuelles, d'images et de vidéos, et adapté à une large gamme de tâches avec une efficacité accrue."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental est le dernier modèle AI multimodal expérimental de Google, offrant une amélioration de la qualité par rapport aux versions précédentes, en particulier pour les connaissances générales, le code et les longs contextes."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 est le dernier modèle d'IA multimodal expérimental de Google, offrant une qualité améliorée par rapport aux versions précédentes."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus possède la capacité de comprendre le contenu vidéo et plusieurs images, adapté aux tâches multimodales."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview possède de puissantes capacités de raisonnement complexe, se distinguant dans les domaines du raisonnement logique, des mathématiques et de la programmation."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash propose des fonctionnalités et des améliorations de nouvelle génération, y compris une vitesse exceptionnelle, l'utilisation d'outils natifs, la génération multimodale et une fenêtre de contexte de 1M tokens."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental est le dernier modèle AI multimodal expérimental de Google, offrant une amélioration de la qualité par rapport aux versions précédentes, en particulier pour les connaissances générales, le code et les longs contextes."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash propose des capacités de traitement multimodal optimisées, adaptées à divers scénarios de tâches complexes."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro combine les dernières technologies d'optimisation pour offrir une capacité de traitement de données multimodales plus efficace."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 est un modèle efficace lancé par Google, couvrant une variété de scénarios d'application allant des petites applications au traitement de données complexes."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 poursuit le concept de conception légère et efficace."
},
"google/gemma-2-2b-it": {
"description": "Modèle d'optimisation des instructions léger de Google."
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 est un modèle efficace lancé par Google, couvrant une variété de scénarios d'application allant des petites applications au traitement de données complexes."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 est une série de modèles de texte open source allégés de Google."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo, adapté à diverses tâches de génération et de compréhension de texte, pointe actuellement vers gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, un modèle de génération de texte à haute capacité, adapté aux tâches complexes."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo, adapté à diverses tâches de génération et de compréhension de texte, pointe actuellement vers gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o est un modèle dynamique, mis à jour en temps réel pour rester à jour avec la dernière version. Il combine une compréhension linguistique puissante et des capacités de génération, adapté aux scénarios d'application à grande échelle, y compris le service client, l'éducation et le support technique."
},
+ "gpt-4o-audio-preview": {
+ "description": "Modèle audio GPT-4o, prenant en charge les entrées et sorties audio."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini est le dernier modèle lancé par OpenAI après le GPT-4 Omni, prenant en charge les entrées multimodales et produisant des sorties textuelles. En tant que leur modèle compact le plus avancé, il est beaucoup moins cher que d'autres modèles de pointe récents et coûte plus de 60 % de moins que le GPT-3.5 Turbo. Il maintient une intelligence de pointe tout en offrant un rapport qualité-prix significatif. Le GPT-4o mini a obtenu un score de 82 % au test MMLU et se classe actuellement au-dessus du GPT-4 en termes de préférences de chat."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "Version mini en temps réel de GPT-4o, prenant en charge les entrées et sorties audio et textuelles en temps réel."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "Version en temps réel de GPT-4o, prenant en charge les entrées et sorties audio et textuelles en temps réel."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "Version en temps réel de GPT-4o, prenant en charge les entrées et sorties audio et textuelles en temps réel."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "Version en temps réel de GPT-4o, prenant en charge les entrées et sorties audio et textuelles en temps réel."
+ },
"grok-2-1212": {
"description": "Ce modèle a été amélioré en termes de précision, de respect des instructions et de capacités multilingues."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "Dernier modèle FunctionCall de l'architecture MOE Hunyuan, formé sur des données FunctionCall de haute qualité, avec une fenêtre contextuelle atteignant 32K, se classant parmi les meilleurs sur plusieurs dimensions d'évaluation."
},
+ "hunyuan-large": {
+ "description": "Le modèle Hunyuan-large a un nombre total de paramètres d'environ 389B, avec environ 52B de paramètres activés, ce qui en fait le modèle MoE open source de l'architecture Transformer avec le plus grand nombre de paramètres et les meilleures performances dans l'industrie."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Expert dans le traitement des tâches de longs documents telles que le résumé de documents et les questions-réponses sur des documents, tout en ayant également la capacité de traiter des tâches de génération de texte général. Il excelle dans l'analyse et la génération de longs textes, capable de répondre efficacement aux besoins de traitement de contenus longs complexes et détaillés."
+ },
"hunyuan-lite": {
"description": "Mise à niveau vers une structure MOE, avec une fenêtre contextuelle de 256k, en tête de nombreux modèles open source dans les évaluations NLP, code, mathématiques, industrie, etc."
},
+ "hunyuan-lite-vision": {
+ "description": "Le dernier modèle multimodal 7B de Hunyuan, avec une fenêtre contextuelle de 32K, prend en charge les dialogues multimodaux en chinois et en anglais, la reconnaissance d'objets d'images, la compréhension de documents et de tableaux, ainsi que les mathématiques multimodales, surpassant les modèles concurrents de 7B sur plusieurs dimensions d'évaluation."
+ },
"hunyuan-pro": {
"description": "Modèle de long texte MOE-32K avec un milliard de paramètres. Atteint un niveau de performance absolument supérieur sur divers benchmarks, capable de traiter des instructions complexes et de raisonner, avec des capacités mathématiques avancées, prenant en charge les appels de fonction, optimisé pour des domaines tels que la traduction multilingue, le droit financier et médical."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Utilise une stratégie de routage améliorée tout en atténuant les problèmes d'équilibrage de charge et de convergence des experts. Pour les longs textes, l'indice de recherche atteint 99,9 %. MOE-256K franchit de nouvelles étapes en termes de longueur et d'efficacité, élargissant considérablement la longueur d'entrée possible."
},
+ "hunyuan-standard-vision": {
+ "description": "Le dernier modèle multimodal de Hunyuan, prenant en charge les réponses multilingues, avec des capacités équilibrées en chinois et en anglais."
+ },
"hunyuan-turbo": {
"description": "Version préliminaire du nouveau modèle de langage de génération Hunyuan, utilisant une nouvelle structure de modèle d'experts mixtes (MoE), offrant une efficacité d'inférence plus rapide et de meilleures performances par rapport à Hunyuan-Pro."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Version fixe de hunyuan-turbo du 20 novembre 2024, une version intermédiaire entre hunyuan-turbo et hunyuan-turbo-latest."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "Optimisations de cette version : mise à l'échelle des instructions de données, augmentation significative de la capacité de généralisation du modèle ; amélioration significative des capacités en mathématiques, en code et en raisonnement logique ; optimisation des capacités de compréhension des mots dans le texte ; optimisation de la qualité de génération de contenu dans la création de texte."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "Optimisation de l'expérience générale, y compris la compréhension NLP, la création de texte, les conversations informelles, les questions-réponses, la traduction, et les domaines spécifiques ; amélioration de l'humanité simulée, optimisation de l'intelligence émotionnelle du modèle ; amélioration de la capacité du modèle à clarifier activement en cas d'ambiguïté d'intention ; amélioration de la capacité à traiter les questions de décomposition de mots ; amélioration de la qualité et de l'interactivité de la création ; amélioration de l'expérience multi-tours."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "Le nouveau modèle phare de langage visuel de Hunyuan de nouvelle génération, utilisant une toute nouvelle structure de modèle d'experts hybrides (MoE), avec des améliorations complètes par rapport à la génération précédente dans les capacités de reconnaissance de base, de création de contenu, de questions-réponses, et d'analyse et de raisonnement liés à la compréhension d'images et de textes."
+ },
"hunyuan-vision": {
"description": "Dernier modèle multimodal Hunyuan, prenant en charge l'entrée d'images et de textes pour générer du contenu textuel."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "Notre dernière série de modèles, offrant des performances d'inférence exceptionnelles, prenant en charge une longueur de contexte de 1M et des capacités améliorées de suivi des instructions et d'appel d'outils."
},
+ "internlm3-latest": {
+ "description": "Notre dernière série de modèles, avec des performances d'inférence exceptionnelles, en tête des modèles open source de même niveau. Par défaut, elle pointe vers notre dernière version du modèle InternLM3."
+ },
+ "jina-deepsearch-v1": {
+ "description": "La recherche approfondie combine la recherche sur le web, la lecture et le raisonnement pour mener des enquêtes complètes. Vous pouvez la considérer comme un agent qui prend en charge vos tâches de recherche - elle effectuera une recherche approfondie et itérative avant de fournir une réponse. Ce processus implique une recherche continue, un raisonnement et une résolution de problèmes sous différents angles. Cela diffère fondamentalement des grands modèles standard qui génèrent des réponses directement à partir de données pré-entraînées et des systèmes RAG traditionnels qui dépendent d'une recherche superficielle unique."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM est un modèle de langage expérimental, spécifique à des tâches, formé pour respecter les principes des sciences de l'apprentissage, capable de suivre des instructions systématiques dans des contextes d'enseignement et d'apprentissage, agissant comme un mentor expert, entre autres."
},
"lite": {
"description": "Spark Lite est un modèle de langage léger, offrant une latence extrêmement faible et une capacité de traitement efficace, entièrement gratuit et ouvert, prenant en charge la recherche en temps réel. Sa capacité de réponse rapide le rend exceptionnel pour les applications d'inférence sur des appareils à faible puissance de calcul et pour le réglage des modèles, offrant aux utilisateurs un excellent rapport coût-efficacité et une expérience intelligente, en particulier dans les scénarios de questions-réponses, de génération de contenu et de recherche."
},
- "llama-3.1-70b-instruct": {
- "description": "Le modèle Llama 3.1 70B Instruct, avec 70B de paramètres, offre des performances exceptionnelles dans la génération de texte et les tâches d'instructions."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B offre une capacité de raisonnement AI plus puissante, adaptée aux applications complexes, prenant en charge un traitement de calcul intensif tout en garantissant efficacité et précision."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B est un modèle à haute performance, offrant une capacité de génération de texte rapide, particulièrement adapté aux scénarios d'application nécessitant une efficacité à grande échelle et un rapport coût-efficacité."
},
- "llama-3.1-8b-instruct": {
- "description": "Le modèle Llama 3.1 8B Instruct, avec 8B de paramètres, prend en charge l'exécution efficace des tâches d'instructions visuelles, offrant d'excellentes capacités de génération de texte."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Le modèle Llama 3.1 Sonar Huge Online, avec 405B de paramètres, prend en charge une longueur de contexte d'environ 127 000 jetons, conçu pour des applications de chat en ligne complexes."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Le modèle Llama 3.1 Sonar Large Chat, avec 70B de paramètres, prend en charge une longueur de contexte d'environ 127 000 jetons, adapté aux tâches de chat hors ligne complexes."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Le modèle Llama 3.1 Sonar Large Online, avec 70B de paramètres, prend en charge une longueur de contexte d'environ 127 000 jetons, adapté aux tâches de chat à haute capacité et diversifiées."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Le modèle Llama 3.1 Sonar Small Chat, avec 8B de paramètres, est conçu pour le chat hors ligne, prenant en charge une longueur de contexte d'environ 127 000 jetons."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Le modèle Llama 3.1 Sonar Small Online, avec 8B de paramètres, prend en charge une longueur de contexte d'environ 127 000 jetons, conçu pour le chat en ligne, capable de traiter efficacement diverses interactions textuelles."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 est conçu pour traiter des tâches combinant des données visuelles et textuelles. Il excelle dans des tâches telles que la description d'images et les questions-réponses visuelles, comblant le fossé entre la génération de langage et le raisonnement visuel."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 est le modèle de langage open source multilingue le plus avancé de la série Llama, offrant des performances comparables à celles du modèle 405B à un coût très bas. Basé sur une architecture Transformer, il améliore son utilité et sa sécurité grâce à un ajustement supervisé (SFT) et un apprentissage par renforcement avec retour humain (RLHF). Sa version optimisée pour les instructions est spécialement conçue pour les dialogues multilingues et surpasse de nombreux modèles de chat open source et fermés sur plusieurs benchmarks industriels. La date limite des connaissances est décembre 2023."
+ },
"llama-3.3-70b-versatile": {
"description": "Le modèle de langage multilingue Llama 3.3 de Meta (LLM) est un modèle génératif pré-entraîné et affiné par instructions avec 70B (entrée/sortie de texte). Le modèle Llama 3.3 affiné par instructions est optimisé pour les cas d'utilisation de dialogue multilingue et surpasse de nombreux modèles de chat open-source et fermés disponibles sur des benchmarks industriels courants."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 est conçu pour traiter des tâches qui combinent des données visuelles et textuelles. Il excelle dans des tâches comme la description d'image et le questionnement visuel, comblant le fossé entre génération de langage et raisonnement visuel."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 est le modèle de langage open source multilingue le plus avancé de la série Llama, offrant une expérience comparable aux performances du modèle 405B à un coût très bas. Basé sur une architecture Transformer, il améliore l'utilité et la sécurité grâce à un ajustement supervisé (SFT) et un apprentissage par renforcement avec retour humain (RLHF). Sa version optimisée pour les instructions est spécialement conçue pour les dialogues multilingues, surpassant de nombreux modèles de chat open source et fermés sur plusieurs benchmarks industriels. Date limite de connaissance : décembre 2023."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Le modèle de langage multilingue Meta Llama 3.3 (LLM) est un modèle génératif pré-entraîné et ajusté par instruction de 70B (entrée/sortie de texte). Le modèle de texte pur ajusté par instruction Llama 3.3 est optimisé pour les cas d'utilisation de dialogue multilingue et surpasse de nombreux modèles de chat open source et fermés sur des benchmarks industriels courants."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 est conçu pour traiter des tâches qui combinent des données visuelles et textuelles. Il excelle dans des tâches comme la description d'image et le questionnement visuel, comblant le fossé entre génération de langage et raisonnement visuel."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "Le modèle Llama 3.1 Turbo 405B offre un support de contexte de très grande capacité pour le traitement de grandes données, se distinguant dans les applications d'intelligence artificielle à très grande échelle."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 est le modèle de pointe lancé par Meta, prenant en charge jusqu'à 405B de paramètres, applicable aux dialogues complexes, à la traduction multilingue et à l'analyse de données."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B offre un support de dialogue efficace en plusieurs langues."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 est conçu pour traiter des tâches combinant des données visuelles et textuelles. Il excelle dans des tâches telles que la description d'images et les questions-réponses visuelles, comblant le fossé entre la génération de langage et le raisonnement visuel."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 est le modèle de langage open source multilingue le plus avancé de la série Llama, offrant des performances comparables à celles du modèle 405B à un coût très bas. Basé sur une architecture Transformer, il améliore son utilité et sa sécurité grâce à un ajustement supervisé (SFT) et un apprentissage par renforcement avec retour humain (RLHF). Sa version optimisée pour les instructions est spécialement conçue pour les dialogues multilingues et surpasse de nombreux modèles de chat open source et fermés sur plusieurs benchmarks industriels. La date limite des connaissances est décembre 2023."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 est le modèle de langage open source multilingue le plus avancé de la série Llama, offrant des performances comparables à celles du modèle 405B à un coût très bas. Basé sur une architecture Transformer, il améliore son utilité et sa sécurité grâce à un ajustement supervisé (SFT) et un apprentissage par renforcement avec retour humain (RLHF). Sa version optimisée pour les instructions est spécialement conçue pour les dialogues multilingues et surpasse de nombreux modèles de chat open source et fermés sur plusieurs benchmarks industriels. La date limite des connaissances est décembre 2023."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct est le modèle le plus grand et le plus puissant du modèle Llama 3.1 Instruct. C'est un modèle de génération de données de dialogue et de raisonnement hautement avancé, qui peut également servir de base pour un pré-entraînement ou un ajustement fin spécialisé dans des domaines spécifiques. Les modèles de langage multilingues (LLMs) fournis par Llama 3.1 sont un ensemble de modèles génératifs pré-entraînés et ajustés par instructions, comprenant des tailles de 8B, 70B et 405B (entrée/sortie de texte). Les modèles de texte ajustés par instructions de Llama 3.1 (8B, 70B, 405B) sont optimisés pour des cas d'utilisation de dialogue multilingue et ont surpassé de nombreux modèles de chat open source disponibles dans des benchmarks industriels courants. Llama 3.1 est conçu pour des usages commerciaux et de recherche dans plusieurs langues. Les modèles de texte ajustés par instructions conviennent aux chats de type assistant, tandis que les modèles pré-entraînés peuvent s'adapter à diverses tâches de génération de langage naturel. Le modèle Llama 3.1 prend également en charge l'amélioration d'autres modèles en utilisant sa sortie, y compris la génération de données synthétiques et le raffinement. Llama 3.1 est un modèle de langage autoregressif utilisant une architecture de transformateur optimisée. Les versions ajustées utilisent un ajustement fin supervisé (SFT) et un apprentissage par renforcement avec retour humain (RLHF) pour répondre aux préférences humaines en matière d'utilité et de sécurité."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 est un modèle de langage ouvert (LLM) destiné aux développeurs, chercheurs et entreprises, conçu pour les aider à construire, expérimenter et étendre de manière responsable leurs idées d'IA générative. En tant que partie intégrante d'un système de base pour l'innovation de la communauté mondiale, il est particulièrement adapté aux appareils à capacité de calcul et de ressources limitées, ainsi qu'à des temps d'entraînement plus rapides."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "LLM avancé, prenant en charge la génération de données synthétiques, la distillation de connaissances et le raisonnement, adapté aux chatbots, à la programmation et aux tâches spécifiques."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "Permet des dialogues complexes, avec une excellente compréhension du contexte, des capacités de raisonnement et de génération de texte."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "Modèle de pointe avancé, doté de compréhension linguistique, d'excellentes capacités de raisonnement et de génération de texte."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "Modèle visuel-linguistique de pointe, spécialisé dans le raisonnement de haute qualité à partir d'images."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "Modèle de langage de pointe de petite taille, doté de compréhension linguistique, d'excellentes capacités de raisonnement et de génération de texte."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "Modèle de langage de pointe de petite taille, doté de compréhension linguistique, d'excellentes capacités de raisonnement et de génération de texte."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "Modèle visuel-linguistique de pointe, spécialisé dans le raisonnement de haute qualité à partir d'images."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "LLM avancé, spécialisé dans le raisonnement, les mathématiques, le bon sens et les appels de fonction."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 est un modèle de langage proposé par Microsoft AI, qui excelle dans les domaines des dialogues complexes, du multilinguisme, du raisonnement et des assistants intelligents."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K est un modèle doté d'une capacité de traitement de contexte ultra-long, adapté à la génération de textes très longs, répondant aux besoins de tâches de génération complexes, capable de traiter jusqu'à 128 000 tokens, idéal pour la recherche, l'académie et la génération de documents volumineux."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Le modèle visuel Kimi (y compris moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, etc.) est capable de comprendre le contenu des images, y compris le texte des images, les couleurs des images et les formes des objets."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K offre une capacité de traitement de contexte de longueur moyenne, capable de traiter 32 768 tokens, particulièrement adapté à la génération de divers documents longs et de dialogues complexes, utilisé dans la création de contenu, la génération de rapports et les systèmes de dialogue."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Le modèle visuel Kimi (y compris moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, etc.) est capable de comprendre le contenu des images, y compris le texte des images, les couleurs des images et les formes des objets."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K est conçu pour des tâches de génération de courts textes, avec des performances de traitement efficaces, capable de traiter 8 192 tokens, idéal pour des dialogues courts, des prises de notes et une génération rapide de contenu."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Le modèle visuel Kimi (y compris moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, etc.) est capable de comprendre le contenu des images, y compris le texte des images, les couleurs des images et les formes des objets."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B est une version améliorée de Nous Hermes 2, intégrant les derniers ensembles de données développés en interne."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B est un modèle de langage de grande taille personnalisé par NVIDIA, conçu pour améliorer le degré d'aide des réponses générées par LLM aux requêtes des utilisateurs."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B est un modèle de langage à grande échelle personnalisé par NVIDIA, conçu pour améliorer l'aide fournie par les réponses générées par LLM aux requêtes des utilisateurs. Ce modèle a excellé dans des tests de référence tels que Arena Hard, AlpacaEval 2 LC et GPT-4-Turbo MT-Bench, se classant premier dans les trois tests d'alignement automatique au 1er octobre 2024. Le modèle utilise RLHF (en particulier REINFORCE), Llama-3.1-Nemotron-70B-Reward et HelpSteer2-Preference pour l'entraînement sur la base du modèle Llama-3.1-70B-Instruct."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "Modèle de langage unique, offrant une précision et une efficacité inégalées."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct est un modèle de langage de grande taille personnalisé par NVIDIA, conçu pour améliorer l'utilité des réponses générées par LLM."
+ },
"o1": {
"description": "Axé sur le raisonnement avancé et la résolution de problèmes complexes, y compris les tâches mathématiques et scientifiques. Idéal pour les applications nécessitant une compréhension approfondie du contexte et des flux de travail d'agent."
},
- "o1-2024-12-17": {
- "description": "o1 est le nouveau modèle de raisonnement d'OpenAI, prenant en charge les entrées multimodales et produisant du texte, adapté aux tâches complexes nécessitant des connaissances générales étendues. Ce modèle dispose d'un contexte de 200K et d'une date limite de connaissances en octobre 2023."
- },
"o1-mini": {
"description": "o1-mini est un modèle de raisonnement rapide et économique conçu pour les applications de programmation, de mathématiques et de sciences. Ce modèle dispose d'un contexte de 128K et d'une date limite de connaissance en octobre 2023."
},
"o1-preview": {
"description": "o1 est le nouveau modèle de raisonnement d'OpenAI, adapté aux tâches complexes nécessitant une vaste connaissance générale. Ce modèle dispose d'un contexte de 128K et d'une date limite de connaissance en octobre 2023."
},
+ "o3-mini": {
+ "description": "o3-mini est notre dernier modèle d'inférence compact, offrant une grande intelligence avec les mêmes objectifs de coût et de latence que o1-mini."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba est un modèle de langage Mamba 2 axé sur la génération de code, offrant un soutien puissant pour des tâches avancées de codage et de raisonnement."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini est le dernier modèle d'OpenAI lancé après GPT-4 Omni, prenant en charge les entrées d'images et de texte et produisant du texte en sortie. En tant que leur modèle compact le plus avancé, il est beaucoup moins cher que d'autres modèles de pointe récents et coûte plus de 60 % de moins que GPT-3.5 Turbo. Il maintient une intelligence de pointe tout en offrant un rapport qualité-prix significatif. GPT-4o mini a obtenu un score de 82 % au test MMLU et se classe actuellement au-dessus de GPT-4 en termes de préférences de chat."
},
- "openai/o1": {
- "description": "o1 est le nouveau modèle de raisonnement d'OpenAI, prenant en charge les entrées multimodales et produisant du texte, adapté aux tâches complexes nécessitant des connaissances générales étendues. Ce modèle dispose d'un contexte de 200K et d'une date limite de connaissances en octobre 2023."
- },
"openai/o1-mini": {
"description": "o1-mini est un modèle de raisonnement rapide et économique conçu pour les applications de programmation, de mathématiques et de sciences. Ce modèle dispose d'un contexte de 128K et d'une date limite de connaissance en octobre 2023."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K est doté d'une capacité de traitement de contexte très étendue, capable de gérer jusqu'à 128K d'informations contextuelles, particulièrement adapté pour l'analyse complète et le traitement des relations logiques à long terme dans des contenus longs, offrant une logique fluide et cohérente ainsi qu'un soutien varié pour les références dans des communications textuelles complexes."
},
+ "qvq-72b-preview": {
+ "description": "Le modèle QVQ est un modèle de recherche expérimental développé par l'équipe Qwen, axé sur l'amélioration des capacités de raisonnement visuel, en particulier dans le domaine du raisonnement mathématique."
+ },
"qwen-coder-plus-latest": {
"description": "Modèle de code Qwen universel."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Modèle de langage visuel à très grande échelle Tongyi Qianwen. Par rapport à la version améliorée, il améliore encore les capacités de raisonnement visuel et de suivi des instructions, offrant un niveau de perception visuelle et de cognition plus élevé."
},
+ "qwen-vl-ocr-latest": {
+ "description": "Le modèle OCR Qwen est un modèle spécialisé dans l'extraction de texte, se concentrant sur la capacité d'extraction de texte à partir d'images de documents, tableaux, questions d'examen, écriture manuscrite, etc. Il peut reconnaître plusieurs langues, actuellement supportées : chinois, anglais, français, japonais, coréen, allemand, russe, italien, vietnamien, arabe."
+ },
"qwen-vl-plus-latest": {
"description": "Version améliorée du modèle de langage visuel à grande échelle Tongyi Qianwen. Amélioration significative des capacités de reconnaissance des détails et de reconnaissance de texte, prenant en charge des résolutions d'image de plus d'un million de pixels et des rapports d'aspect de n'importe quelle taille."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 est une toute nouvelle série de modèles de langage de grande taille, offrant des capacités de compréhension et de génération plus puissantes."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "LLM orienté vers le chinois et l'anglais, ciblant des domaines tels que la langue, la programmation, les mathématiques et le raisonnement."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "LLM avancé, prenant en charge la génération de code, le raisonnement et la correction, couvrant les langages de programmation courants."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "Modèle de code puissant de taille moyenne, prenant en charge une longueur de contexte de 32K, spécialisé dans la programmation multilingue."
+ },
"qwen2": {
"description": "Qwen2 est le nouveau modèle de langage à grande échelle d'Alibaba, offrant d'excellentes performances pour des besoins d'application diversifiés."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "Le modèle de 14B de Tongyi Qwen 2.5, open source."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "Le modèle de 72B de Qwen2.5 est ouvert au public."
+ },
"qwen2.5-32b-instruct": {
"description": "Le modèle de 32B de Tongyi Qwen 2.5, open source."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Le modèle Qwen-Math possède de puissantes capacités de résolution de problèmes mathématiques."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "Amélioration globale des capacités de suivi des instructions, mathématiques, résolution de problèmes et code, amélioration des capacités de reconnaissance, support de divers formats pour un positionnement précis des éléments visuels, compréhension de fichiers vidéo longs (jusqu'à 10 minutes) et localisation d'événements en temps réel, capable de comprendre l'ordre temporel et la vitesse, supportant le contrôle d'agents OS ou Mobile basé sur des capacités d'analyse et de localisation, avec une forte capacité d'extraction d'informations clés et de sortie au format Json. Cette version est la version 72B, la plus puissante de cette série."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "Amélioration globale des capacités de suivi des instructions, mathématiques, résolution de problèmes et code, amélioration des capacités de reconnaissance, support de divers formats pour un positionnement précis des éléments visuels, compréhension de fichiers vidéo longs (jusqu'à 10 minutes) et localisation d'événements en temps réel, capable de comprendre l'ordre temporel et la vitesse, supportant le contrôle d'agents OS ou Mobile basé sur des capacités d'analyse et de localisation, avec une forte capacité d'extraction d'informations clés et de sortie au format Json. Cette version est la version 72B, la plus puissante de cette série."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 est le nouveau modèle de langage à grande échelle de Alibaba, offrant d'excellentes performances pour répondre à des besoins d'application diversifiés."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro est un LLM hautement intelligent lancé par Upstage, axé sur la capacité de suivi des instructions sur un seul GPU, avec un score IFEval supérieur à 80. Actuellement, il supporte l'anglais, et la version officielle est prévue pour novembre 2024, avec une extension du support linguistique et de la longueur du contexte."
},
+ "sonar": {
+ "description": "Produit de recherche léger basé sur le contexte de recherche, plus rapide et moins cher que Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "Produit de recherche avancé prenant en charge le contexte de recherche, avec des requêtes avancées et un suivi."
+ },
+ "sonar-reasoning": {
+ "description": "Nouveau produit API soutenu par le modèle de raisonnement DeepSeek."
+ },
"step-1-128k": {
"description": "Équilibre entre performance et coût, adapté à des scénarios généraux."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "Ce modèle possède de puissantes capacités de compréhension vidéo."
},
+ "step-1o-vision-32k": {
+ "description": "Ce modèle possède de puissantes capacités de compréhension d'image. Par rapport à la série de modèles step-1v, il offre des performances visuelles supérieures."
+ },
"step-1v-32k": {
"description": "Prend en charge les entrées visuelles, améliorant l'expérience d'interaction multimodale."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Prend en charge des interactions contextuelles à grande échelle, adapté aux scénarios de dialogue complexes."
},
+ "step-2-mini": {
+ "description": "Un modèle de grande taille ultra-rapide basé sur la nouvelle architecture d'attention auto-développée MFA, atteignant des résultats similaires à ceux de step1 à un coût très bas, tout en maintenant un débit plus élevé et un temps de réponse plus rapide. Capable de traiter des tâches générales, avec des compétences particulières en matière de codage."
+ },
+ "taichu2_mm": {
+ "description": "Intègre des capacités de compréhension d'images, de transfert de connaissances et d'attribution logique, se distinguant dans le domaine des questions-réponses textuelles et visuelles."
+ },
"taichu_llm": {
"description": "Le modèle de langage Taichu Zidong possède une forte capacité de compréhension linguistique ainsi que des compétences en création de texte, questions-réponses, programmation, calcul mathématique, raisonnement logique, analyse des sentiments, et résumé de texte. Il combine de manière innovante le pré-entraînement sur de grandes données avec des connaissances riches provenant de multiples sources, en perfectionnant continuellement la technologie algorithmique et en intégrant de nouvelles connaissances sur le vocabulaire, la structure, la grammaire et le sens à partir de vastes ensembles de données textuelles, offrant aux utilisateurs des informations et des services plus pratiques ainsi qu'une expérience plus intelligente."
},
+ "text-embedding-3-large": {
+ "description": "Le modèle de vectorisation le plus puissant, adapté aux tâches en anglais et non-anglais."
+ },
+ "text-embedding-3-small": {
+ "description": "Un modèle d'Embedding de nouvelle génération, efficace et économique, adapté à la recherche de connaissances, aux applications RAG, etc."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) offre une capacité de calcul améliorée grâce à des stratégies et une architecture de modèle efficaces."
},
+ "tts-1": {
+ "description": "Le dernier modèle de synthèse vocale, optimisé pour la vitesse dans des scénarios en temps réel."
+ },
+ "tts-1-hd": {
+ "description": "Le dernier modèle de synthèse vocale, optimisé pour la qualité."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) est adapté aux tâches d'instructions détaillées, offrant d'excellentes capacités de traitement du langage."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet élève les normes de l'industrie, surpassant les modèles concurrents et Claude 3 Opus, avec d'excellentes performances dans une large gamme d'évaluations, tout en offrant la vitesse et le coût de nos modèles de niveau intermédiaire."
},
+ "whisper-1": {
+ "description": "Modèle de reconnaissance vocale général, prenant en charge la reconnaissance vocale multilingue, la traduction vocale et la reconnaissance de langue."
+ },
"wizardlm2": {
"description": "WizardLM 2 est un modèle de langage proposé par Microsoft AI, particulièrement performant dans les domaines des dialogues complexes, du multilinguisme, du raisonnement et des assistants intelligents."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Modèle pour des tâches visuelles complexes, offrant des capacités de compréhension et d'analyse d'images de haute performance."
+ },
+ "yi-vision-v2": {
+ "description": "Modèle pour des tâches visuelles complexes, offrant des capacités de compréhension et d'analyse de haute performance basées sur plusieurs images."
}
}
diff --git a/locales/fr-FR/providers.json b/locales/fr-FR/providers.json
index 35e168b2749a1..9db702957d933 100644
--- a/locales/fr-FR/providers.json
+++ b/locales/fr-FR/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure propose une variété de modèles IA avancés, y compris GPT-3.5 et la dernière série GPT-4, prenant en charge divers types de données et tâches complexes, tout en s'engageant à fournir des solutions IA sécurisées, fiables et durables."
},
+ "azureai": {
+ "description": "Azure propose une variété de modèles d'IA avancés, y compris GPT-3.5 et la dernière série GPT-4, prenant en charge divers types de données et des tâches complexes, s'engageant à fournir des solutions d'IA sécurisées, fiables et durables."
+ },
"baichuan": {
"description": "Baichuan Intelligent est une entreprise spécialisée dans le développement de grands modèles d'intelligence artificielle, dont les modèles excellent dans les tâches en chinois telles que l'encyclopédie de connaissances, le traitement de longs textes et la création, surpassant les modèles dominants étrangers. Baichuan Intelligent possède également des capacités multimodales de premier plan, se distinguant dans plusieurs évaluations autorisées. Ses modèles incluent Baichuan 4, Baichuan 3 Turbo et Baichuan 3 Turbo 128k, chacun optimisé pour différents scénarios d'application, offrant des solutions à bon rapport qualité-prix."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek est une entreprise spécialisée dans la recherche et l'application des technologies d'intelligence artificielle, dont le dernier modèle, DeepSeek-V2.5, combine des capacités de dialogue général et de traitement de code, réalisant des améliorations significatives dans l'alignement des préférences humaines, les tâches d'écriture et le suivi des instructions."
},
+ "doubao": {
+ "description": "Un grand modèle développé en interne par ByteDance. Validé par la pratique dans plus de 50 scénarios d'affaires au sein de ByteDance, avec un volume d'utilisation quotidien de plusieurs trillions de tokens, il offre diverses capacités multimodales, créant ainsi une expérience commerciale riche grâce à des performances de modèle de haute qualité."
+ },
"fireworksai": {
"description": "Fireworks AI est un fournisseur de services de modèles linguistiques avancés, axé sur les appels de fonction et le traitement multimodal. Son dernier modèle, Firefunction V2, basé sur Llama-3, est optimisé pour les appels de fonction, les dialogues et le suivi des instructions. Le modèle de langage visuel FireLLaVA-13B prend en charge les entrées mixtes d'images et de texte. D'autres modèles notables incluent la série Llama et la série Mixtral, offrant un support efficace pour le suivi et la génération d'instructions multilingues."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "Organisation open source dédiée à la recherche et au développement d'outils pour les grands modèles. Fournit à tous les développeurs d'IA une plateforme open source efficace et facile à utiliser, rendant les technologies de pointe en matière de grands modèles et d'algorithmes accessibles."
},
+ "jina": {
+ "description": "Jina AI, fondée en 2020, est une entreprise leader dans le domaine de l'IA de recherche. Notre plateforme de recherche de base comprend des modèles vectoriels, des réarrangeurs et de petits modèles de langage, aidant les entreprises à construire des applications de recherche génératives et multimodales fiables et de haute qualité."
+ },
+ "lmstudio": {
+ "description": "LM Studio est une application de bureau pour développer et expérimenter des LLM sur votre ordinateur."
+ },
"minimax": {
"description": "MiniMax est une entreprise de technologie d'intelligence artificielle générale fondée en 2021, dédiée à la co-création d'intelligence avec les utilisateurs. MiniMax a développé de manière autonome différents modèles de grande taille, y compris un modèle de texte MoE à un trillion de paramètres, un modèle vocal et un modèle d'image. Elle a également lancé des applications telles que Conch AI."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI est une plateforme offrant des services API pour divers grands modèles de langage et la génération d'images IA, flexible, fiable et rentable. Elle prend en charge les derniers modèles open source tels que Llama3, Mistral, et fournit des solutions API complètes, conviviales et évolutives pour le développement d'applications IA, adaptées à la croissance rapide des startups IA."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ fournit des conteneurs pour l'inférence de microservices accélérés par GPU auto-hébergés, prenant en charge le déploiement de modèles d'IA pré-entraînés et personnalisés dans le cloud, les centres de données, les PC personnels RTX™ AI et les stations de travail."
+ },
"ollama": {
"description": "Les modèles proposés par Ollama couvrent largement des domaines tels que la génération de code, les calculs mathématiques, le traitement multilingue et les interactions conversationnelles, répondant à des besoins diversifiés pour le déploiement en entreprise et la localisation."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "L'Institut d'automatisation de l'Académie chinoise des sciences et l'Institut de recherche en intelligence artificielle de Wuhan ont lancé une nouvelle génération de grands modèles multimodaux, prenant en charge des tâches de questions-réponses complètes, de création de texte, de génération d'images, de compréhension 3D, d'analyse de signaux, avec des capacités cognitives, de compréhension et de création renforcées, offrant une toute nouvelle expérience interactive."
},
+ "tencentcloud": {
+ "description": "La capacité atomique du moteur de connaissance (LLM Knowledge Engine Atomic Power) est une capacité de question-réponse complète développée sur la base du moteur de connaissance, destinée aux entreprises et aux développeurs. Elle offre la possibilité de créer et de développer des applications de modèles de manière flexible. Vous pouvez assembler votre service de modèle exclusif en utilisant plusieurs capacités atomiques, en appelant des services tels que l'analyse de documents, la séparation, l'embedding, la réécriture multi-tours, etc., pour personnaliser les affaires AI spécifiques à votre entreprise."
+ },
"togetherai": {
"description": "Together AI s'engage à réaliser des performances de pointe grâce à des modèles IA innovants, offrant une large capacité de personnalisation, y compris un support d'évolutivité rapide et un processus de déploiement intuitif, répondant à divers besoins d'entreprise."
},
"upstage": {
"description": "Upstage se concentre sur le développement de modèles IA pour divers besoins commerciaux, y compris Solar LLM et Document AI, visant à réaliser une intelligence générale artificielle (AGI) pour le travail. Créez des agents de dialogue simples via l'API Chat, et prenez en charge les appels de fonction, la traduction, l'intégration et les applications spécifiques à un domaine."
},
+ "vllm": {
+ "description": "vLLM est une bibliothèque rapide et facile à utiliser pour l'inférence et les services LLM."
+ },
+ "volcengine": {
+ "description": "La plateforme de développement des services de grands modèles lancée par ByteDance, offrant des services d'appel de modèles riches en fonctionnalités, sécurisés et compétitifs en termes de prix. Elle propose également des fonctionnalités de bout en bout telles que les données de modèle, le réglage fin, l'inférence et l'évaluation, garantissant ainsi le succès de votre développement d'applications AI."
+ },
"wenxin": {
"description": "Plateforme de développement et de services d'applications AI natives et de modèles de grande envergure, tout-en-un pour les entreprises, offrant la chaîne d'outils la plus complète et facile à utiliser pour le développement de modèles d'intelligence artificielle générative et le développement d'applications."
},
diff --git a/locales/fr-FR/setting.json b/locales/fr-FR/setting.json
index 7a26e2e4a1add..a988ba7925009 100644
--- a/locales/fr-FR/setting.json
+++ b/locales/fr-FR/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Activer la limite de tokens par réponse"
},
+ "enableReasoningEffort": {
+ "title": "Activer l'ajustement de l'intensité de raisonnement"
+ },
"frequencyPenalty": {
- "desc": "Plus la valeur est élevée, plus il est probable de réduire les mots répétés",
- "title": "Pénalité de fréquence"
+ "desc": "Plus la valeur est élevée, plus le vocabulaire est riche et varié ; plus la valeur est basse, plus le vocabulaire est simple et direct",
+ "title": "Richesse du vocabulaire"
},
"maxTokens": {
"desc": "Nombre maximal de tokens utilisés par interaction",
@@ -212,19 +215,31 @@
"desc": "Modèle {{provider}}",
"title": "Modèle"
},
+ "params": {
+ "title": "Paramètres avancés"
+ },
"presencePenalty": {
- "desc": "Plus la valeur est élevée, plus il est probable d'explorer de nouveaux sujets",
- "title": "Pénalité de présence"
+ "desc": "Plus la valeur est élevée, plus il y a tendance à utiliser des expressions différentes, évitant la répétition des concepts ; plus la valeur est basse, plus il y a tendance à utiliser des concepts ou des narrations répétitifs, rendant l'expression plus cohérente",
+ "title": "Diversité de l'expression"
+ },
+ "reasoningEffort": {
+ "desc": "Plus la valeur est élevée, plus la capacité de raisonnement est forte, mais cela peut augmenter le temps de réponse et la consommation de jetons",
+ "options": {
+ "high": "Élevé",
+ "low": "Bas",
+ "medium": "Moyen"
+ },
+ "title": "Intensité de raisonnement"
},
"temperature": {
- "desc": "Plus la valeur est élevée, plus la réponse est aléatoire",
- "title": "Aléatoire",
- "titleWithValue": "Aléatoire {{value}}"
+ "desc": "Plus la valeur est élevée, plus les réponses sont créatives et imaginatives ; plus la valeur est basse, plus les réponses sont rigoureuses",
+ "title": "Niveau de créativité",
+ "warning": "Une valeur de créativité trop élevée peut entraîner des sorties illisibles"
},
"title": "Paramètres du modèle",
"topP": {
- "desc": "Similaire à l'aléatoire, mais ne doit pas être modifié en même temps que l'aléatoire",
- "title": "Échantillonnage topP"
+ "desc": "Considère combien de possibilités, plus la valeur est élevée, plus il accepte de réponses possibles ; plus la valeur est basse, plus il a tendance à choisir la réponse la plus probable. Il n'est pas recommandé de modifier cela en même temps que le niveau de créativité",
+ "title": "Ouverture d'esprit"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Paramètres généraux",
"experiment": "Expérience",
"llm": "Modèle de langue",
+ "provider": "Fournisseur de services d'IA",
"sync": "Synchronisation cloud",
"system-agent": "Agent système",
"tts": "Service vocal"
diff --git a/locales/it-IT/changelog.json b/locales/it-IT/changelog.json
index 7f7549af6c6f9..fa7ed31c4bcf6 100644
--- a/locales/it-IT/changelog.json
+++ b/locales/it-IT/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "Visualizza tutti i registri delle modifiche",
"description": "Tieni traccia delle nuove funzionalità e miglioramenti di {{appName}}",
"pagination": {
- "older": "Visualizza le modifiche precedenti",
- "prev": "Pagina precedente"
+ "next": "Pagina successiva",
+ "older": "Visualizza le modifiche precedenti"
},
"readDetails": "Leggi i dettagli",
"title": "Registro delle modifiche",
diff --git a/locales/it-IT/common.json b/locales/it-IT/common.json
index 5e40a98e1f221..437462deec26c 100644
--- a/locales/it-IT/common.json
+++ b/locales/it-IT/common.json
@@ -275,6 +275,7 @@
},
"temp": "Temporaneo",
"terms": "Termini di servizio",
+ "update": "Aggiornamento",
"updateAgent": "Aggiorna informazioni agente",
"upgradeVersion": {
"action": "Aggiorna",
@@ -285,6 +286,7 @@
"anonymousNickName": "Utente Anonimo",
"billing": "Gestione fatturazione",
"cloud": "Prova {{name}}",
+ "community": "Versione comunitaria",
"data": "Archiviazione dati",
"defaultNickname": "Utente Community",
"discord": "Supporto della community",
@@ -294,7 +296,6 @@
"help": "Centro assistenza",
"moveGuide": "Il pulsante delle impostazioni è stato spostato qui",
"plans": "Piani di abbonamento",
- "preview": "Anteprima",
"profile": "Gestione account",
"setting": "Impostazioni app",
"usages": "Statistiche di utilizzo"
diff --git a/locales/it-IT/components.json b/locales/it-IT/components.json
index 2276ea5f47f44..4fe1203c1d892 100644
--- a/locales/it-IT/components.json
+++ b/locales/it-IT/components.json
@@ -76,6 +76,7 @@
"custom": "Modello personalizzato: di default supporta sia la chiamata di funzioni che il riconoscimento visivo. Verifica l'effettiva disponibilità di tali funzionalità.",
"file": "Questo modello supporta il caricamento e il riconoscimento di file.",
"functionCall": "Questo modello supporta la chiamata di funzioni.",
+ "reasoning": "Questo modello supporta un pensiero profondo",
"tokens": "Questo modello supporta un massimo di {{tokens}} token per sessione.",
"vision": "Questo modello supporta il riconoscimento visivo."
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "Nessun modello attivo. Vai alle impostazioni per attivarne uno.",
"provider": "Provider"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "A causa delle restrizioni di sicurezza del browser, è necessario configurare il cross-origin per utilizzare Ollama correttamente.",
+ "linux": {
+ "env": "Aggiungi `Environment` nella sezione [Service] e aggiungi la variabile d'ambiente OLLAMA_ORIGINS:",
+ "reboot": "Ricarica systemd e riavvia Ollama",
+ "systemd": "Usa systemd per modificare il servizio ollama:"
+ },
+ "macos": "Apri l'applicazione 'Terminale', incolla il seguente comando e premi invio per eseguirlo",
+ "reboot": "Riavvia il servizio Ollama dopo il completamento dell'esecuzione",
+ "title": "Configura Ollama per consentire l'accesso cross-origin",
+ "windows": "Su Windows, fai clic su 'Pannello di controllo' e accedi alla modifica delle variabili d'ambiente di sistema. Crea una nuova variabile d'ambiente chiamata 'OLLAMA_ORIGINS' per il tuo account utente, con valore *, quindi fai clic su 'OK/Applica' per salvare"
+ },
+ "install": {
+ "description": "Assicurati di aver avviato Ollama. Se non hai scaricato Ollama, visita il sito ufficiale <1>per scaricare",
+ "docker": "Se preferisci utilizzare Docker, Ollama offre anche un'immagine Docker ufficiale, puoi scaricarla con il seguente comando:",
+ "linux": {
+ "command": "Installa con il seguente comando:",
+ "manual": "In alternativa, puoi fare riferimento alla <1>guida all'installazione manuale di Linux per installare manualmente"
+ },
+ "title": "Installa e avvia l'app Ollama localmente",
+ "windowsTab": "Windows (versione anteprima)"
+ }
+ },
+ "Thinking": {
+ "thinking": "Pensando profondamente...",
+ "thought": "Ho riflettuto a lungo (tempo impiegato {{duration}} secondi)",
+ "thoughtWithDuration": "Ho riflettuto a lungo"
}
}
diff --git a/locales/it-IT/discover.json b/locales/it-IT/discover.json
index ac7805a8e64f8..e59f6cd2ea536 100644
--- a/locales/it-IT/discover.json
+++ b/locales/it-IT/discover.json
@@ -126,6 +126,10 @@
"title": "Freschezza del tema"
},
"range": "Intervallo",
+ "reasoning_effort": {
+ "desc": "Questa impostazione controlla l'intensità del ragionamento del modello prima di generare una risposta. Un'intensità bassa privilegia la velocità di risposta e risparmia Token, mentre un'intensità alta fornisce un ragionamento più completo, ma consuma più Token e riduce la velocità di risposta. Il valore predefinito è medio, bilanciando l'accuratezza del ragionamento e la velocità di risposta.",
+ "title": "Intensità del ragionamento"
+ },
"temperature": {
"desc": "Questa impostazione influisce sulla diversità delle risposte del modello. Valori più bassi portano a risposte più prevedibili e tipiche, mentre valori più alti incoraggiano risposte più varie e insolite. Quando il valore è impostato a 0, il modello fornisce sempre la stessa risposta per un dato input.",
"title": "Casualità"
diff --git a/locales/it-IT/modelProvider.json b/locales/it-IT/modelProvider.json
index a225b1c382eff..aa3b37ab93e6d 100644
--- a/locales/it-IT/modelProvider.json
+++ b/locales/it-IT/modelProvider.json
@@ -19,6 +19,24 @@
"title": "Chiave API"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Versione API di Azure, seguendo il formato YYYY-MM-DD, consulta [l'ultima versione](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "Ottieni elenco",
+ "title": "Versione API di Azure"
+ },
+ "endpoint": {
+ "desc": "Trova l'endpoint di inferenza del modello Azure AI nella panoramica del progetto Azure AI",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Endpoint di Azure AI"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Trova la chiave API nella panoramica del progetto Azure AI",
+ "placeholder": "Chiave Azure",
+ "title": "Chiave"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "Inserisci l'ID chiave di accesso AWS",
@@ -63,6 +81,46 @@
"title": "ID account Cloudflare / indirizzo API"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Inserisci la tua API Key",
+ "title": "API Key"
+ },
+ "basicTitle": "Informazioni di base",
+ "configTitle": "Informazioni di configurazione",
+ "confirm": "Crea",
+ "createSuccess": "Creazione avvenuta con successo",
+ "description": {
+ "placeholder": "Descrizione del fornitore (opzionale)",
+ "title": "Descrizione del fornitore"
+ },
+ "id": {
+ "desc": "Identificatore unico del fornitore di servizi, non modificabile dopo la creazione",
+ "format": "Può contenere solo numeri, lettere minuscole, trattini (-) e underscore (_) ",
+ "placeholder": "Si consiglia di utilizzare solo lettere minuscole, ad esempio openai, non modificabile dopo la creazione",
+ "required": "Inserisci l'ID del fornitore",
+ "title": "ID del fornitore"
+ },
+ "logo": {
+ "required": "Carica un logo del fornitore valido",
+ "title": "Logo del fornitore"
+ },
+ "name": {
+ "placeholder": "Inserisci il nome visualizzato del fornitore",
+ "required": "Inserisci il nome del fornitore",
+ "title": "Nome del fornitore"
+ },
+ "proxyUrl": {
+ "required": "Inserisci l'indirizzo del proxy",
+ "title": "Indirizzo proxy"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Seleziona il tipo di SDK",
+ "title": "Formato della richiesta"
+ },
+ "title": "Crea fornitore AI personalizzato"
+ },
"github": {
"personalAccessToken": {
"desc": "Inserisci il tuo PAT di Github, clicca [qui](https://github.com/settings/tokens) per crearne uno",
@@ -77,6 +135,23 @@
"title": "Token HuggingFace"
}
},
+ "list": {
+ "title": {
+ "disabled": "Fornitore non attivato",
+ "enabled": "Fornitore attivato"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Aggiungi fornitore personalizzato",
+ "all": "Tutti",
+ "list": {
+ "disabled": "Non attivato",
+ "enabled": "Attivato"
+ },
+ "notFound": "Nessun risultato trovato",
+ "searchProviders": "Cerca fornitori...",
+ "sort": "Ordinamento personalizzato"
+ },
"ollama": {
"checker": {
"desc": "Verifica se l'indirizzo del proxy è stato compilato correttamente",
@@ -94,33 +169,9 @@
"title": "Download del modello in corso {{model}}"
},
"endpoint": {
- "desc": "Inserisci l'indirizzo del proxy dell'interfaccia Ollama. Lascia vuoto se non specificato localmente",
+ "desc": "Deve includere http(s)://, può rimanere vuoto se non specificato localmente",
"title": "Indirizzo del proxy dell'interfaccia"
},
- "setup": {
- "cors": {
- "description": "A causa delle restrizioni di sicurezza del browser, è necessario configurare il cross-origin resource sharing (CORS) per consentire l'utilizzo di Ollama.",
- "linux": {
- "env": "Nella sezione [Service], aggiungi `Environment` e inserisci la variabile di ambiente OLLAMA_ORIGINS:",
- "reboot": "Dopo aver completato l'esecuzione, riavvia il servizio Ollama.",
- "systemd": "Per modificare il servizio ollama, chiama systemd:"
- },
- "macos": "Apri l'applicazione 'Terminale', incolla il comando seguente e premi Invio per eseguirlo",
- "reboot": "Riavvia il servizio Ollama una volta completata l'esecuzione",
- "title": "Configura Ollama per consentire l'accesso cross-origin",
- "windows": "Su Windows, fai clic su 'Pannello di controllo', accedi alle variabili di ambiente di sistema. Crea una nuova variabile di ambiente chiamata 'OLLAMA_ORIGINS' per il tuo account utente, con valore *, quindi fai clic su 'OK/Applica' per salvare le modifiche"
- },
- "install": {
- "description": "Assicurati di aver avviato Ollama. Se non l'hai ancora scaricato, visita il sito ufficiale per <1>scaricarlo",
- "docker": "Se preferisci utilizzare Docker, Ollama fornisce anche un'immagine Docker ufficiale che puoi scaricare tramite il seguente comando:",
- "linux": {
- "command": "Per installare, utilizza il seguente comando:",
- "manual": "Oppure, puoi consultare la <1>Guida all'installazione manuale di Linux per installare manualmente"
- },
- "title": "Installa e avvia l'applicazione Ollama localmente",
- "windowsTab": "Windows (Versione di anteprima)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Annulla download",
@@ -131,25 +182,145 @@
"title": "Scarica il modello Ollama specificato"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Inserisci l'Access Key della piattaforma Qianfan di Baidu",
- "placeholder": "Access Key Qianfan",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "La tua chiave e l'indirizzo proxy saranno crittografati utilizzando l'algoritmo di crittografia <1>AES-GCM",
+ "apiKey": {
+ "desc": "Inserisci la tua {{name}} API Key",
+ "placeholder": "{{name}} API Key",
+ "title": "API Key"
+ },
+ "baseURL": {
+ "desc": "Deve contenere http(s)://",
+ "invalid": "Inserisci un URL valido",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "Indirizzo proxy API"
+ },
+ "checker": {
+ "button": "Controlla",
+ "desc": "Verifica se l'API Key e l'indirizzo proxy sono stati inseriti correttamente",
+ "pass": "Controllo superato",
+ "title": "Verifica connettività"
+ },
+ "fetchOnClient": {
+ "desc": "La modalità di richiesta client avvierà direttamente la richiesta di sessione dal browser, migliorando la velocità di risposta",
+ "title": "Utilizza la modalità di richiesta client"
+ },
+ "helpDoc": "Guida alla configurazione",
+ "waitingForMore": "Altri modelli sono in fase di <1>implementazione, resta sintonizzato"
},
- "checker": {
- "desc": "Verifica se l'AccessKey / SecretAccess è stato inserito correttamente"
+ "createNew": {
+ "title": "Crea modello AI personalizzato"
+ },
+ "item": {
+ "config": "Configura modello",
+ "customModelCards": {
+ "addNew": "Crea e aggiungi modello {{id}}",
+ "confirmDelete": "Stai per eliminare questo modello personalizzato, una volta eliminato non sarà recuperabile, procedi con cautela."
+ },
+ "delete": {
+ "confirm": "Confermi di voler eliminare il modello {{displayName}}?",
+ "success": "Eliminazione avvenuta con successo",
+ "title": "Elimina modello"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Campo effettivamente richiesto in Azure OpenAI",
+ "placeholder": "Inserisci il nome di distribuzione del modello in Azure",
+ "title": "Nome di distribuzione del modello"
+ },
+ "deployName": {
+ "extra": "Questo campo verrà utilizzato come ID del modello quando si invia la richiesta",
+ "placeholder": "Inserisci il nome o l'ID effettivo del modello distribuito",
+ "title": "Nome di distribuzione del modello"
+ },
+ "displayName": {
+ "placeholder": "Inserisci il nome visualizzato del modello, ad esempio ChatGPT, GPT-4, ecc.",
+ "title": "Nome visualizzato del modello"
+ },
+ "files": {
+ "extra": "L'attuale implementazione del caricamento file è solo una soluzione temporanea, limitata a tentativi personali. Attendere implementazioni complete per il caricamento file.",
+ "title": "Supporto per il caricamento file"
+ },
+ "functionCall": {
+ "extra": "Questa configurazione abiliterà solo la capacità del modello di utilizzare strumenti, consentendo così di aggiungere plugin di tipo strumento al modello. Tuttavia, se il modello supporta realmente l'uso degli strumenti dipende interamente dal modello stesso; si prega di testarne l'usabilità",
+ "title": "Supporto all'uso degli strumenti"
+ },
+ "id": {
+ "extra": "Non modificabile dopo la creazione, verrà utilizzato come ID del modello durante la chiamata all'AI",
+ "placeholder": "Inserisci l'ID del modello, ad esempio gpt-4o o claude-3.5-sonnet",
+ "title": "ID del modello"
+ },
+ "modalTitle": "Configurazione modello personalizzato",
+ "reasoning": {
+ "extra": "Questa configurazione attiverà solo la capacità di pensiero profondo del modello; l'effetto specifico dipende interamente dal modello stesso. Si prega di testare autonomamente se il modello possiede una capacità di pensiero profondo utilizzabile.",
+ "title": "Supporto per il pensiero profondo"
+ },
+ "tokens": {
+ "extra": "Imposta il numero massimo di token supportati dal modello",
+ "title": "Finestra di contesto massima",
+ "unlimited": "Illimitato"
+ },
+ "vision": {
+ "extra": "Questa configurazione abiliterà solo la configurazione di caricamento immagini nell'app, la disponibilità di riconoscimento dipende interamente dal modello stesso, testare autonomamente la disponibilità di riconoscimento visivo di questo modello.",
+ "title": "Supporto per riconoscimento visivo"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/Immagine",
+ "inputCharts": "${{amount}}/M caratteri",
+ "inputMinutes": "${{amount}}/minuti",
+ "inputTokens": "Ingresso ${{amount}}/M",
+ "outputTokens": "Uscita ${{amount}}/M"
+ },
+ "releasedAt": "Rilasciato il {{releasedAt}}"
},
- "secretKey": {
- "desc": "Inserisci il Secret Key della piattaforma Qianfan di Baidu",
- "placeholder": "Secret Key Qianfan",
- "title": "Secret Key"
+ "list": {
+ "addNew": "Aggiungi modello",
+ "disabled": "Non attivato",
+ "disabledActions": {
+ "showMore": "Mostra tutto"
+ },
+ "empty": {
+ "desc": "Si prega di creare un modello personalizzato o di importare un modello per iniziare a utilizzarlo",
+ "title": "Nessun modello disponibile"
+ },
+ "enabled": "Attivato",
+ "enabledActions": {
+ "disableAll": "Disattiva tutto",
+ "enableAll": "Attiva tutto",
+ "sort": "Ordinamento modelli personalizzato"
+ },
+ "enabledEmpty": "Nessun modello attivato, attiva i modelli desiderati dall'elenco qui sotto~",
+ "fetcher": {
+ "clear": "Cancella i modelli recuperati",
+ "fetch": "Recupera l'elenco dei modelli",
+ "fetching": "Recupero dell'elenco dei modelli in corso...",
+ "latestTime": "Ultimo aggiornamento: {{time}}",
+ "noLatestTime": "Nessun elenco recuperato finora"
+ },
+ "resetAll": {
+ "conform": "Sei sicuro di voler ripristinare tutte le modifiche al modello corrente? Dopo il ripristino, l'elenco dei modelli correnti tornerà allo stato predefinito",
+ "success": "Ripristino avvenuto con successo",
+ "title": "Ripristina tutte le modifiche"
+ },
+ "search": "Cerca modelli...",
+ "searchResult": "Trovati {{count}} modelli",
+ "title": "Elenco dei modelli",
+ "total": "Totale di {{count}} modelli disponibili"
},
- "unlock": {
- "customRegion": "Regione di servizio personalizzata",
- "description": "Inserisci il tuo AccessKey / SecretKey per iniziare la sessione. L'app non registrerà la tua configurazione di autenticazione",
- "title": "Utilizza le informazioni di autenticazione personalizzate di Wenxin Yiyan"
- }
+ "searchNotFound": "Nessun risultato trovato"
+ },
+ "sortModal": {
+ "success": "Ordinamento aggiornato con successo",
+ "title": "Ordinamento personalizzato",
+ "update": "Aggiorna"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "Stai per eliminare questo fornitore AI, una volta eliminato non sarà recuperabile, confermi di voler eliminare?",
+ "deleteSuccess": "Eliminazione avvenuta con successo",
+ "tooltip": "Aggiorna la configurazione di base del fornitore",
+ "updateSuccess": "Aggiornamento avvenuto con successo"
},
"zeroone": {
"title": "01.AI ZeroOne"
diff --git a/locales/it-IT/models.json b/locales/it-IT/models.json
index 1d04ac93169a9..9fa9b1a585729 100644
--- a/locales/it-IT/models.json
+++ b/locales/it-IT/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K enfatizza la sicurezza semantica e l'orientamento alla responsabilità, progettato specificamente per scenari applicativi con elevati requisiti di sicurezza dei contenuti, garantendo l'accuratezza e la robustezza dell'esperienza utente."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 utilizza la ricerca ad albero per costruire catene di pensiero e introduce un meccanismo di riflessione, addestrato tramite apprendimento rinforzato, dotando il modello della capacità di auto-riflessione e correzione degli errori."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro è un modello avanzato di elaborazione del linguaggio naturale lanciato da 360, con eccellenti capacità di generazione e comprensione del testo, in particolare nel campo della generazione e creazione, capace di gestire compiti complessi di conversione linguistica e interpretazione di ruoli."
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1 utilizza la ricerca ad albero per costruire catene di pensiero e introduce un meccanismo di riflessione, addestrato tramite apprendimento rinforzato, dotando il modello della capacità di auto-riflessione e correzione degli errori."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra è la versione più potente della serie di modelli Spark, migliorando la comprensione e la sintesi del contenuto testuale mentre aggiorna il collegamento alla ricerca online. È una soluzione completa per migliorare la produttività lavorativa e rispondere con precisione alle esigenze, rappresentando un prodotto intelligente all'avanguardia nel settore."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "Il modello con le migliori capacità in patria, supera i modelli principali esteri in compiti cinesi come enciclopedie, testi lunghi e creazione di contenuti. Possiede anche capacità multimodali leader del settore, con prestazioni eccellenti in vari benchmark di valutazione."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Il modello di distillazione DeepSeek-R1 basato su Qwen2.5-Math-1.5B ottimizza le prestazioni di inferenza attraverso l'apprendimento rinforzato e dati di avvio a freddo, aggiornando il benchmark multi-task del modello open source."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Il modello di distillazione DeepSeek-R1 basato su Qwen2.5-14B ottimizza le prestazioni di inferenza attraverso l'apprendimento rinforzato e dati di avvio a freddo, aggiornando il benchmark multi-task del modello open source."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "La serie DeepSeek-R1 ottimizza le prestazioni di inferenza attraverso l'apprendimento rinforzato e dati di avvio a freddo, aggiornando il benchmark multi-task del modello open source, superando il livello di OpenAI-o1-mini."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "Il modello di distillazione DeepSeek-R1 basato su Qwen2.5-Math-7B ottimizza le prestazioni di inferenza attraverso l'apprendimento rinforzato e dati di avvio a freddo, aggiornando il benchmark multi-task del modello open source."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite offre un'estrema velocità di risposta, un miglior rapporto qualità-prezzo e opzioni più flessibili per diversi scenari dei clienti. Supporta inferenze e fine-tuning con una finestra di contesto di 128k."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Modello di linguaggio di grande scala ultra avanzato sviluppato da Baidu, che rispetto a ERNIE 3.5 ha subito un aggiornamento completo delle capacità del modello, ampiamente applicabile a scenari di compiti complessi in vari settori; supporta l'integrazione automatica con il plugin di ricerca di Baidu, garantendo l'aggiornamento delle informazioni nelle risposte."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Modello di linguaggio di grande scala di punta sviluppato da Baidu, con prestazioni eccellenti in vari scenari di compiti complessi; supporta l'integrazione automatica con il plugin di ricerca di Baidu, garantendo l'aggiornamento delle informazioni nelle risposte. Rispetto a ERNIE 4.0, offre prestazioni superiori."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "Il modello linguistico ultra grande di Baidu, auto-sviluppato, offre eccellenti prestazioni generali, ampiamente utilizzabile in scenari complessi di vari settori; supporta l'integrazione automatica dei plugin di ricerca di Baidu, garantendo l'attualità delle informazioni nelle risposte. Rispetto a ERNIE 4.0, offre prestazioni superiori."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct è uno dei più recenti modelli linguistici di grandi dimensioni rilasciati da Alibaba Cloud. Questo modello da 7B ha capacità notevolmente migliorate in codifica e matematica. Il modello offre anche supporto multilingue, coprendo oltre 29 lingue, tra cui cinese e inglese. Ha mostrato miglioramenti significativi nel seguire istruzioni, comprendere dati strutturati e generare output strutturati (soprattutto JSON)."
},
+ "MiniMax-Text-01": {
+ "description": "Nella serie di modelli MiniMax-01, abbiamo fatto un'innovazione audace: per la prima volta abbiamo implementato su larga scala un meccanismo di attenzione lineare, rendendo l'architettura Transformer tradizionale non più l'unica opzione. Questo modello ha un numero di parametri che raggiunge i 456 miliardi, con un'attivazione singola di 45,9 miliardi. Le prestazioni complessive del modello sono paragonabili a quelle dei migliori modelli internazionali, mentre è in grado di gestire in modo efficiente contesti globali lunghi fino a 4 milioni di token, 32 volte rispetto a GPT-4o e 20 volte rispetto a Claude-3.5-Sonnet."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO è un modello altamente flessibile, progettato per offrire un'esperienza creativa eccezionale."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 ha dimostrato prestazioni eccezionali in una varietà di compiti visivi linguistici, tra cui comprensione di documenti e grafici, comprensione di testo in scena, OCR, risoluzione di problemi scientifici e matematici."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 ha dimostrato prestazioni eccezionali in una varietà di compiti visivi linguistici, tra cui comprensione di documenti e grafici, comprensione di testo in scena, OCR, risoluzione di problemi scientifici e matematici."
- },
"Phi-3-medium-128k-instruct": {
"description": "Stesso modello Phi-3-medium, ma con una dimensione di contesto più grande per RAG o prompting a pochi colpi."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat è la versione open source del modello pre-addestrato GLM-4 della serie sviluppata da Zhipu AI. Questo modello ha dimostrato prestazioni eccellenti in vari aspetti, tra cui semantica, matematica, ragionamento, codice e conoscenza. Oltre a supportare conversazioni multi-turno, GLM-4-9B-Chat offre anche funzionalità avanzate come navigazione web, esecuzione di codice, chiamate a strumenti personalizzati (Function Call) e ragionamento su testi lunghi. Il modello supporta 26 lingue, tra cui cinese, inglese, giapponese, coreano e tedesco. Ha mostrato prestazioni eccellenti in vari benchmark, come AlignBench-v2, MT-Bench, MMLU e C-Eval. Questo modello supporta una lunghezza di contesto massima di 128K, rendendolo adatto per ricerche accademiche e applicazioni commerciali."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 è un modello di inferenza guidato dall'apprendimento per rinforzo (RL) che affronta i problemi di ripetitività e leggibilità nel modello. Prima dell'RL, DeepSeek-R1 ha introdotto dati di cold start, ottimizzando ulteriormente le prestazioni di inferenza. Si comporta in modo comparabile a OpenAI-o1 in compiti matematici, di codifica e di inferenza, e migliora l'efficacia complessiva grazie a metodi di addestramento ben progettati."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 è un modello di linguaggio con 6710 miliardi di parametri, basato su un'architettura di esperti misti (MoE) che utilizza attenzione multilivello (MLA) e la strategia di bilanciamento del carico senza perdite ausiliarie, ottimizzando l'efficienza di inferenza e addestramento. Pre-addestrato su 14,8 trilioni di token di alta qualità e successivamente affinato tramite supervisione e apprendimento per rinforzo, DeepSeek-V3 supera altri modelli open source, avvicinandosi ai modelli chiusi di punta."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma è una delle serie di modelli open source leggeri e all'avanguardia sviluppati da Google. È un modello linguistico di grandi dimensioni con solo decoder, supporta l'inglese e offre pesi aperti, varianti pre-addestrate e varianti con fine-tuning per istruzioni. Il modello Gemma è adatto per vari compiti di generazione di testi, tra cui domande e risposte, riassunti e ragionamento. Questo modello da 9B è stato addestrato su 80 trilioni di token. La sua dimensione relativamente ridotta consente di implementarlo in ambienti con risorse limitate, come laptop, desktop o la propria infrastruttura cloud, rendendo così accessibili modelli AI all'avanguardia a un pubblico più ampio e promuovendo l'innovazione."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 è una famiglia di modelli linguistici di grandi dimensioni multilingue sviluppata da Meta, che include varianti pre-addestrate e con fine-tuning per istruzioni con dimensioni di 8B, 70B e 405B. Questo modello di fine-tuning per istruzioni da 8B è ottimizzato per scenari di dialogo multilingue e ha dimostrato prestazioni eccellenti in vari benchmark di settore. L'addestramento del modello ha utilizzato oltre 150 trilioni di token di dati pubblici e ha impiegato tecniche come il fine-tuning supervisionato e l'apprendimento per rinforzo basato su feedback umano per migliorare l'utilità e la sicurezza del modello. Llama 3.1 supporta la generazione di testi e di codice, con una data di scadenza delle conoscenze fissata a dicembre 2023."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview è un modello di elaborazione del linguaggio naturale innovativo, in grado di gestire in modo efficiente compiti complessi di generazione di dialoghi e comprensione del contesto."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview è un modello di ricerca sviluppato dal team Qwen, focalizzato sulle capacità di inferenza visiva, con vantaggi unici nella comprensione di scenari complessi e nella risoluzione di problemi matematici legati alla visione."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview è l'ultimo modello di ricerca sperimentale di Qwen, focalizzato sul miglioramento delle capacità di ragionamento dell'IA. Esplorando meccanismi complessi come la mescolanza linguistica e il ragionamento ricorsivo, i principali vantaggi includono potenti capacità di analisi del ragionamento, abilità matematiche e di programmazione. Tuttavia, ci sono anche problemi di cambio linguistico, cicli di ragionamento, considerazioni di sicurezza e differenze in altre capacità."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct è l'ultima versione della serie di modelli linguistici di grandi dimensioni specifici per il codice rilasciata da Alibaba Cloud. Questo modello, basato su Qwen2.5, ha migliorato significativamente le capacità di generazione, ragionamento e riparazione del codice grazie all'addestramento su 55 trilioni di token. Ha potenziato non solo le capacità di codifica, ma ha anche mantenuto i vantaggi nelle abilità matematiche e generali. Il modello fornisce una base più completa per applicazioni pratiche come agenti di codice."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math si concentra sulla risoluzione di problemi nel campo della matematica, fornendo risposte professionali a domande di alta difficoltà."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 è l'ultima serie del modello Qwen, supporta un contesto di 128k, e rispetto ai modelli open source attualmente migliori, Qwen2-72B supera significativamente i modelli leader attuali in comprensione del linguaggio naturale, conoscenza, codice, matematica e capacità multilingue."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 è l'ultima serie del modello Qwen, in grado di superare i modelli open source ottimali di dimensioni simili e anche modelli di dimensioni maggiori. Qwen2 7B ha ottenuto vantaggi significativi in vari test, in particolare nella comprensione del codice e del cinese."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B è un potente modello di linguaggio visivo, supporta l'elaborazione multimodale di immagini e testo, in grado di riconoscere con precisione il contenuto delle immagini e generare descrizioni o risposte correlate."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct è un grande modello linguistico con 14 miliardi di parametri, con prestazioni eccellenti, ottimizzato per scenari in cinese e multilingue, supporta applicazioni di domande e risposte intelligenti, generazione di contenuti e altro."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct è un grande modello linguistico con 7 miliardi di parametri, supporta chiamate di funzione e interazioni senza soluzione di continuità con sistemi esterni, aumentando notevolmente flessibilità e scalabilità. Ottimizzato per scenari in cinese e multilingue, supporta applicazioni di domande e risposte intelligenti, generazione di contenuti e altro."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct è un modello di istruzioni per la programmazione basato su un pre-addestramento su larga scala, con potenti capacità di comprensione e generazione del codice, in grado di gestire in modo efficiente vari compiti di programmazione, particolarmente adatto per la scrittura intelligente di codice, la generazione di script automatizzati e la risoluzione di problemi di programmazione."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct è un grande modello linguistico progettato per la generazione di codice, la comprensione del codice e scenari di sviluppo efficienti, con una scala di 32 miliardi di parametri all'avanguardia nel settore, in grado di soddisfare esigenze di programmazione diversificate."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "Il grande modello multimodale TeleMM è un modello di comprensione multimodale sviluppato autonomamente da China Telecom, in grado di gestire input di diverse modalità, come testo e immagini, supportando funzioni di comprensione delle immagini e analisi dei grafici, fornendo servizi di comprensione multimodale agli utenti. Il modello è in grado di interagire con gli utenti in modo multimodale, comprendere accuratamente il contenuto dell'input, rispondere a domande, assistere nella creazione e fornire in modo efficiente supporto informativo e ispirazione multimodale. Ha mostrato prestazioni eccellenti in compiti multimodali come percezione fine e ragionamento logico."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large è il modello MoE con architettura Transformer open source più grande del settore, con un totale di 389 miliardi di parametri e 52 miliardi di parametri attivi."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct è un modello linguistico di grandi dimensioni con fine-tuning per istruzioni nella serie Qwen2, con una dimensione di 72B parametri. Questo modello si basa sull'architettura Transformer, utilizzando funzioni di attivazione SwiGLU, bias QKV di attenzione e attenzione a query di gruppo. È in grado di gestire input di grandi dimensioni. Ha dimostrato prestazioni eccellenti in comprensione linguistica, generazione, capacità multilingue, codifica, matematica e ragionamento in vari benchmark, superando la maggior parte dei modelli open source e mostrando competitività paragonabile a modelli proprietari in alcuni compiti."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct è uno dei più recenti modelli linguistici di grandi dimensioni rilasciati da Alibaba Cloud. Questo modello da 72B ha capacità notevolmente migliorate in codifica e matematica. Il modello offre anche supporto multilingue, coprendo oltre 29 lingue, tra cui cinese e inglese. Ha mostrato miglioramenti significativi nel seguire istruzioni, comprendere dati strutturati e generare output strutturati (soprattutto JSON)."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Ottimizzato per scenari di dialogo con personaggi cinesi, offre capacità di generazione di dialoghi fluida e conforme alle espressioni cinesi."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Il modello open source di chiamata di funzione di Fireworks offre capacità di esecuzione di istruzioni eccezionali e caratteristiche personalizzabili."
+ "abab7-chat-preview": {
+ "description": "Rispetto alla serie di modelli abab6.5, ci sono stati notevoli miglioramenti nelle capacità di elaborazione di testi lunghi, matematica e scrittura."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Firefunction-v2, l'ultima offerta di Fireworks, è un modello di chiamata di funzione ad alte prestazioni, sviluppato su Llama-3 e ottimizzato per scenari come chiamate di funzione, dialogo e seguimento di istruzioni."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 è un modello linguistico di grandi dimensioni all'avanguardia, ottimizzato tramite apprendimento rinforzato e dati di cold start, con prestazioni eccezionali nel ragionamento, nella matematica e nella programmazione."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b è un modello di linguaggio visivo in grado di ricevere input sia visivi che testuali, addestrato su dati di alta qualità, adatto per compiti multimodali."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Un potente modello linguistico Mixture-of-Experts (MoE) fornito da Deepseek, con un totale di 671B di parametri, attivando 37B di parametri per ogni token."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Il modello di istruzioni Llama 3 70B è ottimizzato per dialoghi multilingue e comprensione del linguaggio naturale, superando le prestazioni della maggior parte dei modelli concorrenti."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Il modello di istruzioni Llama 3 70B (versione HF) è allineato con i risultati dell'implementazione ufficiale, adatto per compiti di seguimento di istruzioni di alta qualità."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Il modello di istruzioni Llama 3 8B è ottimizzato per dialoghi e compiti multilingue, offrendo prestazioni eccellenti e alta efficienza."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Modello di ragionamento visivo di Meta con 11 miliardi di parametri. Questo modello è ottimizzato per il riconoscimento visivo, il ragionamento visivo, la descrizione delle immagini e la risposta a domande generali riguardanti le immagini. Questo modello è in grado di comprendere dati visivi, come grafici e tabelle, e colmare il divario tra visione e linguaggio generando descrizioni testuali dei dettagli delle immagini."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Il modello di istruzioni Llama 3.2 1B è un modello multilingue leggero lanciato da Meta. Questo modello è progettato per migliorare l'efficienza, offrendo miglioramenti significativi in termini di latenza e costi rispetto a modelli più grandi. I casi d'uso esemplari di questo modello includono recupero e sintesi."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Il modello di istruzioni Llama 3.2 3B è un modello multilingue leggero lanciato da Meta. Questo modello è progettato per migliorare l'efficienza, offrendo miglioramenti significativi in termini di latenza e costi rispetto a modelli più grandi. I casi d'uso esemplari di questo modello includono query e riscrittura di suggerimenti, nonché supporto alla scrittura."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Modello di ragionamento visivo di Meta con 90 miliardi di parametri. Questo modello è ottimizzato per il riconoscimento visivo, il ragionamento visivo, la descrizione delle immagini e la risposta a domande generali riguardanti le immagini. Questo modello è in grado di comprendere dati visivi, come grafici e tabelle, e colmare il divario tra visione e linguaggio generando descrizioni testuali dei dettagli delle immagini."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct è la versione aggiornata di dicembre di Llama 3.1 70B. Questo modello è stato migliorato rispetto a Llama 3.1 70B (rilasciato a luglio 2024), potenziando le capacità di chiamata degli strumenti, il supporto per testi multilingue, le abilità matematiche e di programmazione. Il modello raggiunge livelli di eccellenza nel ragionamento, nella matematica e nel rispetto delle istruzioni, offrendo prestazioni simili a quelle di 3.1 405B, con vantaggi significativi in termini di velocità e costi."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "Modello con 24B di parametri, dotato di capacità all'avanguardia comparabili a modelli di dimensioni maggiori."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Il modello di istruzioni Mixtral MoE 8x22B, con parametri su larga scala e architettura multi-esperto, supporta in modo completo l'elaborazione efficiente di compiti complessi."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Il modello di istruzioni Mixtral MoE 8x7B, con architettura multi-esperto, offre un'elevata efficienza nel seguire e eseguire istruzioni."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Il modello di istruzioni Mixtral MoE 8x7B (versione HF) ha prestazioni coerenti con l'implementazione ufficiale, adatto per vari scenari di compiti efficienti."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "Il modello MythoMax L2 13B combina tecnologie di fusione innovative, specializzandosi in narrazione e interpretazione di ruoli."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "Il modello QwQ è un modello di ricerca sperimentale sviluppato dal team Qwen, focalizzato sul potenziamento delle capacità di ragionamento dell'IA."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "La versione 72B del modello Qwen-VL è il risultato dell'ultima iterazione di Alibaba, rappresentando quasi un anno di innovazione."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 è una serie di modelli linguistici solo decoder sviluppata dal team Qwen di Alibaba Cloud. Questi modelli offrono dimensioni diverse, tra cui 0.5B, 1.5B, 3B, 7B, 14B, 32B e 72B, e ci sono varianti base e di istruzione."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct è l'ultima versione della serie di modelli linguistici di grandi dimensioni specifici per il codice rilasciata da Alibaba Cloud. Questo modello, basato su Qwen2.5, ha migliorato significativamente le capacità di generazione, ragionamento e riparazione del codice grazie all'addestramento su 55 trilioni di token. Ha potenziato non solo le capacità di codifica, ma ha anche mantenuto i vantaggi nelle abilità matematiche e generali. Il modello fornisce una base più completa per applicazioni pratiche come agenti di codice."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "Il modello StarCoder 15.5B supporta compiti di programmazione avanzati, con capacità multilingue potenziate, adatto per la generazione e comprensione di codice complesso."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "Il modello StarCoder 7B è addestrato su oltre 80 linguaggi di programmazione, con eccellenti capacità di completamento del codice e comprensione del contesto."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Il modello Yi-Large offre capacità eccezionali di elaborazione multilingue, utilizzabile per vari compiti di generazione e comprensione del linguaggio."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus è il modello più potente di Anthropic per gestire compiti altamente complessi. Eccelle in prestazioni, intelligenza, fluidità e comprensione."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku è il modello di nuova generazione più veloce di Anthropic. Rispetto a Claude 3 Haiku, Claude 3.5 Haiku ha migliorato le proprie capacità e ha superato il modello più grande della generazione precedente, Claude 3 Opus, in molti test di intelligenza."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet offre capacità superiori rispetto a Opus e una velocità maggiore rispetto a Sonnet, mantenendo lo stesso prezzo di Sonnet. Sonnet è particolarmente abile in programmazione, scienza dei dati, elaborazione visiva e compiti di agenzia."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet offre un equilibrio ideale tra intelligenza e velocità per i carichi di lavoro aziendali. Fornisce la massima utilità a un prezzo inferiore, affidabile e adatto per distribuzioni su larga scala."
},
- "code-raccoon-v1": {
- "description": "Code Raccoon è un assistente intelligente per lo sviluppo software basato su un grande modello linguistico di SenseTime, copre fasi come analisi dei requisiti software, progettazione dell'architettura, scrittura del codice e test del software, soddisfacendo le esigenze degli utenti nella scrittura di codice e nell'apprendimento della programmazione. Code Raccoon supporta oltre 90 linguaggi di programmazione principali come Python, Java, JavaScript, C++, Go, SQL e IDE principali come VS Code, IntelliJ IDEA. Nelle applicazioni pratiche, Code Raccoon può aiutare gli sviluppatori a migliorare l'efficienza della programmazione di oltre il 50%."
- },
"codegeex-4": {
"description": "CodeGeeX-4 è un potente assistente di programmazione AI, supporta domande intelligenti e completamento del codice in vari linguaggi di programmazione, migliorando l'efficienza dello sviluppo."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ è un modello di linguaggio di grandi dimensioni ad alte prestazioni, progettato per scenari aziendali reali e applicazioni complesse."
},
+ "dall-e-2": {
+ "description": "Seconda generazione del modello DALL·E, supporta la generazione di immagini più realistiche e accurate, con una risoluzione quattro volte superiore rispetto alla prima generazione."
+ },
+ "dall-e-3": {
+ "description": "L'ultimo modello DALL·E, rilasciato a novembre 2023. Supporta la generazione di immagini più realistiche e accurate, con una maggiore capacità di dettaglio."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct offre capacità di elaborazione di istruzioni altamente affidabili, supportando applicazioni in vari settori."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 è un potente e conveniente modello linguistico ibrido (MoE). È stato pre-addestrato su un corpus di alta qualità di 81 trilioni di token e ha ulteriormente migliorato le capacità del modello attraverso il fine-tuning supervisionato (SFT) e l'apprendimento per rinforzo (RL). Rispetto a DeepSeek 67B, DeepSeek-V2 offre prestazioni superiori, risparmiando il 42,5% dei costi di addestramento, riducendo il 93,3% della cache KV e aumentando il massimo throughput di generazione di 5,76 volte. Questo modello supporta una lunghezza di contesto di 128k e ha dimostrato prestazioni eccellenti nei benchmark standard e nelle valutazioni di generazione aperta."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 è un modello di inferenza guidato da apprendimento rinforzato (RL) che affronta i problemi di ripetitività e leggibilità nel modello. Prima dell'RL, DeepSeek-R1 ha introdotto dati di cold start, ottimizzando ulteriormente le prestazioni di inferenza. Si comporta in modo comparabile a OpenAI-o1 in compiti matematici, di codifica e di inferenza, e migliora l'efficacia complessiva attraverso metodi di addestramento accuratamente progettati."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "Il modello di distillazione DeepSeek-R1 ottimizza le prestazioni di inferenza attraverso l'apprendimento rinforzato e dati di avvio a freddo, aggiornando il benchmark multi-task del modello open source."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B è un modello di distillazione sviluppato sulla base di Llama-3.1-8B. Questo modello è stato messo a punto utilizzando campioni generati da DeepSeek-R1, mostrando eccellenti capacità di inferenza. Ha ottenuto buoni risultati in vari test di benchmark, raggiungendo un'accuratezza dell'89,1% in MATH-500, una percentuale di passaggio del 50,4% in AIME 2024 e un punteggio di 1205 su CodeForces, dimostrando forti capacità matematiche e di programmazione come modello di dimensioni 8B."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Il modello di distillazione DeepSeek-R1 ottimizza le prestazioni di inferenza attraverso l'apprendimento rinforzato e dati di avvio a freddo, aggiornando il benchmark multi-task del modello open source."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Il modello di distillazione DeepSeek-R1 ottimizza le prestazioni di inferenza attraverso l'apprendimento rinforzato e dati di avvio a freddo, aggiornando il benchmark multi-task del modello open source."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B è un modello ottenuto tramite distillazione della conoscenza basato su Qwen2.5-32B. Questo modello è stato messo a punto utilizzando 800.000 campioni selezionati generati da DeepSeek-R1, mostrando prestazioni eccezionali in vari campi come matematica, programmazione e ragionamento. Ha ottenuto risultati eccellenti in vari test di benchmark, raggiungendo un'accuratezza del 94,3% in MATH-500, dimostrando una forte capacità di ragionamento matematico."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B è un modello ottenuto tramite distillazione della conoscenza basato su Qwen2.5-Math-7B. Questo modello è stato messo a punto utilizzando 800.000 campioni selezionati generati da DeepSeek-R1, mostrando eccellenti capacità di inferenza. Ha ottenuto risultati eccezionali in vari test di benchmark, raggiungendo un'accuratezza del 92,8% in MATH-500, una percentuale di passaggio del 55,5% in AIME 2024 e un punteggio di 1189 su CodeForces, dimostrando forti capacità matematiche e di programmazione come modello di dimensioni 7B."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 combina le eccellenti caratteristiche delle versioni precedenti, migliorando le capacità generali e di codifica."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 è un modello linguistico a esperti misti (MoE) con 6710 miliardi di parametri, che utilizza attenzione latente multi-testa (MLA) e architettura DeepSeekMoE, combinando strategie di bilanciamento del carico senza perdite ausiliarie per ottimizzare l'efficienza di inferenza e addestramento. Pre-addestrato su 14,8 trilioni di token di alta qualità e successivamente affinato supervisionato e tramite apprendimento rinforzato, DeepSeek-V3 supera le prestazioni di altri modelli open source, avvicinandosi ai modelli closed source leader."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B è un modello avanzato addestrato per dialoghi ad alta complessità."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "LLM altamente efficiente, specializzato in ragionamento, matematica e programmazione."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 è un modello linguistico visivo a esperti misti (MoE) sviluppato sulla base di DeepSeekMoE-27B, che utilizza un'architettura MoE con attivazione sparsa, raggiungendo prestazioni eccezionali attivando solo 4,5 miliardi di parametri. Questo modello eccelle in vari compiti, tra cui domande visive, riconoscimento ottico dei caratteri, comprensione di documenti/tabelle/grafici e localizzazione visiva."
+ },
"deepseek-chat": {
"description": "Un nuovo modello open source che integra capacità generali e di codifica, mantenendo non solo le capacità conversazionali generali del modello Chat originale, ma anche la potente capacità di elaborazione del codice del modello Coder, allineandosi meglio alle preferenze umane. Inoltre, DeepSeek-V2.5 ha ottenuto notevoli miglioramenti in vari aspetti, come i compiti di scrittura e il rispetto delle istruzioni."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 è un modello di codice open source di esperti misti, eccelle nei compiti di codice, paragonabile a GPT4-Turbo."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 è un modello di inferenza guidato da apprendimento rinforzato (RL) che affronta i problemi di ripetitività e leggibilità nel modello. Prima dell'RL, DeepSeek-R1 ha introdotto dati di cold start, ottimizzando ulteriormente le prestazioni di inferenza. Si comporta in modo comparabile a OpenAI-o1 in compiti matematici, di codifica e di inferenza, e migliora l'efficacia complessiva attraverso metodi di addestramento accuratamente progettati."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 - il modello più grande e intelligente del pacchetto DeepSeek - è stato distillato nell'architettura Llama 70B. Basato su test di benchmark e valutazioni umane, questo modello è più intelligente del Llama 70B originale, mostrando prestazioni eccezionali in compiti che richiedono precisione matematica e fattuale."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "Il modello della serie DeepSeek-R1-Distill è stato ottenuto tramite la tecnologia di distillazione della conoscenza, ottimizzando i campioni generati da DeepSeek-R1 su modelli open source come Qwen e Llama."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "Il modello della serie DeepSeek-R1-Distill è stato ottenuto tramite la tecnologia di distillazione della conoscenza, ottimizzando i campioni generati da DeepSeek-R1 su modelli open source come Qwen e Llama."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "Il modello della serie DeepSeek-R1-Distill è stato ottenuto tramite la tecnologia di distillazione della conoscenza, ottimizzando i campioni generati da DeepSeek-R1 su modelli open source come Qwen e Llama."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "Il modello della serie DeepSeek-R1-Distill è stato ottenuto tramite la tecnologia di distillazione della conoscenza, ottimizzando i campioni generati da DeepSeek-R1 su modelli open source come Qwen e Llama."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "Il modello della serie DeepSeek-R1-Distill è stato ottenuto tramite la tecnologia di distillazione della conoscenza, ottimizzando i campioni generati da DeepSeek-R1 su modelli open source come Qwen e Llama."
+ },
+ "deepseek-reasoner": {
+ "description": "Modello di ragionamento lanciato da DeepSeek. Prima di fornire la risposta finale, il modello genera una catena di pensiero per migliorare l'accuratezza della risposta finale."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 è un modello di linguaggio Mixture-of-Experts efficiente, adatto per esigenze di elaborazione economica."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B è il modello di codice progettato di DeepSeek, offre potenti capacità di generazione di codice."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 è un modello MoE sviluppato internamente da Hangzhou DeepSeek Artificial Intelligence Technology Research Co., Ltd., con risultati eccezionali in molteplici valutazioni, posizionandosi al primo posto tra i modelli open source nelle classifiche principali. Rispetto al modello V2.5, la velocità di generazione è aumentata di 3 volte, offrendo un'esperienza utente più rapida e fluida."
+ },
"deepseek/deepseek-chat": {
"description": "Un nuovo modello open source che integra capacità generali e di codice, mantenendo non solo le capacità di dialogo generali del modello Chat originale e la potente capacità di elaborazione del codice del modello Coder, ma allineandosi anche meglio alle preferenze umane. Inoltre, DeepSeek-V2.5 ha ottenuto notevoli miglioramenti in vari aspetti, come compiti di scrittura e seguire istruzioni."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 ha notevolmente migliorato le capacità di ragionamento del modello con pochissimi dati etichettati. Prima di fornire la risposta finale, il modello genera una catena di pensiero per migliorare l'accuratezza della risposta finale."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 ha notevolmente migliorato le capacità di ragionamento del modello con pochissimi dati etichettati. Prima di fornire la risposta finale, il modello genera una catena di pensiero per migliorare l'accuratezza della risposta finale."
+ },
"emohaa": {
"description": "Emohaa è un modello psicologico, con capacità di consulenza professionale, aiuta gli utenti a comprendere i problemi emotivi."
},
+ "ernie-3.5-128k": {
+ "description": "Il modello di linguaggio di grandi dimensioni di punta sviluppato internamente da Baidu, coprendo un'enorme quantità di dati in cinese e inglese, con forti capacità generali, in grado di soddisfare la maggior parte delle esigenze di domande e risposte, generazione creativa e scenari di applicazione di plugin; supporta l'integrazione automatica con il plugin di ricerca di Baidu, garantendo l'aggiornamento delle informazioni nelle risposte."
+ },
+ "ernie-3.5-8k": {
+ "description": "Il modello di linguaggio di grandi dimensioni di punta sviluppato internamente da Baidu, coprendo un'enorme quantità di dati in cinese e inglese, con forti capacità generali, in grado di soddisfare la maggior parte delle esigenze di domande e risposte, generazione creativa e scenari di applicazione di plugin; supporta l'integrazione automatica con il plugin di ricerca di Baidu, garantendo l'aggiornamento delle informazioni nelle risposte."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "Il modello di linguaggio di grandi dimensioni di punta sviluppato internamente da Baidu, coprendo un'enorme quantità di dati in cinese e inglese, con forti capacità generali, in grado di soddisfare la maggior parte delle esigenze di domande e risposte, generazione creativa e scenari di applicazione di plugin; supporta l'integrazione automatica con il plugin di ricerca di Baidu, garantendo l'aggiornamento delle informazioni nelle risposte."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "Il modello di linguaggio di grandi dimensioni di punta sviluppato internamente da Baidu, con un aggiornamento completo delle capacità rispetto a ERNIE 3.5, ampiamente applicabile a scenari di compiti complessi in vari campi; supporta l'integrazione automatica con il plugin di ricerca di Baidu, garantendo l'aggiornamento delle informazioni nelle risposte."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "Il modello di linguaggio di grandi dimensioni di punta sviluppato internamente da Baidu, con un aggiornamento completo delle capacità rispetto a ERNIE 3.5, ampiamente applicabile a scenari di compiti complessi in vari campi; supporta l'integrazione automatica con il plugin di ricerca di Baidu, garantendo l'aggiornamento delle informazioni nelle risposte."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "Il modello di linguaggio di grandi dimensioni di punta sviluppato internamente da Baidu, con prestazioni complessive eccezionali, ampiamente applicabile a scenari di compiti complessi in vari campi; supporta l'integrazione automatica con il plugin di ricerca di Baidu, garantendo l'aggiornamento delle informazioni nelle risposte. Rispetto a ERNIE 4.0, offre prestazioni migliori."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "Il modello di linguaggio di grandi dimensioni di punta sviluppato internamente da Baidu, con prestazioni complessive eccezionali, ampiamente applicabile a scenari di compiti complessi in vari campi; supporta l'integrazione automatica con il plugin di ricerca di Baidu, garantendo l'aggiornamento delle informazioni nelle risposte. Rispetto a ERNIE 4.0, offre prestazioni migliori."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "Il modello di linguaggio di grandi dimensioni di punta sviluppato internamente da Baidu, con prestazioni complessive eccezionali, ampiamente applicabile a scenari di compiti complessi in vari campi; supporta l'integrazione automatica con il plugin di ricerca di Baidu, garantendo l'aggiornamento delle informazioni nelle risposte. Rispetto a ERNIE 4.0, offre prestazioni migliori."
+ },
+ "ernie-char-8k": {
+ "description": "Un modello di linguaggio di grandi dimensioni sviluppato internamente da Baidu, adatto per scenari di applicazione come NPC nei giochi, dialoghi di assistenza clienti e interpretazione di ruoli nei dialoghi, con uno stile di personaggio più distintivo e coerente, capacità di seguire istruzioni più forti e prestazioni di inferenza migliori."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Un modello di linguaggio di grandi dimensioni sviluppato internamente da Baidu, adatto per scenari di applicazione come NPC nei giochi, dialoghi di assistenza clienti e interpretazione di ruoli nei dialoghi, con uno stile di personaggio più distintivo e coerente, capacità di seguire istruzioni più forti e prestazioni di inferenza migliori."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite è un modello di linguaggio di grandi dimensioni sviluppato internamente da Baidu, che bilancia prestazioni eccellenti del modello e prestazioni di inferenza, adatto per l'uso con schede di accelerazione AI a bassa potenza."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Un modello di linguaggio di grandi dimensioni leggero sviluppato internamente da Baidu, che bilancia prestazioni eccellenti del modello e prestazioni di inferenza, con risultati migliori rispetto a ERNIE Lite, adatto per l'uso con schede di accelerazione AI a bassa potenza."
+ },
+ "ernie-novel-8k": {
+ "description": "Un modello di linguaggio di grandi dimensioni sviluppato internamente da Baidu, con un evidente vantaggio nella capacità di continuare romanzi, utilizzabile anche in scenari come cortometraggi e film."
+ },
+ "ernie-speed-128k": {
+ "description": "Il modello di linguaggio di grandi dimensioni ad alte prestazioni sviluppato internamente da Baidu, rilasciato nel 2024, con capacità generali eccellenti, adatto come modello di base per la messa a punto, per affrontare meglio i problemi specifici, mantenendo eccellenti prestazioni di inferenza."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "Il modello di linguaggio di grandi dimensioni ad alte prestazioni sviluppato internamente da Baidu, rilasciato nel 2024, con capacità generali eccellenti, con risultati migliori rispetto a ERNIE Speed, adatto come modello di base per la messa a punto, per affrontare meglio i problemi specifici, mantenendo eccellenti prestazioni di inferenza."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny è un modello di linguaggio di grandi dimensioni ad alte prestazioni sviluppato internamente da Baidu, con i costi di distribuzione e messa a punto più bassi della serie Wencin."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Tuning) offre prestazioni stabili e ottimizzabili, è la scelta ideale per soluzioni a compiti complessi."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro supporta fino a 2 milioni di token, è la scelta ideale per modelli multimodali di medie dimensioni, adatta a un supporto multifunzionale per compiti complessi."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash offre funzionalità e miglioramenti di nuova generazione, tra cui velocità eccezionale, utilizzo di strumenti nativi, generazione multimodale e una finestra di contesto di 1M token."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash offre funzionalità e miglioramenti di nuova generazione, tra cui velocità eccezionale, utilizzo di strumenti nativi, generazione multimodale e una finestra di contesto di 1M token."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp è il nuovo modello AI multimodale sperimentale di Google, dotato di funzionalità di nuova generazione, velocità straordinaria, chiamate di strumenti native e generazione multimodale."
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "Un modello Gemini 2.0 Flash ottimizzato per obiettivi di costo-efficacia e bassa latenza."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Exp è il più recente modello AI multimodale sperimentale di Google, dotato di caratteristiche di nuova generazione, velocità eccezionale, chiamate a strumenti nativi e generazione multimodale."
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Exp è il più recente modello AI multimodale sperimentale di Google, dotato di funzionalità di nuova generazione, velocità eccezionale, chiamate a strumenti nativi e generazione multimodale."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 è il modello AI multimodale sperimentale più recente di Google, dotato di capacità di elaborazione rapida, supporta input testuali, visivi e video, ed è adatto per un'ampia gamma di compiti con un'efficiente scalabilità."
- },
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 è il più recente modello AI multimodale sperimentale di Google, dotato di capacità di elaborazione rapida, supporta input di testo, immagini e video, ed è adatto per un'ampia gamma di compiti con un'efficace scalabilità."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental è il più recente modello AI multimodale sperimentale di Google, con un miglioramento della qualità rispetto alle versioni precedenti, in particolare per quanto riguarda la conoscenza del mondo, il codice e i contesti lunghi."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 è il modello AI multimodale sperimentale più recente di Google, con miglioramenti significativi della qualità rispetto alle versioni precedenti."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus ha la capacità di comprendere contenuti video e più immagini, adatto per compiti multimodali."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview possiede potenti capacità di ragionamento complesso, eccellendo nei campi del ragionamento logico, della matematica e della programmazione."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash offre funzionalità e miglioramenti di nuova generazione, tra cui velocità eccezionale, utilizzo di strumenti nativi, generazione multimodale e una finestra di contesto di 1M token."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental è il più recente modello AI multimodale sperimentale di Google, con un miglioramento della qualità rispetto alle versioni precedenti, in particolare per quanto riguarda la conoscenza del mondo, il codice e i contesti lunghi."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash offre capacità di elaborazione multimodale ottimizzate, adatte a vari scenari di compiti complessi."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro combina le più recenti tecnologie di ottimizzazione, offrendo una capacità di elaborazione dei dati multimodali più efficiente."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 è un modello efficiente lanciato da Google, coprendo una varietà di scenari applicativi, dalle piccole applicazioni all'elaborazione di dati complessi."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 continua il concetto di design leggero ed efficiente."
},
"google/gemma-2-2b-it": {
"description": "Modello di ottimizzazione delle istruzioni leggero di Google"
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 è un modello efficiente lanciato da Google, coprendo una varietà di scenari applicativi, dalle piccole applicazioni all'elaborazione di dati complessi."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 è una serie di modelli di testo open source leggeri di Google."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo, adatto a una varietà di compiti di generazione e comprensione del testo, attualmente punta a gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, modello di generazione di testo ad alta capacità, adatto per compiti complessi."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo, adatto a una varietà di compiti di generazione e comprensione del testo, attualmente punta a gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o è un modello dinamico che si aggiorna in tempo reale per mantenere sempre l'ultima versione. Combina una potente comprensione del linguaggio e capacità di generazione, rendendolo adatto a scenari di applicazione su larga scala, inclusi assistenza clienti, istruzione e supporto tecnico."
},
+ "gpt-4o-audio-preview": {
+ "description": "Modello GPT-4o Audio, supporta input e output audio."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini è il modello più recente lanciato da OpenAI dopo il GPT-4 Omni, supporta input visivi e testuali e produce output testuali. Come il loro modello di punta in formato ridotto, è molto più economico rispetto ad altri modelli all'avanguardia recenti e costa oltre il 60% in meno rispetto a GPT-3.5 Turbo. Mantiene un'intelligenza all'avanguardia, offrendo un rapporto qualità-prezzo significativo. GPT-4o mini ha ottenuto un punteggio dell'82% nel test MMLU e attualmente è classificato più in alto di GPT-4 per preferenze di chat."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "Versione in tempo reale di GPT-4o-mini, supporta input e output audio e testuali in tempo reale."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "Versione in tempo reale di GPT-4o, supporta input e output audio e testuali in tempo reale."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "Versione in tempo reale di GPT-4o, supporta input e output audio e testuali in tempo reale."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "Versione in tempo reale di GPT-4o, supporta input e output audio e testuali in tempo reale."
+ },
"grok-2-1212": {
"description": "Questo modello ha migliorato l'accuratezza, il rispetto delle istruzioni e le capacità multilingue."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "Ultimo modello FunctionCall con architettura MOE di Hunyuan, addestrato su dati di alta qualità per le chiamate di funzione, con una finestra di contesto di 32K, è in testa in vari indicatori di valutazione."
},
+ "hunyuan-large": {
+ "description": "Il modello Hunyuan-large ha un numero totale di parametri di circa 389B, con circa 52B di parametri attivati, ed è il modello MoE open source con la più grande scala di parametri e le migliori prestazioni nel settore, basato su architettura Transformer."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Specializzato nel gestire compiti di testi lunghi come riassunti di documenti e domande e risposte sui documenti, possiede anche capacità di generazione di testi generali. Eccelle nell'analisi e nella generazione di testi lunghi, in grado di affrontare efficacemente esigenze complesse e dettagliate di elaborazione di contenuti lunghi."
+ },
"hunyuan-lite": {
"description": "Aggiornato a una struttura MOE, con una finestra di contesto di 256k, è in testa a molti modelli open source in vari set di valutazione su NLP, codice, matematica e settori."
},
+ "hunyuan-lite-vision": {
+ "description": "Il modello multimodale Hunyuan più recente da 7B, con una finestra contestuale di 32K, supporta dialoghi multimodali in cinese e inglese, riconoscimento di oggetti nelle immagini, comprensione di documenti e tabelle, matematica multimodale, e supera i modelli concorrenti da 7B in vari indicatori di valutazione."
+ },
"hunyuan-pro": {
"description": "Modello di testo lungo MOE-32K con un miliardo di parametri. Raggiunge livelli di eccellenza in vari benchmark, con capacità di istruzioni complesse e ragionamento, supporta le chiamate di funzione, ottimizzato per traduzione multilingue, finanza, diritto e medicina."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Utilizza una strategia di routing migliore, alleviando i problemi di bilanciamento del carico e convergenza degli esperti. Per i testi lunghi, l'indice di recupero è del 99,9%. MOE-256K supera ulteriormente in lunghezza ed efficacia, ampliando notevolmente la lunghezza massima di input."
},
+ "hunyuan-standard-vision": {
+ "description": "Il modello multimodale più recente di Hunyuan, supporta risposte in più lingue, con capacità equilibrate in cinese e inglese."
+ },
"hunyuan-turbo": {
"description": "Anteprima della nuova generazione di modelli di linguaggio di Hunyuan, utilizza una nuova struttura di modello ibrido di esperti (MoE), con una maggiore efficienza di inferenza e prestazioni superiori rispetto a hunyuan-pro."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Versione fissa di hunyuan-turbo del 20 novembre 2024, una versione intermedia tra hunyuan-turbo e hunyuan-turbo-latest."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "Ottimizzazione di questa versione: scaling delle istruzioni sui dati, notevole aumento della capacità di generalizzazione del modello; notevole miglioramento delle capacità matematiche, di codifica e di ragionamento logico; ottimizzazione delle capacità di comprensione del testo e delle parole; ottimizzazione della qualità della generazione dei contenuti di creazione del testo."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "Ottimizzazione dell'esperienza generale, inclusi comprensione NLP, creazione di testi, conversazione, domande e risposte, traduzione, e altro; miglioramento dell'umanizzazione, ottimizzazione dell'intelligenza emotiva del modello; potenziamento della capacità del modello di chiarire attivamente in caso di ambiguità; miglioramento della gestione di problemi di analisi di parole e frasi; aumento della qualità e dell'interattività della creazione; miglioramento dell'esperienza multi-turno."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "Il nuovo modello di punta di linguaggio visivo di Hunyuan, adotta una nuova struttura di modello esperto misto (MoE), con miglioramenti complessivi nelle capacità di riconoscimento di base, creazione di contenuti, domande e risposte, analisi e ragionamento rispetto alla generazione precedente."
+ },
"hunyuan-vision": {
"description": "Ultimo modello multimodale di Hunyuan, supporta l'input di immagini e testo per generare contenuti testuali."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "La nostra ultima serie di modelli, con prestazioni di ragionamento eccezionali, supporta una lunghezza di contesto di 1M e offre una migliore capacità di seguire istruzioni e chiamare strumenti."
},
+ "internlm3-latest": {
+ "description": "La nostra ultima serie di modelli, con prestazioni di inferenza eccezionali, è leader tra i modelli open source della stessa classe. Punta di default ai modelli della serie InternLM3 appena rilasciati."
+ },
+ "jina-deepsearch-v1": {
+ "description": "La ricerca approfondita combina la ricerca online, la lettura e il ragionamento, consentendo indagini complete. Puoi considerarlo come un agente che accetta il tuo compito di ricerca - eseguirà una ricerca approfondita e iterativa prima di fornire una risposta. Questo processo implica una continua ricerca, ragionamento e risoluzione dei problemi da diverse angolazioni. Questo è fondamentalmente diverso dai modelli di grandi dimensioni standard che generano risposte direttamente dai dati pre-addestrati e dai tradizionali sistemi RAG che si basano su ricerche superficiali una tantum."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM è un modello linguistico sperimentale, specifico per compiti, addestrato per rispettare i principi della scienza dell'apprendimento, in grado di seguire istruzioni sistematiche in contesti di insegnamento e apprendimento, fungendo da tutor esperto."
},
"lite": {
"description": "Spark Lite è un modello di linguaggio di grandi dimensioni leggero, con latenza estremamente bassa e capacità di elaborazione efficiente, completamente gratuito e aperto, supporta funzionalità di ricerca online in tempo reale. La sua caratteristica di risposta rapida lo rende eccellente per applicazioni di inferenza su dispositivi a bassa potenza e per il fine-tuning dei modelli, offrendo agli utenti un'ottima efficienza dei costi e un'esperienza intelligente, soprattutto nei contesti di domande e risposte, generazione di contenuti e ricerca."
},
- "llama-3.1-70b-instruct": {
- "description": "Il modello Llama 3.1 70B Instruct, con 70B parametri, offre prestazioni eccezionali in generazione di testi di grandi dimensioni e compiti di istruzione."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B offre capacità di ragionamento AI più potenti, adatto per applicazioni complesse, supporta un'elaborazione computazionale elevata garantendo efficienza e precisione."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B è un modello ad alte prestazioni, offre capacità di generazione di testo rapida, particolarmente adatto per scenari applicativi che richiedono efficienza su larga scala e costi contenuti."
},
- "llama-3.1-8b-instruct": {
- "description": "Il modello Llama 3.1 8B Instruct, con 8B parametri, supporta l'esecuzione efficiente di compiti di istruzione, offrendo capacità di generazione testuale di alta qualità."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Il modello Llama 3.1 Sonar Huge Online, con 405B parametri, supporta una lunghezza di contesto di circa 127.000 token, progettato per applicazioni di chat online complesse."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Il modello Llama 3.1 Sonar Large Chat, con 70B parametri, supporta una lunghezza di contesto di circa 127.000 token, adatto per compiti di chat offline complessi."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Il modello Llama 3.1 Sonar Large Online, con 70B parametri, supporta una lunghezza di contesto di circa 127.000 token, adatto per compiti di chat ad alta capacità e diversificati."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Il modello Llama 3.1 Sonar Small Chat, con 8B parametri, è progettato per chat offline, supportando una lunghezza di contesto di circa 127.000 token."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Il modello Llama 3.1 Sonar Small Online, con 8B parametri, supporta una lunghezza di contesto di circa 127.000 token, progettato per chat online, in grado di gestire interazioni testuali in modo efficiente."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 è progettato per gestire compiti che combinano dati visivi e testuali. Eccelle in compiti come la descrizione delle immagini e le domande visive, colmando il divario tra generazione del linguaggio e ragionamento visivo."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 è il modello di linguaggio open source multilingue più avanzato della serie Llama, che offre prestazioni paragonabili a un modello da 405B a un costo estremamente ridotto. Basato su una struttura Transformer, migliora l'utilità e la sicurezza attraverso il fine-tuning supervisionato (SFT) e l'apprendimento per rinforzo con feedback umano (RLHF). La sua versione ottimizzata per le istruzioni è progettata per dialoghi multilingue e supera molti modelli di chat open source e chiusi in vari benchmark di settore. La data di scadenza delle conoscenze è dicembre 2023."
+ },
"llama-3.3-70b-versatile": {
"description": "Meta Llama 3.3 è un modello linguistico di grandi dimensioni multilingue (LLM) da 70B (input/output testuale) con pre-addestramento e aggiustamento delle istruzioni. Il modello di testo puro di Llama 3.3 è ottimizzato per casi d'uso di dialogo multilingue e supera molti modelli di chat open-source e chiusi nei benchmark di settore comuni."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 è progettato per gestire compiti che combinano dati visivi e testuali. Eccelle in compiti come la descrizione di immagini e le domande visive, colmando il divario tra generazione del linguaggio e ragionamento visivo."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 è il modello di linguaggio open source multilingue più avanzato della serie Llama, che offre prestazioni paragonabili a un modello da 405B a costi molto bassi. Basato su architettura Transformer, migliorato tramite fine-tuning supervisionato (SFT) e apprendimento rinforzato con feedback umano (RLHF) per aumentarne l'utilità e la sicurezza. La sua versione ottimizzata per le istruzioni è progettata per dialoghi multilingue, superando molti modelli di chat open source e chiusi in vari benchmark di settore. Data di scadenza delle conoscenze: dicembre 2023."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Il modello di linguaggio di grandi dimensioni multilingue Meta Llama 3.3 (LLM) è un modello generativo pre-addestrato e regolato per istruzioni da 70B (input/output di testo). Il modello di testo puro di Llama 3.3 regolato per istruzioni è ottimizzato per casi d'uso di dialogo multilingue e supera molti modelli di chat open source e chiusi disponibili su benchmark di settore comuni."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 è progettato per gestire compiti che combinano dati visivi e testuali. Eccelle in compiti come la descrizione di immagini e le domande visive, colmando il divario tra generazione del linguaggio e ragionamento visivo."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "Il modello Llama 3.1 Turbo 405B offre un supporto di contesto di capacità estremamente grande per l'elaborazione di big data, eccellendo nelle applicazioni di intelligenza artificiale su larga scala."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 è il modello leader lanciato da Meta, supporta fino a 405B parametri, applicabile a conversazioni complesse, traduzione multilingue e analisi dei dati."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B offre supporto per dialoghi multilingue ad alta efficienza."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 è progettato per gestire compiti che combinano dati visivi e testuali. Si distingue in compiti come la descrizione delle immagini e il question answering visivo, colmando il divario tra generazione del linguaggio e ragionamento visivo."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 è il modello di linguaggio open source multilingue più avanzato della serie Llama, che offre prestazioni paragonabili a un modello da 405B a un costo estremamente ridotto. Basato su una struttura Transformer, migliora l'utilità e la sicurezza attraverso il fine-tuning supervisionato (SFT) e l'apprendimento per rinforzo con feedback umano (RLHF). La sua versione ottimizzata per le istruzioni è progettata per dialoghi multilingue e supera molti modelli di chat open source e chiusi in vari benchmark di settore. La data di scadenza delle conoscenze è dicembre 2023."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 è il modello di linguaggio open source multilingue più avanzato della serie Llama, che offre prestazioni paragonabili a un modello da 405B a un costo estremamente ridotto. Basato su una struttura Transformer, migliora l'utilità e la sicurezza attraverso il fine-tuning supervisionato (SFT) e l'apprendimento per rinforzo con feedback umano (RLHF). La sua versione ottimizzata per le istruzioni è progettata per dialoghi multilingue e supera molti modelli di chat open source e chiusi in vari benchmark di settore. La data di scadenza delle conoscenze è dicembre 2023."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct è il modello più grande e potente della serie Llama 3.1 Instruct, un modello avanzato per la generazione di dati e il ragionamento conversazionale, utilizzabile anche come base per un pre-addestramento o un fine-tuning specializzato in determinati settori. I modelli di linguaggio di grandi dimensioni (LLMs) multilingue forniti da Llama 3.1 sono un insieme di modelli generativi pre-addestrati e ottimizzati per le istruzioni, che includono dimensioni di 8B, 70B e 405B (input/output di testo). I modelli di testo ottimizzati per le istruzioni di Llama 3.1 (8B, 70B, 405B) sono stati progettati per casi d'uso conversazionali multilingue e hanno superato molti modelli di chat open source disponibili in benchmark di settore comuni. Llama 3.1 è progettato per usi commerciali e di ricerca in diverse lingue. I modelli di testo ottimizzati per le istruzioni sono adatti a chat simili a assistenti, mentre i modelli pre-addestrati possono adattarsi a vari compiti di generazione di linguaggio naturale. I modelli Llama 3.1 supportano anche l'uso della loro output per migliorare altri modelli, inclusa la generazione di dati sintetici e il raffinamento. Llama 3.1 è un modello di linguaggio autoregressivo basato su un'architettura di trasformatore ottimizzata. Le versioni ottimizzate utilizzano il fine-tuning supervisionato (SFT) e l'apprendimento per rinforzo con feedback umano (RLHF) per allinearsi alle preferenze umane in termini di utilità e sicurezza."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 è un modello di linguaggio di grandi dimensioni (LLM) open source progettato per sviluppatori, ricercatori e aziende, per aiutarli a costruire, sperimentare e scalare responsabilmente le loro idee di AI generativa. Come parte di un sistema di base per l'innovazione della comunità globale, è particolarmente adatto per dispositivi a bassa potenza e risorse limitate, oltre a garantire tempi di addestramento più rapidi."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "LLM avanzato, supporta la generazione di dati sintetici, la distillazione della conoscenza e il ragionamento, adatto per chatbot, programmazione e compiti specifici."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "Abilita conversazioni complesse, con eccellenti capacità di comprensione del contesto, ragionamento e generazione di testo."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "Modello all'avanguardia, dotato di comprensione del linguaggio, eccellenti capacità di ragionamento e generazione di testo."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "Modello visivo-linguistico all'avanguardia, specializzato nel ragionamento di alta qualità a partire dalle immagini."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "Modello linguistico all'avanguardia di piccole dimensioni, dotato di comprensione del linguaggio, eccellenti capacità di ragionamento e generazione di testo."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "Modello linguistico all'avanguardia di piccole dimensioni, dotato di comprensione del linguaggio, eccellenti capacità di ragionamento e generazione di testo."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "Modello visivo-linguistico all'avanguardia, specializzato nel ragionamento di alta qualità a partire dalle immagini."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "LLM avanzato, specializzato in ragionamento, matematica, conoscenze generali e chiamate di funzione."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 è un modello linguistico fornito da Microsoft AI, particolarmente efficace in conversazioni complesse, multilingue, ragionamento e assistenti intelligenti."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K è un modello con capacità di elaborazione di contesti ultra lunghi, adatto per generare testi molto lunghi, soddisfacendo le esigenze di compiti complessi, in grado di gestire contenuti fino a 128.000 token, particolarmente adatto per applicazioni di ricerca, accademiche e generazione di documenti di grandi dimensioni."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Il modello visivo Kimi (inclusi moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, ecc.) è in grado di comprendere il contenuto delle immagini, inclusi testo, colori e forme degli oggetti."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K offre capacità di elaborazione di contesti di lunghezza media, in grado di gestire 32.768 token, particolarmente adatto per generare vari documenti lunghi e dialoghi complessi, utilizzato in creazione di contenuti, generazione di report e sistemi di dialogo."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Il modello visivo Kimi (inclusi moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, ecc.) è in grado di comprendere il contenuto delle immagini, inclusi testo, colori e forme degli oggetti."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K è progettato per generare compiti di testo brevi, con prestazioni di elaborazione efficienti, in grado di gestire 8.192 token, particolarmente adatto per dialoghi brevi, appunti e generazione rapida di contenuti."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Il modello visivo Kimi (inclusi moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, ecc.) è in grado di comprendere il contenuto delle immagini, inclusi testo, colori e forme degli oggetti."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B è una versione aggiornata di Nous Hermes 2, contenente i più recenti dataset sviluppati internamente."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B è un modello linguistico di grandi dimensioni personalizzato da NVIDIA, progettato per migliorare il grado di aiuto delle risposte generate da LLM alle domande degli utenti."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B è un modello linguistico di grandi dimensioni personalizzato da NVIDIA, progettato per migliorare l'utilità delle risposte generate dai LLM alle domande degli utenti. Questo modello ha ottenuto risultati eccellenti nei benchmark come Arena Hard, AlpacaEval 2 LC e GPT-4-Turbo MT-Bench, classificandosi al primo posto in tutti e tre i benchmark di allineamento automatico fino al 1 ottobre 2024. Il modello è stato addestrato utilizzando RLHF (in particolare REINFORCE), Llama-3.1-Nemotron-70B-Reward e HelpSteer2-Preference come suggerimenti, basandosi sul modello Llama-3.1-70B-Instruct."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "Modello linguistico unico, offre prestazioni di accuratezza ed efficienza senza pari."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct è un modello linguistico di grandi dimensioni personalizzato da NVIDIA, progettato per migliorare l'utilità delle risposte generate da LLM."
+ },
"o1": {
"description": "Focalizzato su inferenze avanzate e risoluzione di problemi complessi, inclusi compiti matematici e scientifici. È particolarmente adatto per applicazioni che richiedono una comprensione profonda del contesto e flussi di lavoro agenti."
},
- "o1-2024-12-17": {
- "description": "o1 è il nuovo modello di inferenza di OpenAI, che supporta input testuali e visivi e produce output testuali, adatto a compiti complessi che richiedono una vasta conoscenza generale. Questo modello ha un contesto di 200K e una data di scadenza della conoscenza di ottobre 2023."
- },
"o1-mini": {
"description": "o1-mini è un modello di inferenza rapido ed economico progettato per applicazioni di programmazione, matematica e scienza. Questo modello ha un contesto di 128K e una data di cutoff della conoscenza di ottobre 2023."
},
"o1-preview": {
"description": "o1 è il nuovo modello di inferenza di OpenAI, adatto a compiti complessi che richiedono una vasta conoscenza generale. Questo modello ha un contesto di 128K e una data di cutoff della conoscenza di ottobre 2023."
},
+ "o3-mini": {
+ "description": "o3-mini è il nostro ultimo modello di inferenza compatto, che offre un'intelligenza elevata con gli stessi obiettivi di costo e latenza di o1-mini."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba è un modello linguistico Mamba 2 focalizzato sulla generazione di codice, offre un forte supporto per compiti avanzati di codifica e ragionamento."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini è il modello più recente di OpenAI, lanciato dopo GPT-4 Omni, che supporta input visivi e testuali e produce output testuali. Come il loro modello di piccole dimensioni più avanzato, è molto più economico rispetto ad altri modelli all'avanguardia recenti e costa oltre il 60% in meno rispetto a GPT-3.5 Turbo. Mantiene un'intelligenza all'avanguardia, offrendo un notevole rapporto qualità-prezzo. GPT-4o mini ha ottenuto un punteggio dell'82% nel test MMLU e attualmente è classificato più in alto di GPT-4 per preferenze di chat."
},
- "openai/o1": {
- "description": "o1 è il nuovo modello di inferenza di OpenAI, che supporta input testuali e visivi e produce output testuali, adatto a compiti complessi che richiedono una vasta conoscenza generale. Questo modello ha un contesto di 200K e una data di scadenza della conoscenza di ottobre 2023."
- },
"openai/o1-mini": {
"description": "o1-mini è un modello di inferenza rapido ed economico progettato per applicazioni di programmazione, matematica e scienza. Questo modello ha un contesto di 128K e una data di cutoff della conoscenza di ottobre 2023."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K è dotato di una capacità di elaborazione del contesto eccezionale, in grado di gestire fino a 128K di informazioni contestuali, particolarmente adatto per l'analisi completa e la gestione di associazioni logiche a lungo termine in contenuti lunghi, fornendo una logica fluida e coerente e un supporto variegato per le citazioni in comunicazioni testuali complesse."
},
+ "qvq-72b-preview": {
+ "description": "Il modello QVQ è un modello di ricerca sperimentale sviluppato dal team Qwen, focalizzato sul miglioramento delle capacità di ragionamento visivo, in particolare nel campo del ragionamento matematico."
+ },
"qwen-coder-plus-latest": {
"description": "Modello di codice Qwen di Tongyi."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Modello di linguaggio visivo Qwen di grande scala. Rispetto alla versione potenziata, migliora ulteriormente la capacità di ragionamento visivo e di aderenza alle istruzioni, offrendo un livello superiore di percezione visiva e cognizione."
},
+ "qwen-vl-ocr-latest": {
+ "description": "Qwen OCR è un modello specializzato nell'estrazione di testo, focalizzato sulla capacità di estrazione di testo da immagini di documenti, tabelle, domande d'esame, scrittura a mano, ecc. È in grado di riconoscere vari testi, supportando attualmente le seguenti lingue: cinese, inglese, francese, giapponese, coreano, tedesco, russo, italiano, vietnamita, arabo."
+ },
"qwen-vl-plus-latest": {
"description": "Versione potenziata del modello di linguaggio visivo Qwen. Migliora notevolmente la capacità di riconoscimento dei dettagli e di riconoscimento del testo, supportando risoluzioni superiori a un milione di pixel e immagini di qualsiasi rapporto di aspetto."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 è una nuova serie di modelli di linguaggio di grandi dimensioni, con capacità di comprensione e generazione più forti."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "LLM orientato al cinese e all'inglese, focalizzato su linguaggio, programmazione, matematica, ragionamento e altro."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "LLM avanzato, supporta la generazione di codice, il ragionamento e la correzione, coprendo i linguaggi di programmazione più diffusi."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "Potente modello di codice di medie dimensioni, supporta una lunghezza di contesto di 32K, specializzato in programmazione multilingue."
+ },
"qwen2": {
"description": "Qwen2 è la nuova generazione di modelli di linguaggio su larga scala di Alibaba, supporta prestazioni eccellenti per esigenze applicative diversificate."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "Modello da 14B di Tongyi Qwen 2.5, open source."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "Il modello da 72B di Qwen2.5 è open source."
+ },
"qwen2.5-32b-instruct": {
"description": "Modello da 32B di Tongyi Qwen 2.5, open source."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Il modello Qwen-Math ha potenti capacità di risoluzione di problemi matematici."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "Miglioramento complessivo nella seguire istruzioni, matematica, risoluzione di problemi e codice, con capacità di riconoscimento universale migliorate, supporto per formati diversi per il posizionamento preciso degli elementi visivi, comprensione di file video lunghi (fino a 10 minuti) e localizzazione di eventi in tempo reale, capacità di comprendere sequenze temporali e velocità, supporto per il controllo di agenti OS o Mobile basato su capacità di analisi e localizzazione, forte capacità di estrazione di informazioni chiave e output in formato Json, questa versione è la 72B, la versione più potente della serie."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "Miglioramento complessivo nella seguire istruzioni, matematica, risoluzione di problemi e codice, con capacità di riconoscimento universale migliorate, supporto per formati diversi per il posizionamento preciso degli elementi visivi, comprensione di file video lunghi (fino a 10 minuti) e localizzazione di eventi in tempo reale, capacità di comprendere sequenze temporali e velocità, supporto per il controllo di agenti OS o Mobile basato su capacità di analisi e localizzazione, forte capacità di estrazione di informazioni chiave e output in formato Json, questa versione è la 72B, la versione più potente della serie."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 è la nuova generazione di modelli linguistici su larga scala di Alibaba, che supporta esigenze applicative diversificate con prestazioni eccellenti."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro è un LLM altamente intelligente lanciato da Upstage, focalizzato sulla capacità di seguire istruzioni su singolo GPU, con un punteggio IFEval superiore a 80. Attualmente supporta l'inglese, con una versione ufficiale prevista per novembre 2024, che espanderà il supporto linguistico e la lunghezza del contesto."
},
+ "sonar": {
+ "description": "Prodotto di ricerca leggero basato sul contesto di ricerca, più veloce e più economico rispetto a Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "Prodotto di ricerca avanzata che supporta il contesto di ricerca, query avanzate e follow-up."
+ },
+ "sonar-reasoning": {
+ "description": "Nuovo prodotto API supportato dal modello di ragionamento DeepSeek."
+ },
"step-1-128k": {
"description": "Equilibrio tra prestazioni e costi, adatto per scenari generali."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "Questo modello possiede potenti capacità di comprensione video."
},
+ "step-1o-vision-32k": {
+ "description": "Questo modello possiede una potente capacità di comprensione delle immagini. Rispetto ai modelli della serie step-1v, offre prestazioni visive superiori."
+ },
"step-1v-32k": {
"description": "Supporta input visivi, migliorando l'esperienza di interazione multimodale."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Supporta interazioni di contesto su larga scala, adatto per scenari di dialogo complessi."
},
+ "step-2-mini": {
+ "description": "Un modello di grandi dimensioni ad alta velocità basato sulla nuova architettura di attenzione auto-sviluppata MFA, in grado di raggiungere risultati simili a quelli di step1 a un costo molto basso, mantenendo al contempo una maggiore capacità di elaborazione e tempi di risposta più rapidi. È in grado di gestire compiti generali, con competenze particolari nella programmazione."
+ },
+ "taichu2_mm": {
+ "description": "Integra capacità di comprensione delle immagini, trasferimento di conoscenze, attribuzione logica, ecc., e si distingue nel campo delle domande e risposte basate su testo e immagini."
+ },
"taichu_llm": {
"description": "Il modello linguistico Taichu di Zīdōng ha una straordinaria capacità di comprensione del linguaggio e abilità in creazione di testi, domande di conoscenza, programmazione, calcoli matematici, ragionamento logico, analisi del sentimento e sintesi di testi. Combina in modo innovativo il pre-addestramento su grandi dati con una ricca conoscenza multi-sorgente, affinando continuamente la tecnologia degli algoritmi e assorbendo costantemente nuove conoscenze da dati testuali massivi, migliorando continuamente le prestazioni del modello. Fornisce agli utenti informazioni e servizi più convenienti e un'esperienza più intelligente."
},
+ "text-embedding-3-large": {
+ "description": "Il modello di vettorizzazione più potente, adatto per compiti in inglese e non inglese."
+ },
+ "text-embedding-3-small": {
+ "description": "Modello di Embedding di nuova generazione, efficiente ed economico, adatto per la ricerca di conoscenza, applicazioni RAG e altri scenari."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) offre capacità di calcolo potenziate attraverso strategie e architetture di modelli efficienti."
},
+ "tts-1": {
+ "description": "L'ultimo modello di sintesi vocale, ottimizzato per la velocità in scenari in tempo reale."
+ },
+ "tts-1-hd": {
+ "description": "L'ultimo modello di sintesi vocale, ottimizzato per la qualità."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) è adatto per compiti di istruzione dettagliati, offrendo eccellenti capacità di elaborazione linguistica."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet ha elevato gli standard del settore, superando le prestazioni dei modelli concorrenti e di Claude 3 Opus, dimostrando eccellenza in una vasta gamma di valutazioni, mantenendo al contempo la velocità e i costi dei nostri modelli di livello medio."
},
+ "whisper-1": {
+ "description": "Modello di riconoscimento vocale universale, supporta il riconoscimento vocale multilingue, la traduzione vocale e il riconoscimento linguistico."
+ },
"wizardlm2": {
"description": "WizardLM 2 è un modello di linguaggio fornito da Microsoft AI, particolarmente efficace in dialoghi complessi, multilingue, ragionamento e assistenti intelligenti."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Modello per compiti visivi complessi, offre elevate prestazioni nella comprensione e analisi delle immagini."
+ },
+ "yi-vision-v2": {
+ "description": "Modello per compiti visivi complessi, che offre capacità di comprensione e analisi ad alte prestazioni basate su più immagini."
}
}
diff --git a/locales/it-IT/providers.json b/locales/it-IT/providers.json
index 89f96835eed51..60ccf19b062ef 100644
--- a/locales/it-IT/providers.json
+++ b/locales/it-IT/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure offre una varietà di modelli AI avanzati, tra cui GPT-3.5 e l'ultima serie GPT-4, supportando diversi tipi di dati e compiti complessi, con un impegno per soluzioni AI sicure, affidabili e sostenibili."
},
+ "azureai": {
+ "description": "Azure offre una varietà di modelli AI avanzati, tra cui GPT-3.5 e l'ultima serie GPT-4, supportando diversi tipi di dati e compiti complessi, impegnandosi per soluzioni AI sicure, affidabili e sostenibili."
+ },
"baichuan": {
"description": "Baichuan Intelligence è un'azienda focalizzata sulla ricerca e sviluppo di modelli di intelligenza artificiale di grandi dimensioni, i cui modelli eccellono in compiti in cinese come enciclopedie di conoscenza, elaborazione di testi lunghi e creazione di contenuti, superando i modelli mainstream esteri. Baichuan Intelligence ha anche capacità multimodali leader nel settore, mostrando prestazioni eccezionali in diverse valutazioni autorevoli. I suoi modelli includono Baichuan 4, Baichuan 3 Turbo e Baichuan 3 Turbo 128k, ottimizzati per diversi scenari applicativi, offrendo soluzioni ad alto rapporto qualità-prezzo."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek è un'azienda focalizzata sulla ricerca e applicazione della tecnologia AI, il cui ultimo modello DeepSeek-V2.5 combina capacità di dialogo generico e elaborazione del codice, realizzando miglioramenti significativi nell'allineamento delle preferenze umane, nei compiti di scrittura e nel rispetto delle istruzioni."
},
+ "doubao": {
+ "description": "Il grande modello sviluppato internamente da ByteDance. Validato attraverso oltre 50 scenari aziendali interni, con un utilizzo quotidiano di trilioni di token che affinano continuamente il modello, offre diverse capacità multimodali, creando esperienze aziendali ricche con risultati di alta qualità."
+ },
"fireworksai": {
"description": "Fireworks AI è un fornitore leader di servizi di modelli linguistici avanzati, focalizzato su chiamate funzionali e elaborazione multimodale. Il suo ultimo modello Firefunction V2, basato su Llama-3, è ottimizzato per chiamate di funzione, dialogo e rispetto delle istruzioni. Il modello di linguaggio visivo FireLLaVA-13B supporta input misti di immagini e testo. Altri modelli notevoli includono la serie Llama e la serie Mixtral, offrendo supporto efficiente per il rispetto e la generazione di istruzioni multilingue."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "Un'organizzazione open source dedicata alla ricerca e allo sviluppo di strumenti per modelli di grandi dimensioni. Fornisce a tutti gli sviluppatori di AI una piattaforma open source efficiente e facile da usare, rendendo le tecnologie e gli algoritmi all'avanguardia accessibili a tutti."
},
+ "jina": {
+ "description": "Jina AI, fondata nel 2020, è una delle principali aziende di ricerca AI. La nostra piattaforma di base per la ricerca include modelli vettoriali, riordinatori e piccoli modelli linguistici, per aiutare le aziende a costruire applicazioni di ricerca generativa e multimodale affidabili e di alta qualità."
+ },
+ "lmstudio": {
+ "description": "LM Studio è un'applicazione desktop per sviluppare e sperimentare LLM sul tuo computer."
+ },
"minimax": {
"description": "MiniMax è un'azienda di tecnologia dell'intelligenza artificiale generale fondata nel 2021, dedicata alla co-creazione di intelligenza con gli utenti. MiniMax ha sviluppato modelli generali di diverse modalità, tra cui un modello di testo MoE con trilioni di parametri, un modello vocale e un modello visivo. Ha anche lanciato applicazioni come Conch AI."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI è una piattaforma che offre API per vari modelli di linguaggio di grandi dimensioni e generazione di immagini AI, flessibile, affidabile e conveniente. Supporta i più recenti modelli open source come Llama3 e Mistral, fornendo soluzioni API complete, user-friendly e scalabili per lo sviluppo di applicazioni AI, adatte alla rapida crescita delle startup AI."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ fornisce contenitori per l'inferenza di microservizi accelerati da GPU self-hosted, supportando il deployment di modelli AI pre-addestrati e personalizzati su cloud, data center, PC RTX™ AI e workstation."
+ },
"ollama": {
"description": "I modelli forniti da Ollama coprono ampiamente aree come generazione di codice, operazioni matematiche, elaborazione multilingue e interazioni conversazionali, supportando esigenze diversificate per implementazioni aziendali e localizzate."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "L'Istituto di Automazione dell'Accademia Cinese delle Scienze e l'Istituto di Ricerca sull'Intelligenza Artificiale di Wuhan hanno lanciato una nuova generazione di modelli di grandi dimensioni multimodali, supportando domande e risposte a più turni, creazione di testi, generazione di immagini, comprensione 3D, analisi dei segnali e altre attività di domanda e risposta complete, con capacità cognitive, di comprensione e di creazione più forti, offrendo un'esperienza interattiva completamente nuova."
},
+ "tencentcloud": {
+ "description": "La potenza atomica del motore di conoscenza (LLM Knowledge Engine Atomic Power) è una capacità completa di domande e risposte sviluppata sulla base del motore di conoscenza, rivolta a imprese e sviluppatori, che offre la possibilità di costruire e sviluppare applicazioni modello in modo flessibile. Puoi assemblare il tuo servizio modello esclusivo utilizzando diverse capacità atomiche, richiamando servizi di analisi documentale, suddivisione, embedding, riscrittura multipla e altro, per personalizzare il tuo business AI esclusivo."
+ },
"togetherai": {
"description": "Together AI si impegna a raggiungere prestazioni leader attraverso modelli AI innovativi, offrendo ampie capacità di personalizzazione, inclusi supporto per scalabilità rapida e processi di distribuzione intuitivi, per soddisfare le varie esigenze aziendali."
},
"upstage": {
"description": "Upstage si concentra sullo sviluppo di modelli AI per varie esigenze commerciali, inclusi Solar LLM e document AI, con l'obiettivo di realizzare un'intelligenza artificiale generale artificiale (AGI) per il lavoro. Crea semplici agenti di dialogo tramite Chat API e supporta chiamate funzionali, traduzioni, embedding e applicazioni specifiche del settore."
},
+ "vllm": {
+ "description": "vLLM è una libreria veloce e facile da usare per l'inferenza e i servizi LLM."
+ },
+ "volcengine": {
+ "description": "La piattaforma di sviluppo dei servizi di modelli di grandi dimensioni lanciata da ByteDance, offre servizi di invocazione di modelli ricchi di funzionalità, sicuri e competitivi in termini di prezzo, fornendo anche dati sui modelli, messa a punto, inferenza, valutazione e altre funzionalità end-to-end, garantendo in modo completo lo sviluppo e l'implementazione delle vostre applicazioni AI."
+ },
"wenxin": {
"description": "Piattaforma di sviluppo e servizi per modelli di grandi dimensioni e applicazioni AI native, a livello aziendale, che offre la catena di strumenti completa e facile da usare per lo sviluppo di modelli di intelligenza artificiale generativa e per l'intero processo di sviluppo delle applicazioni."
},
diff --git a/locales/it-IT/setting.json b/locales/it-IT/setting.json
index 97ae4ea80332c..89163725a522e 100644
--- a/locales/it-IT/setting.json
+++ b/locales/it-IT/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Abilita limite di risposta singola"
},
+ "enableReasoningEffort": {
+ "title": "Attiva la regolazione dell'intensità del ragionamento"
+ },
"frequencyPenalty": {
- "desc": "Più alto è il valore, più probabile è la riduzione delle parole ripetute",
- "title": "Penalità di frequenza"
+ "desc": "Maggiore è il valore, più ricca e varia sarà la scelta delle parole; minore è il valore, più semplici e dirette saranno le parole",
+ "title": "Ricchezza del vocabolario"
},
"maxTokens": {
"desc": "Numero massimo di token utilizzati per interazione singola",
@@ -212,19 +215,31 @@
"desc": "Modello {{provider}}",
"title": "Modello"
},
+ "params": {
+ "title": "Parametri avanzati"
+ },
"presencePenalty": {
- "desc": "Più alto è il valore, più probabile è l'estensione a nuovi argomenti",
- "title": "Freschezza dell'argomento"
+ "desc": "Maggiore è il valore, maggiore sarà la tendenza a esprimere in modi diversi, evitando ripetizioni; minore è il valore, maggiore sarà la tendenza a utilizzare concetti o narrazioni ripetute, rendendo l'espressione più coerente",
+ "title": "Divergenza espressiva"
+ },
+ "reasoningEffort": {
+ "desc": "Maggiore è il valore, più forte è la capacità di ragionamento, ma potrebbe aumentare il tempo di risposta e il consumo di Token",
+ "options": {
+ "high": "Alto",
+ "low": "Basso",
+ "medium": "Medio"
+ },
+ "title": "Intensità del ragionamento"
},
"temperature": {
- "desc": "Più alto è il valore, più casuale è la risposta",
- "title": "Casualità",
- "titleWithValue": "Casualità {{value}}"
+ "desc": "Maggiore è il valore, più creativi e fantasiosi saranno le risposte; minore è il valore, più rigorose saranno le risposte",
+ "title": "Attività Creativa",
+ "warning": "Un valore troppo alto per l'attività creativa potrebbe generare output illeggibili"
},
- "title": "Impostazioni del modello",
+ "title": "Impostazioni del Modello",
"topP": {
- "desc": "Simile alla casualità, ma non modificare insieme alla casualità",
- "title": "Campionamento principale"
+ "desc": "Considera quante possibilità, maggiore è il valore, più risposte potenziali vengono accettate; minore è il valore, più si tende a scegliere la risposta più probabile. Non si consiglia di modificarlo insieme all'attività creativa",
+ "title": "Apertura Mentale"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Impostazioni comuni",
"experiment": "实验",
"llm": "Modello linguistico",
+ "provider": "Fornitore di servizi AI",
"sync": "云端同步",
"system-agent": "Assistente di sistema",
"tts": "Servizio vocale"
diff --git a/locales/ja-JP/changelog.json b/locales/ja-JP/changelog.json
index 386c567b042d8..f8106c072c9e8 100644
--- a/locales/ja-JP/changelog.json
+++ b/locales/ja-JP/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "すべての更新ログを見る",
"description": "{{appName}}の新機能と改善を継続的に追跡",
"pagination": {
- "older": "履歴の変更を表示",
- "prev": "前のページ"
+ "next": "次のページ",
+ "older": "履歴の変更を表示"
},
"readDetails": "詳細を読む",
"title": "更新ログ",
diff --git a/locales/ja-JP/common.json b/locales/ja-JP/common.json
index f324aa5140b5f..0beea193d9871 100644
--- a/locales/ja-JP/common.json
+++ b/locales/ja-JP/common.json
@@ -275,6 +275,7 @@
},
"temp": "一時的",
"terms": "利用規約",
+ "update": "更新",
"updateAgent": "エージェント情報を更新",
"upgradeVersion": {
"action": "アップグレード",
@@ -285,6 +286,7 @@
"anonymousNickName": "匿名ユーザー",
"billing": "請求管理",
"cloud": "{{name}} を体験",
+ "community": "コミュニティ版",
"data": "データストレージ",
"defaultNickname": "コミュニティユーザー",
"discord": "コミュニティサポート",
@@ -294,7 +296,6 @@
"help": "ヘルプセンター",
"moveGuide": "設定ボタンがこちらに移動しました",
"plans": "サブスクリプションプラン",
- "preview": "プレビュー",
"profile": "アカウント管理",
"setting": "アプリ設定",
"usages": "利用量統計"
diff --git a/locales/ja-JP/components.json b/locales/ja-JP/components.json
index 87d1dbe7b76ba..8e6787012af06 100644
--- a/locales/ja-JP/components.json
+++ b/locales/ja-JP/components.json
@@ -76,6 +76,7 @@
"custom": "カスタムモデル、デフォルトでは関数呼び出しとビジョン認識の両方をサポートしています。上記機能の有効性を確認してください。",
"file": "このモデルはファイルのアップロードと認識をサポートしています。",
"functionCall": "このモデルは関数呼び出し（Function Call）をサポートしています。",
+ "reasoning": "このモデルは深い思考をサポートしています",
"tokens": "このモデルは1つのセッションあたり最大{{tokens}}トークンをサポートしています。",
"vision": "このモデルはビジョン認識をサポートしています。"
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "有効なモデルがありません。設定に移動して有効にしてください。",
"provider": "プロバイダー"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "ブラウザのセキュリティ制限により、Ollamaを正常に使用するにはクロスオリジン設定が必要です。",
+ "linux": {
+ "env": "[Service] セクションに `Environment` を追加し、OLLAMA_ORIGINS 環境変数を設定します：",
+ "reboot": "systemdをリロードし、Ollamaを再起動します",
+ "systemd": "systemdを呼び出してollamaサービスを編集します："
+ },
+ "macos": "「ターミナル」アプリを開き、以下のコマンドを貼り付けてEnterを押して実行します",
+ "reboot": "実行が完了したらOllamaサービスを再起動してください",
+ "title": "Ollamaのクロスオリジンアクセスを許可する設定",
+ "windows": "Windowsでは、「コントロールパネル」をクリックし、システム環境変数を編集します。ユーザーアカウント用に「OLLAMA_ORIGINS」という名前の環境変数を新規作成し、値を * に設定して「OK/適用」をクリックして保存します"
+ },
+ "install": {
+ "description": "Ollamaが起動していることを確認してください。まだOllamaをダウンロードしていない場合は、公式サイト<1>からダウンロードしてください。",
+ "docker": "Dockerを使用することを好む場合、Ollamaは公式のDockerイメージも提供しています。以下のコマンドでプルできます：",
+ "linux": {
+ "command": "以下のコマンドでインストールします：",
+ "manual": "または、<1>Linux手動インストールガイドを参照して自分でインストールすることもできます。"
+ },
+ "title": "ローカルにOllamaアプリをインストールして起動する",
+ "windowsTab": "Windows（プレビュー版)"
+ }
+ },
+ "Thinking": {
+ "thinking": "深く考えています...",
+ "thought": "深く考えました（所要時間 {{duration}} 秒）",
+ "thoughtWithDuration": "深く考えました"
}
}
diff --git a/locales/ja-JP/discover.json b/locales/ja-JP/discover.json
index c7bb66f3a6c84..291122a9bb483 100644
--- a/locales/ja-JP/discover.json
+++ b/locales/ja-JP/discover.json
@@ -126,6 +126,10 @@
"title": "トピックの新鮮さ"
},
"range": "範囲",
+ "reasoning_effort": {
+ "desc": "この設定は、モデルが回答を生成する前の推論の強度を制御するために使用されます。低強度は応答速度を優先し、トークンを節約しますが、高強度はより完全な推論を提供しますが、より多くのトークンを消費し、応答速度が低下します。デフォルト値は中で、推論の正確性と応答速度のバランスを取ります。",
+ "title": "推論強度"
+ },
"temperature": {
"desc": "この設定は、モデルの応答の多様性に影響を与えます。低い値はより予測可能で典型的な応答をもたらし、高い値はより多様で珍しい応答を奨励します。値が0に設定されると、モデルは与えられた入力に対して常に同じ応答を返します。",
"title": "ランダム性"
diff --git a/locales/ja-JP/modelProvider.json b/locales/ja-JP/modelProvider.json
index dfc773218fe7f..3034659fa95be 100644
--- a/locales/ja-JP/modelProvider.json
+++ b/locales/ja-JP/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API Key"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "AzureのAPIバージョン。YYYY-MM-DD形式に従い、[最新バージョン](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)を参照してください。",
+ "fetch": "リストを取得",
+ "title": "Azure APIバージョン"
+ },
+ "endpoint": {
+ "desc": "Azure AIプロジェクトの概要からAzure AIモデル推論エンドポイントを見つけます。",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Azure AIエンドポイント"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Azure AIプロジェクトの概要からAPIキーを見つけます。",
+ "placeholder": "Azureキー",
+ "title": "キー"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "AWS Access Key Id を入力してください",
@@ -63,6 +81,46 @@
"title": "Cloudflare アカウント ID / API アドレス"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "あなたの API キーを入力してください",
+ "title": "API キー"
+ },
+ "basicTitle": "基本情報",
+ "configTitle": "設定情報",
+ "confirm": "新規作成",
+ "createSuccess": "新規作成に成功しました",
+ "description": {
+ "placeholder": "サービスプロバイダーの紹介（任意）",
+ "title": "サービスプロバイダーの紹介"
+ },
+ "id": {
+ "desc": "サービスプロバイダーの一意の識別子であり、作成後は変更できません",
+ "format": "数字、小文字のアルファベット、ハイフン（-）、およびアンダースコア（_）のみを含むことができます",
+ "placeholder": "小文字で入力してください（例: openai）。作成後は変更できません",
+ "required": "サービスプロバイダー ID を入力してください",
+ "title": "サービスプロバイダー ID"
+ },
+ "logo": {
+ "required": "正しいサービスプロバイダーのロゴをアップロードしてください",
+ "title": "サービスプロバイダーのロゴ"
+ },
+ "name": {
+ "placeholder": "サービスプロバイダーの表示名を入力してください",
+ "required": "サービスプロバイダー名を入力してください",
+ "title": "サービスプロバイダー名"
+ },
+ "proxyUrl": {
+ "required": "プロキシURLを入力してください",
+ "title": "プロキシアドレス"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "SDK タイプを選択してください",
+ "title": "リクエスト形式"
+ },
+ "title": "カスタム AI サービスプロバイダーの作成"
+ },
"github": {
"personalAccessToken": {
"desc": "あなたのGithub PATを入力してください。[こちら](https://github.com/settings/tokens)をクリックして作成します",
@@ -77,6 +135,23 @@
"title": "HuggingFace トークン"
}
},
+ "list": {
+ "title": {
+ "disabled": "サービスプロバイダーは無効です",
+ "enabled": "サービスプロバイダーは有効です"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "カスタムサービスプロバイダーを追加",
+ "all": "すべて",
+ "list": {
+ "disabled": "未使用",
+ "enabled": "使用中"
+ },
+ "notFound": "検索結果が見つかりません",
+ "searchProviders": "サービスプロバイダーを検索...",
+ "sort": "カスタムソート"
+ },
"ollama": {
"checker": {
"desc": "プロキシアドレスが正しく入力されているかをテストします",
@@ -94,33 +169,9 @@
"title": "モデル{{model}}をダウンロード中"
},
"endpoint": {
- "desc": "Ollamaプロキシインターフェースアドレスを入力してください。ローカルで追加の指定がない場合は空白のままにしてください",
+ "desc": "http(s)://を含める必要があります。ローカルで特に指定がない場合は空白のままで構いません",
"title": "プロキシインターフェースアドレス"
},
- "setup": {
- "cors": {
- "description": "ブラウザのセキュリティ制限により、Ollama を正常に使用するにはクロスオリジンリクエストを許可する必要があります。",
- "linux": {
- "env": "[Service] セクションに `Environment` を追加し、OLLAMA_ORIGINS 環境変数を設定してください：",
- "reboot": "systemd を再読み込みして Ollama を再起動します。",
- "systemd": "systemd を呼び出して ollama サービスを編集します："
- },
- "macos": "「ターミナル」アプリを開き、以下のコマンドを貼り付けて実行し、Enter キーを押してください",
- "reboot": "Ollama サービスを再起動するには、実行後に再起動してください",
- "title": "Ollama の CORS アクセスを許可する設定",
- "windows": "Windows 上では、「コントロールパネル」をクリックしてシステム環境変数を編集します。ユーザーアカウントに「OLLAMA_ORIGINS」という名前の環境変数を作成し、値を * に設定し、「OK/適用」をクリックして保存します"
- },
- "install": {
- "description": "Ollamaを有効にしていることを確認してください。Ollamaをまだダウンロードしていない場合は、公式サイト<1>からダウンロードしてください。",
- "docker": "もしDockerを使用することを好む場合、Ollamaは公式Dockerイメージも提供しています。以下のコマンドを使用して取得できます：",
- "linux": {
- "command": "以下のコマンドを使用してインストール：",
- "manual": "または、<1>Linuxマニュアルインストールガイドを参照して手動でインストールすることもできます"
- },
- "title": "ローカルでOllamaアプリをインストールして起動する",
- "windowsTab": "Windows（プレビュー版）"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "ダウンロードをキャンセル",
@@ -131,25 +182,145 @@
"title": "指定されたOllamaモデルをダウンロード"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "百度千帆プラットフォームのAccess Keyを入力してください",
- "placeholder": "Qianfan Access Key",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "あなたのキーとプロキシアドレスなどは <1>AES-GCM 暗号化アルゴリズムを使用して暗号化されます",
+ "apiKey": {
+ "desc": "あなたの {{name}} API キーを入力してください",
+ "placeholder": "{{name}} API キー",
+ "title": "API キー"
+ },
+ "baseURL": {
+ "desc": "http(s):// を含める必要があります",
+ "invalid": "有効なURLを入力してください",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "API プロキシアドレス"
+ },
+ "checker": {
+ "button": "チェック",
+ "desc": "API キーとプロキシアドレスが正しく入力されているかテストします",
+ "pass": "チェックに合格しました",
+ "title": "接続性チェック"
+ },
+ "fetchOnClient": {
+ "desc": "クライアントリクエストモードはブラウザから直接セッションリクエストを発起し、応答速度を向上させます",
+ "title": "クライアントリクエストモードを使用"
+ },
+ "helpDoc": "設定ガイド",
+ "waitingForMore": "さらに多くのモデルが <1>接続予定 です。お楽しみに"
},
- "checker": {
- "desc": "AccessKey / SecretAccessが正しく入力されているかテストします"
+ "createNew": {
+ "title": "カスタム AI モデルの作成"
+ },
+ "item": {
+ "config": "モデルを設定",
+ "customModelCards": {
+ "addNew": "{{id}} モデルを作成して追加",
+ "confirmDelete": "このカスタムモデルを削除しようとしています。削除後は復元できませんので、慎重に操作してください。"
+ },
+ "delete": {
+ "confirm": "モデル {{displayName}} を削除してもよろしいですか？",
+ "success": "削除に成功しました",
+ "title": "モデルを削除"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Azure OpenAI で実際にリクエストされるフィールド",
+ "placeholder": "Azure でのモデルデプロイ名を入力してください",
+ "title": "モデルデプロイ名"
+ },
+ "deployName": {
+ "extra": "リクエストを送信する際に、このフィールドがモデルIDとして使用されます。",
+ "placeholder": "モデルの実際のデプロイ名またはIDを入力してください。",
+ "title": "モデルデプロイ名"
+ },
+ "displayName": {
+ "placeholder": "モデルの表示名を入力してください（例: ChatGPT、GPT-4 など）",
+ "title": "モデル表示名"
+ },
+ "files": {
+ "extra": "現在のファイルアップロード実装は一つのハック手法に過ぎず、自己責任での試行に限られます。完全なファイルアップロード機能は今後の実装をお待ちください",
+ "title": "ファイルアップロードをサポート"
+ },
+ "functionCall": {
+ "extra": "この設定は、モデルがツールを使用する機能を有効にし、モデルにツールタイプのプラグインを追加できるようにします。ただし、実際にツールを使用できるかどうかはモデル自体に依存するため、使用可能性を自分でテストしてください",
+ "title": "ツール使用のサポート"
+ },
+ "id": {
+ "extra": "作成後は変更できません。AIを呼び出す際にモデルIDとして使用されます。",
+ "placeholder": "モデルIDを入力してください。例：gpt-4o または claude-3.5-sonnet",
+ "title": "モデル ID"
+ },
+ "modalTitle": "カスタムモデル設定",
+ "reasoning": {
+ "extra": "この設定は、モデルの深い思考能力を有効にするだけです。具体的な効果はモデル自体に依存しますので、このモデルが利用可能な深い思考能力を持っているかどうかはご自身でテストしてください。",
+ "title": "深い思考をサポート"
+ },
+ "tokens": {
+ "extra": "モデルがサポートする最大トークン数を設定する",
+ "title": "最大コンテキストウィンドウ",
+ "unlimited": "無制限"
+ },
+ "vision": {
+ "extra": "この設定はアプリ内の画像アップロード設定のみを有効にします。認識のサポートはモデル自体に依存しますので、そのモデルの視覚認識機能の可用性を自分でテストしてください",
+ "title": "視覚認識をサポート"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/画像",
+ "inputCharts": "${{amount}}/M 文字",
+ "inputMinutes": "${{amount}}/分",
+ "inputTokens": "入力 ${{amount}}/M",
+ "outputTokens": "出力 ${{amount}}/M"
+ },
+ "releasedAt": "リリース日: {{releasedAt}}"
},
- "secretKey": {
- "desc": "百度千帆プラットフォームのSecret Keyを入力してください",
- "placeholder": "Qianfan Secret Key",
- "title": "Secret Key"
+ "list": {
+ "addNew": "モデルを追加",
+ "disabled": "無効",
+ "disabledActions": {
+ "showMore": "すべて表示"
+ },
+ "empty": {
+ "desc": "カスタムモデルを作成するか、モデルを取得してから使用を開始してください",
+ "title": "利用可能なモデルはありません"
+ },
+ "enabled": "有効",
+ "enabledActions": {
+ "disableAll": "すべて無効にする",
+ "enableAll": "すべて有効にする",
+ "sort": "カスタムモデルの並べ替え"
+ },
+ "enabledEmpty": "有効なモデルはありません。下のリストからお気に入りのモデルを有効にしてください〜",
+ "fetcher": {
+ "clear": "取得したモデルをクリア",
+ "fetch": "モデルリストを取得",
+ "fetching": "モデルリストを取得中...",
+ "latestTime": "最終更新日時：{{time}}",
+ "noLatestTime": "まだリストを取得していません"
+ },
+ "resetAll": {
+ "conform": "現在のモデルのすべての変更をリセットしてもよろしいですか？リセット後、現在のモデルリストはデフォルトの状態に戻ります",
+ "success": "リセットに成功しました",
+ "title": "すべての変更をリセット"
+ },
+ "search": "モデルを検索...",
+ "searchResult": "{{count}} 個のモデルが見つかりました",
+ "title": "モデルリスト",
+ "total": "利用可能なモデルは合計 {{count}} 件です"
},
- "unlock": {
- "customRegion": "カスタムサービス地域",
- "description": "AccessKey / SecretKeyを入力することでセッションを開始できます。アプリはあなたの認証設定を記録しません",
- "title": "カスタム文心一言認証情報を使用"
- }
+ "searchNotFound": "検索結果が見つかりませんでした"
+ },
+ "sortModal": {
+ "success": "ソートが更新されました",
+ "title": "カスタムソート",
+ "update": "更新"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "この AI サービスプロバイダーを削除しようとしています。削除後は復元できません。削除してもよろしいですか？",
+ "deleteSuccess": "削除に成功しました",
+ "tooltip": "サービスプロバイダーの基本設定を更新",
+ "updateSuccess": "更新に成功しました"
},
"zeroone": {
"title": "01.AI 零一万物"
diff --git a/locales/ja-JP/models.json b/locales/ja-JP/models.json
index 93390a59b51b8..4e33b09c5bd8e 100644
--- a/locales/ja-JP/models.json
+++ b/locales/ja-JP/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8Kは意味の安全性と責任指向を強調し、コンテンツの安全性に高い要求を持つアプリケーションシーンのために設計されており、ユーザー体験の正確性と堅牢性を確保します。"
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1は、ツリーサーチを使用して思考の連鎖を構築し、反省メカニズムを導入し、強化学習で訓練されたモデルであり、自己反省と誤り訂正の能力を備えています。"
+ },
"360gpt2-pro": {
"description": "360GPT2 Proは360社が発表した高級自然言語処理モデルで、卓越したテキスト生成と理解能力を備え、特に生成と創作の分野で優れたパフォーマンスを発揮し、複雑な言語変換や役割演技タスクを処理できます。"
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1は、木探索を使用して思考の連鎖を構築し、反省メカニズムを導入し、強化学習で訓練され、自己反省と誤り訂正の能力を備えています。"
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultraは星火大モデルシリーズの中で最も強力なバージョンで、ネットワーク検索のリンクをアップグレードし、テキストコンテンツの理解と要約能力を向上させています。これは、オフィスの生産性を向上させ、要求に正確に応えるための全方位のソリューションであり、業界をリードするインテリジェントな製品です。"
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "モデル能力は国内で第一であり、知識百科、長文、生成創作などの中国語タスクで海外の主流モデルを超えています。また、業界をリードするマルチモーダル能力を持ち、多くの権威ある評価基準で優れたパフォーマンスを示しています。"
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Qwen2.5-Math-1.5Bに基づくDeepSeek-R1蒸留モデルで、強化学習とコールドスタートデータを通じて推論性能を最適化し、オープンソースモデルがマルチタスクの基準を刷新しました。"
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Qwen2.5-14Bに基づくDeepSeek-R1蒸留モデルで、強化学習とコールドスタートデータを通じて推論性能を最適化し、オープンソースモデルがマルチタスクの基準を刷新しました。"
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1シリーズは、強化学習とコールドスタートデータを通じて推論性能を最適化し、オープンソースモデルがマルチタスクの基準を刷新し、OpenAI-o1-miniのレベルを超えました。"
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "Qwen2.5-Math-7Bに基づくDeepSeek-R1蒸留モデルで、強化学習とコールドスタートデータを通じて推論性能を最適化し、オープンソースモデルがマルチタスクの基準を刷新しました。"
+ },
"Doubao-lite-128k": {
"description": "Doubao-liteは、極めて高速な応答速度と優れたコストパフォーマンスを備え、顧客のさまざまなシーンに柔軟な選択肢を提供します。128kコンテキストウィンドウの推論と微調整をサポートしています。"
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "百度が独自に開発したフラッグシップの超大規模言語モデルで、ERNIE 3.5に比べてモデル能力が全面的にアップグレードされ、さまざまな分野の複雑なタスクシナリオに広く適用されます。百度検索プラグインとの自動接続をサポートし、質問応答情報のタイムリーさを保証します。"
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "百度が独自に開発したフラッグシップの超大規模言語モデルで、総合的な効果が優れており、さまざまな分野の複雑なタスクシーンに広く適用可能です。百度検索プラグインとの自動接続をサポートし、問答情報のタイムリーさを保証します。ERNIE 4.0に比べて性能がさらに優れています。"
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "百度が自主開発したフラッグシップの超大規模な言語モデルで、総合的なパフォーマンスが優れており、各分野の複雑なタスクシナリオに広く適応します；百度検索プラグインとの自動連携をサポートし、質問応答情報のタイムリーさを保証します。ERNIE 4.0に比べてパフォーマンスが向上しています。"
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-InstructはAlibaba Cloudが発表した最新の大規模言語モデルシリーズの一つです。この7Bモデルはコーディングや数学などの分野で顕著な能力の改善を持っています。このモデルは29以上の言語をカバーする多言語サポートも提供しており、中国語、英語などが含まれています。モデルは指示の遵守、構造化データの理解、特にJSONのような構造化出力の生成において顕著な向上を示しています。"
},
+ "MiniMax-Text-01": {
+ "description": "MiniMax-01シリーズモデルでは、大胆な革新を行いました：初めて大規模に線形注意メカニズムを実現し、従来のTransformerアーキテクチャが唯一の選択肢ではなくなりました。このモデルのパラメータ数は4560億に達し、単回のアクティベーションは459億です。モデルの総合性能は海外のトップモデルに匹敵し、世界最長の400万トークンのコンテキストを効率的に処理でき、GPT-4oの32倍、Claude-3.5-Sonnetの20倍です。"
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPOは非常に柔軟なマルチモデル統合で、卓越した創造的体験を提供することを目的としています。"
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2はさまざまな視覚と言語タスクで卓越した性能を発揮しており、文書や図表の理解、シーンテキストの理解、OCR、科学および数学の問題解決などを含みます。"
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2はさまざまな視覚と言語タスクで卓越した性能を発揮しており、文書や図表の理解、シーンテキストの理解、OCR、科学および数学の問題解決などを含みます。"
- },
"Phi-3-medium-128k-instruct": {
"description": "同じPhi-3-mediumモデルですが、RAGまたは少数ショットプロンプティング用により大きなコンテキストサイズを持っています。"
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chatは智譜AIが提供するGLM-4シリーズの事前訓練モデルのオープンバージョンです。このモデルは意味、数学、推論、コード、知識などの複数の側面で優れたパフォーマンスを示します。多輪対話をサポートするだけでなく、GLM-4-9B-Chatはウェブブラウジング、コード実行、カスタムツール呼び出し（Function Call）、長文推論などの高度な機能も備えています。モデルは中国語、英語、日本語、韓国語、ドイツ語など26の言語をサポートしています。多くのベンチマークテストで、GLM-4-9B-Chatは優れた性能を示し、AlignBench-v2、MT-Bench、MMLU、C-Evalなどでの評価が行われています。このモデルは最大128Kのコンテキスト長をサポートし、学術研究や商業アプリケーションに適しています。"
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1は、強化学習（RL）駆動の推論モデルで、モデル内の繰り返しと可読性の問題を解決します。RLの前に、DeepSeek-R1はコールドスタートデータを導入し、推論性能をさらに最適化しました。数学、コード、推論タスクにおいてOpenAI-o1と同等の性能を発揮し、精巧に設計されたトレーニング手法によって全体的な効果を向上させています。"
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3は、6710億パラメータを持つ混合専門家（MoE）言語モデルで、多頭潜在注意力（MLA）とDeepSeekMoEアーキテクチャを採用し、無補助損失の負荷バランス戦略を組み合わせて推論とトレーニングの効率を最適化しています。14.8兆の高品質トークンで事前トレーニングを行い、監視付き微調整と強化学習を経て、DeepSeek-V3は他のオープンソースモデルを超え、先進的なクローズドモデルに近づいています。"
+ },
"Pro/google/gemma-2-9b-it": {
"description": "GemmaはGoogleが開発した軽量で最先端のオープンモデルシリーズの一つです。これはデコーダーのみの大規模言語モデルで、英語をサポートし、オープンウェイト、事前訓練バリアント、指示微調整バリアントを提供します。Gemmaモデルは質問応答、要約、推論などのさまざまなテキスト生成タスクに適しています。この9Bモデルは8兆トークンで訓練されました。その比較的小さな規模により、リソースが限られた環境（ノートパソコン、デスクトップ、または自分のクラウドインフラストラクチャなど）でのデプロイが可能になり、より多くの人々が最先端のAIモデルにアクセスできるようになり、革新を促進します。"
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1はMetaが開発した多言語大規模言語モデルファミリーで、8B、70B、405Bの3つのパラメータ規模の事前訓練および指示微調整バリアントを含みます。この8B指示微調整モデルは多言語対話シーンに最適化されており、複数の業界ベンチマークテストで優れたパフォーマンスを示しています。モデルの訓練には150兆トークン以上の公開データが使用され、監視微調整や人間のフィードバック強化学習などの技術が採用され、モデルの有用性と安全性が向上しています。Llama 3.1はテキスト生成とコード生成をサポートし、知識のカットオフ日は2023年12月です。"
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Previewは、複雑な対話生成と文脈理解タスクを効率的に処理できる革新的な自然言語処理モデルです。"
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Previewは、Qwenチームによって開発された視覚推論能力に特化した研究モデルであり、複雑なシーン理解と視覚関連の数学問題を解決する上で独自の利点を持っています。"
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-PreviewはQwenの最新の実験的研究モデルで、AIの推論能力を向上させることに特化しています。言語の混合、再帰的推論などの複雑なメカニズムを探求することで、主な利点は強力な推論分析能力、数学およびプログラミング能力です。同時に、言語切り替えの問題、推論のループ、安全性の考慮、その他の能力の違いも存在します。"
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-InstructはAlibaba Cloudが発表したコード特化型大規模言語モデルシリーズの最新バージョンです。このモデルはQwen2.5を基に、55兆トークンの訓練を通じて、コード生成、推論、修正能力を大幅に向上させました。コーディング能力を強化するだけでなく、数学および一般的な能力の利点も維持しています。このモデルはコードエージェントなどの実際のアプリケーションに対して、より包括的な基盤を提供します。"
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Mathは、数学分野の問題解決に特化しており、高難度の問題に対して専門的な解答を提供します。"
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2はQwenモデルの最新シリーズで、128kのコンテキストをサポートしています。現在の最適なオープンソースモデルと比較して、Qwen2-72Bは自然言語理解、知識、コード、数学、そして多言語などの能力において、現在のリーディングモデルを大幅に上回っています。"
},
"Qwen2-7B-Instruct": {
"description": "Qwen2はQwenモデルの最新シリーズで、同等の規模の最適なオープンソースモデルやそれ以上の規模のモデルを超えることができ、Qwen2 7Bは複数の評価で顕著な優位性を示し、特にコードと中国語理解において優れています。"
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72Bは、強力な視覚言語モデルであり、画像とテキストのマルチモーダル処理をサポートし、画像の内容を正確に認識し、関連する説明や回答を生成できます。"
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instructは、140億パラメータの大規模言語モデルで、優れたパフォーマンスを発揮し、中国語と多言語シーンを最適化し、インテリジェントQ&A、コンテンツ生成などのアプリケーションをサポートします。"
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instructは、70億パラメータの大規模言語モデルで、関数呼び出しと外部システムとのシームレスなインタラクションをサポートし、柔軟性と拡張性を大幅に向上させます。中国語と多言語シーンを最適化し、インテリジェントQ&A、コンテンツ生成などのアプリケーションをサポートします。"
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instructは、大規模な事前学習に基づくプログラミング指示モデルであり、強力なコード理解と生成能力を持ち、さまざまなプログラミングタスクを効率的に処理でき、特にスマートコード作成、自動化スクリプト生成、プログラミング問題の解決に適しています。"
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instructは、コード生成、コード理解、効率的な開発シーンのために設計された大規模言語モデルで、業界をリードする32Bパラメータ規模を採用しており、多様なプログラミングニーズに応えます。"
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "TeleMM多モーダル大モデルは中国電信が自主開発した多モーダル理解大モデルで、テキスト、画像などの多様なモーダル入力を処理し、画像理解、グラフ分析などの機能をサポートし、ユーザーにクロスモーダルの理解サービスを提供します。モデルはユーザーと多モーダルでインタラクションし、入力内容を正確に理解し、質問に答え、創作を支援し、効率的に多モーダル情報とインスピレーションのサポートを提供します。細粒度の認識、論理推論などの多モーダルタスクで優れたパフォーマンスを示しています。"
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Largeは業界最大のオープンソースTransformerアーキテクチャMoEモデルで、3890億の総パラメータ数と520億のアクティブパラメータ数を持っています。"
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-InstructはQwen2シリーズの指示微調整大規模言語モデルで、パラメータ規模は72Bです。このモデルはTransformerアーキテクチャに基づき、SwiGLU活性化関数、注意QKVバイアス、グループクエリ注意などの技術を採用しています。大規模な入力を処理することができます。このモデルは言語理解、生成、多言語能力、コーディング、数学、推論などの複数のベンチマークテストで優れたパフォーマンスを示し、ほとんどのオープンソースモデルを超え、特定のタスクでは専有モデルと同等の競争力を示しています。"
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-InstructはAlibaba Cloudが発表した最新の大規模言語モデルシリーズの一つです。この72Bモデルはコーディングや数学などの分野で顕著な能力の改善を持っています。このモデルは29以上の言語をカバーする多言語サポートも提供しており、中国語、英語などが含まれています。モデルは指示の遵守、構造化データの理解、特にJSONのような構造化出力の生成において顕著な向上を示しています。"
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "中国語のキャラクター対話シーンに最適化されており、流暢で中国語の表現習慣に合った対話生成能力を提供します。"
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Fireworksのオープンソース関数呼び出しモデルは、卓越した指示実行能力とオープンでカスタマイズ可能な特性を提供します。"
+ "abab7-chat-preview": {
+ "description": "abab6.5シリーズモデルに比べて、長文、数学、執筆などの能力が大幅に向上しています。"
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Fireworks社の最新のFirefunction-v2は、Llama-3を基に開発された高性能な関数呼び出しモデルであり、多くの最適化を経て、特に関数呼び出し、対話、指示のフォローなどのシナリオに適しています。"
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1は、強化学習とコールドスタートデータの最適化を経た最先端の大規模言語モデルで、優れた推論、数学、プログラミング性能を持っています。"
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13bは、画像とテキストの入力を同時に受け取ることができる視覚言語モデルであり、高品質なデータで訓練されており、多モーダルタスクに適しています。"
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Deepseekが提供する強力なMixture-of-Experts (MoE)言語モデルで、総パラメータ数は671Bであり、各トークンは37Bのパラメータを活性化します。"
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Llama 3 70B指示モデルは、多言語対話と自然言語理解に最適化されており、ほとんどの競合モデルを上回る性能を持っています。"
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Llama 3 70B指示モデル（HFバージョン）は、公式実装結果と一致し、高品質な指示フォロータスクに適しています。"
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Llama 3 8B指示モデルは、対話や多言語タスクに最適化されており、卓越した効率を発揮します。"
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Metaの11Bパラメータ指示調整画像推論モデルです。このモデルは視覚認識、画像推論、画像説明、および画像に関する一般的な質問への回答に最適化されています。このモデルは、グラフや図表などの視覚データを理解し、画像の詳細をテキストで記述することで、視覚と言語の間のギャップを埋めることができます。"
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Llama 3.2 1B指示モデルはMetaが発表した軽量な多言語モデルです。このモデルは効率を向上させることを目的としており、より大規模なモデルと比較して遅延とコストの面で大きな改善を提供します。このモデルの使用例には、情報検索や要約が含まれます。"
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Llama 3.2 3B指示モデルはMetaが発表した軽量な多言語モデルです。このモデルは効率を向上させることを目的としており、より大規模なモデルと比較して遅延とコストの面で大きな改善を提供します。このモデルの使用例には、問い合わせやプロンプトのリライト、執筆支援が含まれます。"
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Metaの90Bパラメータ指示調整画像推論モデルです。このモデルは視覚認識、画像推論、画像説明、および画像に関する一般的な質問への回答に最適化されています。このモデルは、グラフや図表などの視覚データを理解し、画像の詳細をテキストで記述することで、視覚と言語の間のギャップを埋めることができます。"
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instructは、Llama 3.1 70Bの12月の更新版です。このモデルは、2024年7月にリリースされたLlama 3.1 70Bを基に改良され、ツール呼び出し、多言語テキストサポート、数学およびプログラミング能力が強化されています。このモデルは、推論、数学、指示遵守の面で業界の最前線に達しており、3.1 405Bと同等の性能を提供しつつ、速度とコストにおいて顕著な利点を持っています。"
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "24Bパラメータモデルで、より大規模なモデルと同等の最先端の能力を備えています。"
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Mixtral MoE 8x22B指示モデルは、大規模なパラメータと多専門家アーキテクチャを持ち、複雑なタスクの高効率処理を全方位でサポートします。"
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Mixtral MoE 8x7B指示モデルは、多専門家アーキテクチャを提供し、高効率の指示フォローと実行をサポートします。"
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Mixtral MoE 8x7B指示モデル（HFバージョン）は、公式実装と一致し、さまざまな高効率タスクシナリオに適しています。"
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "MythoMax L2 13Bモデルは、新しい統合技術を組み合わせており、物語やキャラクターの役割に優れています。"
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "QwQモデルはQwenチームによって開発された実験的な研究モデルで、AIの推論能力を強化することに焦点を当てています。"
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "Qwen-VLモデルの72Bバージョンは、アリババの最新のイテレーションの成果であり、近年の革新を代表しています。"
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5はAlibaba Cloud Qwenチームによって開発された一連のデコーダーのみを含む言語モデルです。これらのモデルは、0.5B、1.5B、3B、7B、14B、32B、72Bなど、さまざまなサイズを提供し、ベース版と指示版の2種類のバリエーションがあります。"
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B InstructはAlibaba Cloudが発表したコード特化型大規模言語モデルシリーズの最新バージョンです。このモデルはQwen2.5を基に、55兆トークンの訓練を通じて、コード生成、推論、修正能力を大幅に向上させました。コーディング能力を強化するだけでなく、数学および一般的な能力の利点も維持しています。このモデルはコードエージェントなどの実際のアプリケーションに対して、より包括的な基盤を提供します。"
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "StarCoder 15.5Bモデルは、高度なプログラミングタスクをサポートし、多言語能力を強化し、複雑なコード生成と理解に適しています。"
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "StarCoder 7Bモデルは、80以上のプログラミング言語に特化して訓練されており、優れたプログラミング補完能力と文脈理解を持っています。"
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Yi-Largeモデルは、卓越した多言語処理能力を持ち、さまざまな言語生成と理解タスクに使用できます。"
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opusは、Anthropicが高度に複雑なタスクを処理するために開発した最も強力なモデルです。性能、知能、流暢さ、理解力において卓越したパフォーマンスを発揮します。"
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haikuは、Anthropicの最も高速な次世代モデルです。Claude 3 Haikuと比較して、Claude 3.5 Haikuはすべてのスキルで向上しており、多くの知能ベンチマークテストで前世代の最大モデルClaude 3 Opusを超えています。"
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 SonnetはOpusを超える能力を提供し、Sonnetよりも速い速度を持ちながら、Sonnetと同じ価格を維持します。Sonnetは特にプログラミング、データサイエンス、視覚処理、代理タスクに優れています。"
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnetは、企業のワークロードに理想的なバランスを提供し、より低価格で最大の効用を提供し、信頼性が高く、大規模な展開に適しています。"
},
- "code-raccoon-v1": {
- "description": "コード小浣熊は、商湯の大規模言語モデルに基づくソフトウェアインテリジェント開発アシスタントで、ソフトウェア要件分析、アーキテクチャ設計、コード作成、ソフトウェアテストなどのプロセスをカバーし、ユーザーのコード作成やプログラミング学習などのさまざまなニーズに応えます。コード小浣熊は、Python、Java、JavaScript、C++、Go、SQLなど90以上の主流プログラミング言語と、VS Code、IntelliJ IDEAなどの主流IDEをサポートしています。実際のアプリケーションでは、コード小浣熊は開発者のプログラミング効率を50%以上向上させることができます。"
- },
"codegeex-4": {
"description": "CodeGeeX-4は強力なAIプログラミングアシスタントで、さまざまなプログラミング言語のインテリジェントな質問応答とコード補完をサポートし、開発効率を向上させます。"
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+は、リアルな企業シーンと複雑なアプリケーションのために設計された高性能な大規模言語モデルです。"
},
+ "dall-e-2": {
+ "description": "第二世代DALL·Eモデル、よりリアルで正確な画像生成をサポートし、解像度は第一世代の4倍です"
+ },
+ "dall-e-3": {
+ "description": "最新のDALL·Eモデル、2023年11月にリリース。よりリアルで正確な画像生成をサポートし、詳細表現力が向上しています"
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instructは、高い信頼性の指示処理能力を提供し、多業界アプリケーションをサポートします。"
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2は強力でコスト効率の高い混合専門家（MoE）言語モデルです。8.1兆トークンの高品質コーパスで事前訓練され、監視微調整（SFT）と強化学習（RL）を通じてモデルの能力をさらに向上させました。DeepSeek 67Bと比較して、DeepSeek-V2は性能が向上し、42.5%の訓練コストを節約し、93.3%のKVキャッシュを削減し、最大生成スループットを5.76倍に向上させました。このモデルは128kのコンテキスト長をサポートし、標準ベンチマークテストおよびオープン生成評価で優れたパフォーマンスを示しています。"
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1は、強化学習（RL）駆動の推論モデルであり、モデル内の繰り返しと可読性の問題を解決します。RLの前に、DeepSeek-R1はコールドスタートデータを導入し、推論性能をさらに最適化しました。数学、コード、推論タスクにおいてOpenAI-o1と同等のパフォーマンスを発揮し、精巧に設計されたトレーニング手法によって全体的な効果を向上させました。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "DeepSeek-R1蒸留モデルで、強化学習とコールドスタートデータを通じて推論性能を最適化し、オープンソースモデルがマルチタスクの基準を刷新しました。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8Bは、Llama-3.1-8Bに基づいて開発された蒸留モデルです。このモデルは、DeepSeek-R1が生成したサンプルを使用して微調整され、優れた推論能力を示しています。複数のベンチマークテストで良好なパフォーマンスを示し、特にMATH-500では89.1%の正確性を達成し、AIME 2024では50.4%の合格率を達成し、CodeForcesでは1205のスコアを獲得し、8B規模のモデルとして強力な数学とプログラミング能力を示しています。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "DeepSeek-R1蒸留モデルで、強化学習とコールドスタートデータを通じて推論性能を最適化し、オープンソースモデルがマルチタスクの基準を刷新しました。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "DeepSeek-R1蒸留モデルで、強化学習とコールドスタートデータを通じて推論性能を最適化し、オープンソースモデルがマルチタスクの基準を刷新しました。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32Bは、Qwen2.5-32Bに基づいて知識蒸留によって得られたモデルです。このモデルは、DeepSeek-R1が生成した80万の選りすぐりのサンプルを使用して微調整され、数学、プログラミング、推論などの複数の分野で卓越した性能を示しています。AIME 2024、MATH-500、GPQA Diamondなどの複数のベンチマークテストで優れた成績を収めており、特にMATH-500では94.3%の正確性を達成し、強力な数学的推論能力を示しています。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7Bは、Qwen2.5-Math-7Bに基づいて知識蒸留によって得られたモデルです。このモデルは、DeepSeek-R1が生成した80万の選りすぐりのサンプルを使用して微調整され、優れた推論能力を示しています。複数のベンチマークテストで優れた成績を収めており、特にMATH-500では92.8%の正確性を達成し、AIME 2024では55.5%の合格率を達成し、CodeForcesでは1189のスコアを獲得し、7B規模のモデルとして強力な数学とプログラミング能力を示しています。"
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5は以前のバージョンの優れた特徴を集約し、汎用性とコーディング能力を強化しました。"
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3は、6710億パラメータを持つ混合専門家（MoE）言語モデルであり、多頭潜在注意（MLA）とDeepSeekMoEアーキテクチャを採用し、補助損失なしの負荷バランス戦略を組み合わせて、推論とトレーニングの効率を最適化します。14.8兆の高品質トークンで事前トレーニングを行い、監視微調整と強化学習を経て、DeepSeek-V3は他のオープンソースモデルを超え、先進的なクローズドソースモデルに近づきました。"
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67Bは、高い複雑性の対話のために訓練された先進的なモデルです。"
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "最先端の効率的なLLMで、推論、数学、プログラミングに優れています。"
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2は、DeepSeekMoE-27Bに基づいて開発された混合専門家（MoE）視覚言語モデルであり、スパースアクティベーションのMoEアーキテクチャを採用し、わずか4.5Bパラメータを活性化することで卓越した性能を実現しています。このモデルは、視覚的質問応答、光学文字認識、文書/表/グラフ理解、視覚的定位などの複数のタスクで優れたパフォーマンスを発揮します。"
+ },
"deepseek-chat": {
"description": "一般的な対話能力と強力なコード処理能力を兼ね備えた新しいオープンソースモデルであり、元のChatモデルの対話能力とCoderモデルのコード処理能力を保持しつつ、人間の好みにより良く整合しています。さらに、DeepSeek-V2.5は、執筆タスクや指示に従う能力など、さまざまな面で大幅な向上を実現しました。"
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2は、オープンソースの混合エキスパートコードモデルであり、コードタスクにおいて優れた性能を発揮し、GPT4-Turboに匹敵します。"
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1は、強化学習（RL）駆動の推論モデルであり、モデル内の繰り返しと可読性の問題を解決します。RLの前に、DeepSeek-R1はコールドスタートデータを導入し、推論性能をさらに最適化しました。数学、コード、推論タスクにおいてOpenAI-o1と同等のパフォーマンスを発揮し、精巧に設計されたトレーニング手法によって全体的な効果を向上させました。"
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1——DeepSeekスイートの中でより大きく、より賢いモデル——がLlama 70Bアーキテクチャに蒸留されました。ベンチマークテストと人間評価に基づき、このモデルは元のLlama 70Bよりも賢く、特に数学と事実の正確性が求められるタスクで優れたパフォーマンスを示します。"
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "DeepSeek-R1-Distillシリーズモデルは、知識蒸留技術を通じて、DeepSeek-R1が生成したサンプルをQwen、Llamaなどのオープンソースモデルに微調整して得られたものです。"
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "DeepSeek-R1-Distillシリーズモデルは、知識蒸留技術を通じて、DeepSeek-R1が生成したサンプルをQwen、Llamaなどのオープンソースモデルに微調整して得られたものです。"
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "DeepSeek-R1-Distillシリーズモデルは、知識蒸留技術を通じて、DeepSeek-R1が生成したサンプルをQwen、Llamaなどのオープンソースモデルに微調整して得られたものです。"
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "DeepSeek-R1-Distillシリーズモデルは、知識蒸留技術を通じて、DeepSeek-R1が生成したサンプルをQwen、Llamaなどのオープンソースモデルに微調整して得られたものです。"
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "DeepSeek-R1-Distillシリーズモデルは、知識蒸留技術を通じて、DeepSeek-R1が生成したサンプルをQwen、Llamaなどのオープンソースモデルに微調整して得られたものです。"
+ },
+ "deepseek-reasoner": {
+ "description": "DeepSeekが提供する推論モデルです。最終的な回答を出力する前に、モデルは思考の連鎖を出力し、最終的な答えの正確性を高めます。"
+ },
"deepseek-v2": {
"description": "DeepSeek V2は、高効率なMixture-of-Experts言語モデルであり、経済的な処理ニーズに適しています。"
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236Bは、DeepSeekの設計コードモデルであり、強力なコード生成能力を提供します。"
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3は、杭州深度求索人工知能基礎技術研究有限公司が独自に開発したMoEモデルで、複数の評価で優れた成績を収め、主流のランキングでオープンソースモデルの首位に立っています。V3はV2.5モデルに比べて生成速度が3倍向上し、ユーザーにより迅速でスムーズな使用体験を提供します。"
+ },
"deepseek/deepseek-chat": {
"description": "汎用性とコード能力を融合させた新しいオープンソースモデルで、元のChatモデルの汎用対話能力とCoderモデルの強力なコード処理能力を保持しつつ、人間の好みにより良く整合しています。さらに、DeepSeek-V2.5は執筆タスク、指示の遵守などの多くの面で大幅な向上を実現しました。"
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1は、わずかなラベル付きデータしかない状況で、モデルの推論能力を大幅に向上させました。最終的な回答を出力する前に、モデルは思考の連鎖を出力し、最終的な答えの正確性を向上させます。"
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1は、わずかなラベル付きデータしかない状況で、モデルの推論能力を大幅に向上させました。最終的な回答を出力する前に、モデルは思考の連鎖を出力し、最終的な答えの正確性を向上させます。"
+ },
"emohaa": {
"description": "Emohaaは心理モデルで、専門的な相談能力を持ち、ユーザーが感情問題を理解するのを助けます。"
},
+ "ernie-3.5-128k": {
+ "description": "百度が独自に開発したフラッグシップの大規模言語モデルで、膨大な中英文コーパスをカバーし、強力な汎用能力を持ち、ほとんどの対話質問応答、創作生成、プラグインアプリケーションシーンの要求を満たすことができます。百度検索プラグインとの自動接続をサポートし、質問応答情報のタイムリーさを保証します。"
+ },
+ "ernie-3.5-8k": {
+ "description": "百度が独自に開発したフラッグシップの大規模言語モデルで、膨大な中英文コーパスをカバーし、強力な汎用能力を持ち、ほとんどの対話質問応答、創作生成、プラグインアプリケーションシーンの要求を満たすことができます。百度検索プラグインとの自動接続をサポートし、質問応答情報のタイムリーさを保証します。"
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "百度が独自に開発したフラッグシップの大規模言語モデルで、膨大な中英文コーパスをカバーし、強力な汎用能力を持ち、ほとんどの対話質問応答、創作生成、プラグインアプリケーションシーンの要求を満たすことができます。百度検索プラグインとの自動接続をサポートし、質問応答情報のタイムリーさを保証します。"
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "百度が独自に開発したフラッグシップの超大規模言語モデルで、ERNIE 3.5に比べてモデル能力が全面的にアップグレードされ、さまざまな分野の複雑なタスクシーンに広く適用されます。百度検索プラグインとの自動接続をサポートし、質問応答情報のタイムリーさを保証します。"
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "百度が独自に開発したフラッグシップの超大規模言語モデルで、ERNIE 3.5に比べてモデル能力が全面的にアップグレードされ、さまざまな分野の複雑なタスクシーンに広く適用されます。百度検索プラグインとの自動接続をサポートし、質問応答情報のタイムリーさを保証します。"
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "百度が独自に開発したフラッグシップの超大規模言語モデルで、総合的なパフォーマンスが優れており、さまざまな分野の複雑なタスクシーンに広く適用されます。百度検索プラグインとの自動接続をサポートし、質問応答情報のタイムリーさを保証します。ERNIE 4.0に比べてパフォーマンスがさらに優れています。"
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "百度が独自に開発したフラッグシップの超大規模言語モデルで、総合的なパフォーマンスが優れており、さまざまな分野の複雑なタスクシーンに広く適用されます。百度検索プラグインとの自動接続をサポートし、質問応答情報のタイムリーさを保証します。ERNIE 4.0に比べてパフォーマンスがさらに優れています。"
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "百度が独自に開発したフラッグシップの超大規模言語モデルで、総合的なパフォーマンスが優れており、さまざまな分野の複雑なタスクシーンに広く適用されます。百度検索プラグインとの自動接続をサポートし、質問応答情報のタイムリーさを保証します。ERNIE 4.0に比べてパフォーマンスがさらに優れています。"
+ },
+ "ernie-char-8k": {
+ "description": "百度が独自に開発した垂直シーン向けの大規模言語モデルで、ゲームのNPC、カスタマーサービスの対話、対話キャラクターの役割演技などのアプリケーションシーンに適しており、キャラクターのスタイルがより鮮明で一貫しており、指示に従う能力が強く、推論性能が優れています。"
+ },
+ "ernie-char-fiction-8k": {
+ "description": "百度が独自に開発した垂直シーン向けの大規模言語モデルで、ゲームのNPC、カスタマーサービスの対話、対話キャラクターの役割演技などのアプリケーションシーンに適しており、キャラクターのスタイルがより鮮明で一貫しており、指示に従う能力が強く、推論性能が優れています。"
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Liteは、百度が独自に開発した軽量級の大規模言語モデルで、優れたモデル効果と推論性能を兼ね備え、低計算能力のAIアクセラレータカードでの推論使用に適しています。"
+ },
+ "ernie-lite-pro-128k": {
+ "description": "百度が独自に開発した軽量級の大規模言語モデルで、優れたモデル効果と推論性能を兼ね備え、ERNIE Liteよりも優れた効果を持ち、低計算能力のAIアクセラレータカードでの推論使用に適しています。"
+ },
+ "ernie-novel-8k": {
+ "description": "百度が独自に開発した汎用大規模言語モデルで、小説の続編作成能力に明らかな優位性があり、短編劇や映画などのシーンにも使用できます。"
+ },
+ "ernie-speed-128k": {
+ "description": "百度が2024年に最新リリースした自社開発の高性能大規模言語モデルで、汎用能力が優れており、基盤モデルとして微調整に適しており、特定のシーンの問題をより良く処理し、優れた推論性能を持っています。"
+ },
+ "ernie-speed-pro-128k": {
+ "description": "百度が2024年に最新リリースした自社開発の高性能大規模言語モデルで、汎用能力が優れており、ERNIE Speedよりも優れた効果を持ち、基盤モデルとして微調整に適しており、特定のシーンの問題をより良く処理し、優れた推論性能を持っています。"
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tinyは、百度が独自に開発した超高性能の大規模言語モデルで、文心シリーズモデルの中でデプロイと微調整コストが最も低いです。"
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001（チューニング）は、安定した調整可能な性能を提供し、複雑なタスクのソリューションに理想的な選択肢です。"
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Proは、最大200万トークンをサポートする中型マルチモーダルモデルの理想的な選択肢であり、複雑なタスクに対する多面的なサポートを提供します。"
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flashは、卓越した速度、ネイティブツールの使用、マルチモーダル生成、1Mトークンのコンテキストウィンドウを含む次世代の機能と改善を提供します。"
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flashは、卓越した速度、ネイティブツールの使用、マルチモーダル生成、1Mトークンのコンテキストウィンドウを含む次世代の機能と改善を提供します。"
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp は、Google の最新の実験的なマルチモーダル AI モデルで、次世代の機能、卓越した速度、ネイティブツールの呼び出し、マルチモーダル生成を備えています。"
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "コスト効率と低遅延を目指して最適化されたGemini 2.0 Flashモデルです。"
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Expは、Googleの最新の実験的なマルチモーダルAIモデルであり、次世代の機能、卓越した速度、ネイティブツールの呼び出し、マルチモーダル生成を備えています。"
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Expは、Googleの最新の実験的なマルチモーダルAIモデルで、次世代の機能、卓越した速度、ネイティブツールの呼び出し、そしてマルチモーダル生成を備えています。"
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114は、Googleの最新の実験的なマルチモーダルAIモデルで、迅速な処理能力を備え、テキスト、画像、動画の入力をサポートし、様々なタスクに効率的に対応できます。"
- },
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121は、Googleの最新の実験的なマルチモーダルAIモデルで、高速処理能力を備え、テキスト、画像、動画の入力をサポートし、さまざまなタスクに対して効率的に拡張できます。"
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimentalは、Googleの最新の実験的なマルチモーダルAIモデルで、歴史的なバージョンと比較して品質が向上しています。特に、世界の知識、コード、長いコンテキストにおいて顕著です。"
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206は、Googleの最新の実験的なマルチモーダルAIモデルで、過去のバージョンと比較して品質が向上しています。"
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plusは動画コンテンツや複数の画像を理解する能力を持ち、マルチモーダルタスクに適しています。"
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Previewは、強力な複雑な推論能力を備え、論理推論、数学、プログラミングなどの分野で優れたパフォーマンスを発揮します。"
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flashは、卓越した速度、ネイティブツールの使用、マルチモーダル生成、1Mトークンのコンテキストウィンドウを含む次世代の機能と改善を提供します。"
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimentalは、Googleの最新の実験的なマルチモーダルAIモデルで、歴史的なバージョンと比較して品質が向上しています。特に、世界の知識、コード、長いコンテキストにおいて顕著です。"
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flashは、最適化されたマルチモーダル処理能力を提供し、さまざまな複雑なタスクシナリオに適しています。"
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Proは、最新の最適化技術を組み合わせて、より効率的なマルチモーダルデータ処理能力を実現します。"
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2はGoogleが提供する効率的なモデルで、小型アプリケーションから複雑なデータ処理まで、さまざまなアプリケーションシナリオをカバーしています。"
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2は、軽量化と高効率のデザイン理念を継承しています。"
},
"google/gemma-2-2b-it": {
"description": "Googleの軽量指示調整モデル"
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2はGoogleが提供する効率的なモデルで、小型アプリケーションから複雑なデータ処理まで、さまざまなアプリケーションシナリオをカバーしています。"
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2は、Googleの軽量オープンソーステキストモデルシリーズです。"
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turboは、さまざまなテキスト生成と理解タスクに適しており、現在はgpt-3.5-turbo-0125を指しています。"
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k、高容量テキスト生成モデル、複雑なタスクに適しています。"
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turboは、さまざまなテキスト生成と理解タスクに適しており、現在はgpt-3.5-turbo-0125を指しています。"
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4oは動的モデルで、リアルタイムで更新され、常に最新バージョンを保持します。 powerfulな言語理解と生成能力を組み合わせており、カスタマーサービス、教育、技術サポートなどの大規模なアプリケーションに適しています。"
},
+ "gpt-4o-audio-preview": {
+ "description": "GPT-4o Audio モデル、音声の入力と出力をサポート"
+ },
"gpt-4o-mini": {
"description": "GPT-4o miniは、OpenAIがGPT-4 Omniの後に発表した最新のモデルで、画像とテキストの入力をサポートし、テキストを出力します。最先端の小型モデルとして、最近の他の先進モデルよりもはるかに安価で、GPT-3.5 Turboよりも60%以上安価です。最先端の知能を維持しつつ、コストパフォーマンスが大幅に向上しています。GPT-4o miniはMMLUテストで82%のスコアを獲得し、現在チャットの好みではGPT-4よりも高い評価を得ています。"
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "GPT-4o-miniリアルタイムバージョン、音声とテキストのリアルタイム入力と出力をサポート"
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "GPT-4oリアルタイムバージョン、音声とテキストのリアルタイム入力と出力をサポート"
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "GPT-4oリアルタイムバージョン、音声とテキストのリアルタイム入力と出力をサポート"
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "GPT-4oリアルタイムバージョン、音声とテキストのリアルタイム入力と出力をサポート"
+ },
"grok-2-1212": {
"description": "このモデルは、精度、指示の遵守、そして多言語能力において改善されています。"
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "混元の最新のMOEアーキテクチャFunctionCallモデルで、高品質のFunctionCallデータトレーニングを経て、コンテキストウィンドウは32Kに達し、複数の次元の評価指標でリーダーシップを発揮しています。"
},
+ "hunyuan-large": {
+ "description": "Hunyuan-largeモデルの総パラメータ数は約389B、活性化パラメータ数は約52Bで、現在業界で最大のパラメータ規模を持ち、最も優れた効果を持つTransformerアーキテクチャのオープンソースMoEモデル。"
+ },
+ "hunyuan-large-longcontext": {
+ "description": "文書要約や文書問答などの長文タスクを得意とし、一般的なテキスト生成タスクの処理能力も備えている。長文の分析と生成において優れたパフォーマンスを発揮し、複雑で詳細な長文内容の処理要求に効果的に対応できる。"
+ },
"hunyuan-lite": {
"description": "MOE構造にアップグレードされ、コンテキストウィンドウは256kで、NLP、コード、数学、業界などの多くの評価セットで多くのオープンソースモデルをリードしています。"
},
+ "hunyuan-lite-vision": {
+ "description": "混元最新の7Bマルチモーダルモデル、コンテキストウィンドウ32K、中英文シーンのマルチモーダル対話、画像物体認識、文書表理解、マルチモーダル数学などをサポートし、複数の次元で評価指標が7B競合モデルを上回る。"
+ },
"hunyuan-pro": {
"description": "万億規模のパラメータを持つMOE-32K長文モデルです。さまざまなベンチマークで絶対的なリーダーシップを達成し、複雑な指示や推論、複雑な数学能力を備え、functioncallをサポートし、多言語翻訳、金融、法律、医療などの分野で重点的に最適化されています。"
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "より優れたルーティング戦略を採用し、負荷分散と専門家の収束の問題を緩和しました。長文に関しては、大海捞針指標が99.9%に達しています。MOE-256Kは長さと効果の面でさらに突破し、入力可能な長さを大幅に拡張しました。"
},
+ "hunyuan-standard-vision": {
+ "description": "混元最新のマルチモーダルモデルで、多言語での応答をサポートし、中英文能力が均衡している。"
+ },
"hunyuan-turbo": {
"description": "混元の新世代大規模言語モデルのプレビュー版で、全く新しい混合専門家モデル（MoE）構造を採用し、hunyuan-proに比べて推論効率が向上し、パフォーマンスも強化されています。"
},
+ "hunyuan-turbo-20241120": {
+ "description": "hunyuan-turbo 2024年11月20日の固定バージョンで、hunyuan-turboとhunyuan-turbo-latestの間に位置するバージョン。"
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "このバージョンの最適化：データ指令のスケーリングにより、モデルの汎用的な一般化能力を大幅に向上；数学、コード、論理推論能力を大幅に向上；テキスト理解と語彙理解に関連する能力を最適化；テキスト作成の内容生成の質を最適化。"
+ },
+ "hunyuan-turbo-latest": {
+ "description": "汎用体験の最適化、NLP理解、テキスト作成、雑談、知識問答、翻訳、分野などを含む；擬人性を向上させ、モデルの感情知能を最適化；意図が曖昧な時のモデルの能動的な明確化能力を向上；語彙解析に関する問題の処理能力を向上；創作の質とインタラクティブ性を向上；多段階体験を向上。"
+ },
+ "hunyuan-turbo-vision": {
+ "description": "混元の次世代視覚言語フラッグシップ大モデルで、全く新しい混合専門家モデル（MoE）構造を採用し、画像とテキストの理解に関連する基礎認識、コンテンツ作成、知識問答、分析推論などの能力が前世代モデルに比べて全面的に向上。"
+ },
"hunyuan-vision": {
"description": "混元の最新のマルチモーダルモデルで、画像とテキストの入力をサポートし、テキストコンテンツを生成します。"
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "私たちの最新のモデルシリーズで、卓越した推論性能を持ち、1Mのコンテキスト長をサポートし、より強力な指示追従とツール呼び出し能力を備えています。"
},
+ "internlm3-latest": {
+ "description": "私たちの最新のモデルシリーズは、卓越した推論性能を持ち、同等のオープンソースモデルの中でリーダーシップを発揮しています。デフォルトで最新のInternLM3シリーズモデルを指します。"
+ },
+ "jina-deepsearch-v1": {
+ "description": "深層検索は、ウェブ検索、読解、推論を組み合わせて、包括的な調査を行います。これは、あなたの研究タスクを受け入れる代理人として考えることができ、広範な検索を行い、何度も反復してから答えを提供します。このプロセスには、継続的な研究、推論、さまざまな視点からの問題解決が含まれます。これは、事前に訓練されたデータから直接答えを生成する標準的な大規模モデルや、一度きりの表面的な検索に依存する従来のRAGシステムとは根本的に異なります。"
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLMは、学習科学の原則に従って訓練された実験的なタスク特化型言語モデルで、教育や学習のシーンでシステムの指示に従い、専門的なメンターとして機能します。"
},
"lite": {
"description": "Spark Liteは軽量な大規模言語モデルで、非常に低い遅延と高い処理能力を備えています。完全に無料でオープンであり、リアルタイムのオンライン検索機能をサポートしています。その迅速な応答特性により、低算力デバイスでの推論アプリケーションやモデルの微調整において優れたパフォーマンスを発揮し、特に知識問答、コンテンツ生成、検索シーンにおいて優れたコストパフォーマンスとインテリジェントな体験を提供します。"
},
- "llama-3.1-70b-instruct": {
- "description": "Llama 3.1 70B Instructモデルは、70Bパラメータを持ち、大規模なテキスト生成と指示タスクで卓越した性能を提供します。"
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70Bは、より強力なAI推論能力を提供し、複雑なアプリケーションに適しており、非常に多くの計算処理をサポートし、高効率と精度を保証します。"
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8Bは、高効率モデルであり、迅速なテキスト生成能力を提供し、大規模な効率とコスト効果が求められるアプリケーションシナリオに非常に適しています。"
},
- "llama-3.1-8b-instruct": {
- "description": "Llama 3.1 8B Instructモデルは、8Bパラメータを持ち、画面指示タスクの高効率な実行をサポートし、優れたテキスト生成能力を提供します。"
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Llama 3.1 Sonar Huge Onlineモデルは、405Bパラメータを持ち、約127,000トークンのコンテキスト長をサポートし、複雑なオンラインチャットアプリケーション用に設計されています。"
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Llama 3.1 Sonar Large Chatモデルは、70Bパラメータを持ち、約127,000トークンのコンテキスト長をサポートし、複雑なオフラインチャットタスクに適しています。"
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Llama 3.1 Sonar Large Onlineモデルは、70Bパラメータを持ち、約127,000トークンのコンテキスト長をサポートし、高容量で多様なチャットタスクに適しています。"
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Llama 3.1 Sonar Small Chatモデルは、8Bパラメータを持ち、オフラインチャット用に設計されており、約127,000トークンのコンテキスト長をサポートします。"
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Llama 3.1 Sonar Small Onlineモデルは、8Bパラメータを持ち、約127,000トークンのコンテキスト長をサポートし、オンラインチャット用に設計されており、さまざまなテキストインタラクションを効率的に処理できます。"
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2は、視覚データとテキストデータを組み合わせたタスクを処理することを目的としています。画像の説明や視覚的質問応答などのタスクで優れたパフォーマンスを発揮し、言語生成と視覚推論の間のギャップを埋めます。"
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3は、Llamaシリーズの最先端の多言語オープンソース大規模言語モデルで、非常に低コストで405Bモデルに匹敵する性能を体験できます。Transformer構造に基づき、監視付き微調整（SFT）と人間のフィードバックによる強化学習（RLHF）を通じて有用性と安全性を向上させています。その指示調整バージョンは多言語対話に最適化されており、複数の業界ベンチマークで多くのオープンソースおよびクローズドチャットモデルを上回る性能を発揮します。知識のカットオフ日は2023年12月です。"
+ },
"llama-3.3-70b-versatile": {
"description": "Meta Llama 3.3は、70B（テキスト入力/テキスト出力）の事前学習と指示調整による生成モデルを持つ多言語大規模言語モデル（LLM）です。Llama 3.3の指示調整済みのプレーンテキストモデルは、多言語の対話ユースケースに最適化されており、一般的な業界ベンチマークで多くの利用可能なオープンソースおよびクローズドチャットモデルを上回っています。"
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2は視覚データとテキストデータを組み合わせたタスクを処理することを目的としています。画像の説明や視覚的質問応答などのタスクで優れた性能を発揮し、言語生成と視覚推論の間のギャップを埋めます。"
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3はLlamaシリーズの最先端の多言語オープンソース大規模言語モデルで、非常に低コストで405Bモデルに匹敵する性能を体験できます。Transformer構造に基づき、監視付き微調整（SFT）と人間のフィードバック強化学習（RLHF）を通じて有用性と安全性を向上させています。その指示調整バージョンは多言語対話に最適化されており、複数の業界ベンチマークで多くのオープンソースおよびクローズドチャットモデルを上回る性能を発揮します。知識のカットオフ日は2023年12月です"
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Meta Llama 3.3の多言語大規模言語モデル（LLM）は、70B（テキスト入力/テキスト出力）の事前訓練と指示調整生成モデルです。Llama 3.3の指示調整された純粋なテキストモデルは、多言語対話のユースケースに最適化されており、一般的な業界ベンチマークで多くの利用可能なオープンソースおよびクローズドチャットモデルを上回っています。"
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2は視覚データとテキストデータを組み合わせたタスクを処理することを目的としています。画像の説明や視覚的質問応答などのタスクで優れた性能を発揮し、言語生成と視覚推論の間のギャップを埋めます。"
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "405BのLlama 3.1 Turboモデルは、大規模データ処理のために超大容量のコンテキストサポートを提供し、超大規模な人工知能アプリケーションで優れたパフォーマンスを発揮します。"
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1はMetaが提供する先進的なモデルで、最大405Bのパラメータをサポートし、複雑な対話、多言語翻訳、データ分析の分野で利用できます。"
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70Bは多言語の高効率な対話サポートを提供します。"
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2は、視覚とテキストデータを組み合わせたタスクを処理することを目的としています。画像の説明や視覚的な質問応答などのタスクで優れたパフォーマンスを発揮し、言語生成と視覚推論の間のギャップを超えています。"
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3は、Llamaシリーズの最先端の多言語オープンソース大規模言語モデルで、非常に低コストで405Bモデルに匹敵する性能を体験できます。Transformer構造に基づき、監視付き微調整（SFT）と人間のフィードバックによる強化学習（RLHF）を通じて有用性と安全性を向上させています。その指示調整バージョンは多言語対話に最適化されており、複数の業界ベンチマークで多くのオープンソースおよびクローズドチャットモデルを上回る性能を発揮します。知識のカットオフ日は2023年12月です。"
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3は、Llamaシリーズの最先端の多言語オープンソース大規模言語モデルで、非常に低コストで405Bモデルに匹敵する性能を体験できます。Transformer構造に基づき、監視付き微調整（SFT）と人間のフィードバックによる強化学習（RLHF）を通じて有用性と安全性を向上させています。その指示調整バージョンは多言語対話に最適化されており、複数の業界ベンチマークで多くのオープンソースおよびクローズドチャットモデルを上回る性能を発揮します。知識のカットオフ日は2023年12月です。"
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instructは、Llama 3.1 Instructモデルの中で最大かつ最も強力なモデルであり、高度に進化した対話推論および合成データ生成モデルです。また、特定の分野での専門的な継続的な事前トレーニングや微調整の基盤としても使用できます。Llama 3.1が提供する多言語大規模言語モデル（LLMs）は、8B、70B、405Bのサイズ（テキスト入力/出力）を含む、事前トレーニングされた指示調整された生成モデルのセットです。Llama 3.1の指示調整されたテキストモデル（8B、70B、405B）は、多言語対話のユースケースに最適化されており、一般的な業界ベンチマークテストで多くの利用可能なオープンソースチャットモデルを上回っています。Llama 3.1は、さまざまな言語の商業および研究用途に使用されることを目的としています。指示調整されたテキストモデルは、アシスタントのようなチャットに適しており、事前トレーニングモデルはさまざまな自然言語生成タスクに適応できます。Llama 3.1モデルは、他のモデルを改善するためにその出力を利用することもサポートしており、合成データ生成や洗練にも対応しています。Llama 3.1は、最適化されたトランスフォーマーアーキテクチャを使用した自己回帰型言語モデルです。調整されたバージョンは、監視付き微調整（SFT）と人間のフィードバックを伴う強化学習（RLHF）を使用して、人間の助けや安全性に対する好みに適合させています。"
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3は、開発者、研究者、企業向けのオープンな大規模言語モデル（LLM）であり、生成AIのアイデアを構築、実験、責任を持って拡張するのを支援することを目的としています。世界的なコミュニティの革新の基盤システムの一部として、計算能力とリソースが限られたエッジデバイスや、より迅速なトレーニング時間に非常に適しています。"
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "高度なLLMで、合成データ生成、知識蒸留、推論をサポートし、チャットボット、プログラミング、特定の分野のタスクに適しています。"
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "複雑な対話を可能にし、卓越した文脈理解、推論能力、テキスト生成能力を備えています。"
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "高度な最先端モデルで、言語理解、卓越した推論能力、テキスト生成能力を備えています。"
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "最先端の視覚-言語モデルで、画像から高品質な推論を行うのが得意です。"
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "最先端の小型言語モデルで、言語理解、卓越した推論能力、テキスト生成能力を備えています。"
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "最先端の小型言語モデルで、言語理解、卓越した推論能力、テキスト生成能力を備えています。"
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "最先端の視覚-言語モデルで、画像から高品質な推論を行うのが得意です。"
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "高度なLLMで、推論、数学、常識、関数呼び出しに優れています。"
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2はMicrosoft AIが提供する言語モデルで、複雑な対話、多言語、推論、インテリジェントアシスタントの分野で特に優れた性能を発揮します。"
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128Kは、超長いコンテキスト処理能力を持つモデルであり、超長文の生成に適しており、複雑な生成タスクのニーズを満たし、最大128,000トークンの内容を処理でき、研究、学術、大型文書生成などのアプリケーションシーンに非常に適しています。"
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Kimi視覚モデル（moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-previewなどを含む）は、画像の内容を理解でき、画像の文字、色、物体の形状などを含みます。"
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32Kは、中程度の長さのコンテキスト処理能力を提供し、32,768トークンを処理でき、さまざまな長文や複雑な対話の生成に特に適しており、コンテンツ作成、報告書生成、対話システムなどの分野で使用されます。"
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Kimi視覚モデル（moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-previewなどを含む）は、画像の内容を理解でき、画像の文字、色、物体の形状などを含みます。"
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8Kは、短文生成タスクのために設計されており、高効率な処理性能を持ち、8,192トークンを処理でき、短い対話、速記、迅速なコンテンツ生成に非常に適しています。"
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Kimi視覚モデル（moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-previewなどを含む）は、画像の内容を理解でき、画像の文字、色、物体の形状などを含みます。"
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8BはNous Hermes 2のアップグレード版で、最新の内部開発データセットを含んでいます。"
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70BはNVIDIAがカスタマイズした大型言語モデルで、LLMが生成した応答がユーザーの問い合わせをサポートする程度を向上させることを目的としています。"
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70BはNVIDIAによってカスタマイズされた大規模言語モデルで、LLMが生成する応答がユーザーのクエリにどれだけ役立つかを向上させることを目的としています。このモデルはArena Hard、AlpacaEval 2 LC、GPT-4-Turbo MT-Benchなどのベンチマークテストで優れたパフォーマンスを示し、2024年10月1日現在、すべての自動整合ベンチマークテストで1位にランクされています。このモデルはRLHF（特にREINFORCE）、Llama-3.1-Nemotron-70B-Reward、HelpSteer2-Preferenceプロンプトを使用してLlama-3.1-70B-Instructモデルの基盤の上で訓練されています。"
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "独自の言語モデルで、比類のない精度と効率を提供します。"
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instructは、NVIDIAがカスタマイズした大規模言語モデルで、LLMが生成する応答の有用性を向上させることを目的としています。"
+ },
"o1": {
"description": "高度な推論と複雑な問題の解決に焦点を当てており、数学や科学のタスクを含みます。深いコンテキスト理解とエージェントワークフローを必要とするアプリケーションに非常に適しています。"
},
- "o1-2024-12-17": {
- "description": "o1はOpenAIの新しい推論モデルで、画像とテキストの入力をサポートし、テキストを出力します。広範な一般知識を必要とする複雑なタスクに適しています。このモデルは200Kのコンテキストを持ち、2023年10月の知識のカットオフがあります。"
- },
"o1-mini": {
"description": "o1-miniは、プログラミング、数学、科学のアプリケーションシーンに特化して設計された迅速で経済的な推論モデルです。このモデルは128Kのコンテキストを持ち、2023年10月の知識のカットオフがあります。"
},
"o1-preview": {
"description": "o1はOpenAIの新しい推論モデルで、広範な一般知識を必要とする複雑なタスクに適しています。このモデルは128Kのコンテキストを持ち、2023年10月の知識のカットオフがあります。"
},
+ "o3-mini": {
+ "description": "o3-miniは、o1-miniと同じコストと遅延目標で高い知能を提供する最新の小型推論モデルです。"
+ },
"open-codestral-mamba": {
"description": "Codestral Mambaは、コード生成に特化したMamba 2言語モデルであり、高度なコードおよび推論タスクを強力にサポートします。"
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o miniはOpenAIがGPT-4 Omniの後に発表した最新モデルで、画像とテキストの入力をサポートし、テキストを出力します。彼らの最先端の小型モデルとして、最近の他の最前線モデルよりもはるかに安価で、GPT-3.5 Turboよりも60%以上安価です。最先端の知能を維持しつつ、顕著なコストパフォーマンスを誇ります。GPT-4o miniはMMLUテストで82%のスコアを獲得し、現在チャットの好みでGPT-4よりも高い評価を得ています。"
},
- "openai/o1": {
- "description": "o1はOpenAIの新しい推論モデルで、画像とテキストの入力をサポートし、テキストを出力します。広範な一般知識を必要とする複雑なタスクに適しています。このモデルは200Kのコンテキストを持ち、2023年10月の知識のカットオフがあります。"
- },
"openai/o1-mini": {
"description": "o1-miniは、プログラミング、数学、科学のアプリケーションシーンに特化して設計された迅速で経済的な推論モデルです。このモデルは128Kのコンテキストを持ち、2023年10月の知識のカットオフがあります。"
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128Kは特大のコンテキスト処理能力を備え、最大128Kのコンテキスト情報を処理できます。特に、全体を通じての分析や長期的な論理的関連性の処理が必要な長文コンテンツに適しており、複雑なテキストコミュニケーションにおいて滑らかで一貫した論理と多様な引用サポートを提供します。"
},
+ "qvq-72b-preview": {
+ "description": "QVQモデルはQwenチームによって開発された実験的研究モデルで、視覚推論能力の向上に特化しており、特に数学推論の分野で優れた性能を発揮。"
+ },
"qwen-coder-plus-latest": {
"description": "通義千問コードモデル。"
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "通義千問の超大規模視覚言語モデル。強化版に比べて、視覚推論能力と指示遵守能力をさらに向上させ、より高い視覚認識と認知レベルを提供します。"
},
+ "qwen-vl-ocr-latest": {
+ "description": "通義千問OCRは、文書、表、試験問題、手書き文字などの画像から文字を抽出する専用モデルです。多様な文字を認識でき、現在サポートされている言語は中国語、英語、フランス語、日本語、韓国語、ドイツ語、ロシア語、イタリア語、ベトナム語、アラビア語です。"
+ },
"qwen-vl-plus-latest": {
"description": "通義千問の大規模視覚言語モデルの強化版。詳細認識能力と文字認識能力を大幅に向上させ、100万ピクセル以上の解像度と任意のアスペクト比の画像をサポートします。"
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2は全く新しい大型言語モデルシリーズで、より強力な理解と生成能力を備えています。"
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "中国語と英語に対応したLLMで、言語、プログラミング、数学、推論などの分野に特化しています。"
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "高度なLLMで、コード生成、推論、修正をサポートし、主流のプログラミング言語をカバーしています。"
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "強力な中型コードモデルで、32Kのコンテキスト長をサポートし、多言語プログラミングに優れています。"
+ },
"qwen2": {
"description": "Qwen2は、Alibabaの新世代大規模言語モデルであり、優れた性能で多様なアプリケーションニーズをサポートします。"
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "通義千問2.5の対外オープンソースの14B規模のモデルです。"
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "通義千問2.5が公開した72B規模のモデルです。"
+ },
"qwen2.5-32b-instruct": {
"description": "通義千問2.5の対外オープンソースの32B規模のモデルです。"
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Qwen-Mathモデルは、強力な数学の問題解決能力を持っています。"
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "指示に従い、数学、問題解決、コード全体の向上、万物認識能力の向上を実現し、多様な形式で視覚要素を直接的に正確に特定し、長い動画ファイル（最大10分）を理解し、秒単位のイベント時刻を特定でき、時間の前後や速さを理解し、解析と特定能力に基づいてOSやモバイルのエージェントを操作し、重要な情報抽出能力とJson形式出力能力が強化されています。このバージョンは72Bバージョンで、本シリーズの中で最も強力なバージョンです。"
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "指示に従い、数学、問題解決、コード全体の向上、万物認識能力の向上を実現し、多様な形式で視覚要素を直接的に正確に特定し、長い動画ファイル（最大10分）を理解し、秒単位のイベント時刻を特定でき、時間の前後や速さを理解し、解析と特定能力に基づいてOSやモバイルのエージェントを操作し、重要な情報抽出能力とJson形式出力能力が強化されています。このバージョンは72Bバージョンで、本シリーズの中で最も強力なバージョンです。"
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5はAlibabaの次世代大規模言語モデルで、優れた性能を持ち、多様なアプリケーションのニーズをサポートします。"
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar ProはUpstageが発表した高インテリジェンスLLMで、単一GPUの指示追従能力に特化しており、IFEvalスコアは80以上です。現在は英語をサポートしており、正式版は2024年11月にリリース予定で、言語サポートとコンテキスト長を拡張します。"
},
+ "sonar": {
+ "description": "検索コンテキストに基づく軽量検索製品で、Sonar Proよりも速く、安価です。"
+ },
+ "sonar-pro": {
+ "description": "検索コンテキストをサポートする高度な検索製品で、高度なクエリとフォローアップをサポートします。"
+ },
+ "sonar-reasoning": {
+ "description": "DeepSeek推論モデルによってサポートされる新しいAPI製品です。"
+ },
"step-1-128k": {
"description": "性能とコストのバランスを取り、一般的なシナリオに適しています。"
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "このモデルは、強力なビデオ理解能力を備えています。"
},
+ "step-1o-vision-32k": {
+ "description": "このモデルは強力な画像理解能力を持っています。step-1vシリーズモデルと比較して、より優れた視覚性能を発揮します。"
+ },
"step-1v-32k": {
"description": "視覚入力をサポートし、多モーダルインタラクション体験を強化します。"
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "大規模なコンテキストインタラクションをサポートし、複雑な対話シナリオに適しています。"
},
+ "step-2-mini": {
+ "description": "新世代の自社開発のAttentionアーキテクチャMFAに基づく超高速大モデルで、非常に低コストでstep1と同様の効果を達成しつつ、より高いスループットと迅速な応答遅延を維持しています。一般的なタスクを処理でき、コード能力において特長を持っています。"
+ },
+ "taichu2_mm": {
+ "description": "画像理解、知識移転、論理帰納などの能力を融合し、画像とテキストの質問応答分野で優れたパフォーマンスを発揮します。"
+ },
"taichu_llm": {
"description": "紫東太初言語大モデルは、強力な言語理解能力とテキスト創作、知識問答、コードプログラミング、数学計算、論理推論、感情分析、テキスト要約などの能力を備えています。革新的に大データの事前学習と多源の豊富な知識を組み合わせ、アルゴリズム技術を継続的に磨き、膨大なテキストデータから語彙、構造、文法、意味などの新しい知識を吸収し、モデルの効果を進化させています。ユーザーにより便利な情報とサービス、よりインテリジェントな体験を提供します。"
},
+ "text-embedding-3-large": {
+ "description": "最も強力なベクトル化モデル、英語および非英語のタスクに適しています"
+ },
+ "text-embedding-3-small": {
+ "description": "効率的で経済的な次世代埋め込みモデル、知識検索やRAGアプリケーションなどのシーンに適しています"
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B)は、高効率の戦略とモデルアーキテクチャを通じて、強化された計算能力を提供します。"
},
+ "tts-1": {
+ "description": "最新のテキスト音声合成モデル、リアルタイムシーン向けに速度を最適化"
+ },
+ "tts-1-hd": {
+ "description": "最新のテキスト音声合成モデル、品質を最適化"
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B)は、精密な指示タスクに適しており、優れた言語処理能力を提供します。"
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnetは業界標準を向上させ、競合モデルやClaude 3 Opusを超える性能を持ち、広範な評価で優れた結果を示し、我々の中程度のモデルの速度とコストを兼ね備えています。"
},
+ "whisper-1": {
+ "description": "汎用音声認識モデル、多言語音声認識、音声翻訳、言語認識をサポート"
+ },
"wizardlm2": {
"description": "WizardLM 2は、Microsoft AIが提供する言語モデルであり、複雑な対話、多言語、推論、インテリジェントアシスタントの分野で特に優れた性能を発揮します。"
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "複雑な視覚タスクモデルであり、高性能な画像理解と分析能力を提供します。"
+ },
+ "yi-vision-v2": {
+ "description": "複雑な視覚タスクモデルで、複数の画像に基づく高性能な理解と分析能力を提供します。"
}
}
diff --git a/locales/ja-JP/providers.json b/locales/ja-JP/providers.json
index 8c74c0daf8ce7..e4441ce542554 100644
--- a/locales/ja-JP/providers.json
+++ b/locales/ja-JP/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azureは、GPT-3.5や最新のGPT-4シリーズを含む多様な先進AIモデルを提供し、さまざまなデータタイプや複雑なタスクをサポートし、安全で信頼性が高く持続可能なAIソリューションに取り組んでいます。"
},
+ "azureai": {
+ "description": "Azureは、GPT-3.5や最新のGPT-4シリーズを含む多様な先進的AIモデルを提供し、さまざまなデータタイプや複雑なタスクをサポートし、安全で信頼性が高く持続可能なAIソリューションに取り組んでいます。"
+ },
"baichuan": {
"description": "百川智能は、人工知能大モデルの研究開発に特化した企業であり、そのモデルは国内の知識百科、長文処理、生成創作などの中国語タスクで卓越したパフォーマンスを示し、海外の主流モデルを超えています。百川智能は、業界をリードする多モーダル能力を持ち、複数の権威ある評価で優れたパフォーマンスを示しています。そのモデルには、Baichuan 4、Baichuan 3 Turbo、Baichuan 3 Turbo 128kなどが含まれ、異なるアプリケーションシーンに最適化され、高コストパフォーマンスのソリューションを提供しています。"
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeekは、人工知能技術の研究と応用に特化した企業であり、最新のモデルDeepSeek-V2.5は、汎用対話とコード処理能力を融合させ、人間の好みの整合、ライティングタスク、指示の遵守などの面で顕著な向上を実現しています。"
},
+ "doubao": {
+ "description": "バイトダンスが開発した独自の大規模モデルです。バイトダンス内部の50以上のビジネスシーンでの実践を通じて検証され、毎日数兆トークンの大規模な使用量で磨かれ、多様なモーダル能力を提供し、高品質なモデル効果で企業に豊かなビジネス体験を提供します。"
+ },
"fireworksai": {
"description": "Fireworks AIは、先進的な言語モデルサービスのリーダーであり、機能呼び出しと多モーダル処理に特化しています。最新のモデルFirefunction V2はLlama-3に基づいており、関数呼び出し、対話、指示の遵守に最適化されています。視覚言語モデルFireLLaVA-13Bは、画像とテキストの混合入力をサポートしています。他の注目すべきモデルには、LlamaシリーズやMixtralシリーズがあり、高効率の多言語指示遵守と生成サポートを提供しています。"
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "大規模モデルの研究と開発ツールチェーンに特化したオープンソース組織です。すべてのAI開発者に対して、高効率で使いやすいオープンプラットフォームを提供し、最先端の大規模モデルとアルゴリズム技術を身近に感じられるようにします。"
},
+ "jina": {
+ "description": "Jina AIは2020年に設立され、検索AIのリーディングカンパニーです。私たちの検索基盤プラットフォームには、ベクトルモデル、リランキングモデル、小型言語モデルが含まれており、企業が信頼性が高く高品質な生成AIおよびマルチモーダル検索アプリケーションを構築するのを支援します。"
+ },
+ "lmstudio": {
+ "description": "LM Studioは、あなたのコンピュータ上でLLMを開発し、実験するためのデスクトップアプリケーションです。"
+ },
"minimax": {
"description": "MiniMaxは2021年に設立された汎用人工知能テクノロジー企業であり、ユーザーと共に知能を共創することに取り組んでいます。MiniMaxは、さまざまなモードの汎用大モデルを独自に開発しており、トリリオンパラメータのMoEテキスト大モデル、音声大モデル、画像大モデルを含んでいます。また、海螺AIなどのアプリケーションも展開しています。"
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AIは、さまざまな大規模言語モデルとAI画像生成のAPIサービスを提供するプラットフォームであり、柔軟で信頼性が高く、コスト効率に優れています。Llama3、Mistralなどの最新のオープンソースモデルをサポートし、生成的AIアプリケーションの開発に向けた包括的でユーザーフレンドリーかつ自動スケーリングのAPIソリューションを提供し、AIスタートアップの急成長を支援します。"
},
+ "nvidia": {
+ "description": "NVIDIA NIM™は、自己ホスティングのGPU加速推論マイクロサービスに使用できるコンテナを提供し、クラウド、データセンター、RTX™ AIパーソナルコンピュータ、ワークステーション上で事前トレーニング済みおよびカスタムAIモデルを展開することをサポートします。"
+ },
"ollama": {
"description": "Ollamaが提供するモデルは、コード生成、数学演算、多言語処理、対話インタラクションなどの分野を広くカバーし、企業向けおよびローカライズされた展開の多様なニーズに対応しています。"
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "中科院自動化研究所と武漢人工知能研究院が新世代の多モーダル大モデルを発表し、多輪問答、テキスト創作、画像生成、3D理解、信号分析などの包括的な問答タスクをサポートし、より強力な認知、理解、創作能力を持ち、新しいインタラクティブな体験を提供します。"
},
+ "tencentcloud": {
+ "description": "知識エンジン原子能力（LLM Knowledge Engine Atomic Power）は、知識エンジンに基づいて開発された知識問答の全体的な能力であり、企業や開発者向けに、柔軟にモデルアプリケーションを構築・開発する能力を提供します。複数の原子能力を使用して、専用のモデルサービスを構築し、文書解析、分割、埋め込み、多段階の書き換えなどのサービスを組み合わせて、企業専用のAIビジネスをカスタマイズできます。"
+ },
"togetherai": {
"description": "Together AIは、革新的なAIモデルを通じて先進的な性能を実現することに取り組んでおり、迅速なスケーリングサポートや直感的な展開プロセスを含む広範なカスタマイズ能力を提供し、企業のさまざまなニーズに応えています。"
},
"upstage": {
"description": "Upstageは、さまざまなビジネスニーズに応じたAIモデルの開発に特化しており、Solar LLMや文書AIを含み、人造一般知能（AGI）の実現を目指しています。Chat APIを通じてシンプルな対話エージェントを作成し、機能呼び出し、翻訳、埋め込み、特定分野のアプリケーションをサポートします。"
},
+ "vllm": {
+ "description": "vLLMは、LLM推論とサービスのための迅速で使いやすいライブラリです。"
+ },
+ "volcengine": {
+ "description": "バイトダンスが提供する大規模モデルサービスの開発プラットフォームで、機能が豊富で安全性が高く、価格競争力のあるモデル呼び出しサービスを提供します。また、モデルデータ、ファインチューニング、推論、評価などのエンドツーエンド機能を提供し、AIアプリケーションの開発を全面的にサポートします。"
+ },
"wenxin": {
"description": "企業向けのワンストップ大規模モデルとAIネイティブアプリケーションの開発およびサービスプラットフォームで、最も包括的で使いやすい生成的人工知能モデルの開発とアプリケーション開発の全プロセスツールチェーンを提供します。"
},
diff --git a/locales/ja-JP/setting.json b/locales/ja-JP/setting.json
index 43c5adf188611..42722a8956746 100644
--- a/locales/ja-JP/setting.json
+++ b/locales/ja-JP/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "単一応答制限を有効にする"
},
+ "enableReasoningEffort": {
+ "title": "推論強度調整を有効にする"
+ },
"frequencyPenalty": {
- "desc": "値が大きいほど、単語の繰り返しを減らす可能性が高くなります",
- "title": "頻度ペナルティ"
+ "desc": "値が大きいほど、言葉がより豊かで多様になります。値が小さいほど、言葉はより素朴でシンプルになります。",
+ "title": "語彙の豊かさ"
},
"maxTokens": {
"desc": "1 回の対話で使用される最大トークン数",
@@ -212,19 +215,31 @@
"desc": "{{provider}}モデル",
"title": "モデル"
},
+ "params": {
+ "title": "高度なパラメータ"
+ },
"presencePenalty": {
- "desc": "値が大きいほど、新しいトピックに拡張する可能性が高くなります",
- "title": "トピックの新鮮度"
+ "desc": "値が大きいほど、異なる表現方法を好み、概念の繰り返しを避けます。値が小さいほど、繰り返しの概念や表現を使用する傾向が強く、一貫性のある表現になります。",
+ "title": "表現の多様性"
+ },
+ "reasoningEffort": {
+ "desc": "値が大きいほど推論能力が高まりますが、応答時間とトークン消費が増加する可能性があります",
+ "options": {
+ "high": "高",
+ "low": "低",
+ "medium": "中"
+ },
+ "title": "推論強度"
},
"temperature": {
- "desc": "値が大きいほど、応答がよりランダムになります",
- "title": "ランダム性",
- "titleWithValue": "ランダム性 {{value}}"
+ "desc": "数値が大きいほど、回答はより創造的で想像力に富む；数値が小さいほど、回答はより厳密になる",
+ "title": "創造性の活性度",
+ "warning": "創造性の活性度が高すぎると、出力に乱れが生じる可能性があります"
},
"title": "モデル設定",
"topP": {
- "desc": "ランダム性と同様ですが、ランダム性と一緒に変更しないでください",
- "title": "トップ P サンプリング"
+ "desc": "どれだけの可能性を考慮するか。値が大きいほど、より多くの可能な回答を受け入れる；値が小さいほど、最も可能性の高い回答を選ぶ傾向がある。創造性の活性度と一緒に変更することは推奨しません",
+ "title": "思考の開放度"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "一般設定",
"experiment": "実験",
"llm": "言語モデル",
+ "provider": "AIサービスプロバイダー",
"sync": "クラウド同期",
"system-agent": "システムアシスタント",
"tts": "音声サービス"
diff --git a/locales/ko-KR/changelog.json b/locales/ko-KR/changelog.json
index cc70e7a8b689f..08dc7225fecd3 100644
--- a/locales/ko-KR/changelog.json
+++ b/locales/ko-KR/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "모든 업데이트 로그 보기",
"description": "{{appName}}의 새로운 기능과 개선 사항을 지속적으로 추적하세요",
"pagination": {
- "older": "이전 변경 사항 보기",
- "prev": "이전 페이지"
+ "next": "다음 페이지",
+ "older": "이전 변경 사항 보기"
},
"readDetails": "자세히 읽기",
"title": "업데이트 로그",
diff --git a/locales/ko-KR/common.json b/locales/ko-KR/common.json
index 8d98cb52963a7..217386a3ab6d5 100644
--- a/locales/ko-KR/common.json
+++ b/locales/ko-KR/common.json
@@ -275,6 +275,7 @@
},
"temp": "임시",
"terms": "이용 약관",
+ "update": "업데이트",
"updateAgent": "에이전트 정보 업데이트",
"upgradeVersion": {
"action": "업그레이드",
@@ -285,6 +286,7 @@
"anonymousNickName": "익명 사용자",
"billing": "결제 관리",
"cloud": "체험 {{name}}",
+ "community": "커뮤니티 버전",
"data": "데이터 저장",
"defaultNickname": "커뮤니티 사용자",
"discord": "커뮤니티 지원",
@@ -294,7 +296,6 @@
"help": "도움말 센터",
"moveGuide": "설정 버튼을 여기로 이동했습니다",
"plans": "요금제",
- "preview": "미리보기",
"profile": "계정 관리",
"setting": "앱 설정",
"usages": "사용량 통계"
diff --git a/locales/ko-KR/components.json b/locales/ko-KR/components.json
index 42c28e5bc1771..a78b9ed9bf31c 100644
--- a/locales/ko-KR/components.json
+++ b/locales/ko-KR/components.json
@@ -76,6 +76,7 @@
"custom": "사용자 정의 모델, 기본적으로 함수 호출 및 시각 인식을 모두 지원하며, 실제 기능을 확인하세요",
"file": "이 모델은 파일 업로드 및 인식을 지원합니다",
"functionCall": "이 모델은 함수 호출을 지원합니다",
+ "reasoning": "이 모델은 깊이 있는 사고를 지원합니다.",
"tokens": "이 모델은 단일 세션당 최대 {{tokens}} 토큰을 지원합니다",
"vision": "이 모델은 시각 인식을 지원합니다"
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "활성화된 모델이 없습니다. 설정으로 이동하여 활성화하세요",
"provider": "제공자"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "브라우저 보안 제한으로 인해 Ollama를 사용하기 위해서는 교차 출처 구성이 필요합니다.",
+ "linux": {
+ "env": "[Service] 섹션에 `Environment`를 추가하고 OLLAMA_ORIGINS 환경 변수를 추가하세요:",
+ "reboot": "systemd를 재로드하고 Ollama를 재시작하세요",
+ "systemd": "systemd를 호출하여 ollama 서비스를 편집하세요:"
+ },
+ "macos": "터미널 애플리케이션을 열고 아래 명령어를 붙여넣은 후 Enter 키를 눌러 실행하세요",
+ "reboot": "작업이 완료된 후 Ollama 서비스를 재시작하세요",
+ "title": "Ollama의 교차 출처 접근 허용 구성",
+ "windows": "Windows에서 '제어판'을 클릭하고 시스템 환경 변수를 편집하세요. 사용자 계정에 'OLLAMA_ORIGINS'라는 이름의 환경 변수를 새로 만들고 값으로 *를 입력한 후 '확인/적용'을 클릭하여 저장하세요."
+ },
+ "install": {
+ "description": "Ollama가 이미 실행 중인지 확인하세요. Ollama를 다운로드하지 않았다면 공식 웹사이트<1>에서 다운로드하세요.",
+ "docker": "Docker를 사용하는 것을 선호하는 경우, Ollama는 공식 Docker 이미지를 제공합니다. 아래 명령어로 가져올 수 있습니다:",
+ "linux": {
+ "command": "아래 명령어로 설치하세요:",
+ "manual": "또는 <1>Linux 수동 설치 가이드를 참조하여 직접 설치할 수 있습니다."
+ },
+ "title": "로컬에 Ollama 애플리케이션 설치 및 실행",
+ "windowsTab": "Windows (미리보기 버전)"
+ }
+ },
+ "Thinking": {
+ "thinking": "심층적으로 생각 중...",
+ "thought": "심층적으로 생각했습니다 (소요 시간 {{duration}} 초)",
+ "thoughtWithDuration": "심층적으로 생각했습니다"
}
}
diff --git a/locales/ko-KR/discover.json b/locales/ko-KR/discover.json
index ba698e083b070..91c64a86bc513 100644
--- a/locales/ko-KR/discover.json
+++ b/locales/ko-KR/discover.json
@@ -126,6 +126,10 @@
"title": "주제 신선도"
},
"range": "범위",
+ "reasoning_effort": {
+ "desc": "이 설정은 모델이 응답을 생성하기 전에 추론 강도를 제어하는 데 사용됩니다. 낮은 강도는 응답 속도를 우선시하고 토큰을 절약하며, 높은 강도는 더 완전한 추론을 제공하지만 더 많은 토큰을 소모하고 응답 속도를 저하시킵니다. 기본값은 중간으로, 추론 정확성과 응답 속도의 균형을 맞춥니다.",
+ "title": "추론 강도"
+ },
"temperature": {
"desc": "이 설정은 모델 응답의 다양성에 영향을 미칩니다. 낮은 값은 더 예측 가능하고 전형적인 응답을 유도하며, 높은 값은 더 다양하고 드문 응답을 장려합니다. 값이 0으로 설정되면 모델은 주어진 입력에 대해 항상 동일한 응답을 제공합니다.",
"title": "무작위성"
diff --git a/locales/ko-KR/modelProvider.json b/locales/ko-KR/modelProvider.json
index 860fa15bdfd03..c387521b03c35 100644
--- a/locales/ko-KR/modelProvider.json
+++ b/locales/ko-KR/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API 키"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Azure API 버전, YYYY-MM-DD 형식을 따릅니다. [최신 버전](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)을 참조하세요.",
+ "fetch": "목록 가져오기",
+ "title": "Azure API 버전"
+ },
+ "endpoint": {
+ "desc": "Azure AI 프로젝트 개요에서 Azure AI 모델 추론 엔드포인트를 찾습니다.",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Azure AI 엔드포인트"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Azure AI 프로젝트 개요에서 API 키를 찾습니다.",
+ "placeholder": "Azure 키",
+ "title": "키"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "AWS 액세스 키 ID를 입력하세요.",
@@ -63,6 +81,46 @@
"title": "클라우드 플레어 계정 ID / API 주소"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "API 키를 입력하세요",
+ "title": "API 키"
+ },
+ "basicTitle": "기본 정보",
+ "configTitle": "설정 정보",
+ "confirm": "새로 만들기",
+ "createSuccess": "생성이 성공적으로 완료되었습니다",
+ "description": {
+ "placeholder": "서비스 제공자 소개 (선택 사항)",
+ "title": "서비스 제공자 소개"
+ },
+ "id": {
+ "desc": "서비스 제공자의 고유 식별자로, 생성 후에는 수정할 수 없습니다.",
+ "format": "숫자, 소문자, 하이픈(-), 및 언더스코어(_)만 포함할 수 있습니다.",
+ "placeholder": "소문자로 입력하세요, 예: openai, 생성 후 수정할 수 없습니다",
+ "required": "서비스 제공자 ID를 입력하세요",
+ "title": "서비스 제공자 ID"
+ },
+ "logo": {
+ "required": "올바른 서비스 제공자 로고를 업로드하세요",
+ "title": "서비스 제공자 로고"
+ },
+ "name": {
+ "placeholder": "서비스 제공자의 표시 이름을 입력하세요",
+ "required": "서비스 제공자 이름을 입력하세요",
+ "title": "서비스 제공자 이름"
+ },
+ "proxyUrl": {
+ "required": "프록시 주소를 입력하세요",
+ "title": "프록시 주소"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "SDK 유형을 선택하세요",
+ "title": "요청 형식"
+ },
+ "title": "사용자 정의 AI 서비스 제공자 생성"
+ },
"github": {
"personalAccessToken": {
"desc": "당신의 Github PAT를 입력하세요. [여기](https://github.com/settings/tokens)를 클릭하여 생성하세요.",
@@ -77,6 +135,23 @@
"title": "HuggingFace 토큰"
}
},
+ "list": {
+ "title": {
+ "disabled": "서비스 제공자가 비활성화되었습니다",
+ "enabled": "서비스 제공자가 활성화되었습니다"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "사용자 정의 서비스 제공자 추가",
+ "all": "모두",
+ "list": {
+ "disabled": "비활성화됨",
+ "enabled": "활성화됨"
+ },
+ "notFound": "검색 결과를 찾을 수 없습니다",
+ "searchProviders": "서비스 제공자 검색...",
+ "sort": "사용자 정의 정렬"
+ },
"ollama": {
"checker": {
"desc": "프록시 주소가 올바르게 입력되었는지 테스트합니다",
@@ -94,33 +169,9 @@
"title": "모델 {{model}} 다운로드 중"
},
"endpoint": {
- "desc": "Ollama 인터페이스 프록시 주소를 입력하세요. 로컬에서 별도로 지정하지 않은 경우 비워둘 수 있습니다",
+ "desc": "http(s)://를 포함해야 하며, 로컬에서 추가로 지정하지 않은 경우 비워둘 수 있습니다.",
"title": "인터페이스 프록시 주소"
},
- "setup": {
- "cors": {
- "description": "브라우저 보안 제한으로 인해 Ollama를 사용하려면 CORS 구성이 필요합니다.",
- "linux": {
- "env": "[Service] 섹션에 `Environment`를 추가하고 OLLAMA_ORIGINS 환경 변수를 추가하십시오:",
- "reboot": "systemd를 다시로드하고 Ollama를 다시 시작하십시오.",
- "systemd": "systemd를 호출하여 ollama 서비스를 편집하십시오: "
- },
- "macos": "「터미널」앱을 열고 다음 명령을 붙여넣고 Enter를 눌러 실행하십시오.",
- "reboot": "작업을 완료한 후 Ollama 서비스를 다시 시작하십시오.",
- "title": "CORS 액세스를 허용하도록 Ollama 구성",
- "windows": "Windows에서는 '제어판'을 클릭하여 시스템 환경 변수를 편집하십시오. 사용자 계정에 'OLLAMA_ORIGINS'이라는 환경 변수를 만들고 값으로 *을 입력한 후 '확인/적용'을 클릭하여 저장하십시오."
- },
- "install": {
- "description": "Ollama가 활성화되어 있는지 확인하고, Ollama를 다운로드하지 않았다면 공식 웹사이트<1>에서 다운로드하십시오.",
- "docker": "Docker를 사용하는 것을 선호하는 경우 Ollama는 공식 Docker 이미지도 제공하며 다음 명령을 사용하여 가져올 수 있습니다:",
- "linux": {
- "command": "다음 명령을 사용하여 설치하십시오:",
- "manual": "또는 <1>Linux 수동 설치 안내를 참조하여 직접 설치할 수도 있습니다."
- },
- "title": "로컬에서 Ollama 애플리케이션을 설치하고 시작하십시오",
- "windowsTab": "Windows (미리보기판)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "다운로드 취소",
@@ -131,25 +182,145 @@
"title": "지정된 Ollama 모델 다운로드"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "바이두 천범 플랫폼의 Access Key를 입력하세요",
- "placeholder": "Qianfan Access Key",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "귀하의 비밀 키와 프록시 주소 등은 <1>AES-GCM 암호화 알고리즘을 사용하여 암호화됩니다",
+ "apiKey": {
+ "desc": "{{name}} API 키를 입력하세요",
+ "placeholder": "{{name}} API 키",
+ "title": "API 키"
+ },
+ "baseURL": {
+ "desc": "http(s)://를 포함해야 합니다",
+ "invalid": "유효한 URL을 입력하세요",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "API 프록시 주소"
+ },
+ "checker": {
+ "button": "검사",
+ "desc": "API 키와 프록시 주소가 올바르게 입력되었는지 테스트합니다",
+ "pass": "검사 통과",
+ "title": "연결성 검사"
+ },
+ "fetchOnClient": {
+ "desc": "클라이언트 요청 모드는 브라우저에서 직접 세션 요청을 시작하여 응답 속도를 높일 수 있습니다",
+ "title": "클라이언트 요청 모드 사용"
+ },
+ "helpDoc": "설정 가이드",
+ "waitingForMore": "더 많은 모델이 <1>계획 중입니다. 기대해 주세요"
},
- "checker": {
- "desc": "AccessKey / SecretAccess가 올바르게 입력되었는지 테스트합니다"
+ "createNew": {
+ "title": "사용자 정의 AI 모델 생성"
+ },
+ "item": {
+ "config": "모델 구성",
+ "customModelCards": {
+ "addNew": "{{id}} 모델 생성 및 추가",
+ "confirmDelete": "해당 사용자 정의 모델을 삭제하려고 합니다. 삭제 후에는 복구할 수 없으니 신중하게 진행하세요."
+ },
+ "delete": {
+ "confirm": "모델 {{displayName}}를 삭제하시겠습니까?",
+ "success": "삭제 성공",
+ "title": "모델 삭제"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Azure OpenAI에서 실제 요청되는 필드",
+ "placeholder": "Azure에서 모델 배포 이름을 입력하세요",
+ "title": "모델 배포 이름"
+ },
+ "deployName": {
+ "extra": "요청을 보낼 때 이 필드가 모델 ID로 사용됩니다.",
+ "placeholder": "모델 실제 배포 이름 또는 ID를 입력하세요.",
+ "title": "모델 배포 이름"
+ },
+ "displayName": {
+ "placeholder": "모델의 표시 이름을 입력하세요, 예: ChatGPT, GPT-4 등",
+ "title": "모델 표시 이름"
+ },
+ "files": {
+ "extra": "현재 파일 업로드 구현은 단지 하나의 해킹 방법일 뿐이며, 스스로 시도하는 것만 가능합니다. 완전한 파일 업로드 기능은 후속 구현을 기다려 주세요.",
+ "title": "파일 업로드 지원"
+ },
+ "functionCall": {
+ "extra": "이 설정은 모델이 도구를 사용할 수 있는 기능을 활성화하며, 이를 통해 모델에 도구형 플러그인을 추가할 수 있습니다. 그러나 실제 도구 사용 지원 여부는 모델 자체에 따라 다르므로 사용 가능성을 직접 테스트해 보시기 바랍니다.",
+ "title": "도구 사용 지원"
+ },
+ "id": {
+ "extra": "생성 후 수정할 수 없으며, AI 호출 시 모델 ID로 사용됩니다.",
+ "placeholder": "모델 ID를 입력하세요, 예: gpt-4o 또는 claude-3.5-sonnet",
+ "title": "모델 ID"
+ },
+ "modalTitle": "사용자 정의 모델 구성",
+ "reasoning": {
+ "extra": "이 설정은 모델의 심층 사고 능력만을 활성화합니다. 구체적인 효과는 모델 자체에 따라 다르므로, 해당 모델이 사용 가능한 심층 사고 능력을 갖추고 있는지 직접 테스트해 보시기 바랍니다.",
+ "title": "심층 사고 지원"
+ },
+ "tokens": {
+ "extra": "모델이 지원하는 최대 토큰 수 설정",
+ "title": "최대 컨텍스트 창",
+ "unlimited": "제한 없음"
+ },
+ "vision": {
+ "extra": "이 설정은 애플리케이션 내에서 이미지 업로드 기능만 활성화합니다. 인식 지원 여부는 모델 자체에 따라 다르므로, 해당 모델의 시각 인식 가능성을 스스로 테스트하세요.",
+ "title": "시각 인식 지원"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/이미지",
+ "inputCharts": "${{amount}}/M 문자",
+ "inputMinutes": "${{amount}}/분",
+ "inputTokens": "입력 ${{amount}}/M",
+ "outputTokens": "출력 ${{amount}}/M"
+ },
+ "releasedAt": "발행일 {{releasedAt}}"
},
- "secretKey": {
- "desc": "바이두 천범 플랫폼의 Secret Key를 입력하세요",
- "placeholder": "Qianfan Secret Key",
- "title": "Secret Key"
+ "list": {
+ "addNew": "모델 추가",
+ "disabled": "비활성화",
+ "disabledActions": {
+ "showMore": "모두 보기"
+ },
+ "empty": {
+ "desc": "사용할 수 있는 모델이 없습니다. 사용자 정의 모델을 생성하거나 모델을 가져온 후 시작하세요.",
+ "title": "사용 가능한 모델이 없습니다."
+ },
+ "enabled": "활성화",
+ "enabledActions": {
+ "disableAll": "모두 비활성화",
+ "enableAll": "모두 활성화",
+ "sort": "사용자 정의 모델 정렬"
+ },
+ "enabledEmpty": "활성화된 모델이 없습니다. 아래 목록에서 원하는 모델을 활성화하세요~",
+ "fetcher": {
+ "clear": "가져온 모델 지우기",
+ "fetch": "모델 목록 가져오기",
+ "fetching": "모델 목록을 가져오는 중...",
+ "latestTime": "마지막 업데이트 시간: {{time}}",
+ "noLatestTime": "아직 목록을 가져오지 않았습니다."
+ },
+ "resetAll": {
+ "conform": "현재 모델의 모든 수정을 초기화하시겠습니까? 초기화 후 현재 모델 목록은 기본 상태로 돌아갑니다.",
+ "success": "초기화 성공",
+ "title": "모든 수정 초기화"
+ },
+ "search": "모델 검색...",
+ "searchResult": "{{count}} 개의 모델이 검색되었습니다",
+ "title": "모델 목록",
+ "total": "사용 가능한 모델 총 {{count}} 개"
},
- "unlock": {
- "customRegion": "사용자 정의 서비스 지역",
- "description": "당신의 AccessKey / SecretKey를 입력하면 대화를 시작할 수 있습니다. 애플리케이션은 당신의 인증 구성을 기록하지 않습니다",
- "title": "사용자 정의 문신 일언 인증 정보 사용"
- }
+ "searchNotFound": "검색 결과를 찾을 수 없습니다"
+ },
+ "sortModal": {
+ "success": "정렬 업데이트 성공",
+ "title": "사용자 정의 정렬",
+ "update": "업데이트"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "해당 AI 서비스 제공자를 삭제하려고 합니다. 삭제 후에는 복구할 수 없으니 확인하시겠습니까?",
+ "deleteSuccess": "삭제 성공",
+ "tooltip": "서비스 제공자 기본 설정 업데이트",
+ "updateSuccess": "업데이트 성공"
},
"zeroone": {
"title": "01.AI Zero One All Things"
diff --git a/locales/ko-KR/models.json b/locales/ko-KR/models.json
index 69dfd7616bbb5..68a2968ee4010 100644
--- a/locales/ko-KR/models.json
+++ b/locales/ko-KR/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K는 의미 안전성과 책임 지향성을 강조하며, 콘텐츠 안전에 대한 높은 요구가 있는 응용 시나리오를 위해 설계되어 사용자 경험의 정확성과 안정성을 보장합니다."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1은 트리 탐색을 사용하여 사고 체인을 구축하고 반성 메커니즘을 도입하였으며, 강화 학습을 통해 훈련되어 자기 반성과 오류 수정 능력을 갖추고 있습니다."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro는 360 회사에서 출시한 고급 자연어 처리 모델로, 뛰어난 텍스트 생성 및 이해 능력을 갖추고 있으며, 특히 생성 및 창작 분야에서 뛰어난 성능을 발휘하여 복잡한 언어 변환 및 역할 연기 작업을 처리할 수 있습니다."
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1은 트리 탐색을 사용하여 사고 체인을 구축하고 반성 메커니즘을 도입하여 강화 학습으로 훈련되며, 모델은 자기 반성과 오류 수정 능력을 갖추고 있습니다."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra는 스타크 대형 모델 시리즈 중 가장 강력한 버전으로, 업그레이드된 네트워크 검색 링크와 함께 텍스트 내용의 이해 및 요약 능력을 향상시킵니다. 사무 생산성을 높이고 정확한 요구에 응답하기 위한 종합 솔루션으로, 업계를 선도하는 스마트 제품입니다."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "모델 능력이 국내 1위이며, 지식 백과, 긴 텍스트, 생성 창작 등 중국어 작업에서 해외 주류 모델을 초월합니다. 또한 업계 선도적인 다중 모달 능력을 갖추고 있으며, 여러 권위 있는 평가 기준에서 우수한 성과를 보입니다."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Qwen2.5-Math-1.5B를 기반으로 한 DeepSeek-R1 증류 모델로, 강화 학습과 콜드 스타트 데이터를 통해 추론 성능을 최적화하며, 오픈 소스 모델로 다중 작업 기준을 갱신합니다."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Qwen2.5-14B를 기반으로 한 DeepSeek-R1 증류 모델로, 강화 학습과 콜드 스타트 데이터를 통해 추론 성능을 최적화하며, 오픈 소스 모델로 다중 작업 기준을 갱신합니다."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1 시리즈는 강화 학습과 콜드 스타트 데이터를 통해 추론 성능을 최적화하며, 오픈 소스 모델로 다중 작업 기준을 갱신하고 OpenAI-o1-mini 수준을 초월합니다."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "Qwen2.5-Math-7B를 기반으로 한 DeepSeek-R1 증류 모델로, 강화 학습과 콜드 스타트 데이터를 통해 추론 성능을 최적화하며, 오픈 소스 모델로 다중 작업 기준을 갱신합니다."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite는 극한의 응답 속도와 더 나은 가격 대비 성능을 자랑하며, 고객의 다양한 상황에 유연한 선택을 제공합니다. 128k 컨텍스트 윈도우의 추론 및 세부 조정을 지원합니다."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "바이두가 자체 개발한 플래그십 초대규모 언어 모델로, ERNIE 3.5에 비해 모델 능력이 전면적으로 업그레이드되었으며, 다양한 분야의 복잡한 작업 시나리오에 널리 적용됩니다. 자동으로 바이두 검색 플러그인과 연결되어 질문 응답 정보의 시의성을 보장합니다."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "바이두가 자체 개발한 플래그십 초대규모 대언어 모델로, 종합적인 효과가 뛰어나며, 다양한 분야의 복잡한 작업 장면에 널리 적용됩니다. 바이두 검색 플러그인과 자동으로 연결되어 질문과 답변 정보의 시의성을 보장합니다. ERNIE 4.0에 비해 성능이 더욱 우수합니다."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "바이두가 개발한 플래그십 대규모 언어 모델로, 다양한 분야의 복잡한 작업 환경에서 뛰어난 종합 효과를 보여줍니다. 바이두 검색 플러그인 자동 연결을 지원하여 질문과 답변 정보의 시의성을 보장합니다. ERNIE 4.0에 비해 성능이 더욱 우수합니다."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct는 Alibaba Cloud에서 발표한 최신 대규모 언어 모델 시리즈 중 하나입니다. 이 7B 모델은 코딩 및 수학 분야에서 상당한 개선된 능력을 가지고 있습니다. 이 모델은 또한 29개 이상의 언어를 포함한 다국어 지원을 제공합니다. 모델은 지침 준수, 구조화된 데이터 이해 및 구조화된 출력 생성(특히 JSON)에서 상당한 향상을 보입니다."
},
+ "MiniMax-Text-01": {
+ "description": "MiniMax-01 시리즈 모델에서는 대담한 혁신을 이루었습니다: 대규모로 선형 주의 메커니즘을 처음으로 구현하였으며, 전통적인 Transformer 아키텍처가 더 이상 유일한 선택이 아닙니다. 이 모델의 파라미터 수는 4560억에 달하며, 단일 활성화는 45.9억입니다. 모델의 종합 성능은 해외 최고의 모델과 견줄 수 있으며, 전 세계에서 가장 긴 400만 토큰의 문맥을 효율적으로 처리할 수 있습니다. 이는 GPT-4o의 32배, Claude-3.5-Sonnet의 20배에 해당합니다."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO는 뛰어난 창의적 경험을 제공하기 위해 설계된 고도로 유연한 다중 모델 통합입니다."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2는 문서 및 차트 이해, 장면 텍스트 이해, OCR, 과학 및 수학 문제 해결을 포함한 다양한 시각 언어 작업에서 뛰어난 성능을 보여줍니다."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2는 문서 및 차트 이해, 장면 텍스트 이해, OCR, 과학 및 수학 문제 해결을 포함한 다양한 시각 언어 작업에서 뛰어난 성능을 보여줍니다."
- },
"Phi-3-medium-128k-instruct": {
"description": "같은 Phi-3-medium 모델이지만 RAG 또는 몇 가지 샷 프롬프트를 위한 더 큰 컨텍스트 크기를 가지고 있습니다."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat은 Zhizhu AI가 출시한 GLM-4 시리즈의 사전 훈련 모델 중 오픈 소스 버전입니다. 이 모델은 의미, 수학, 추론, 코드 및 지식 등 여러 측면에서 뛰어난 성능을 보입니다. GLM-4-9B-Chat은 다중 회전 대화를 지원할 뿐만 아니라 웹 브라우징, 코드 실행, 사용자 정의 도구 호출(Function Call) 및 긴 텍스트 추론과 같은 고급 기능도 갖추고 있습니다. 이 모델은 중국어, 영어, 일본어, 한국어 및 독일어를 포함한 26개 언어를 지원합니다. 여러 벤치마크 테스트에서 GLM-4-9B-Chat은 AlignBench-v2, MT-Bench, MMLU 및 C-Eval 등에서 뛰어난 성능을 보였습니다. 이 모델은 최대 128K의 컨텍스트 길이를 지원하며, 학술 연구 및 상업적 응용에 적합합니다."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1은 강화 학습(RL) 기반의 추론 모델로, 모델 내의 반복성과 가독성 문제를 해결합니다. RL 이전에 DeepSeek-R1은 콜드 스타트 데이터를 도입하여 추론 성능을 더욱 최적화했습니다. 수학, 코드 및 추론 작업에서 OpenAI-o1과 유사한 성능을 보이며, 정교하게 설계된 훈련 방법을 통해 전체적인 효과를 향상시켰습니다."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3는 6710억 개의 매개변수를 가진 혼합 전문가(MoE) 언어 모델로, 다중 헤드 잠재 주의(MLA) 및 DeepSeekMoE 아키텍처를 사용하여 보조 손실 없는 부하 균형 전략을 결합하여 추론 및 훈련 효율성을 최적화합니다. 14.8조 개의 고품질 토큰에서 사전 훈련을 수행하고 감독 미세 조정 및 강화 학습을 통해 DeepSeek-V3는 성능 면에서 다른 오픈 소스 모델을 초월하며, 선도적인 폐쇄형 모델에 근접합니다."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma는 Google이 개발한 경량화된 최첨단 오픈 모델 시리즈 중 하나입니다. 이는 단일 디코더 대규모 언어 모델로, 영어를 지원하며 오픈 가중치, 사전 훈련 변형 및 지침 미세 조정 변형을 제공합니다. Gemma 모델은 질문 응답, 요약 및 추론을 포함한 다양한 텍스트 생성 작업에 적합합니다. 이 9B 모델은 80조 개의 토큰으로 훈련되었습니다. 상대적으로 작은 규모로 인해 노트북, 데스크탑 또는 개인 클라우드 인프라와 같은 자원이 제한된 환경에서 배포할 수 있어 더 많은 사람들이 최첨단 AI 모델에 접근하고 혁신을 촉진할 수 있습니다."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1은 Meta가 개발한 다국어 대규모 언어 모델 가족으로, 8B, 70B 및 405B의 세 가지 파라미터 규모의 사전 훈련 및 지침 미세 조정 변형을 포함합니다. 이 8B 지침 미세 조정 모델은 다국어 대화 시나리오에 최적화되어 있으며, 여러 산업 벤치마크 테스트에서 우수한 성능을 보입니다. 모델 훈련에는 15조 개 이상의 공개 데이터 토큰이 사용되었으며, 감독 미세 조정 및 인간 피드백 강화 학습과 같은 기술을 통해 모델의 유용성과 안전성을 향상시켰습니다. Llama 3.1은 텍스트 생성 및 코드 생성을 지원하며, 지식 마감일은 2023년 12월입니다."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview는 복잡한 대화 생성 및 맥락 이해 작업을 효율적으로 처리할 수 있는 혁신적인 자연어 처리 모델입니다."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview는 Qwen 팀이 개발한 시각적 추론 능력에 중점을 둔 연구 모델로, 복잡한 장면 이해 및 시각 관련 수학 문제 해결에서 독특한 장점을 가지고 있습니다."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview는 Qwen의 최신 실험적 연구 모델로, AI 추론 능력을 향상시키는 데 중점을 두고 있습니다. 언어 혼합, 재귀 추론 등 복잡한 메커니즘을 탐구하며, 주요 장점으로는 강력한 추론 분석 능력, 수학 및 프로그래밍 능력이 포함됩니다. 동시에 언어 전환 문제, 추론 루프, 안전성 고려 및 기타 능력 차이와 같은 문제도 존재합니다."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct는 Alibaba Cloud에서 발표한 코드 특화 대규모 언어 모델 시리즈의 최신 버전입니다. 이 모델은 Qwen2.5를 기반으로 하여 55조 개의 토큰으로 훈련되어 코드 생성, 추론 및 수정 능력을 크게 향상시켰습니다. 이 모델은 코딩 능력을 강화할 뿐만 아니라 수학 및 일반 능력의 장점도 유지합니다. 모델은 코드 에이전트와 같은 실제 응용 프로그램에 더 포괄적인 기반을 제공합니다."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math는 수학 분야의 문제 해결에 중점을 두고 있으며, 고난이도 문제에 대한 전문적인 해답을 제공합니다."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2는 Qwen 모델의 최신 시리즈로, 128k 컨텍스트를 지원합니다. 현재 최상의 오픈 소스 모델과 비교할 때, Qwen2-72B는 자연어 이해, 지식, 코드, 수학 및 다국어 등 여러 능력에서 현재 선도하는 모델을 현저히 초월합니다."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2는 Qwen 모델의 최신 시리즈로, 동등한 규모의 최적 오픈 소스 모델은 물론 더 큰 규모의 모델을 초월할 수 있습니다. Qwen2 7B는 여러 평가에서 현저한 우위를 차지하였으며, 특히 코드 및 중국어 이해에서 두드러진 성과를 보였습니다."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B는 강력한 시각 언어 모델로, 이미지와 텍스트의 다중 모드 처리를 지원하며, 이미지 내용을 정확하게 인식하고 관련 설명이나 답변을 생성할 수 있습니다."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct는 140억 매개변수를 가진 대형 언어 모델로, 성능이 우수하며 중국어 및 다국어 시나리오를 최적화하여 스마트 Q&A, 콘텐츠 생성 등의 응용을 지원합니다."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct는 70억 매개변수를 가진 대형 언어 모델로, 함수 호출 및 외부 시스템과의 원활한 상호작용을 지원하여 유연성과 확장성을 크게 향상시킵니다. 중국어 및 다국어 시나리오를 최적화하여 스마트 Q&A, 콘텐츠 생성 등의 응용을 지원합니다."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct는 대규모 사전 훈련된 프로그래밍 지침 모델로, 강력한 코드 이해 및 생성 능력을 갖추고 있으며, 다양한 프로그래밍 작업을 효율적으로 처리할 수 있습니다. 특히 스마트 코드 작성, 자동화 스크립트 생성 및 프로그래밍 문제 해결에 적합합니다."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct는 코드 생성, 코드 이해 및 효율적인 개발 시나리오를 위해 설계된 대형 언어 모델로, 업계 최고의 32B 매개변수 규모를 채택하여 다양한 프로그래밍 요구를 충족합니다."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "TeleMM 다중 모달 대모델은 중국 전신이 독자적으로 개발한 다중 모달 이해 대모델로, 텍스트, 이미지 등 다양한 모달 입력을 처리할 수 있으며, 이미지 이해, 차트 분석 등의 기능을 지원하여 사용자에게 교차 모달 이해 서비스를 제공합니다. 이 모델은 사용자와 다중 모달 상호작용을 통해 입력 내용을 정확하게 이해하고 질문에 답하며 창작을 도와주고, 효율적으로 다중 모달 정보와 영감을 제공합니다. 세밀한 인식, 논리 추론 등 다중 모달 작업에서 뛰어난 성능을 보입니다."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large는 업계에서 가장 큰 오픈 소스 Transformer 아키텍처 MoE 모델로, 총 3890억 개의 매개변수와 520억 개의 활성 매개변수를 가지고 있습니다."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct는 Qwen2 시리즈의 지침 미세 조정 대규모 언어 모델로, 파라미터 규모는 72B입니다. 이 모델은 Transformer 아키텍처를 기반으로 하며, SwiGLU 활성화 함수, 주의 QKV 편향 및 그룹 쿼리 주의와 같은 기술을 사용합니다. 이 모델은 대규모 입력을 처리할 수 있습니다. 이 모델은 언어 이해, 생성, 다국어 능력, 코딩, 수학 및 추론 등 여러 벤치마크 테스트에서 뛰어난 성능을 보이며, 대부분의 오픈 소스 모델을 초월하고 특정 작업에서 독점 모델과 동등한 경쟁력을 보여줍니다."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct는 Alibaba Cloud에서 발표한 최신 대규모 언어 모델 시리즈 중 하나입니다. 이 72B 모델은 코딩 및 수학 분야에서 상당한 개선된 능력을 가지고 있습니다. 이 모델은 또한 29개 이상의 언어를 포함한 다국어 지원을 제공합니다. 모델은 지침 준수, 구조화된 데이터 이해 및 구조화된 출력 생성(특히 JSON)에서 상당한 향상을 보입니다."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "중국어 캐릭터 대화 시나리오에 최적화되어 있으며, 유창하고 중국어 표현 습관에 맞는 대화 생성 능력을 제공합니다."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Fireworks 오픈 소스 함수 호출 모델로, 뛰어난 지시 실행 능력과 개방형 커스터마이징 기능을 제공합니다."
+ "abab7-chat-preview": {
+ "description": "abab6.5 시리즈 모델에 비해 긴 글, 수학, 작문 등에서 능력이 크게 향상되었습니다."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Fireworks 회사의 최신 Firefunction-v2는 Llama-3를 기반으로 개발된 뛰어난 함수 호출 모델로, 많은 최적화를 통해 함수 호출, 대화 및 지시 따르기 등의 시나리오에 특히 적합합니다."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1은 최첨단 대형 언어 모델로, 강화 학습과 콜드 스타트 데이터를 최적화하여 뛰어난 추론, 수학 및 프로그래밍 성능을 제공합니다."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b는 이미지와 텍스트 입력을 동시에 수용할 수 있는 비주얼 언어 모델로, 고품질 데이터로 훈련되어 다중 모달 작업에 적합합니다."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Deepseek에서 제공하는 강력한 Mixture-of-Experts (MoE) 언어 모델로, 총 매개변수 수는 671B이며, 각 토큰은 37B 매개변수를 활성화합니다."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Llama 3 70B 지시 모델은 다국어 대화 및 자연어 이해를 위해 최적화되어 있으며, 대부분의 경쟁 모델보다 성능이 우수합니다."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Llama 3 70B 지시 모델(HF 버전)은 공식 구현 결과와 일치하며, 고품질의 지시 따르기 작업에 적합합니다."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Llama 3 8B 지시 모델은 대화 및 다국어 작업을 위해 최적화되어 있으며, 뛰어난 성능과 효율성을 제공합니다."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Meta의 11B 파라미터 지시 조정 이미지 추론 모델입니다. 이 모델은 시각 인식, 이미지 추론, 이미지 설명 및 이미지에 대한 일반적인 질문에 답변하기 위해 최적화되었습니다. 이 모델은 차트 및 그래프와 같은 시각 데이터를 이해할 수 있으며, 이미지 세부 사항을 설명하는 텍스트를 생성하여 시각과 언어 간의 격차를 메웁니다."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Llama 3.2 1B 지시 모델은 Meta가 출시한 경량 다국어 모델입니다. 이 모델은 효율성을 높이기 위해 설계되었으며, 더 큰 모델에 비해 지연 시간과 비용에서 상당한 개선을 제공합니다. 이 모델의 예시 사용 사례에는 검색 및 요약이 포함됩니다."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Llama 3.2 3B 지시 모델은 Meta가 출시한 경량 다국어 모델입니다. 이 모델은 효율성을 높이기 위해 설계되었으며, 더 큰 모델에 비해 지연 시간과 비용에서 상당한 개선을 제공합니다. 이 모델의 예시 사용 사례에는 쿼리 및 프롬프트 재작성, 작문 지원이 포함됩니다."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Meta의 90B 파라미터 지시 조정 이미지 추론 모델입니다. 이 모델은 시각 인식, 이미지 추론, 이미지 설명 및 이미지에 대한 일반적인 질문에 답변하기 위해 최적화되었습니다. 이 모델은 차트 및 그래프와 같은 시각 데이터를 이해할 수 있으며, 이미지 세부 사항을 설명하는 텍스트를 생성하여 시각과 언어 간의 격차를 메웁니다."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct는 Llama 3.1 70B의 12월 업데이트 버전입니다. 이 모델은 Llama 3.1 70B(2024년 7월 출시)를 기반으로 개선되어 도구 호출, 다국어 텍스트 지원, 수학 및 프로그래밍 능력을 강화했습니다. 이 모델은 추론, 수학 및 지시 준수에서 업계 최고 수준에 도달했으며, 3.1 405B와 유사한 성능을 제공하면서 속도와 비용에서 상당한 이점을 가지고 있습니다."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "24B 매개변수 모델로, 더 큰 모델과 동등한 최첨단 능력을 갖추고 있습니다."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Mixtral MoE 8x22B 지시 모델은 대규모 매개변수와 다수의 전문가 아키텍처를 통해 복잡한 작업의 효율적인 처리를 전방위적으로 지원합니다."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Mixtral MoE 8x7B 지시 모델은 다수의 전문가 아키텍처를 통해 효율적인 지시 따르기 및 실행을 제공합니다."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Mixtral MoE 8x7B 지시 모델(HF 버전)은 성능이 공식 구현과 일치하며, 다양한 효율적인 작업 시나리오에 적합합니다."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "MythoMax L2 13B 모델은 혁신적인 통합 기술을 결합하여 서사 및 역할 수행에 강점을 보입니다."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "QwQ 모델은 Qwen 팀이 개발한 실험적 연구 모델로, AI 추론 능력을 향상시키는 데 중점을 두고 있습니다."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "Qwen-VL 모델의 72B 버전은 알리바바의 최신 반복 결과로, 거의 1년간의 혁신을 대표합니다."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5는 Alibaba Cloud Qwen 팀이 개발한 일련의 디코더 전용 언어 모델입니다. 이러한 모델은 0.5B, 1.5B, 3B, 7B, 14B, 32B 및 72B와 같은 다양한 크기를 제공하며, 기본 버전과 지시 버전 두 가지 변형이 있습니다."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct는 Alibaba Cloud에서 발표한 코드 특화 대규모 언어 모델 시리즈의 최신 버전입니다. 이 모델은 Qwen2.5를 기반으로 하여 55조 개의 토큰으로 훈련되어 코드 생성, 추론 및 수정 능력을 크게 향상시켰습니다. 이 모델은 코딩 능력을 강화할 뿐만 아니라 수학 및 일반 능력의 장점도 유지합니다. 모델은 코드 에이전트와 같은 실제 응용 프로그램에 더 포괄적인 기반을 제공합니다."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "StarCoder 15.5B 모델은 고급 프로그래밍 작업을 지원하며, 다국어 능력이 강화되어 복잡한 코드 생성 및 이해에 적합합니다."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "StarCoder 7B 모델은 80개 이상의 프로그래밍 언어를 대상으로 훈련되어 뛰어난 프로그래밍 완성 능력과 문맥 이해를 제공합니다."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Yi-Large 모델은 뛰어난 다국어 처리 능력을 갖추고 있으며, 다양한 언어 생성 및 이해 작업에 사용될 수 있습니다."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus는 Anthropic이 복잡한 작업을 처리하기 위해 개발한 가장 강력한 모델입니다. 성능, 지능, 유창성 및 이해력에서 뛰어난 성과를 보입니다."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku는 Anthropic의 가장 빠른 차세대 모델입니다. Claude 3 Haiku와 비교하여 Claude 3.5 Haiku는 모든 기술에서 향상되었으며, 많은 지능 벤치마크 테스트에서 이전 세대의 가장 큰 모델인 Claude 3 Opus를 초월했습니다."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet은 Opus를 초월하는 능력과 Sonnet보다 더 빠른 속도를 제공하며, Sonnet과 동일한 가격을 유지합니다. Sonnet은 프로그래밍, 데이터 과학, 비주얼 처리 및 에이전트 작업에 특히 강합니다."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet은 기업 작업 부하에 이상적인 균형을 제공하며, 더 낮은 가격으로 최대 효용을 제공합니다. 신뢰성이 높고 대규모 배포에 적합합니다."
},
- "code-raccoon-v1": {
- "description": "코드 라쿤은 상상 대형 언어 모델을 기반으로 한 소프트웨어 지능 개발 도우미로, 소프트웨어 요구 분석, 아키텍처 설계, 코드 작성, 소프트웨어 테스트 등 다양한 단계를 포괄하여 사용자 코드 작성 및 프로그래밍 학습 등 다양한 요구를 충족합니다. 코드 라쿤은 Python, Java, JavaScript, C++, Go, SQL 등 90개 이상의 주요 프로그래밍 언어와 VS Code, IntelliJ IDEA 등 주요 IDE를 지원합니다. 실제 응용에서 코드 라쿤은 개발자의 프로그래밍 효율성을 50% 이상 향상시킬 수 있습니다."
- },
"codegeex-4": {
"description": "CodeGeeX-4는 강력한 AI 프로그래밍 도우미로, 다양한 프로그래밍 언어에 대한 스마트 Q&A 및 코드 완성을 지원하여 개발 효율성을 높입니다."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+는 실제 기업 환경 및 복잡한 응용을 위해 설계된 고성능 대형 언어 모델입니다."
},
+ "dall-e-2": {
+ "description": "2세대 DALL·E 모델로, 더 사실적이고 정확한 이미지 생성을 지원하며, 해상도는 1세대의 4배입니다."
+ },
+ "dall-e-3": {
+ "description": "최신 DALL·E 모델로, 2023년 11월에 출시되었습니다. 더 사실적이고 정확한 이미지 생성을 지원하며, 세부 표현력이 강화되었습니다."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct는 높은 신뢰성을 가진 지시 처리 능력을 제공하며, 다양한 산업 응용을 지원합니다."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2는 강력하고 경제적인 혼합 전문가(MoE) 언어 모델입니다. 81조 개의 고품질 토큰 데이터셋에서 사전 훈련되었으며, 감독 미세 조정(SFT) 및 강화 학습(RL)을 통해 모델 능력을 더욱 향상시켰습니다. DeepSeek 67B와 비교할 때, DeepSeek-V2는 성능이 더 강력하면서도 42.5%의 훈련 비용을 절감하고 93.3%의 KV 캐시를 줄이며 최대 생성 처리량을 5.76배 향상시켰습니다. 이 모델은 128k의 컨텍스트 길이를 지원하며, 표준 벤치마크 테스트와 오픈 생성 평가에서 모두 뛰어난 성능을 보입니다."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1은 강화 학습(RL) 기반의 추론 모델로, 모델 내의 반복성과 가독성 문제를 해결합니다. RL 이전에 DeepSeek-R1은 콜드 스타트 데이터를 도입하여 추론 성능을 더욱 최적화했습니다. 수학, 코드 및 추론 작업에서 OpenAI-o1과 유사한 성능을 보이며, 정교하게 설계된 훈련 방법을 통해 전체적인 효과를 향상시켰습니다."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "DeepSeek-R1 증류 모델로, 강화 학습과 콜드 스타트 데이터를 통해 추론 성능을 최적화하며, 오픈 소스 모델로 다중 작업 기준을 갱신합니다."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B는 Llama-3.1-8B를 기반으로 개발된 증류 모델입니다. 이 모델은 DeepSeek-R1이 생성한 샘플을 사용하여 미세 조정되었으며, 뛰어난 추론 능력을 보여줍니다. 여러 기준 테스트에서 좋은 성적을 거두었으며, MATH-500에서 89.1%의 정확도를 달성하고, AIME 2024에서 50.4%의 통과율을 기록했으며, CodeForces에서 1205의 점수를 얻어 8B 규모의 모델로서 강력한 수학 및 프로그래밍 능력을 보여줍니다."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "DeepSeek-R1 증류 모델로, 강화 학습과 콜드 스타트 데이터를 통해 추론 성능을 최적화하며, 오픈 소스 모델로 다중 작업 기준을 갱신합니다."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "DeepSeek-R1 증류 모델로, 강화 학습과 콜드 스타트 데이터를 통해 추론 성능을 최적화하며, 오픈 소스 모델로 다중 작업 기준을 갱신합니다."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B는 Qwen2.5-32B를 기반으로 지식 증류를 통해 얻은 모델입니다. 이 모델은 DeepSeek-R1이 생성한 80만 개의 선별된 샘플을 사용하여 미세 조정되었으며, 수학, 프로그래밍 및 추론 등 여러 분야에서 뛰어난 성능을 보여줍니다. AIME 2024, MATH-500, GPQA Diamond 등 여러 기준 테스트에서 우수한 성적을 거두었으며, MATH-500에서 94.3%의 정확도를 달성하여 강력한 수학 추론 능력을 보여줍니다."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B는 Qwen2.5-Math-7B를 기반으로 지식 증류를 통해 얻은 모델입니다. 이 모델은 DeepSeek-R1이 생성한 80만 개의 선별된 샘플을 사용하여 미세 조정되었으며, 뛰어난 추론 능력을 보여줍니다. 여러 기준 테스트에서 우수한 성적을 거두었으며, MATH-500에서 92.8%의 정확도를 달성하고, AIME 2024에서 55.5%의 통과율을 기록했으며, CodeForces에서 1189의 점수를 얻어 7B 규모의 모델로서 강력한 수학 및 프로그래밍 능력을 보여줍니다."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5는 이전 버전의 우수한 기능을 집약하여 일반 및 인코딩 능력을 강화했습니다."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3는 6710억 개의 매개변수를 가진 혼합 전문가(MoE) 언어 모델로, 다중 헤드 잠재 주의(MLA) 및 DeepSeekMoE 아키텍처를 채택하여 보조 손실 없는 부하 균형 전략을 결합하여 추론 및 훈련 효율성을 최적화합니다. 14.8조 개의 고품질 토큰에서 사전 훈련을 수행하고 감독 미세 조정 및 강화 학습을 통해 DeepSeek-V3는 성능 면에서 다른 오픈 소스 모델을 초월하며, 선도적인 폐쇄형 모델에 근접합니다."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B는 고복잡성 대화를 위해 훈련된 고급 모델입니다."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "추론, 수학 및 프로그래밍에 능숙한 최첨단 효율 LLM입니다."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2는 DeepSeekMoE-27B를 기반으로 개발된 혼합 전문가(MoE) 비주얼 언어 모델로, 희소 활성화 MoE 아키텍처를 사용하여 4.5B 매개변수만 활성화된 상태에서 뛰어난 성능을 발휘합니다. 이 모델은 비주얼 질문 응답, 광학 문자 인식, 문서/표/차트 이해 및 비주얼 위치 지정 등 여러 작업에서 우수한 성과를 보입니다."
+ },
"deepseek-chat": {
"description": "일반 및 코드 능력을 융합한 새로운 오픈 소스 모델로, 기존 Chat 모델의 일반 대화 능력과 Coder 모델의 강력한 코드 처리 능력을 유지하면서 인간의 선호에 더 잘 맞춰졌습니다. 또한, DeepSeek-V2.5는 작문 작업, 지시 따르기 등 여러 측면에서 큰 향상을 이루었습니다."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2는 오픈 소스 혼합 전문가 코드 모델로, 코드 작업에서 뛰어난 성능을 발휘하며, GPT4-Turbo와 경쟁할 수 있습니다."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1은 강화 학습(RL) 기반의 추론 모델로, 모델 내의 반복성과 가독성 문제를 해결합니다. RL 이전에 DeepSeek-R1은 콜드 스타트 데이터를 도입하여 추론 성능을 더욱 최적화했습니다. 수학, 코드 및 추론 작업에서 OpenAI-o1과 유사한 성능을 보이며, 정교하게 설계된 훈련 방법을 통해 전체적인 효과를 향상시켰습니다."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 - DeepSeek 패키지에서 더 크고 더 스마트한 모델이 Llama 70B 아키텍처로 증류되었습니다. 기준 테스트와 인공지능 평가에 따르면, 이 모델은 원래 Llama 70B보다 더 스마트하며, 특히 수학 및 사실 정확성이 필요한 작업에서 뛰어난 성능을 보입니다."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "DeepSeek-R1-Distill 시리즈 모델은 지식 증류 기술을 통해 DeepSeek-R1이 생성한 샘플을 Qwen, Llama 등 오픈 소스 모델에 미세 조정하여 얻은 것입니다."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "DeepSeek-R1-Distill 시리즈 모델은 지식 증류 기술을 통해 DeepSeek-R1이 생성한 샘플을 Qwen, Llama 등 오픈 소스 모델에 미세 조정하여 얻은 것입니다."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "DeepSeek-R1-Distill 시리즈 모델은 지식 증류 기술을 통해 DeepSeek-R1이 생성한 샘플을 Qwen, Llama 등 오픈 소스 모델에 미세 조정하여 얻은 것입니다."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "DeepSeek-R1-Distill 시리즈 모델은 지식 증류 기술을 통해 DeepSeek-R1이 생성한 샘플을 Qwen, Llama 등 오픈 소스 모델에 미세 조정하여 얻은 것입니다."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "DeepSeek-R1-Distill 시리즈 모델은 지식 증류 기술을 통해 DeepSeek-R1이 생성한 샘플을 Qwen, Llama 등 오픈 소스 모델에 미세 조정하여 얻은 것입니다."
+ },
+ "deepseek-reasoner": {
+ "description": "DeepSeek에서 제공하는 추론 모델입니다. 최종 답변을 출력하기 전에 모델은 먼저 사고 과정을 출력하여 최종 답변의 정확성을 높입니다."
+ },
"deepseek-v2": {
"description": "DeepSeek V2는 경제적이고 효율적인 처리 요구에 적합한 Mixture-of-Experts 언어 모델입니다."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B는 DeepSeek의 설계 코드 모델로, 강력한 코드 생성 능력을 제공합니다."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3는 항저우 심도 탐색 인공지능 기초 기술 연구 회사에서 자체 개발한 MoE 모델로, 여러 평가에서 뛰어난 성적을 거두며, 주류 순위에서 오픈 소스 모델 1위를 차지하고 있습니다. V3는 V2.5 모델에 비해 생성 속도가 3배 향상되어 사용자에게 더 빠르고 원활한 사용 경험을 제공합니다."
+ },
"deepseek/deepseek-chat": {
"description": "일반 및 코드 능력을 통합한 새로운 오픈 소스 모델로, 기존 Chat 모델의 일반 대화 능력과 Coder 모델의 강력한 코드 처리 능력을 유지하면서 인간의 선호에 더 잘 맞춰졌습니다. 또한, DeepSeek-V2.5는 작문 작업, 지시 따르기 등 여러 분야에서 큰 향상을 이루었습니다."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1은 극히 적은 주석 데이터로 모델의 추론 능력을 크게 향상시킵니다. 최종 답변을 출력하기 전에 모델은 먼저 사고의 연쇄 내용을 출력하여 최종 답변의 정확성을 높입니다."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1은 극히 적은 주석 데이터로 모델의 추론 능력을 크게 향상시킵니다. 최종 답변을 출력하기 전에 모델은 먼저 사고의 연쇄 내용을 출력하여 최종 답변의 정확성을 높입니다."
+ },
"emohaa": {
"description": "Emohaa는 심리 모델로, 전문 상담 능력을 갖추고 있어 사용자가 감정 문제를 이해하는 데 도움을 줍니다."
},
+ "ernie-3.5-128k": {
+ "description": "바이두가 자체 개발한 플래그십 대규모 언어 모델로, 방대한 중영문 자료를 포함하고 있으며, 강력한 일반 능력을 가지고 있어 대부분의 대화 질문 답변, 창작 생성, 플러그인 응용 시나리오 요구를 충족할 수 있습니다. 바이두 검색 플러그인과 자동으로 연결되어 질문 답변 정보의 시의성을 보장합니다."
+ },
+ "ernie-3.5-8k": {
+ "description": "바이두가 자체 개발한 플래그십 대규모 언어 모델로, 방대한 중영문 자료를 포함하고 있으며, 강력한 일반 능력을 가지고 있어 대부분의 대화 질문 답변, 창작 생성, 플러그인 응용 시나리오 요구를 충족할 수 있습니다. 바이두 검색 플러그인과 자동으로 연결되어 질문 답변 정보의 시의성을 보장합니다."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "바이두가 자체 개발한 플래그십 대규모 언어 모델로, 방대한 중영문 자료를 포함하고 있으며, 강력한 일반 능력을 가지고 있어 대부분의 대화 질문 답변, 창작 생성, 플러그인 응용 시나리오 요구를 충족할 수 있습니다. 바이두 검색 플러그인과 자동으로 연결되어 질문 답변 정보의 시의성을 보장합니다."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "바이두가 자체 개발한 플래그십 초대규모 언어 모델로, ERNIE 3.5에 비해 모델 능력이 전면 업그레이드되었으며, 다양한 분야의 복잡한 작업 시나리오에 널리 적용됩니다. 바이두 검색 플러그인과 자동으로 연결되어 질문 답변 정보의 시의성을 보장합니다."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "바이두가 자체 개발한 플래그십 초대규모 언어 모델로, ERNIE 3.5에 비해 모델 능력이 전면 업그레이드되었으며, 다양한 분야의 복잡한 작업 시나리오에 널리 적용됩니다. 바이두 검색 플러그인과 자동으로 연결되어 질문 답변 정보의 시의성을 보장합니다."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "바이두가 자체 개발한 플래그십 초대규모 언어 모델로, 종합적인 성능이 뛰어나며, 다양한 분야의 복잡한 작업 시나리오에 널리 적용됩니다. 바이두 검색 플러그인과 자동으로 연결되어 질문 답변 정보의 시의성을 보장합니다. ERNIE 4.0에 비해 성능이 더 우수합니다."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "바이두가 자체 개발한 플래그십 초대규모 언어 모델로, 종합적인 성능이 뛰어나며, 다양한 분야의 복잡한 작업 시나리오에 널리 적용됩니다. 바이두 검색 플러그인과 자동으로 연결되어 질문 답변 정보의 시의성을 보장합니다. ERNIE 4.0에 비해 성능이 더 우수합니다."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "바이두가 자체 개발한 플래그십 초대규모 언어 모델로, 종합적인 성능이 뛰어나며, 다양한 분야의 복잡한 작업 시나리오에 널리 적용됩니다. 바이두 검색 플러그인과 자동으로 연결되어 질문 답변 정보의 시의성을 보장합니다. ERNIE 4.0에 비해 성능이 더 우수합니다."
+ },
+ "ernie-char-8k": {
+ "description": "바이두가 자체 개발한 수직 장면 대형 언어 모델로, 게임 NPC, 고객 서비스 대화, 대화 역할극 등 응용 시나리오에 적합하며, 캐릭터 스타일이 더 뚜렷하고 일관되며, 지시 따르기 능력이 더 강하고 추론 성능이 우수합니다."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "바이두가 자체 개발한 수직 장면 대형 언어 모델로, 게임 NPC, 고객 서비스 대화, 대화 역할극 등 응용 시나리오에 적합하며, 캐릭터 스타일이 더 뚜렷하고 일관되며, 지시 따르기 능력이 더 강하고 추론 성능이 우수합니다."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite는 바이두가 자체 개발한 경량 대형 언어 모델로, 우수한 모델 효과와 추론 성능을 겸비하여 저전력 AI 가속 카드 추론에 적합합니다."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "바이두가 자체 개발한 경량 대형 언어 모델로, 우수한 모델 효과와 추론 성능을 겸비하여 ERNIE Lite보다 더 우수하며, 저전력 AI 가속 카드 추론에 적합합니다."
+ },
+ "ernie-novel-8k": {
+ "description": "바이두가 자체 개발한 일반 대형 언어 모델로, 소설 연속 작성 능력에서 뚜렷한 장점을 가지고 있으며, 단편극, 영화 등 시나리오에서도 사용할 수 있습니다."
+ },
+ "ernie-speed-128k": {
+ "description": "바이두가 2024년에 최신 출시한 고성능 대형 언어 모델로, 일반 능력이 우수하여 특정 시나리오 문제를 더 잘 처리하기 위해 기초 모델로 미세 조정하는 데 적합하며, 뛰어난 추론 성능을 가지고 있습니다."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "바이두가 2024년에 최신 출시한 고성능 대형 언어 모델로, 일반 능력이 우수하여 ERNIE Speed보다 더 우수하며, 특정 시나리오 문제를 더 잘 처리하기 위해 기초 모델로 미세 조정하는 데 적합하며, 뛰어난 추론 성능을 가지고 있습니다."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny는 바이두가 자체 개발한 초고성능 대형 언어 모델로, 문신 시리즈 모델 중 배포 및 미세 조정 비용이 가장 낮습니다."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Tuning)은 안정적이고 조정 가능한 성능을 제공하며, 복잡한 작업 솔루션의 이상적인 선택입니다."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro는 최대 200만 개의 토큰을 지원하며, 중형 다중 모달 모델의 이상적인 선택으로 복잡한 작업에 대한 다각적인 지원을 제공합니다."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash는 뛰어난 속도, 원주율 도구 사용, 다중 모달 생성 및 1M 토큰 문맥 창을 포함한 차세대 기능과 개선 사항을 제공합니다."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash는 뛰어난 속도, 원주율 도구 사용, 다중 모달 생성 및 1M 토큰 문맥 창을 포함한 차세대 기능과 개선 사항을 제공합니다."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp는 Google의 최신 실험적 다중 모달 AI 모델로, 차세대 기능, 뛰어난 속도, 네이티브 도구 호출 및 다중 모달 생성을 갖추고 있습니다."
},
- "gemini-2.0-flash-thinking-exp-1219": {
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "비용 효율성과 낮은 지연 시간 등을 목표로 최적화된 Gemini 2.0 Flash 모델입니다."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
"description": "Gemini 2.0 Flash Exp는 Google의 최신 실험적 다중 모드 AI 모델로, 차세대 기능, 뛰어난 속도, 네이티브 도구 호출 및 다중 모드 생성을 제공합니다."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114는 Google의 최신 실험적 다중 모달 AI 모델로, 빠른 처리 능력을 갖추고 있으며, 텍스트, 이미지 및 비디오 입력을 지원하여 다양한 작업에 효율적으로 확장할 수 있습니다."
+ "gemini-2.0-flash-thinking-exp-1219": {
+ "description": "Gemini 2.0 Flash Exp는 Google의 최신 실험적 다중 모드 AI 모델로, 차세대 기능, 뛰어난 속도, 네이티브 도구 호출 및 다중 모드 생성을 제공합니다."
},
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121은 Google의 최신 실험적 다중 모달 AI 모델로, 빠른 처리 능력을 갖추고 있으며 텍스트, 이미지 및 비디오 입력을 지원하여 다양한 작업에 효율적으로 확장할 수 있습니다."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental은 Google의 최신 실험적 다중 모달 AI 모델로, 이전 버전과 비교하여 품질이 향상되었습니다. 특히 세계 지식, 코드 및 긴 문맥에 대해 개선되었습니다."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206은 Google의 최신 실험적 다중 모달 AI 모델로, 이전 버전에 비해 품질이 개선되었습니다."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus는 비디오 콘텐츠 및 다수의 이미지에 대한 이해 능력을 갖추고 있어 다중 모드 작업에 적합합니다."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview는 강력한 복잡한 추론 능력을 갖추고 있으며, 논리 추론, 수학, 프로그래밍 등 분야에서 우수한 성능을 발휘합니다."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash는 뛰어난 속도, 원주율 도구 사용, 다중 모달 생성 및 1M 토큰 문맥 창을 포함한 차세대 기능과 개선 사항을 제공합니다."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental은 Google의 최신 실험적 다중 모달 AI 모델로, 이전 버전과 비교하여 품질이 향상되었습니다. 특히 세계 지식, 코드 및 긴 문맥에 대해 개선되었습니다."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash는 최적화된 다중 모달 처리 능력을 제공하며, 다양한 복잡한 작업 시나리오에 적합합니다."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro는 최신 최적화 기술을 결합하여 더 효율적인 다중 모달 데이터 처리 능력을 제공합니다."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2는 Google에서 출시한 효율적인 모델로, 소형 애플리케이션부터 복잡한 데이터 처리까지 다양한 응용 시나리오를 포함합니다."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2는 경량화와 효율적인 설계를 이어갑니다."
},
"google/gemma-2-2b-it": {
"description": "Google의 경량 지시 조정 모델"
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2는 Google에서 출시한 효율적인 모델로, 소형 애플리케이션부터 복잡한 데이터 처리까지 다양한 응용 시나리오를 포함합니다."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2는 Google의 경량화된 오픈 소스 텍스트 모델 시리즈입니다."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo는 다양한 텍스트 생성 및 이해 작업에 적합하며, 현재 gpt-3.5-turbo-0125를 가리킵니다."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, 고용량 텍스트 생성 모델로 복잡한 작업에 적합합니다."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo는 다양한 텍스트 생성 및 이해 작업에 적합하며, 현재 gpt-3.5-turbo-0125를 가리킵니다."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o는 동적 모델로, 최신 버전을 유지하기 위해 실시간으로 업데이트됩니다. 강력한 언어 이해 및 생성 능력을 결합하여 고객 서비스, 교육 및 기술 지원을 포함한 대규모 애플리케이션에 적합합니다."
},
+ "gpt-4o-audio-preview": {
+ "description": "GPT-4o 오디오 모델로, 오디오 입력 및 출력을 지원합니다."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini는 OpenAI가 GPT-4 Omni 이후에 출시한 최신 모델로, 텍스트와 이미지를 입력받아 텍스트를 출력합니다. 이 모델은 최신의 소형 모델로, 최근의 다른 최첨단 모델보다 훨씬 저렴하며, GPT-3.5 Turbo보다 60% 이상 저렴합니다. 최첨단의 지능을 유지하면서도 뛰어난 가성비를 자랑합니다. GPT-4o mini는 MMLU 테스트에서 82%의 점수를 기록했으며, 현재 채팅 선호도에서 GPT-4보다 높은 순위를 차지하고 있습니다."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "GPT-4o-mini 실시간 버전으로, 오디오 및 텍스트의 실시간 입력 및 출력을 지원합니다."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "GPT-4o 실시간 버전으로, 오디오 및 텍스트의 실시간 입력 및 출력을 지원합니다."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "GPT-4o 실시간 버전으로, 오디오 및 텍스트의 실시간 입력 및 출력을 지원합니다."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "GPT-4o 실시간 버전으로, 오디오 및 텍스트의 실시간 입력 및 출력을 지원합니다."
+ },
"grok-2-1212": {
"description": "이 모델은 정확성, 지시 준수 및 다국어 능력에서 개선되었습니다."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "혼원 최신 MOE 구조의 FunctionCall 모델로, 고품질 FunctionCall 데이터 훈련을 거쳤으며, 컨텍스트 윈도우는 32K에 도달하고 여러 차원의 평가 지표에서 선두에 있습니다."
},
+ "hunyuan-large": {
+ "description": "Hunyuan-large 모델의 총 매개변수 수는 약 389B, 활성화 매개변수 수는 약 52B로, 현재 업계에서 매개변수 규모가 가장 크고 성능이 가장 우수한 Transformer 구조의 오픈 소스 MoE 모델입니다."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "문서 요약 및 문서 질문 응답과 같은 긴 문서 작업을 잘 처리하며, 일반 텍스트 생성 작업도 수행할 수 있는 능력을 갖추고 있습니다. 긴 텍스트의 분석 및 생성에서 뛰어난 성능을 보이며, 복잡하고 상세한 긴 문서 내용 처리 요구에 효과적으로 대응할 수 있습니다."
+ },
"hunyuan-lite": {
"description": "MOE 구조로 업그레이드되었으며, 컨텍스트 윈도우는 256k로 설정되어 NLP, 코드, 수학, 산업 등 여러 평가 집합에서 많은 오픈 소스 모델을 선도하고 있습니다."
},
+ "hunyuan-lite-vision": {
+ "description": "혼원 최신 7B 다중 모달 모델, 컨텍스트 윈도우 32K, 중문 및 영문 환경에서의 다중 모달 대화, 이미지 객체 인식, 문서 표 이해, 다중 모달 수학 등을 지원하며, 여러 차원에서 평가 지표가 7B 경쟁 모델보다 우수합니다."
+ },
"hunyuan-pro": {
"description": "조 단위 매개변수 규모의 MOE-32K 긴 문서 모델입니다. 다양한 벤치마크에서 절대적인 선두 수준에 도달하며, 복잡한 지시 및 추론, 복잡한 수학 능력을 갖추고 있으며, functioncall을 지원하고 다국어 번역, 금융, 법률, 의료 등 분야에서 최적화된 응용을 제공합니다."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "더 나은 라우팅 전략을 채택하여 부하 균형 및 전문가 수렴 문제를 완화했습니다. 긴 문서의 경우, 대해잡기 지표가 99.9%에 도달했습니다. MOE-256K는 길이와 효과에서 더욱 발전하여 입력 길이를 크게 확장했습니다."
},
+ "hunyuan-standard-vision": {
+ "description": "혼원 최신 다중 모달 모델, 다국어 응답 지원, 중문 및 영문 능력이 균형 잡혀 있습니다."
+ },
"hunyuan-turbo": {
"description": "혼원 최신 세대 대형 언어 모델의 미리보기 버전으로, 새로운 혼합 전문가 모델(MoE) 구조를 채택하여 hunyuan-pro보다 추론 효율이 더 빠르고 성능이 더 뛰어납니다."
},
+ "hunyuan-turbo-20241120": {
+ "description": "hunyuan-turbo 2024년 11월 20일 고정 버전, hunyuan-turbo와 hunyuan-turbo-latest 사이의 버전."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "이번 버전 최적화: 데이터 지시 스케일링, 모델의 일반화 능력 대폭 향상; 수학, 코드, 논리 추론 능력 대폭 향상; 텍스트 이해 및 단어 이해 관련 능력 최적화; 텍스트 창작 내용 생성 질 최적화."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "일반적인 경험 최적화, NLP 이해, 텍스트 창작, 대화, 지식 질문 응답, 번역, 분야 등; 인간화 향상, 모델의 감정 지능 최적화; 의도가 모호할 때 모델의 능동적인 명확화 능력 향상; 단어 및 구문 분석 관련 문제 처리 능력 향상; 창작의 질과 상호작용성 향상; 다중 회차 경험 향상."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "혼원 차세대 비주얼 언어 플래그십 대형 모델, 새로운 혼합 전문가 모델(MoE) 구조를 채택하여, 이미지 및 텍스트 이해 관련 기본 인식, 콘텐츠 창작, 지식 질문 응답, 분석 추론 등의 능력이 이전 세대 모델에 비해 전반적으로 향상되었습니다."
+ },
"hunyuan-vision": {
"description": "혼원 최신 다중 모달 모델로, 이미지와 텍스트 입력을 지원하여 텍스트 콘텐츠를 생성합니다."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "우리가 최신으로 선보이는 모델 시리즈로, 뛰어난 추론 성능을 자랑하며 1M의 컨텍스트 길이와 더 강력한 지시 따르기 및 도구 호출 기능을 지원합니다."
},
+ "internlm3-latest": {
+ "description": "우리의 최신 모델 시리즈는 뛰어난 추론 성능을 가지고 있으며, 동급 오픈 소스 모델 중에서 선두를 달리고 있습니다. 기본적으로 최신 출시된 InternLM3 시리즈 모델을 가리킵니다."
+ },
+ "jina-deepsearch-v1": {
+ "description": "딥 서치는 웹 검색, 독서 및 추론을 결합하여 포괄적인 조사를 수행합니다. 연구 작업을 수용하는 에이전트로 생각할 수 있으며, 광범위한 검색을 수행하고 여러 번 반복한 후에야 답변을 제공합니다. 이 과정은 지속적인 연구, 추론 및 다양한 각도에서 문제를 해결하는 것을 포함합니다. 이는 사전 훈련된 데이터에서 직접 답변을 생성하는 표준 대형 모델 및 일회성 표면 검색에 의존하는 전통적인 RAG 시스템과 근본적으로 다릅니다."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM은 학습 과학 원칙에 맞춰 훈련된 실험적이고 특정 작업에 특화된 언어 모델로, 교육 및 학습 환경에서 시스템 지침을 따르며 전문가 멘토 역할을 수행합니다."
},
"lite": {
"description": "Spark Lite는 경량 대형 언어 모델로, 매우 낮은 지연 시간과 효율적인 처리 능력을 갖추고 있으며, 완전히 무료로 제공되고 실시간 온라인 검색 기능을 지원합니다. 빠른 응답 특성 덕분에 저전력 장치에서의 추론 응용 및 모델 미세 조정에서 뛰어난 성능을 발휘하며, 사용자에게 뛰어난 비용 효율성과 스마트한 경험을 제공합니다. 특히 지식 질문 응답, 콘텐츠 생성 및 검색 시나리오에서 두각을 나타냅니다."
},
- "llama-3.1-70b-instruct": {
- "description": "Llama 3.1 70B Instruct 모델은 70B 매개변수를 갖추고 있으며, 대규모 텍스트 생성 및 지시 작업에서 뛰어난 성능을 제공합니다."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B는 더 강력한 AI 추론 능력을 제공하며, 복잡한 응용 프로그램에 적합하고, 많은 계산 처리를 지원하며 효율성과 정확성을 보장합니다."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B는 효율적인 모델로, 빠른 텍스트 생성 능력을 제공하며, 대규모 효율성과 비용 효과성이 필요한 응용 프로그램에 매우 적합합니다."
},
- "llama-3.1-8b-instruct": {
- "description": "Llama 3.1 8B Instruct 모델은 8B 매개변수를 갖추고 있으며, 화면 지시 작업의 효율적인 실행을 지원하고 우수한 텍스트 생성 능력을 제공합니다."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Llama 3.1 Sonar Huge Online 모델은 405B 매개변수를 갖추고 있으며, 약 127,000개의 토큰의 컨텍스트 길이를 지원하여 복잡한 온라인 채팅 애플리케이션을 위해 설계되었습니다."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Llama 3.1 Sonar Large Chat 모델은 70B 매개변수를 갖추고 있으며, 약 127,000개의 토큰의 컨텍스트 길이를 지원하여 복잡한 오프라인 채팅 작업에 적합합니다."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Llama 3.1 Sonar Large Online 모델은 70B 매개변수를 갖추고 있으며, 약 127,000개의 토큰의 컨텍스트 길이를 지원하여 대용량 및 다양한 채팅 작업에 적합합니다."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Llama 3.1 Sonar Small Chat 모델은 8B 매개변수를 갖추고 있으며, 오프라인 채팅을 위해 설계되어 약 127,000개의 토큰의 컨텍스트 길이를 지원합니다."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Llama 3.1 Sonar Small Online 모델은 8B 매개변수를 갖추고 있으며, 약 127,000개의 토큰의 컨텍스트 길이를 지원하여 온라인 채팅을 위해 설계되었습니다."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2는 시각 및 텍스트 데이터를 결합한 작업을 처리하기 위해 설계되었습니다. 이미지 설명 및 시각적 질문 응답과 같은 작업에서 뛰어난 성능을 보이며, 언어 생성과 시각적 추론 간의 간극을 넘습니다."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3은 Llama 시리즈에서 가장 진보된 다국어 오픈 소스 대형 언어 모델로, 매우 낮은 비용으로 405B 모델의 성능을 경험할 수 있습니다. Transformer 구조를 기반으로 하며, 감독 미세 조정(SFT)과 인간 피드백 강화 학습(RLHF)을 통해 유용성과 안전성을 향상시켰습니다. 이 지시 조정 버전은 다국어 대화를 위해 최적화되어 있으며, 여러 산업 벤치마크에서 많은 오픈 소스 및 폐쇄형 채팅 모델보다 우수한 성능을 보입니다. 지식 마감일은 2023년 12월입니다."
+ },
"llama-3.3-70b-versatile": {
"description": "Meta Llama 3.3 다국어 대형 언어 모델(LLM)은 70B(텍스트 입력/텍스트 출력)에서 사전 학습 및 지침 조정 생성 모델입니다. Llama 3.3의 지침 조정 순수 텍스트 모델은 다국어 대화 사용 사례에 최적화되어 있으며, 많은 오픈 소스 및 폐쇄형 채팅 모델보다 일반 산업 기준에서 우수한 성능을 보입니다."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2는 시각 및 텍스트 데이터를 결합한 작업을 처리하도록 설계되었습니다. 이미지 설명 및 시각적 질문 응답과 같은 작업에서 뛰어난 성능을 발휘하며, 언어 생성과 시각 추론 간의 간극을 메웁니다."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3은 Llama 시리즈에서 가장 진보된 다국어 오픈 소스 대형 언어 모델로, 매우 낮은 비용으로 405B 모델의 성능을 경험할 수 있습니다. Transformer 구조를 기반으로 하며, 감독 미세 조정(SFT) 및 인간 피드백 강화 학습(RLHF)을 통해 유용성과 안전성을 향상시켰습니다. 그 지시 조정 버전은 다국어 대화를 최적화하여 여러 산업 벤치마크에서 많은 오픈 소스 및 폐쇄형 채팅 모델보다 우수한 성능을 보입니다. 지식 마감일은 2023년 12월입니다."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Meta Llama 3.3 다국어 대형 언어 모델(LLM)은 70B(텍스트 입력/텍스트 출력)에서 사전 훈련 및 지시 조정 생성 모델입니다. Llama 3.3 지시 조정의 순수 텍스트 모델은 다국어 대화 사용 사례에 최적화되어 있으며, 일반 산업 기준에서 많은 사용 가능한 오픈 소스 및 폐쇄형 채팅 모델보다 우수한 성능을 보입니다."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2는 시각 및 텍스트 데이터를 결합한 작업을 처리하도록 설계되었습니다. 이미지 설명 및 시각적 질문 응답과 같은 작업에서 뛰어난 성능을 발휘하며, 언어 생성과 시각 추론 간의 간극을 메웁니다."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "405B Llama 3.1 Turbo 모델은 대규모 데이터 처리를 위한 초대용량의 컨텍스트 지원을 제공하며, 초대규모 인공지능 애플리케이션에서 뛰어난 성능을 발휘합니다."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1은 Meta에서 출시한 선도적인 모델로, 최대 405B 매개변수를 지원하며 복잡한 대화, 다국어 번역 및 데이터 분석 분야에 적용됩니다."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B는 다국어의 효율적인 대화 지원을 제공합니다."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2는 시각 및 텍스트 데이터를 결합한 작업을 처리하기 위해 설계되었습니다. 이미지 설명 및 시각적 질문 응답과 같은 작업에서 뛰어난 성능을 보이며, 언어 생성과 시각적 추론 간의 간극을 넘습니다."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3은 Llama 시리즈에서 가장 진보된 다국어 오픈 소스 대형 언어 모델로, 매우 낮은 비용으로 405B 모델의 성능을 경험할 수 있습니다. Transformer 구조를 기반으로 하며, 감독 미세 조정(SFT)과 인간 피드백 강화 학습(RLHF)을 통해 유용성과 안전성을 향상시켰습니다. 이 지시 조정 버전은 다국어 대화를 위해 최적화되어 있으며, 여러 산업 벤치마크에서 많은 오픈 소스 및 폐쇄형 채팅 모델보다 우수한 성능을 보입니다. 지식 마감일은 2023년 12월입니다."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3은 Llama 시리즈에서 가장 진보된 다국어 오픈 소스 대형 언어 모델로, 매우 낮은 비용으로 405B 모델의 성능을 경험할 수 있습니다. Transformer 구조를 기반으로 하며, 감독 미세 조정(SFT)과 인간 피드백 강화 학습(RLHF)을 통해 유용성과 안전성을 향상시켰습니다. 이 지시 조정 버전은 다국어 대화를 위해 최적화되어 있으며, 여러 산업 벤치마크에서 많은 오픈 소스 및 폐쇄형 채팅 모델보다 우수한 성능을 보입니다. 지식 마감일은 2023년 12월입니다."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct는 Llama 3.1 Instruct 모델 중 가장 크고 강력한 모델로, 고도로 발전된 대화 추론 및 합성 데이터 생성 모델입니다. 특정 분야에서 전문적인 지속적 사전 훈련 또는 미세 조정의 기초로도 사용될 수 있습니다. Llama 3.1이 제공하는 다국어 대형 언어 모델(LLMs)은 8B, 70B 및 405B 크기의 사전 훈련된 지시 조정 생성 모델로 구성되어 있습니다(텍스트 입력/출력). Llama 3.1 지시 조정 텍스트 모델(8B, 70B, 405B)은 다국어 대화 사용 사례에 최적화되어 있으며, 일반 산업 벤치마크 테스트에서 많은 오픈 소스 채팅 모델을 초과했습니다. Llama 3.1은 다양한 언어의 상업적 및 연구 용도로 설계되었습니다. 지시 조정 텍스트 모델은 비서와 유사한 채팅에 적합하며, 사전 훈련 모델은 다양한 자연어 생성 작업에 적응할 수 있습니다. Llama 3.1 모델은 또한 모델의 출력을 활용하여 다른 모델을 개선하는 것을 지원하며, 합성 데이터 생성 및 정제에 사용될 수 있습니다. Llama 3.1은 최적화된 변압기 아키텍처를 사용한 자기 회귀 언어 모델입니다. 조정된 버전은 감독 미세 조정(SFT) 및 인간 피드백이 포함된 강화 학습(RLHF)을 사용하여 인간의 도움 및 안전성 선호에 부합하도록 설계되었습니다."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3은 개발자, 연구자 및 기업을 위한 오픈 대형 언어 모델(LLM)로, 생성 AI 아이디어를 구축하고 실험하며 책임감 있게 확장하는 데 도움을 주기 위해 설계되었습니다. 전 세계 커뮤니티 혁신의 기초 시스템의 일환으로, 계산 능력과 자원이 제한된 환경, 엣지 장치 및 더 빠른 훈련 시간에 매우 적합합니다."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "합성 데이터 생성, 지식 증류 및 추론을 지원하는 고급 LLM으로, 챗봇, 프로그래밍 및 특정 분야 작업에 적합합니다."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "복잡한 대화를 가능하게 하며, 뛰어난 맥락 이해, 추론 능력 및 텍스트 생성 능력을 갖추고 있습니다."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "언어 이해, 뛰어난 추론 능력 및 텍스트 생성 능력을 갖춘 고급 최첨단 모델입니다."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "이미지에서 고품질 추론을 수행하는 최첨단 비주얼-언어 모델입니다."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "언어 이해, 뛰어난 추론 능력 및 텍스트 생성 능력을 갖춘 최첨단 소형 언어 모델입니다."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "언어 이해, 뛰어난 추론 능력 및 텍스트 생성 능력을 갖춘 최첨단 소형 언어 모델입니다."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "이미지에서 고품질 추론을 수행하는 최첨단 비주얼-언어 모델입니다."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "추론, 수학, 상식 및 함수 호출에 능숙한 고급 LLM입니다."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2는 Microsoft AI가 제공하는 언어 모델로, 복잡한 대화, 다국어, 추론 및 스마트 어시스턴트 분야에서 특히 뛰어난 성능을 보입니다."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K는 초장기 컨텍스트 처리 능력을 갖춘 모델로, 초장문 생성을 위해 설계되었으며, 복잡한 생성 작업 요구를 충족하고 최대 128,000개의 토큰을 처리할 수 있어, 연구, 학술 및 대형 문서 생성 등 응용 시나리오에 매우 적합합니다."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Kimi 시각 모델(예: moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview 등)은 이미지 내용을 이해할 수 있으며, 이미지 텍스트, 이미지 색상 및 물체 형태 등을 포함합니다."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K는 중간 길이의 컨텍스트 처리 능력을 제공하며, 32,768개의 토큰을 처리할 수 있어, 다양한 장문 및 복잡한 대화 생성을 위해 특히 적합하며, 콘텐츠 생성, 보고서 작성 및 대화 시스템 등 분야에 활용됩니다."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Kimi 시각 모델(예: moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview 등)은 이미지 내용을 이해할 수 있으며, 이미지 텍스트, 이미지 색상 및 물체 형태 등을 포함합니다."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K는 짧은 텍스트 작업 생성을 위해 설계되었으며, 효율적인 처리 성능을 갖추고 있어 8,192개의 토큰을 처리할 수 있으며, 짧은 대화, 속기 및 빠른 콘텐츠 생성에 매우 적합합니다."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Kimi 시각 모델(예: moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview 등)은 이미지 내용을 이해할 수 있으며, 이미지 텍스트, 이미지 색상 및 물체 형태 등을 포함합니다."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B는 Nous Hermes 2의 업그레이드 버전으로, 최신 내부 개발 데이터 세트를 포함하고 있습니다."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B는 NVIDIA가 맞춤 제작한 대형 언어 모델로, LLM 생성된 응답이 사용자 쿼리에 도움이 되는 정도를 높이기 위해 설계되었습니다."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B는 NVIDIA가 맞춤 제작한 대규모 언어 모델로, LLM이 생성한 응답이 사용자 쿼리에 얼마나 도움이 되는지를 향상시키기 위해 설계되었습니다. 이 모델은 Arena Hard, AlpacaEval 2 LC 및 GPT-4-Turbo MT-Bench와 같은 벤치마크 테스트에서 뛰어난 성능을 보였으며, 2024년 10월 1일 기준으로 모든 자동 정렬 벤치마크 테스트에서 1위를 차지했습니다. 이 모델은 RLHF(특히 REINFORCE), Llama-3.1-Nemotron-70B-Reward 및 HelpSteer2-Preference 프롬프트를 사용하여 Llama-3.1-70B-Instruct 모델을 기반으로 훈련되었습니다."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "비교할 수 없는 정확성과 효율성을 제공하는 독특한 언어 모델입니다."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct는 NVIDIA가 맞춤 제작한 대형 언어 모델로, LLM이 생성한 응답의 유용성을 향상시키기 위해 설계되었습니다."
+ },
"o1": {
"description": "고급 추론 및 복잡한 문제 해결에 중점을 두며, 수학 및 과학 작업을 포함합니다. 깊이 있는 컨텍스트 이해와 에이전트 작업 흐름이 필요한 애플리케이션에 매우 적합합니다."
},
- "o1-2024-12-17": {
- "description": "o1은 OpenAI의 새로운 추론 모델로, 이미지와 텍스트 입력을 지원하며 텍스트 출력을 제공합니다. 광범위한 일반 지식이 필요한 복잡한 작업에 적합합니다. 이 모델은 200K의 컨텍스트와 2023년 10월의 지식 기준일을 가지고 있습니다."
- },
"o1-mini": {
"description": "o1-mini는 프로그래밍, 수학 및 과학 응용 프로그램을 위해 설계된 빠르고 경제적인 추론 모델입니다. 이 모델은 128K의 컨텍스트와 2023년 10월의 지식 기준일을 가지고 있습니다."
},
"o1-preview": {
"description": "o1은 OpenAI의 새로운 추론 모델로, 광범위한 일반 지식이 필요한 복잡한 작업에 적합합니다. 이 모델은 128K의 컨텍스트와 2023년 10월의 지식 기준일을 가지고 있습니다."
},
+ "o3-mini": {
+ "description": "o3-mini는 최신 소형 추론 모델로, o1-mini와 동일한 비용과 지연 목표에서 높은 지능을 제공합니다."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba는 코드 생성을 전문으로 하는 Mamba 2 언어 모델로, 고급 코드 및 추론 작업에 강력한 지원을 제공합니다."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini는 OpenAI가 GPT-4 Omni 이후에 출시한 최신 모델로, 이미지와 텍스트 입력을 지원하며 텍스트를 출력합니다. 가장 진보된 소형 모델로, 최근의 다른 최첨단 모델보다 훨씬 저렴하며, GPT-3.5 Turbo보다 60% 이상 저렴합니다. 최첨단 지능을 유지하면서도 뛰어난 가성비를 자랑합니다. GPT-4o mini는 MMLU 테스트에서 82%의 점수를 기록했으며, 현재 채팅 선호도에서 GPT-4보다 높은 순위를 차지하고 있습니다."
},
- "openai/o1": {
- "description": "o1은 OpenAI의 새로운 추론 모델로, 이미지와 텍스트 입력을 지원하며 텍스트 출력을 제공합니다. 광범위한 일반 지식이 필요한 복잡한 작업에 적합합니다. 이 모델은 200K의 컨텍스트와 2023년 10월의 지식 기준일을 가지고 있습니다."
- },
"openai/o1-mini": {
"description": "o1-mini는 프로그래밍, 수학 및 과학 응용 프로그램을 위해 설계된 빠르고 경제적인 추론 모델입니다. 이 모델은 128K의 컨텍스트와 2023년 10월의 지식 기준일을 가지고 있습니다."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K는 매우 큰 컨텍스트 처리 능력을 갖추고 있으며, 최대 128K의 컨텍스트 정보를 처리할 수 있습니다. 특히 전체 분석 및 장기 논리 연관 처리가 필요한 긴 문서 콘텐츠에 적합하며, 복잡한 텍스트 커뮤니케이션에서 매끄럽고 일관된 논리와 다양한 인용 지원을 제공합니다."
},
+ "qvq-72b-preview": {
+ "description": "QVQ 모델은 Qwen 팀이 개발한 실험적 연구 모델로, 시각적 추론 능력 향상에 중점을 두고 있으며, 특히 수학적 추론 분야에서 두드러진 성과를 보입니다."
+ },
"qwen-coder-plus-latest": {
"description": "통의 천문 코드 모델입니다."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "통의천문 초대규모 비주얼 언어 모델. 강화판에 비해 시각적 추론 능력과 지시 준수 능력을 다시 한 번 향상시켜, 더 높은 시각적 인식과 인지 수준을 제공합니다."
},
+ "qwen-vl-ocr-latest": {
+ "description": "통의천문OCR은 문서, 표, 시험지, 손글씨 등 다양한 유형의 이미지에서 텍스트 추출 능력에 중점을 둔 전용 모델입니다. 여러 언어를 인식할 수 있으며, 현재 지원되는 언어는 중국어, 영어, 프랑스어, 일본어, 한국어, 독일어, 러시아어, 이탈리아어, 베트남어, 아랍어입니다."
+ },
"qwen-vl-plus-latest": {
"description": "통의천문 대규모 비주얼 언어 모델 강화판. 세부 사항 인식 능력과 문자 인식 능력을 크게 향상시켰으며, 백만 화소 이상의 해상도와 임의의 가로 세로 비율의 이미지를 지원합니다."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2는 더 강력한 이해 및 생성 능력을 갖춘 새로운 대형 언어 모델 시리즈입니다."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "중국어와 영어를 위한 LLM으로, 언어, 프로그래밍, 수학, 추론 등 다양한 분야를 다룹니다."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "코드 생성, 추론 및 수정 지원을 위한 고급 LLM으로, 주요 프로그래밍 언어를 포함합니다."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "32K 컨텍스트 길이를 지원하는 강력한 중형 코드 모델로, 다국어 프로그래밍에 능숙합니다."
+ },
"qwen2": {
"description": "Qwen2는 Alibaba의 차세대 대규모 언어 모델로, 뛰어난 성능으로 다양한 응용 요구를 지원합니다."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "통의 천문 2.5 외부 오픈 소스 14B 규모 모델입니다."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "통의천문2.5의 외부 오픈 소스 72B 규모 모델입니다."
+ },
"qwen2.5-32b-instruct": {
"description": "통의 천문 2.5 외부 오픈 소스 32B 규모 모델입니다."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Qwen-Math 모델은 강력한 수학 문제 해결 능력을 가지고 있습니다."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "지시 따르기, 수학, 문제 해결, 코드 전반적인 향상, 모든 사물 인식 능력 향상, 다양한 형식의 시각적 요소를 직접 정확하게 위치 지정할 수 있으며, 최대 10분 길이의 긴 비디오 파일을 이해하고 초 단위의 사건 시점을 위치 지정할 수 있습니다. 시간의 선후와 속도를 이해할 수 있으며, 분석 및 위치 지정 능력을 기반으로 OS 또는 모바일 에이전트를 조작할 수 있습니다. 주요 정보 추출 능력과 Json 형식 출력 능력이 뛰어나며, 이 버전은 72B 버전으로, 이 시리즈에서 가장 강력한 버전입니다."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "지시 따르기, 수학, 문제 해결, 코드 전반적인 향상, 모든 사물 인식 능력 향상, 다양한 형식의 시각적 요소를 직접 정확하게 위치 지정할 수 있으며, 최대 10분 길이의 긴 비디오 파일을 이해하고 초 단위의 사건 시점을 위치 지정할 수 있습니다. 시간의 선후와 속도를 이해할 수 있으며, 분석 및 위치 지정 능력을 기반으로 OS 또는 모바일 에이전트를 조작할 수 있습니다. 주요 정보 추출 능력과 Json 형식 출력 능력이 뛰어나며, 이 버전은 72B 버전으로, 이 시리즈에서 가장 강력한 버전입니다."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5는 Alibaba의 차세대 대규모 언어 모델로, 뛰어난 성능으로 다양한 응용 요구를 지원합니다."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro는 Upstage에서 출시한 고지능 LLM으로, 단일 GPU의 지시 추적 능력에 중점을 두고 있으며, IFEval 점수가 80 이상입니다. 현재 영어를 지원하며, 정식 버전은 2024년 11월에 출시될 예정이며, 언어 지원 및 컨텍스트 길이를 확장할 계획입니다."
},
+ "sonar": {
+ "description": "검색 맥락 기반의 경량 검색 제품으로, Sonar Pro보다 더 빠르고 저렴합니다."
+ },
+ "sonar-pro": {
+ "description": "고급 쿼리 및 후속 작업을 지원하는 검색 맥락 기반의 고급 검색 제품입니다."
+ },
+ "sonar-reasoning": {
+ "description": "DeepSeek 추론 모델이 지원하는 새로운 API 제품입니다."
+ },
"step-1-128k": {
"description": "성능과 비용의 균형을 맞추어 일반적인 시나리오에 적합합니다."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "이 모델은 강력한 비디오 이해 능력을 가지고 있습니다."
},
+ "step-1o-vision-32k": {
+ "description": "이 모델은 강력한 이미지 이해 능력을 가지고 있습니다. step-1v 시리즈 모델에 비해 더 강력한 시각 성능을 자랑합니다."
+ },
"step-1v-32k": {
"description": "시각 입력을 지원하여 다중 모달 상호작용 경험을 강화합니다."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "대규모 컨텍스트 상호작용을 지원하며, 복잡한 대화 시나리오에 적합합니다."
},
+ "step-2-mini": {
+ "description": "신세대 자체 개발 Attention 아키텍처인 MFA를 기반으로 한 초고속 대형 모델로, 매우 낮은 비용으로 step1과 유사한 효과를 달성하면서도 더 높은 처리량과 더 빠른 응답 지연을 유지합니다. 일반적인 작업을 처리할 수 있으며, 코드 능력에 있어 특장점을 가지고 있습니다."
+ },
+ "taichu2_mm": {
+ "description": "이미지 이해, 지식 이전, 논리 귀속 등의 능력을 통합하여, 이미지-텍스트 질문 응답 분야에서 뛰어난 성능을 보입니다."
+ },
"taichu_llm": {
"description": "자이동 태초 언어 대모델은 뛰어난 언어 이해 능력과 텍스트 창작, 지식 질문 응답, 코드 프로그래밍, 수학 계산, 논리 추론, 감정 분석, 텍스트 요약 등의 능력을 갖추고 있습니다. 혁신적으로 대규모 데이터 사전 훈련과 다원적 풍부한 지식을 결합하여 알고리즘 기술을 지속적으로 다듬고, 방대한 텍스트 데이터에서 어휘, 구조, 문법, 의미 등의 새로운 지식을 지속적으로 흡수하여 모델 성능을 지속적으로 진화시킵니다. 사용자에게 보다 편리한 정보와 서비스, 그리고 더 지능적인 경험을 제공합니다."
},
+ "text-embedding-3-large": {
+ "description": "가장 강력한 벡터화 모델로, 영어 및 비영어 작업에 적합합니다."
+ },
+ "text-embedding-3-small": {
+ "description": "효율적이고 경제적인 차세대 임베딩 모델로, 지식 검색, RAG 애플리케이션 등 다양한 상황에 적합합니다."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B)는 효율적인 전략과 모델 아키텍처를 통해 향상된 계산 능력을 제공합니다."
},
+ "tts-1": {
+ "description": "최신 텍스트 음성 변환 모델로, 실시간 상황에 최적화된 속도를 제공합니다."
+ },
+ "tts-1-hd": {
+ "description": "최신 텍스트 음성 변환 모델로, 품질을 최적화했습니다."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B)는 세밀한 지시 작업에 적합하며, 뛰어난 언어 처리 능력을 제공합니다."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet는 업계 표준을 향상시켰으며, 경쟁 모델과 Claude 3 Opus를 초월하는 성능을 보여주고, 광범위한 평가에서 뛰어난 성과를 보이며, 중간 수준 모델의 속도와 비용을 갖추고 있습니다."
},
+ "whisper-1": {
+ "description": "범용 음성 인식 모델로, 다국어 음성 인식, 음성 번역 및 언어 인식을 지원합니다."
+ },
"wizardlm2": {
"description": "WizardLM 2는 Microsoft AI에서 제공하는 언어 모델로, 복잡한 대화, 다국어, 추론 및 스마트 어시스턴트 분야에서 특히 뛰어난 성능을 발휘합니다."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "복잡한 시각 작업 모델로, 고성능 이미지 이해 및 분석 능력을 제공합니다."
+ },
+ "yi-vision-v2": {
+ "description": "복잡한 시각적 작업 모델로, 여러 이미지를 기반으로 한 고성능 이해 및 분석 능력을 제공합니다."
}
}
diff --git a/locales/ko-KR/providers.json b/locales/ko-KR/providers.json
index 303489e498a6e..cc7103042d6b1 100644
--- a/locales/ko-KR/providers.json
+++ b/locales/ko-KR/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure는 GPT-3.5 및 최신 GPT-4 시리즈를 포함한 다양한 고급 AI 모델을 제공하며, 다양한 데이터 유형과 복잡한 작업을 지원하고 안전하고 신뢰할 수 있으며 지속 가능한 AI 솔루션을 목표로 하고 있습니다."
},
+ "azureai": {
+ "description": "Azure는 GPT-3.5 및 최신 GPT-4 시리즈를 포함한 다양한 고급 AI 모델을 제공하며, 다양한 데이터 유형과 복잡한 작업을 지원하고 안전하고 신뢰할 수 있으며 지속 가능한 AI 솔루션을 위해 노력합니다."
+ },
"baichuan": {
"description": "百川智能은 인공지능 대형 모델 연구 개발에 집중하는 회사로, 그 모델은 국내 지식 백과, 긴 텍스트 처리 및 생성 창작 등 중국어 작업에서 뛰어난 성능을 보이며, 해외 주류 모델을 초월합니다. 百川智能은 업계 선도적인 다중 모드 능력을 갖추고 있으며, 여러 권위 있는 평가에서 우수한 성능을 보였습니다. 그 모델에는 Baichuan 4, Baichuan 3 Turbo 및 Baichuan 3 Turbo 128k 등이 포함되어 있으며, 각각 다른 응용 시나리오에 최적화되어 비용 효율적인 솔루션을 제공합니다."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek는 인공지능 기술 연구 및 응용에 집중하는 회사로, 최신 모델인 DeepSeek-V2.5는 일반 대화 및 코드 처리 능력을 통합하고 인간의 선호 정렬, 작문 작업 및 지시 따르기 등에서 상당한 향상을 이루었습니다."
},
+ "doubao": {
+ "description": "바이트댄스가 개발한 자체 대형 모델입니다. 바이트댄스 내부의 50개 이상의 비즈니스 시나리오에서 검증되었으며, 매일 수조 개의 토큰 사용량을 지속적으로 다듬어 다양한 모드 기능을 제공하여 우수한 모델 효과로 기업에 풍부한 비즈니스 경험을 제공합니다."
+ },
"fireworksai": {
"description": "Fireworks AI는 기능 호출 및 다중 모드 처리를 전문으로 하는 선도적인 고급 언어 모델 서비스 제공업체입니다. 최신 모델인 Firefunction V2는 Llama-3를 기반으로 하며, 함수 호출, 대화 및 지시 따르기에 최적화되어 있습니다. 비주얼 언어 모델인 FireLLaVA-13B는 이미지와 텍스트 혼합 입력을 지원합니다. 기타 주목할 만한 모델로는 Llama 시리즈와 Mixtral 시리즈가 있으며, 효율적인 다국어 지시 따르기 및 생성 지원을 제공합니다."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "대규모 모델 연구 및 개발 도구 체인에 전념하는 오픈 소스 조직입니다. 모든 AI 개발자에게 효율적이고 사용하기 쉬운 오픈 소스 플랫폼을 제공하여 최첨단 대규모 모델 및 알고리즘 기술을 손쉽게 이용할 수 있도록 합니다."
},
+ "jina": {
+ "description": "Jina AI는 2020년에 설립된 선도적인 검색 AI 회사입니다. 우리의 검색 기반 플랫폼은 기업이 신뢰할 수 있고 고품질의 생성적 AI 및 다중 모드 검색 애플리케이션을 구축할 수 있도록 돕는 벡터 모델, 재배치기 및 소형 언어 모델을 포함하고 있습니다."
+ },
+ "lmstudio": {
+ "description": "LM Studio는 귀하의 컴퓨터에서 LLM을 개발하고 실험하기 위한 데스크탑 애플리케이션입니다."
+ },
"minimax": {
"description": "MiniMax는 2021년에 설립된 일반 인공지능 기술 회사로, 사용자와 함께 지능을 공동 창출하는 데 전념하고 있습니다. MiniMax는 다양한 모드의 일반 대형 모델을 독자적으로 개발하였으며, 여기에는 조 단위의 MoE 텍스트 대형 모델, 음성 대형 모델 및 이미지 대형 모델이 포함됩니다. 또한 해마 AI와 같은 응용 프로그램을 출시하였습니다."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI는 다양한 대형 언어 모델과 AI 이미지 생성을 제공하는 API 서비스 플랫폼으로, 유연하고 신뢰할 수 있으며 비용 효율적입니다. Llama3, Mistral 등 최신 오픈 소스 모델을 지원하며, 생성적 AI 응용 프로그램 개발을 위한 포괄적이고 사용자 친화적이며 자동 확장 가능한 API 솔루션을 제공하여 AI 스타트업의 빠른 발전에 적합합니다."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™은 클라우드, 데이터 센터, RTX™ AI 개인용 컴퓨터 및 워크스테이션에서 사전 훈련된 AI 모델과 사용자 정의 AI 모델을 배포할 수 있도록 지원하는 컨테이너를 제공합니다."
+ },
"ollama": {
"description": "Ollama가 제공하는 모델은 코드 생성, 수학 연산, 다국어 처리 및 대화 상호작용 등 다양한 분야를 포괄하며, 기업급 및 로컬 배포의 다양한 요구를 지원합니다."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "중국과학원 자동화 연구소와 우한 인공지능 연구원이 출시한 차세대 다중 모드 대형 모델은 다중 회차 질문 응답, 텍스트 창작, 이미지 생성, 3D 이해, 신호 분석 등 포괄적인 질문 응답 작업을 지원하며, 더 강력한 인지, 이해 및 창작 능력을 갖추고 있어 새로운 상호작용 경험을 제공합니다."
},
+ "tencentcloud": {
+ "description": "지식 엔진 원자 능력(LLM Knowledge Engine Atomic Power)은 지식 엔진을 기반으로 개발된 지식 질문 응답의 전체 링크 능력으로, 기업 및 개발자를 대상으로 하여 유연한 모델 응용 프로그램 구성 및 개발 능력을 제공합니다. 여러 원자 능력을 통해 귀하만의 모델 서비스를 구성하고, 문서 분석, 분할, 임베딩, 다중 회차 수정 등의 서비스를 호출하여 조합하여 기업 전용 AI 비즈니스를 맞춤화할 수 있습니다."
+ },
"togetherai": {
"description": "Together AI는 혁신적인 AI 모델을 통해 선도적인 성능을 달성하는 데 전념하며, 빠른 확장 지원 및 직관적인 배포 프로세스를 포함한 광범위한 사용자 정의 기능을 제공하여 기업의 다양한 요구를 충족합니다."
},
"upstage": {
"description": "Upstage는 Solar LLM 및 문서 AI를 포함하여 다양한 비즈니스 요구를 위한 AI 모델 개발에 집중하고 있으며, 인공지능 일반 지능(AGI)을 실현하는 것을 목표로 하고 있습니다. Chat API를 통해 간단한 대화 에이전트를 생성하고 기능 호출, 번역, 임베딩 및 특정 분야 응용 프로그램을 지원합니다."
},
+ "vllm": {
+ "description": "vLLM은 LLM 추론 및 서비스를 위한 빠르고 사용하기 쉬운 라이브러리입니다."
+ },
+ "volcengine": {
+ "description": "바이트댄스가 출시한 대형 모델 서비스 개발 플랫폼으로, 기능이 풍부하고 안전하며 가격 경쟁력이 있는 모델 호출 서비스를 제공합니다. 또한 모델 데이터, 세밀 조정, 추론, 평가 등 엔드 투 엔드 기능을 제공하여 귀하의 AI 애플리케이션 개발을 전방위적으로 지원합니다."
+ },
"wenxin": {
"description": "기업용 원스톱 대형 모델 및 AI 네이티브 애플리케이션 개발 및 서비스 플랫폼으로, 가장 포괄적이고 사용하기 쉬운 생성적 인공지능 모델 개발 및 애플리케이션 개발 전체 프로세스 도구 체인을 제공합니다."
},
diff --git a/locales/ko-KR/setting.json b/locales/ko-KR/setting.json
index 5425cd642af19..6456716055009 100644
--- a/locales/ko-KR/setting.json
+++ b/locales/ko-KR/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "단일 응답 제한 활성화"
},
+ "enableReasoningEffort": {
+ "title": "추론 강도 조정 활성화"
+ },
"frequencyPenalty": {
- "desc": "값이 클수록 반복 단어가 줄어듭니다",
- "title": "빈도 패널티"
+ "desc": "값이 클수록 단어 선택이 더 다양하고 풍부해지며, 값이 작을수록 단어 선택이 더 간단하고 소박해집니다.",
+ "title": "어휘 다양성"
},
"maxTokens": {
"desc": "단일 상호 작용에 사용되는 최대 토큰 수",
@@ -212,19 +215,31 @@
"desc": "{{provider}} 모델",
"title": "모델"
},
+ "params": {
+ "title": "고급 매개변수"
+ },
"presencePenalty": {
- "desc": "값이 클수록 새로운 주제로 확장될 가능성이 높아집니다",
- "title": "주제 신선도"
+ "desc": "값이 클수록 다양한 표현 방식으로 기울어져 개념의 반복을 피하고, 값이 작을수록 반복적인 개념이나 서술을 사용하는 경향이 있어 표현이 더 일관됩니다.",
+ "title": "표현의 다양성"
+ },
+ "reasoningEffort": {
+ "desc": "값이 클수록 추론 능력이 강해지지만, 응답 시간과 토큰 소모가 증가할 수 있습니다.",
+ "options": {
+ "high": "높음",
+ "low": "낮음",
+ "medium": "중간"
+ },
+ "title": "추론 강도"
},
"temperature": {
- "desc": "값이 클수록 응답이 더 무작위해집니다",
- "title": "랜덤성",
- "titleWithValue": "랜덤성 {{value}}"
+ "desc": "값이 클수록 답변이 더 창의적이고 상상력이 풍부해지며, 값이 작을수록 답변이 더 엄격해집니다.",
+ "title": "창의성 활성화",
+ "warning": "창의성 활성화 값이 너무 크면 출력이 깨질 수 있습니다."
},
"title": "모델 설정",
"topP": {
- "desc": "랜덤성과 유사하지만 함께 변경하지 마세요",
- "title": "상위 P 샘플링"
+ "desc": "얼마나 많은 가능성을 고려할지, 값이 클수록 더 많은 가능성 있는 답변을 수용하고, 값이 작을수록 가장 가능성이 높은 답변을 선택하는 경향이 있습니다. 창의성 활성화와 함께 변경하는 것은 권장하지 않습니다.",
+ "title": "사고 개방성"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "일반 설정",
"experiment": "실험",
"llm": "언어 모델",
+ "provider": "AI 서비스 제공자",
"sync": "클라우드 동기화",
"system-agent": "시스템 도우미",
"tts": "음성 서비스"
diff --git a/locales/nl-NL/changelog.json b/locales/nl-NL/changelog.json
index 0076f83087cca..24027e61886c2 100644
--- a/locales/nl-NL/changelog.json
+++ b/locales/nl-NL/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "Bekijk alle changelogs",
"description": "Blijf op de hoogte van nieuwe functies en verbeteringen van {{appName}}",
"pagination": {
- "older": "Bekijk eerdere wijzigingen",
- "prev": "Vorige pagina"
+ "next": "Volgende pagina",
+ "older": "Bekijk eerdere wijzigingen"
},
"readDetails": "Lees meer",
"title": "Changelog",
diff --git a/locales/nl-NL/common.json b/locales/nl-NL/common.json
index ec486adb3ba89..b06a3a3474dbf 100644
--- a/locales/nl-NL/common.json
+++ b/locales/nl-NL/common.json
@@ -275,6 +275,7 @@
},
"temp": "tijdelijk",
"terms": "algemene voorwaarden",
+ "update": "Bijwerken",
"updateAgent": "update assistent",
"upgradeVersion": {
"action": "upgraden",
@@ -285,6 +286,7 @@
"anonymousNickName": "anonieme gebruiker",
"billing": "facturatie",
"cloud": "Ervaar {{name}}",
+ "community": "Gemeenschapsversie",
"data": "gegevensopslag",
"defaultNickname": "communitygebruiker",
"discord": "communityondersteuning",
@@ -294,7 +296,6 @@
"help": "helpcentrum",
"moveGuide": "instellingen verplaatst naar hier",
"plans": "abonnementen",
- "preview": "voorbeeldversie",
"profile": "accountbeheer",
"setting": "app-instellingen",
"usages": "gebruiksstatistieken"
diff --git a/locales/nl-NL/components.json b/locales/nl-NL/components.json
index cffd157f3fde6..71913d6d3cfd5 100644
--- a/locales/nl-NL/components.json
+++ b/locales/nl-NL/components.json
@@ -76,6 +76,7 @@
"custom": "Custom model, by default, supports both function call and visual recognition. Please verify the availability of the above capabilities based on actual needs.",
"file": "This model supports file upload for reading and recognition.",
"functionCall": "This model supports function call.",
+ "reasoning": "Dit model ondersteunt diepgaand denken",
"tokens": "This model supports up to {{tokens}} tokens in a single session.",
"vision": "This model supports visual recognition."
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "No enabled model, please go to settings to enable.",
"provider": "Provider"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "Vanwege beveiligingsbeperkingen in de browser moet je cross-origin configuratie voor Ollama instellen om het correct te kunnen gebruiken.",
+ "linux": {
+ "env": "Voeg `Environment` toe onder de [Service] sectie en voeg de OLLAMA_ORIGINS omgevingsvariabele toe:",
+ "reboot": "Herlaad systemd en herstart Ollama",
+ "systemd": "Roep systemd aan om de ollama service te bewerken:"
+ },
+ "macos": "Open de 'Terminal' applicatie, plak de volgende opdracht en druk op enter om uit te voeren",
+ "reboot": "Herstart de Ollama service na het voltooien van de uitvoering",
+ "title": "Configureer Ollama voor cross-origin toegang",
+ "windows": "Op Windows, klik op 'Configuratiescherm', ga naar systeemomgevingsvariabelen bewerken. Maak een nieuwe omgevingsvariabele aan met de naam 'OLLAMA_ORIGINS' voor je gebruikersaccount, met de waarde * en klik op 'OK/Toepassen' om op te slaan"
+ },
+ "install": {
+ "description": "Zorg ervoor dat je Ollama hebt ingeschakeld. Als je Ollama nog niet hebt gedownload, ga dan naar de officiële website <1>om te downloaden",
+ "docker": "Als je de voorkeur geeft aan het gebruik van Docker, biedt Ollama ook een officiële Docker-image aan die je kunt ophalen met de volgende opdracht:",
+ "linux": {
+ "command": "Installeer met de volgende opdracht:",
+ "manual": "Of je kunt de <1>Linux handmatige installatiehandleiding raadplegen voor een handmatige installatie"
+ },
+ "title": "Installeer en start de Ollama applicatie lokaal",
+ "windowsTab": "Windows (previewversie)"
+ }
+ },
+ "Thinking": {
+ "thinking": "Diep in gedachten...",
+ "thought": "Diep nagedacht (tijd gebruikt {{duration}} seconden)",
+ "thoughtWithDuration": "Diep nagedacht"
}
}
diff --git a/locales/nl-NL/discover.json b/locales/nl-NL/discover.json
index 5ff3bd314579b..c7f34b75ee749 100644
--- a/locales/nl-NL/discover.json
+++ b/locales/nl-NL/discover.json
@@ -126,6 +126,10 @@
"title": "Onderwerp versheid"
},
"range": "Bereik",
+ "reasoning_effort": {
+ "desc": "Deze instelling wordt gebruikt om de redeneerkracht van het model te regelen voordat het een antwoord genereert. Lage kracht geeft prioriteit aan de responssnelheid en bespaart tokens, terwijl hoge kracht een completere redenering biedt, maar meer tokens verbruikt en de responssnelheid verlaagt. De standaardwaarde is gemiddeld, wat een balans biedt tussen redeneringsnauwkeurigheid en responssnelheid.",
+ "title": "Redeneerkracht"
+ },
"temperature": {
"desc": "Deze instelling beïnvloedt de diversiteit van de reacties van het model. Lagere waarden leiden tot meer voorspelbare en typische reacties, terwijl hogere waarden meer diverse en ongebruikelijke reacties aanmoedigen. Wanneer de waarde op 0 is ingesteld, geeft het model altijd dezelfde reactie op een gegeven invoer.",
"title": "Willekeurigheid"
diff --git a/locales/nl-NL/modelProvider.json b/locales/nl-NL/modelProvider.json
index 8ed9933a477bc..493d709e1014e 100644
--- a/locales/nl-NL/modelProvider.json
+++ b/locales/nl-NL/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API Key"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "De API-versie van Azure, volgens het formaat YYYY-MM-DD. Raadpleeg de [laatste versie](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "Lijst ophalen",
+ "title": "Azure API-versie"
+ },
+ "endpoint": {
+ "desc": "Vind het Azure AI-model inferentie-eindpunt in het overzicht van het Azure AI-project",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Azure AI-eindpunt"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Vind de API-sleutel in het overzicht van het Azure AI-project",
+ "placeholder": "Azure-sleutel",
+ "title": "Sleutel"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "Voer AWS Access Key Id in",
@@ -63,6 +81,46 @@
"title": "Cloudflare-account ID / API-URL"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Vul je API-sleutel in",
+ "title": "API-sleutel"
+ },
+ "basicTitle": "Basisinformatie",
+ "configTitle": "Configuratie-informatie",
+ "confirm": "Nieuw aanmaken",
+ "createSuccess": "Succesvol aangemaakt",
+ "description": {
+ "placeholder": "Beschrijving van de provider (optioneel)",
+ "title": "Beschrijving van de provider"
+ },
+ "id": {
+ "desc": "Een unieke identificatie voor de dienstverlener, kan na creatie niet meer worden gewijzigd",
+ "format": "Mag alleen cijfers, kleine letters, koppeltekens (-) en onderstrepingstekens (_) bevatten",
+ "placeholder": "Gebruik alleen kleine letters, bijvoorbeeld openai, kan niet worden gewijzigd na aanmaak",
+ "required": "Vul de provider ID in",
+ "title": "Provider ID"
+ },
+ "logo": {
+ "required": "Upload een correcte provider-logo",
+ "title": "Provider-logo"
+ },
+ "name": {
+ "placeholder": "Voer de weergavenaam van de provider in",
+ "required": "Vul de naam van de provider in",
+ "title": "Naam van de provider"
+ },
+ "proxyUrl": {
+ "required": "Vul het proxyadres in",
+ "title": "Proxy-adres"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Selecteer het SDK-type",
+ "title": "Aanvraagformaat"
+ },
+ "title": "Maak een aangepaste AI-provider"
+ },
"github": {
"personalAccessToken": {
"desc": "Vul je Github PAT in, klik [hier](https://github.com/settings/tokens) om er een te maken",
@@ -77,6 +135,23 @@
"title": "HuggingFace Token"
}
},
+ "list": {
+ "title": {
+ "disabled": "Dienstverlener niet ingeschakeld",
+ "enabled": "Dienstverlener ingeschakeld"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Voeg aangepaste provider toe",
+ "all": "Alles",
+ "list": {
+ "disabled": "Niet ingeschakeld",
+ "enabled": "Ingeschakeld"
+ },
+ "notFound": "Geen zoekresultaten gevonden",
+ "searchProviders": "Zoek providers...",
+ "sort": "Aangepaste sortering"
+ },
"ollama": {
"checker": {
"desc": "Test of het proxyadres correct is ingevuld",
@@ -94,33 +169,9 @@
"title": "Model {{model}} wordt gedownload"
},
"endpoint": {
- "desc": "Voer het Ollama interface proxyadres in, laat leeg indien niet specifiek aangegeven",
+ "desc": "Moet http(s):// bevatten, kan leeg gelaten worden als lokaal niet specifiek opgegeven",
"title": "Interface Proxyadres"
},
- "setup": {
- "cors": {
- "description": "Vanwege beveiligingsbeperkingen in de browser moet je CORS-configuratie voor Ollama instellen om het correct te kunnen gebruiken.",
- "linux": {
- "env": "Voeg `Environment` toe onder de [Service] sectie en voeg de OLLAMA_ORIGINS omgevingsvariabele toe:",
- "reboot": "Herlaad systemd en herstart Ollama",
- "systemd": "Roep systemd aan om de ollama service te bewerken:"
- },
- "macos": "Open de 'Terminal' applicatie, plak de volgende opdracht en druk op enter om uit te voeren",
- "reboot": "Herstart de Ollama service na voltooiing",
- "title": "Configureer Ollama voor cross-origin toegang",
- "windows": "Op Windows, klik op 'Configuratiescherm', ga naar systeemomgevingsvariabelen bewerken. Maak een nieuwe omgevingsvariabele met de naam 'OLLAMA_ORIGINS' voor je gebruikersaccount, met de waarde *, en klik op 'OK/Toepassen' om op te slaan"
- },
- "install": {
- "description": "Zorg ervoor dat je Ollama hebt ingeschakeld. Als je Ollama nog niet hebt gedownload, ga dan naar de officiële website <1>om te downloaden",
- "docker": "Als je de voorkeur geeft aan Docker, biedt Ollama ook een officiële Docker-image aan, die je kunt ophalen met de volgende opdracht:",
- "linux": {
- "command": "Installeer met de volgende opdracht:",
- "manual": "Of je kunt de <1>Linux handmatige installatiehandleiding raadplegen voor een handmatige installatie"
- },
- "title": "Installeer en start de Ollama applicatie lokaal",
- "windowsTab": "Windows (previewversie)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Annuleer download",
@@ -131,25 +182,145 @@
"title": "Download het opgegeven Ollama model"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Vul de Access Key van het Baidu Qianfan-platform in",
- "placeholder": "Qianfan Access Key",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "Je sleutel en proxy-adres worden versleuteld met <1>AES-GCM encryptie-algoritme",
+ "apiKey": {
+ "desc": "Vul je {{name}} API-sleutel in",
+ "placeholder": "{{name}} API-sleutel",
+ "title": "API-sleutel"
+ },
+ "baseURL": {
+ "desc": "Moet http(s):// bevatten",
+ "invalid": "Voer een geldige URL in",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "API-proxy-adres"
+ },
+ "checker": {
+ "button": "Controleer",
+ "desc": "Test of de API-sleutel en proxy-adres correct zijn ingevuld",
+ "pass": "Controle geslaagd",
+ "title": "Connectiviteitstest"
+ },
+ "fetchOnClient": {
+ "desc": "Clientaanvraagmodus zal sessieaanvragen rechtstreeks vanuit de browser initiëren, wat de responssnelheid kan verbeteren",
+ "title": "Gebruik clientaanvraagmodus"
+ },
+ "helpDoc": "Configuratiehandleiding",
+ "waitingForMore": "Meer modellen zijn in <1>planning voor integratie, blijf op de hoogte"
},
- "checker": {
- "desc": "Test of de AccessKey / SecretAccess correct is ingevuld"
+ "createNew": {
+ "title": "Maak een aangepast AI-model"
+ },
+ "item": {
+ "config": "Configureer model",
+ "customModelCards": {
+ "addNew": "Maak en voeg {{id}} model toe",
+ "confirmDelete": "Je staat op het punt dit aangepaste model te verwijderen, na verwijdering kan het niet worden hersteld, wees voorzichtig."
+ },
+ "delete": {
+ "confirm": "Bevestig verwijdering van model {{displayName}}?",
+ "success": "Verwijdering geslaagd",
+ "title": "Verwijder model"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Het veld dat daadwerkelijk wordt aangevraagd in Azure OpenAI",
+ "placeholder": "Voer de modelimplementatienaam in Azure in",
+ "title": "Modelimplementatienaam"
+ },
+ "deployName": {
+ "extra": "Dit veld wordt als model-ID verzonden bij het indienen van een verzoek",
+ "placeholder": "Voer de naam of ID van het daadwerkelijk gedeployde model in",
+ "title": "Modeldeploynaam"
+ },
+ "displayName": {
+ "placeholder": "Voer de weergavenaam van het model in, bijvoorbeeld ChatGPT, GPT-4, enz.",
+ "title": "Weergavenaam van het model"
+ },
+ "files": {
+ "extra": "De huidige bestandsuploadimplementatie is slechts een hackoplossing, alleen voor eigen gebruik. Volledige bestandsuploadcapaciteit komt later beschikbaar.",
+ "title": "Ondersteuning voor bestandsupload"
+ },
+ "functionCall": {
+ "extra": "Deze configuratie schakelt alleen de mogelijkheid in voor het model om tools te gebruiken, waardoor het mogelijk is om plug-ins voor tools aan het model toe te voegen. Of het model daadwerkelijk tools kan gebruiken, hangt echter volledig af van het model zelf; test de bruikbaarheid zelf.",
+ "title": "Ondersteuning voor het gebruik van tools"
+ },
+ "id": {
+ "extra": "Kan niet worden gewijzigd na creatie, wordt gebruikt als model-id bij het aanroepen van AI",
+ "placeholder": "Voer model-id in, bijvoorbeeld gpt-4o of claude-3.5-sonnet",
+ "title": "Model ID"
+ },
+ "modalTitle": "Configuratie van aangepast model",
+ "reasoning": {
+ "extra": "Deze configuratie schakelt alleen de mogelijkheid voor diepgaand denken van het model in. Het specifieke effect hangt volledig af van het model zelf, test zelf of dit model in staat is tot bruikbaar diepgaand denken.",
+ "title": "Ondersteuning voor diepgaand denken"
+ },
+ "tokens": {
+ "extra": "Stel het maximale aantal tokens in dat door het model wordt ondersteund",
+ "title": "Maximale contextvenster",
+ "unlimited": "Onbeperkt"
+ },
+ "vision": {
+ "extra": "Deze configuratie zal alleen de afbeeldinguploadcapaciteit in de applicatie inschakelen, of herkenning wordt ondersteund hangt volledig af van het model zelf, test de beschikbaarheid van de visuele herkenningscapaciteit van dit model zelf.",
+ "title": "Ondersteuning voor visuele herkenning"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/Afbeelding",
+ "inputCharts": "${{amount}}/M Tekens",
+ "inputMinutes": "${{amount}}/Minuten",
+ "inputTokens": "Invoer ${{amount}}/M",
+ "outputTokens": "Uitvoer ${{amount}}/M"
+ },
+ "releasedAt": "Uitgebracht op {{releasedAt}}"
},
- "secretKey": {
- "desc": "Vul de Secret Key van het Baidu Qianfan-platform in",
- "placeholder": "Qianfan Secret Key",
- "title": "Secret Key"
+ "list": {
+ "addNew": "Model toevoegen",
+ "disabled": "Niet ingeschakeld",
+ "disabledActions": {
+ "showMore": "Toon alles"
+ },
+ "empty": {
+ "desc": "Maak een aangepast model of haal een model op om te beginnen met gebruiken.",
+ "title": "Geen beschikbare modellen"
+ },
+ "enabled": "Ingeschakeld",
+ "enabledActions": {
+ "disableAll": "Alle uitschakelen",
+ "enableAll": "Alle inschakelen",
+ "sort": "Aangepaste model sortering"
+ },
+ "enabledEmpty": "Geen ingeschakelde modellen, schakel de modellen hieronder in die je leuk vindt~",
+ "fetcher": {
+ "clear": "Verwijder de opgehaalde modellen",
+ "fetch": "Haal modellenlijst op",
+ "fetching": "Bezig met het ophalen van de modellenlijst...",
+ "latestTime": "Laatste update tijd: {{time}}",
+ "noLatestTime": "Lijst nog niet opgehaald"
+ },
+ "resetAll": {
+ "conform": "Weet je zeker dat je alle wijzigingen van het huidige model wilt resetten? Na de reset zal de huidige modellenlijst terugkeren naar de standaardstatus",
+ "success": "Resetten geslaagd",
+ "title": "Reset alle wijzigingen"
+ },
+ "search": "Zoek modellen...",
+ "searchResult": "Gevonden {{count}} modellen",
+ "title": "Modellenlijst",
+ "total": "In totaal {{count}} modellen beschikbaar"
},
- "unlock": {
- "customRegion": "Aangepaste servicegebied",
- "description": "Voer je AccessKey / SecretKey in om de sessie te starten. De applicatie zal je authenticatie-instellingen niet opslaan",
- "title": "Gebruik aangepaste Wenxin Yiyan-authenticatie-informatie"
- }
+ "searchNotFound": "Geen zoekresultaten gevonden"
+ },
+ "sortModal": {
+ "success": "Sortering succesvol bijgewerkt",
+ "title": "Aangepaste sortering",
+ "update": "Bijwerken"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "Je staat op het punt deze AI-provider te verwijderen, na verwijdering kan deze niet worden hersteld, bevestig je verwijdering?",
+ "deleteSuccess": "Verwijdering geslaagd",
+ "tooltip": "Werk basisconfiguratie van provider bij",
+ "updateSuccess": "Bijwerking geslaagd"
},
"zeroone": {
"title": "01.AI Nul Een Alles"
diff --git a/locales/nl-NL/models.json b/locales/nl-NL/models.json
index bd70e7bdffb07..226fded413491 100644
--- a/locales/nl-NL/models.json
+++ b/locales/nl-NL/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K legt de nadruk op semantische veiligheid en verantwoordelijkheid, speciaal ontworpen voor toepassingen met hoge eisen aan inhoudsveiligheid, en zorgt voor nauwkeurigheid en robuustheid in de gebruikerservaring."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 bouwt denkketens op met behulp van boomzoekmethoden en introduceert een reflectiemechanisme, getraind met versterkend leren, waardoor het model in staat is tot zelfreflectie en foutcorrectie."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro is een geavanceerd natuurlijk taalverwerkingsmodel dat is ontwikkeld door 360, met uitstekende tekstgeneratie- en begripcapaciteiten, vooral in de generatieve en creatieve domeinen, en kan complexe taaltransformaties en rolinterpretatietaken aan."
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1 bouwt een denkketen op met behulp van boomzoekmethoden en introduceert een reflectiemechanisme, waarbij het gebruik maakt van versterkend leren om het model in staat te stellen tot zelfreflectie en foutcorrectie."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra is de krachtigste versie in de Spark-grootmodelserie, die de netwerkintegratie heeft geüpgraded en de tekstbegrip- en samenvattingscapaciteiten heeft verbeterd. Het is een allesomvattende oplossing voor het verbeteren van de kantoorproductiviteit en het nauwkeurig reageren op behoeften, en is een toonaangevend intelligent product in de industrie."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "Modelcapaciteiten zijn nationaal de beste, overtreft buitenlandse mainstream modellen in kennisencyclopedie, lange teksten en creatieve generatie in Chinese taken. Beschikt ook over toonaangevende multimodale capaciteiten en presteert uitstekend op verschillende autoritatieve evaluatiebenchmarks."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "DeepSeek-R1 distillatiemodel gebaseerd op Qwen2.5-Math-1.5B, geoptimaliseerd voor inferentieprestaties door versterkend leren en koude startdata, open-source model dat de multi-taak benchmark vernieuwt."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "DeepSeek-R1 distillatiemodel gebaseerd op Qwen2.5-14B, geoptimaliseerd voor inferentieprestaties door versterkend leren en koude startdata, open-source model dat de multi-taak benchmark vernieuwt."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "De DeepSeek-R1 serie optimaliseert inferentieprestaties door versterkend leren en koude startdata, open-source model dat de multi-taak benchmark vernieuwt en de OpenAI-o1-mini niveaus overtreft."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1 distillatiemodel gebaseerd op Qwen2.5-Math-7B, geoptimaliseerd voor inferentieprestaties door versterkend leren en koude startdata, open-source model dat de multi-taak benchmark vernieuwt."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite beschikt over een uitstekende responssnelheid en een goede prijs-kwaliteitverhouding, en biedt klanten flexibele keuzes voor verschillende scenario's. Ondersteunt inferentie en fine-tuning met een contextvenster van 128k."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Het door Baidu ontwikkelde vlaggenschip van een ultra-groot taalmodel, dat in vergelijking met ERNIE 3.5 een algehele upgrade van de modelcapaciteiten heeft gerealiseerd, en breed toepasbaar is in complexe taken in verschillende domeinen; ondersteunt automatische integratie met de Baidu-zoekplug-in om de actualiteit van vraag- en antwoordinformatie te waarborgen."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Het vlaggenschip supergrote taalmodel van Baidu, met uitstekende algehele prestaties, breed toepasbaar in complexe taakscenario's in verschillende domeinen; ondersteunt automatische integratie met Baidu zoekplug-ins, wat de actualiteit van vraag- en antwoordinformatie waarborgt. In vergelijking met ERNIE 4.0 presteert het beter."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "De zelfontwikkelde vlaggenschip super-grote taalmodel van Baidu, dat uitmuntend presteert in diverse complexe taakscenario's in verschillende domeinen; ondersteunt automatische integratie met de Baidu-zoekplug-in, waarborgt de actualiteit van vraag-antwoordinformatie. Overtreft in performance ten opzichte van ERNIE 4.0."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct is een van de nieuwste grote taalmodellen die door Alibaba Cloud is uitgebracht. Dit 7B-model heeft aanzienlijke verbeteringen in coderings- en wiskundige vaardigheden. Het model biedt ook meertalige ondersteuning, met meer dan 29 ondersteunde talen, waaronder Chinees en Engels. Het model heeft aanzienlijke verbeteringen in het volgen van instructies, het begrijpen van gestructureerde gegevens en het genereren van gestructureerde uitvoer (vooral JSON)."
},
+ "MiniMax-Text-01": {
+ "description": "In de MiniMax-01-serie modellen hebben we gedurfde innovaties doorgevoerd: voor het eerst op grote schaal een lineaire aandachtmechanisme geïmplementeerd, waardoor de traditionele Transformer-architectuur niet langer de enige keuze is. Dit model heeft een parameterhoeveelheid van maar liefst 456 miljard, met een enkele activatie van 45,9 miljard. De algehele prestaties van het model zijn vergelijkbaar met die van de beste modellen in het buitenland, terwijl het efficiënt de wereldwijd langste context van 4 miljoen tokens kan verwerken, wat 32 keer de capaciteit van GPT-4o en 20 keer die van Claude-3.5-Sonnet is."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO is een zeer flexibele multi-model combinatie, ontworpen om een uitstekende creatieve ervaring te bieden."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 toont uitstekende prestaties bij diverse visuele taaltaken, waaronder document- en grafiekbegrip, scène-tekstbegrip, OCR, en het oplossen van wetenschappelijke en wiskundige problemen."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 toont uitstekende prestaties bij diverse visuele taaltaken, waaronder document- en grafiekbegrip, scène-tekstbegrip, OCR, en het oplossen van wetenschappelijke en wiskundige problemen."
- },
"Phi-3-medium-128k-instruct": {
"description": "Hetzelfde Phi-3-medium model, maar met een grotere contextgrootte voor RAG of few shot prompting."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat is de open-source versie van het GLM-4-serie voorgetrainde model, gelanceerd door Zhipu AI. Dit model presteert uitstekend in semantiek, wiskunde, redenering, code en kennis. Naast ondersteuning voor meerdaagse gesprekken, beschikt GLM-4-9B-Chat ook over geavanceerde functies zoals webbrowser, code-uitvoering, aangepaste tool-aanroepen (Function Call) en lange tekstredenering. Het model ondersteunt 26 talen, waaronder Chinees, Engels, Japans, Koreaans en Duits. In verschillende benchmarktests toont GLM-4-9B-Chat uitstekende prestaties, zoals AlignBench-v2, MT-Bench, MMLU en C-Eval. Dit model ondersteunt een maximale contextlengte van 128K, geschikt voor academisch onderzoek en commerciële toepassingen."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 is een inferentiemodel aangedreven door versterkend leren (RL), dat de problemen van herhaling en leesbaarheid in modellen aanpakt. Voor RL introduceert DeepSeek-R1 koude startdata, wat de inferentieprestaties verder optimaliseert. Het presteert vergelijkbaar met OpenAI-o1 in wiskunde, code en inferentietaken, en verbetert de algehele effectiviteit door zorgvuldig ontworpen trainingsmethoden."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 is een hybride expert (MoE) taalmodel met 6710 miljard parameters, dat gebruikmaakt van multi-head latent attention (MLA) en de DeepSeekMoE-architectuur, gecombineerd met een load balancing-strategie zonder extra verlies, om de inferentie- en trainingsefficiëntie te optimaliseren. Door voorgetraind te worden op 14,8 biljoen hoogwaardige tokens en vervolgens te worden fijngesteld met supervisie en versterkend leren, overtreft DeepSeek-V3 andere open-source modellen in prestaties en komt het dicht in de buurt van toonaangevende gesloten modellen."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma is een van de lichtgewicht, state-of-the-art open modelseries ontwikkeld door Google. Het is een groot taalmodel met alleen decodering, dat Engels ondersteunt en open gewichten, voorgetrainde varianten en instructie-fijn afgestelde varianten biedt. Het Gemma-model is geschikt voor verschillende tekstgeneratietaken, waaronder vraag-en-antwoord, samenvattingen en redenering. Dit 9B-model is getraind met 8 biljoen tokens. De relatief kleine omvang maakt het mogelijk om in omgevingen met beperkte middelen te worden geïmplementeerd, zoals laptops, desktops of uw eigen cloudinfrastructuur, waardoor meer mensen toegang hebben tot geavanceerde AI-modellen en innovatie wordt bevorderd."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 is een familie van meertalige grote taalmodellen ontwikkeld door Meta, inclusief voorgetrainde en instructie-fijn afgestelde varianten met parameter groottes van 8B, 70B en 405B. Dit 8B instructie-fijn afgestelde model is geoptimaliseerd voor meertalige gespreksscenario's en presteert uitstekend in verschillende industriële benchmarktests. Het model is getraind met meer dan 150 biljoen tokens van openbare gegevens en maakt gebruik van technieken zoals supervisie-fijn afstemming en versterkend leren met menselijke feedback om de bruikbaarheid en veiligheid van het model te verbeteren. Llama 3.1 ondersteunt tekstgeneratie en codegeneratie, met een kennisafkapdatum van december 2023."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview is een innovatief natuurlijk taalverwerkingsmodel dat efficiënt complexe dialooggeneratie en contextbegripstaken kan verwerken."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview is een onderzoeksmodel ontwikkeld door het Qwen-team, dat zich richt op visuele redeneervaardigheden en unieke voordelen heeft in het begrijpen van complexe scènes en het oplossen van visueel gerelateerde wiskundige problemen."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview is het nieuwste experimentele onderzoeksmodel van Qwen, gericht op het verbeteren van AI-redeneringscapaciteiten. Door het verkennen van complexe mechanismen zoals taalmixing en recursieve redenering, zijn de belangrijkste voordelen onder andere krachtige redeneringsanalyses, wiskundige en programmeervaardigheden. Tegelijkertijd zijn er ook problemen met taalwisseling, redeneringscycli, veiligheidskwesties en verschillen in andere capaciteiten."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct is de nieuwste versie van de code-specifieke grote taalmodelreeks die door Alibaba Cloud is uitgebracht. Dit model is aanzienlijk verbeterd in codegeneratie, redenering en herstelcapaciteiten door training met 55 biljoen tokens, gebaseerd op Qwen2.5. Het versterkt niet alleen de coderingscapaciteiten, maar behoudt ook de voordelen van wiskundige en algemene vaardigheden. Het model biedt een meer uitgebreide basis voor praktische toepassingen zoals code-agenten."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math richt zich op het oplossen van wiskundige vraagstukken en biedt professionele antwoorden op moeilijke vragen."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 is de nieuwste serie van het Qwen-model, dat 128k context ondersteunt. In vergelijking met de huidige beste open-source modellen, overtreft Qwen2-72B op het gebied van natuurlijke taalbegrip, kennis, code, wiskunde en meertaligheid aanzienlijk de huidige toonaangevende modellen."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 is de nieuwste serie van het Qwen-model, dat in staat is om de beste open-source modellen van gelijke grootte of zelfs grotere modellen te overtreffen. Qwen2 7B heeft aanzienlijke voordelen behaald in verschillende evaluaties, vooral op het gebied van code en begrip van het Chinees."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B is een krachtig visueel taalmodel dat multimodale verwerking van afbeeldingen en tekst ondersteunt, in staat om afbeeldingsinhoud nauwkeurig te herkennen en relevante beschrijvingen of antwoorden te genereren."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct is een groot taalmodel met 14 miljard parameters, met uitstekende prestaties, geoptimaliseerd voor Chinese en meertalige scenario's, en ondersteunt toepassingen zoals intelligente vraag-en-antwoord en contentgeneratie."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct is een groot taalmodel met 7 miljard parameters, dat function calls ondersteunt en naadloos kan interageren met externe systemen, wat de flexibiliteit en schaalbaarheid aanzienlijk vergroot. Geoptimaliseerd voor Chinese en meertalige scenario's, ondersteunt het toepassingen zoals intelligente vraag-en-antwoord en contentgeneratie."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct is een op grote schaal voorgetraind programmeerinstructiemodel met krachtige codebegrip- en generatiecapaciteiten, dat efficiënt verschillende programmeertaken kan verwerken, vooral geschikt voor slimme codegeneratie, automatiseringsscripts en het beantwoorden van programmeervragen."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct is een groot taalmodel dat speciaal is ontworpen voor codegeneratie, codebegrip en efficiënte ontwikkelingsscenario's, met een toonaangevende parameteromvang van 32B, dat kan voldoen aan diverse programmeerbehoeften."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "Het TeleMM multimodale grote model is een door China Telecom ontwikkeld model voor multimodale begrip, dat verschillende modaliteiten zoals tekst en afbeeldingen kan verwerken, en ondersteunt functies zoals beeldbegrip en grafiekanalyse, en biedt gebruikers cross-modale begripdiensten. Het model kan met gebruikers communiceren in meerdere modaliteiten, de invoer nauwkeurig begrijpen, vragen beantwoorden, helpen bij creatie en efficiënt multimodale informatie en inspiratie bieden. Het presteert uitstekend in multimodale taken zoals fijne perceptie en logische redenering."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large is het grootste open source Transformer-architectuur MoE-model in de industrie, met een totaal van 389 miljard parameters en 52 miljard geactiveerde parameters."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct is een instructie-fijn afgesteld groot taalmodel in de Qwen2-serie, met een parameter grootte van 72B. Dit model is gebaseerd op de Transformer-architectuur en maakt gebruik van technieken zoals de SwiGLU-activeringsfunctie, aandacht QKV-bias en groepsquery-aandacht. Het kan grote invoer verwerken. Dit model presteert uitstekend in taalbegrip, generatie, meertalige capaciteiten, codering, wiskunde en redenering in verschillende benchmarktests, en overtreft de meeste open-source modellen, en toont in sommige taken een concurrentievermogen vergelijkbaar met dat van propriëtaire modellen."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct is een van de nieuwste grote taalmodellen die door Alibaba Cloud is uitgebracht. Dit 72B-model heeft aanzienlijke verbeteringen in coderings- en wiskundige vaardigheden. Het model biedt ook meertalige ondersteuning, met meer dan 29 ondersteunde talen, waaronder Chinees en Engels. Het model heeft aanzienlijke verbeteringen in het volgen van instructies, het begrijpen van gestructureerde gegevens en het genereren van gestructureerde uitvoer (vooral JSON)."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Geoptimaliseerd voor Chinese personagegesprekken, biedt vloeiende en cultureel passende gespreksgeneratiecapaciteiten."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Fireworks open-source functie-aanroepmodel biedt uitstekende instructie-uitvoeringscapaciteiten en aanpasbare functies."
+ "abab7-chat-preview": {
+ "description": "In vergelijking met de abab6.5-serie modellen zijn er aanzienlijke verbeteringen in de capaciteiten op het gebied van lange teksten, wiskunde, schrijven, enzovoort."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Firefunction-v2, ontwikkeld door Fireworks, is een hoogpresterend functie-aanroepmodel, gebaseerd op Llama-3 en geoptimaliseerd voor functie-aanroepen, gesprekken en instructies."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 is een geavanceerd groot taalmodel, geoptimaliseerd met versterkend leren en koude startdata, met uitstekende prestaties in redeneren, wiskunde en programmeren."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b is een visueel taalmodel dat zowel afbeeldingen als tekstinvoer kan verwerken, getraind op hoogwaardige gegevens, geschikt voor multimodale taken."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Een krachtige Mixture-of-Experts (MoE) taalmodel van Deepseek, met een totaal aantal parameters van 671B, waarbij 37B parameters per token worden geactiveerd."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Llama 3 70B instructiemodel, speciaal geoptimaliseerd voor meertalige gesprekken en natuurlijke taalbegrip, presteert beter dan de meeste concurrerende modellen."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Llama 3 70B instructiemodel (HF-versie), consistent met de officiële implementatieresultaten, geschikt voor hoogwaardige instructietaken."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Llama 3 8B instructiemodel, geoptimaliseerd voor gesprekken en meertalige taken, presteert uitstekend en efficiënt."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Meta's 11B-parameter instructie-geoptimaliseerde beeldredeneringsmodel. Dit model is geoptimaliseerd voor visuele herkenning, beeldredenering, afbeeldingsbeschrijving en het beantwoorden van algemene vragen over afbeeldingen. Dit model kan visuele gegevens begrijpen, zoals diagrammen en grafieken, en overbrugt de kloof tussen visuele informatie en tekst door het genereren van tekstbeschrijvingen van afbeeldingsdetails."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Llama 3.2 1B instructiemodel is een lichtgewicht meertalig model geïntroduceerd door Meta. Dit model is ontworpen om de efficiëntie te verhogen, met aanzienlijke verbeteringen in latentie en kosten in vergelijking met grotere modellen. Voorbeelden van gebruikssituaties van dit model zijn ophalen en samenvatten."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Llama 3.2 3B instructiemodel is een lichtgewicht meertalig model geïntroduceerd door Meta. Dit model is ontworpen om de efficiëntie te verhogen, met aanzienlijke verbeteringen in latentie en kosten in vergelijking met grotere modellen. Voorbeelden van gebruikssituaties van dit model zijn het herformuleren van vragen en prompts, evenals schrijfondersteuning."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Meta's 90B-parameter instructie-geoptimaliseerde beeldredeneringsmodel. Dit model is geoptimaliseerd voor visuele herkenning, beeldredenering, afbeeldingsbeschrijving en het beantwoorden van algemene vragen over afbeeldingen. Dit model kan visuele gegevens begrijpen, zoals diagrammen en grafieken, en overbrugt de kloof tussen visuele informatie en tekst door het genereren van tekstbeschrijvingen van afbeeldingsdetails."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct is de update van december voor Llama 3.1 70B. Dit model is verbeterd op basis van Llama 3.1 70B (uitgebracht in juli 2024) en biedt verbeterde toolaanroepen, ondersteuning voor meertalige teksten, wiskunde en programmeervaardigheden. Het model heeft een toonaangevende prestatie bereikt op het gebied van redeneren, wiskunde en het volgen van instructies, en kan prestaties bieden die vergelijkbaar zijn met die van 3.1 405B, met aanzienlijke voordelen op het gebied van snelheid en kosten."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "Een model met 24B parameters, dat geavanceerde mogelijkheden biedt die vergelijkbaar zijn met grotere modellen."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Mixtral MoE 8x22B instructiemodel, met een groot aantal parameters en een multi-expertarchitectuur, biedt uitgebreide ondersteuning voor de efficiënte verwerking van complexe taken."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Mixtral MoE 8x7B instructiemodel, met een multi-expertarchitectuur die efficiënte instructievolging en uitvoering biedt."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Mixtral MoE 8x7B instructiemodel (HF-versie), met prestaties die overeenkomen met de officiële implementatie, geschikt voor verschillende efficiënte taakscenario's."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "MythoMax L2 13B model, dat gebruik maakt van innovatieve samenvoegtechnologie, is goed in verhalen vertellen en rollenspellen."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "Het QwQ-model is een experimenteel onderzoeksmodel ontwikkeld door het Qwen-team, gericht op het verbeteren van de AI-redeneringscapaciteiten."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "De 72B versie van het Qwen-VL model is het nieuwste resultaat van Alibaba's iteraties, dat bijna een jaar aan innovaties vertegenwoordigt."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 is een serie decoder-only taalmodellen ontwikkeld door het Alibaba Qwen-team. Deze modellen zijn beschikbaar in verschillende groottes, waaronder 0.5B, 1.5B, 3B, 7B, 14B, 32B en 72B, met zowel een basisversie als een instructieversie."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct is de nieuwste versie van de code-specifieke grote taalmodelreeks die door Alibaba Cloud is uitgebracht. Dit model is aanzienlijk verbeterd in codegeneratie, redenering en herstelcapaciteiten door training met 55 biljoen tokens, gebaseerd op Qwen2.5. Het versterkt niet alleen de coderingscapaciteiten, maar behoudt ook de voordelen van wiskundige en algemene vaardigheden. Het model biedt een meer uitgebreide basis voor praktische toepassingen zoals code-agenten."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "StarCoder 15.5B model, ondersteunt geavanceerde programmeertaken, met verbeterde meertalige capaciteiten, geschikt voor complexe codegeneratie en -begrip."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "StarCoder 7B model, getraind op meer dan 80 programmeertalen, met uitstekende programmeervulcapaciteiten en contextbegrip."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Yi-Large model, met uitstekende meertalige verwerkingscapaciteiten, geschikt voor verschillende taalgeneratie- en begripstaken."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus is het krachtigste model van Anthropic voor het verwerken van zeer complexe taken. Het excelleert in prestaties, intelligentie, vloeiendheid en begrip."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku is het snelste volgende generatie model van Anthropic. In vergelijking met Claude 3 Haiku heeft Claude 3.5 Haiku verbeteringen in verschillende vaardigheden en overtreft het de grootste modellen van de vorige generatie, Claude 3 Opus, in veel intellectuele benchmarktests."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet biedt mogelijkheden die verder gaan dan Opus en een snellere snelheid dan Sonnet, terwijl het dezelfde prijs als Sonnet behoudt. Sonnet is bijzonder goed in programmeren, datawetenschap, visuele verwerking en agenttaken."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet biedt een ideale balans tussen intelligentie en snelheid voor bedrijfswerkbelastingen. Het biedt maximale bruikbaarheid tegen een lagere prijs, betrouwbaar en geschikt voor grootschalige implementatie."
},
- "code-raccoon-v1": {
- "description": "Code Raccoon is een software-intelligente ontwikkelingsassistent gebaseerd op het SenseTime grote taalmodel, dat softwarebehoefteanalyse, architectuurontwerp, code schrijven, softwaretesten en andere fasen dekt, en voldoet aan de verschillende behoeften van gebruikers voor code schrijven en programmeerleren. Code Raccoon ondersteunt meer dan 90 populaire programmeertalen zoals Python, Java, JavaScript, C++, Go, SQL en populaire IDE's zoals VS Code en IntelliJ IDEA. In de praktijk kan Code Raccoon ontwikkelaars helpen om de programmeerefficiëntie met meer dan 50% te verhogen."
- },
"codegeex-4": {
"description": "CodeGeeX-4 is een krachtige AI-programmeerassistent die slimme vraag- en antwoordmogelijkheden en code-aanvulling ondersteunt voor verschillende programmeertalen, waardoor de ontwikkelingssnelheid wordt verhoogd."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ is een hoogpresterend groot taalmodel, speciaal ontworpen voor echte zakelijke scenario's en complexe toepassingen."
},
+ "dall-e-2": {
+ "description": "De tweede generatie DALL·E model, ondersteunt realistischere en nauwkeurigere beeldgeneratie, met een resolutie die vier keer zo hoog is als die van de eerste generatie."
+ },
+ "dall-e-3": {
+ "description": "Het nieuwste DALL·E model, uitgebracht in november 2023. Ondersteunt realistischere en nauwkeurigere beeldgeneratie met een sterkere detailweergave."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct biedt betrouwbare instructieverwerkingscapaciteiten en ondersteunt toepassingen in verschillende sectoren."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 is een krachtig en kosteneffectief hybride expert (MoE) taalmodel. Het is voorgetraind op een hoogwaardige corpus van 81 biljoen tokens en verder verbeterd door middel van supervisie-fijnafstemming (SFT) en versterkend leren (RL). In vergelijking met DeepSeek 67B bespaart DeepSeek-V2 42,5% van de trainingskosten, vermindert 93,3% van de KV-cache en verhoogt de maximale generatiedoorvoer met 5,76 keer. Dit model ondersteunt een contextlengte van 128k en presteert uitstekend in standaard benchmarktests en open generatieve evaluaties."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 is een op versterkend leren (RL) aangedreven inferentiemodel dat de problemen van herhaling en leesbaarheid in het model oplost. Voor RL introduceerde DeepSeek-R1 koude startdata om de inferentieprestaties verder te optimaliseren. Het presteert vergelijkbaar met OpenAI-o1 in wiskunde, code en inferentietaken, en verbetert de algehele effectiviteit door zorgvuldig ontworpen trainingsmethoden."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "DeepSeek-R1 distillatiemodel, geoptimaliseerd voor inferentieprestaties door versterkend leren en koude startdata, open-source model dat de multi-taak benchmark vernieuwt."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B is een distillatiemodel ontwikkeld op basis van Llama-3.1-8B. Dit model is fijn afgestemd met voorbeelden gegenereerd door DeepSeek-R1 en toont uitstekende inferentiecapaciteiten. Het heeft goed gepresteerd in verschillende benchmarktests, met een nauwkeurigheid van 89,1% op MATH-500, een slaagpercentage van 50,4% op AIME 2024, en een score van 1205 op CodeForces, wat sterke wiskundige en programmeercapaciteiten aantoont voor een model van 8B schaal."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "DeepSeek-R1 distillatiemodel, geoptimaliseerd voor inferentieprestaties door versterkend leren en koude startdata, open-source model dat de multi-taak benchmark vernieuwt."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "DeepSeek-R1 distillatiemodel, geoptimaliseerd voor inferentieprestaties door versterkend leren en koude startdata, open-source model dat de multi-taak benchmark vernieuwt."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B is een model dat is verkregen door kennisdistillatie van Qwen2.5-32B. Dit model is fijn afgestemd met 800.000 zorgvuldig geselecteerde voorbeelden gegenereerd door DeepSeek-R1 en toont uitstekende prestaties in verschillende domeinen zoals wiskunde, programmeren en redeneren. Het heeft uitstekende resultaten behaald in meerdere benchmarktests, waaronder een nauwkeurigheid van 94,3% op MATH-500, wat sterke wiskundige redeneringscapaciteiten aantoont."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B is een model dat is verkregen door kennisdistillatie van Qwen2.5-Math-7B. Dit model is fijn afgestemd met 800.000 zorgvuldig geselecteerde voorbeelden gegenereerd door DeepSeek-R1 en toont uitstekende inferentiecapaciteiten. Het heeft uitstekende resultaten behaald in verschillende benchmarktests, met een nauwkeurigheid van 92,8% op MATH-500, een slaagpercentage van 55,5% op AIME 2024, en een score van 1189 op CodeForces, wat sterke wiskundige en programmeercapaciteiten aantoont voor een model van 7B schaal."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 combineert de uitstekende kenmerken van eerdere versies en versterkt de algemene en coderingscapaciteiten."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 is een hybride expert (MoE) taalmodel met 6710 miljard parameters, dat gebruikmaakt van multi-head latent attention (MLA) en de DeepSeekMoE-architectuur, gecombineerd met een load balancing-strategie zonder extra verlies, om de inferentie- en trainingsefficiëntie te optimaliseren. Door voorgetraind te worden op 14,8 biljoen hoogwaardige tokens en vervolgens te worden fijngetuned met supervisie en versterkend leren, overtreft DeepSeek-V3 andere open-source modellen in prestaties en komt het dicht in de buurt van toonaangevende gesloten modellen."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B is een geavanceerd model dat is getraind voor complexe gesprekken."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "Geavanceerd efficiënt LLM, gespecialiseerd in redeneren, wiskunde en programmeren."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 is een hybride expert (MoE) visueel taalmodel dat is ontwikkeld op basis van DeepSeekMoE-27B, met een MoE-architectuur met spaarzame activatie, die uitstekende prestaties levert met slechts 4,5 miljard geactiveerde parameters. Dit model presteert uitstekend in verschillende taken, waaronder visuele vraag-antwoord, optische tekenherkenning, document/tabel/grafiekbegrip en visuele positionering."
+ },
"deepseek-chat": {
"description": "Een nieuw open-source model dat algemene en code-capaciteiten combineert, behoudt niet alleen de algemene conversatiecapaciteiten van het oorspronkelijke Chat-model en de krachtige codeverwerkingscapaciteiten van het Coder-model, maar is ook beter afgestemd op menselijke voorkeuren. Bovendien heeft DeepSeek-V2.5 aanzienlijke verbeteringen gerealiseerd in schrijfopdrachten, instructievolging en andere gebieden."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 is een open-source hybride expertcode-model, presteert uitstekend in code-taken en is vergelijkbaar met GPT4-Turbo."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 is een op versterkend leren (RL) aangedreven inferentiemodel dat de problemen van herhaling en leesbaarheid in het model oplost. Voor RL introduceerde DeepSeek-R1 koude startdata om de inferentieprestaties verder te optimaliseren. Het presteert vergelijkbaar met OpenAI-o1 in wiskunde, code en inferentietaken, en verbetert de algehele effectiviteit door zorgvuldig ontworpen trainingsmethoden."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 - een groter en slimmer model binnen de DeepSeek suite - is gedistilleerd naar de Llama 70B architectuur. Op basis van benchmarktests en menselijke evaluaties is dit model slimmer dan de originele Llama 70B, vooral in taken die wiskundige en feitelijke nauwkeurigheid vereisen."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "Het DeepSeek-R1-Distill model is verkregen door middel van kennisdistillatie-technologie, waarbij samples gegenereerd door DeepSeek-R1 zijn afgestemd op open-source modellen zoals Qwen en Llama."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "Het DeepSeek-R1-Distill model is verkregen door middel van kennisdistillatie-technologie, waarbij samples gegenereerd door DeepSeek-R1 zijn afgestemd op open-source modellen zoals Qwen en Llama."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "Het DeepSeek-R1-Distill model is verkregen door middel van kennisdistillatie-technologie, waarbij samples gegenereerd door DeepSeek-R1 zijn afgestemd op open-source modellen zoals Qwen en Llama."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "Het DeepSeek-R1-Distill model is verkregen door middel van kennisdistillatie-technologie, waarbij samples gegenereerd door DeepSeek-R1 zijn afgestemd op open-source modellen zoals Qwen en Llama."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "Het DeepSeek-R1-Distill model is verkregen door middel van kennisdistillatie-technologie, waarbij samples gegenereerd door DeepSeek-R1 zijn afgestemd op open-source modellen zoals Qwen en Llama."
+ },
+ "deepseek-reasoner": {
+ "description": "Het redeneer model van DeepSeek. Voordat het model het uiteindelijke antwoord geeft, genereert het eerst een stuk denkproces om de nauwkeurigheid van het uiteindelijke antwoord te verbeteren."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 is een efficiënt Mixture-of-Experts taalmodel, geschikt voor kosteneffectieve verwerkingsbehoeften."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B is het ontwerpcode-model van DeepSeek, biedt krachtige codegeneratiecapaciteiten."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 is een MoE-model dat is ontwikkeld door Hangzhou DeepSeek Artificial Intelligence Technology Research Co., Ltd. Het heeft uitstekende scores in verschillende evaluaties en staat bovenaan de open-source modellen in de mainstream ranglijsten. V3 heeft de generatiesnelheid met 3 keer verbeterd in vergelijking met het V2.5 model, wat zorgt voor een snellere en soepelere gebruikerservaring."
+ },
"deepseek/deepseek-chat": {
"description": "Een nieuw open-source model dat algemene en codeercapaciteiten combineert, niet alleen de algemene gespreksvaardigheden van het oorspronkelijke Chat-model en de krachtige codeverwerkingscapaciteiten van het Coder-model behoudt, maar ook beter is afgestemd op menselijke voorkeuren. Bovendien heeft DeepSeek-V2.5 aanzienlijke verbeteringen gerealiseerd in schrijfopdrachten, instructievolging en meer."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 heeft de redeneringscapaciteiten van het model aanzienlijk verbeterd, zelfs met zeer weinig gelabelde gegevens. Voordat het model het uiteindelijke antwoord geeft, genereert het eerst een denkproces om de nauwkeurigheid van het uiteindelijke antwoord te verbeteren."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 heeft de redeneringscapaciteiten van het model aanzienlijk verbeterd, zelfs met zeer weinig gelabelde gegevens. Voordat het model het uiteindelijke antwoord geeft, genereert het eerst een denkproces om de nauwkeurigheid van het uiteindelijke antwoord te verbeteren."
+ },
"emohaa": {
"description": "Emohaa is een psychologisch model met professionele adviescapaciteiten, dat gebruikers helpt emotionele problemen te begrijpen."
},
+ "ernie-3.5-128k": {
+ "description": "Het vlaggenschip grote taalmodel van Baidu, zelf ontwikkeld, dekt een enorme hoeveelheid Chinese en Engelse corpora, met sterke algemene capaciteiten die voldoen aan de meeste eisen voor dialoogvragen, creatieve generatie en plug-in toepassingsscenario's; ondersteunt automatische integratie met Baidu zoekplug-ins om de actualiteit van vraag- en antwoordinformatie te waarborgen."
+ },
+ "ernie-3.5-8k": {
+ "description": "Het vlaggenschip grote taalmodel van Baidu, zelf ontwikkeld, dekt een enorme hoeveelheid Chinese en Engelse corpora, met sterke algemene capaciteiten die voldoen aan de meeste eisen voor dialoogvragen, creatieve generatie en plug-in toepassingsscenario's; ondersteunt automatische integratie met Baidu zoekplug-ins om de actualiteit van vraag- en antwoordinformatie te waarborgen."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "Het vlaggenschip grote taalmodel van Baidu, zelf ontwikkeld, dekt een enorme hoeveelheid Chinese en Engelse corpora, met sterke algemene capaciteiten die voldoen aan de meeste eisen voor dialoogvragen, creatieve generatie en plug-in toepassingsscenario's; ondersteunt automatische integratie met Baidu zoekplug-ins om de actualiteit van vraag- en antwoordinformatie te waarborgen."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "Het vlaggenschip ultra-grote taalmodel van Baidu, zelf ontwikkeld, heeft een algehele upgrade van modelcapaciteiten in vergelijking met ERNIE 3.5, en is breed toepasbaar in complexe taakscenario's in verschillende domeinen; ondersteunt automatische integratie met Baidu zoekplug-ins om de actualiteit van vraag- en antwoordinformatie te waarborgen."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "Het vlaggenschip ultra-grote taalmodel van Baidu, zelf ontwikkeld, heeft een algehele upgrade van modelcapaciteiten in vergelijking met ERNIE 3.5, en is breed toepasbaar in complexe taakscenario's in verschillende domeinen; ondersteunt automatische integratie met Baidu zoekplug-ins om de actualiteit van vraag- en antwoordinformatie te waarborgen."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "Het vlaggenschip ultra-grote taalmodel van Baidu, zelf ontwikkeld, presteert uitstekend in algehele effectiviteit en is breed toepasbaar in complexe taakscenario's in verschillende domeinen; ondersteunt automatische integratie met Baidu zoekplug-ins om de actualiteit van vraag- en antwoordinformatie te waarborgen. Het presteert beter dan ERNIE 4.0."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "Het vlaggenschip ultra-grote taalmodel van Baidu, zelf ontwikkeld, presteert uitstekend in algehele effectiviteit en is breed toepasbaar in complexe taakscenario's in verschillende domeinen; ondersteunt automatische integratie met Baidu zoekplug-ins om de actualiteit van vraag- en antwoordinformatie te waarborgen. Het presteert beter dan ERNIE 4.0."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "Het vlaggenschip ultra-grote taalmodel van Baidu, zelf ontwikkeld, presteert uitstekend in algehele effectiviteit en is breed toepasbaar in complexe taakscenario's in verschillende domeinen; ondersteunt automatische integratie met Baidu zoekplug-ins om de actualiteit van vraag- en antwoordinformatie te waarborgen. Het presteert beter dan ERNIE 4.0."
+ },
+ "ernie-char-8k": {
+ "description": "Een door Baidu ontwikkeld groot taalmodel voor verticale scenario's, geschikt voor toepassingen zoals game NPC's, klantenservice dialoog, en rollenspellen, met een duidelijkere en consistentere karakterstijl, sterkere instructievolgcapaciteiten en betere inferentieprestaties."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Een door Baidu ontwikkeld groot taalmodel voor verticale scenario's, geschikt voor toepassingen zoals game NPC's, klantenservice dialoog, en rollenspellen, met een duidelijkere en consistentere karakterstijl, sterkere instructievolgcapaciteiten en betere inferentieprestaties."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite is een lichtgewicht groot taalmodel dat door Baidu is ontwikkeld, dat uitstekende modelprestaties en inferentiecapaciteiten combineert, geschikt voor gebruik met AI-versnelling kaarten met lage rekencapaciteit."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Een lichtgewicht groot taalmodel dat door Baidu is ontwikkeld, dat uitstekende modelprestaties en inferentiecapaciteiten combineert, met betere prestaties dan ERNIE Lite, geschikt voor gebruik met AI-versnelling kaarten met lage rekencapaciteit."
+ },
+ "ernie-novel-8k": {
+ "description": "Een algemeen groot taalmodel dat door Baidu is ontwikkeld, met duidelijke voordelen in het vervolgschrijven van romans, en ook toepasbaar in korte toneelstukken, films en andere scenario's."
+ },
+ "ernie-speed-128k": {
+ "description": "Het nieuwste zelfontwikkelde hoge-prestatie grote taalmodel van Baidu, dat uitstekende algemene capaciteiten heeft en geschikt is als basis model voor afstemming, om beter specifieke scenario's aan te pakken, met uitstekende inferentieprestaties."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "Het nieuwste zelfontwikkelde hoge-prestatie grote taalmodel van Baidu, dat uitstekende algemene capaciteiten heeft en betere prestaties levert dan ERNIE Speed, geschikt als basis model voor afstemming, om beter specifieke scenario's aan te pakken, met uitstekende inferentieprestaties."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny is een ultra-presterend groot taalmodel dat de laagste implementatie- en afstemmingskosten heeft binnen de Wenxin modelreeks."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Tuning) biedt stabiele en afstelbare prestaties, ideaal voor oplossingen voor complexe taken."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro ondersteunt tot 2 miljoen tokens en is de ideale keuze voor middelgrote multimodale modellen, geschikt voor veelzijdige ondersteuning van complexe taken."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash biedt next-gen functies en verbeteringen, waaronder uitstekende snelheid, native toolgebruik, multimodale generatie en een contextvenster van 1M tokens."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash biedt next-gen functies en verbeteringen, waaronder uitstekende snelheid, native toolgebruik, multimodale generatie en een contextvenster van 1M tokens."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp is Google's nieuwste experimentele multimodale AI-model, met de volgende generatie functies, uitstekende snelheid, ingebouwde tool-aanroepen en multimodale generatie."
},
- "gemini-2.0-flash-thinking-exp-1219": {
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "Een Gemini 2.0 Flash-model dat is geoptimaliseerd voor kosteneffectiviteit en lage latentie."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
"description": "Gemini 2.0 Flash Exp is Google's nieuwste experimentele multimodale AI-model, met next-gen functies, uitstekende snelheid, native tool-aanroepen en multimodale generatie."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 is Google's nieuwste experimentele multimodale AI-model, met een snelle verwerkingscapaciteit, ondersteunt tekst-, beeld- en video-invoer, geschikt voor efficiënte schaling van diverse taken."
+ "gemini-2.0-flash-thinking-exp-1219": {
+ "description": "Gemini 2.0 Flash Exp is Google's nieuwste experimentele multimodale AI-model, met next-gen functies, uitstekende snelheid, native tool-aanroepen en multimodale generatie."
},
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 is Google's nieuwste experimentele multimodale AI-model, dat snel kan verwerken en ondersteuning biedt voor tekst-, beeld- en video-invoer, geschikt voor efficiënte opschaling van verschillende taken."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental is Google's nieuwste experimentele multimodale AI-model, met aanzienlijke kwaliteitsverbeteringen ten opzichte van eerdere versies, vooral op het gebied van wereldkennis, code en lange context."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 is Google's nieuwste experimentele multimodale AI-model, met een aanzienlijke kwaliteitsverbetering ten opzichte van eerdere versies."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus heeft de capaciteit om video-inhoud en meerdere afbeeldingen te begrijpen, geschikt voor multimodale taken."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview heeft krachtige complexe redeneercapaciteiten en presteert uitstekend in logische redenering, wiskunde en programmeren."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash biedt next-gen functies en verbeteringen, waaronder uitstekende snelheid, native toolgebruik, multimodale generatie en een contextvenster van 1M tokens."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental is Google's nieuwste experimentele multimodale AI-model, met aanzienlijke kwaliteitsverbeteringen ten opzichte van eerdere versies, vooral op het gebied van wereldkennis, code en lange context."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash biedt geoptimaliseerde multimodale verwerkingscapaciteiten, geschikt voor verschillende complexe taakscenario's."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro combineert de nieuwste optimalisatietechnologieën en biedt efficiëntere multimodale gegevensverwerkingscapaciteiten."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 is een efficiënt model van Google, dat een breed scala aan toepassingen dekt, van kleine toepassingen tot complexe gegevensverwerking."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 behoudt het ontwerpprincipe van lichtgewicht en efficiëntie."
},
"google/gemma-2-2b-it": {
"description": "Google's lichtgewicht instructieafstemmingsmodel"
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 is een efficiënt model van Google, dat een breed scala aan toepassingen dekt, van kleine toepassingen tot complexe gegevensverwerking."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 is een lichtgewicht open-source tekstmodelserie van Google."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo, geschikt voor verschillende tekstgeneratie- en begrijptaken, wijst momenteel naar gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, een model voor tekstgeneratie met hoge capaciteit, geschikt voor complexe taken."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo, geschikt voor verschillende tekstgeneratie- en begrijptaken, wijst momenteel naar gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o is een dynamisch model dat in real-time wordt bijgewerkt om de meest actuele versie te behouden. Het combineert krachtige taalbegrip en generatiemogelijkheden, geschikt voor grootschalige toepassingen zoals klantenservice, onderwijs en technische ondersteuning."
},
+ "gpt-4o-audio-preview": {
+ "description": "GPT-4o Audio model, ondersteunt audio-invoer en -uitvoer."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini is het nieuwste model van OpenAI, gelanceerd na GPT-4 Omni, en ondersteunt zowel tekst- als beeldinvoer met tekstuitvoer. Als hun meest geavanceerde kleine model is het veel goedkoper dan andere recente toonaangevende modellen en meer dan 60% goedkoper dan GPT-3.5 Turbo. Het behoudt de meest geavanceerde intelligentie met een aanzienlijke prijs-kwaliteitverhouding. GPT-4o mini behaalde 82% op de MMLU-test en staat momenteel hoger in chatvoorkeuren dan GPT-4."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "GPT-4o-mini realtime versie, ondersteunt audio en tekst realtime invoer en uitvoer."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "GPT-4o realtime versie, ondersteunt audio en tekst realtime invoer en uitvoer."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "GPT-4o realtime versie, ondersteunt audio en tekst realtime invoer en uitvoer."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "GPT-4o realtime versie, ondersteunt audio en tekst realtime invoer en uitvoer."
+ },
"grok-2-1212": {
"description": "Dit model heeft verbeteringen aangebracht in nauwkeurigheid, instructievolging en meertalige capaciteiten."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "Het nieuwste MOE-architectuur FunctionCall-model van Hunyuan, getraind op hoogwaardige FunctionCall-gegevens, met een contextvenster van 32K, en staat voorop in meerdere dimensies van evaluatie-indicatoren."
},
+ "hunyuan-large": {
+ "description": "Het Hunyuan-large model heeft een totaal aantal parameters van ongeveer 389B, met ongeveer 52B actieve parameters, en is het grootste en beste open-source MoE-model met Transformer-architectuur in de industrie."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Uitstekend in het verwerken van lange teksttaken zoals document samenvattingen en documentvragen, en heeft ook de capaciteit om algemene tekstgeneratietaken uit te voeren. Het presteert uitstekend in de analyse en generatie van lange teksten en kan effectief omgaan met complexe en gedetailleerde lange inhoudsverwerkingsbehoeften."
+ },
"hunyuan-lite": {
"description": "Geüpgraded naar een MOE-structuur, met een contextvenster van 256k, en leidt in verschillende evaluatiesets op het gebied van NLP, code, wiskunde en industrie ten opzichte van vele open-source modellen."
},
+ "hunyuan-lite-vision": {
+ "description": "De nieuwste 7B multimodale Hunyuan-model, met een contextvenster van 32K, ondersteunt multimodale gesprekken in het Chinees en het Engels, objectherkenning in afbeeldingen, document- en tabelbegrip, multimodale wiskunde, enz., en scoort op meerdere dimensies beter dan 7B concurrentiemodellen."
+ },
"hunyuan-pro": {
"description": "Een MOE-32K lange tekstmodel met triljoenen parameters. Bereikt een absoluut leidend niveau op verschillende benchmarks, met complexe instructies en redenering, beschikt over complexe wiskundige capaciteiten, ondersteunt function calls, en is geoptimaliseerd voor toepassingen in meertalige vertaling, financiële, juridische en medische gebieden."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Maakt gebruik van een betere routeringsstrategie en verlicht tegelijkertijd de problemen van load balancing en expert convergentie. Voor lange teksten bereikt de naald in een hooiberg-index 99,9%. MOE-256K doorbreekt verder in lengte en effectiviteit, waardoor de invoerlengte aanzienlijk wordt vergroot."
},
+ "hunyuan-standard-vision": {
+ "description": "De nieuwste multimodale Hunyuan-model, ondersteunt meertalige antwoorden, met evenwichtige capaciteiten in het Chinees en het Engels."
+ },
"hunyuan-turbo": {
"description": "Een previewversie van het nieuwe generatie grote taalmodel van Hunyuan, met een nieuwe hybride expertmodel (MoE) structuur, die sneller inferentie-efficiëntie biedt en betere prestaties levert dan hunyan-pro."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Hunyuan-turbo versie van 20 november 2024, een vaste versie die zich tussen hunyuan-turbo en hunyuan-turbo-latest bevindt."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "Deze versie optimaliseert: gegevensinstructiescaling, wat de algemene generalisatiecapaciteit van het model aanzienlijk verbetert; aanzienlijke verbetering van wiskunde-, code- en logische redeneervaardigheden; optimalisatie van tekstbegrip en woordbegrip gerelateerde capaciteiten; optimalisatie van de kwaliteit van tekstcreatie en inhoudsgeneratie."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "Algemene ervaring optimalisatie, inclusief NLP-begrip, tekstcreatie, casual gesprekken, kennisvragen, vertalingen, en domeinspecifieke toepassingen; verbetering van de menselijkheid, optimalisatie van de emotionele intelligentie van het model; verbetering van het vermogen van het model om actief te verduidelijken bij vage intenties; verbetering van de verwerking van vragen over woord- en zinsanalyse; verbetering van de kwaliteit en interactie van creaties; verbetering van de ervaring in meerdere rondes."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "De nieuwe generatie visuele taal vlaggenschipmodel van Hunyuan, met een geheel nieuwe hybride expertmodel (MoE) structuur, biedt aanzienlijke verbeteringen in basisherkenning, inhoudcreatie, kennisvragen, en analytische redeneervaardigheden in vergelijking met de vorige generatie modellen."
+ },
"hunyuan-vision": {
"description": "Het nieuwste multimodale model van Hunyuan, ondersteunt het genereren van tekstinhoud op basis van afbeelding + tekstinvoer."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "Onze nieuwste modelreeks met uitstekende redeneervaardigheden, ondersteunt een contextlengte van 1M en heeft verbeterde instructievolging en toolaanroepmogelijkheden."
},
+ "internlm3-latest": {
+ "description": "Onze nieuwste modelreeks heeft uitstekende inferentieprestaties en leidt de open-source modellen in dezelfde klasse. Standaard gericht op ons recentste InternLM3 model."
+ },
+ "jina-deepsearch-v1": {
+ "description": "Diepe zoekopdrachten combineren webzoekopdrachten, lezen en redeneren voor een uitgebreide verkenning. Je kunt het beschouwen als een agent die jouw onderzoeksopdracht aanneemt - het zal een uitgebreide zoektocht uitvoeren en meerdere iteraties doorlopen voordat het een antwoord geeft. Dit proces omvat voortdurende onderzoek, redeneren en het oplossen van problemen vanuit verschillende invalshoeken. Dit is fundamenteel anders dan het rechtstreeks genereren van antwoorden uit voorgetrainde gegevens door standaard grote modellen en het vertrouwen op eenmalige oppervlakkige zoekopdrachten van traditionele RAG-systemen."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM is een experimenteel, taak-specifiek taalmodel dat is getraind volgens de principes van de leerwetenschap, en kan systeeminstructies volgen in onderwijs- en leeromgevingen, en fungeert als een expertmentor."
},
"lite": {
"description": "Spark Lite is een lichtgewicht groot taalmodel met extreem lage latentie en efficiënte verwerkingscapaciteit. Het is volledig gratis en open, en ondersteunt realtime online zoekfunctionaliteit. De snelle respons maakt het uitermate geschikt voor inferentie op apparaten met lage rekenkracht en modelafstemming, wat gebruikers uitstekende kosteneffectiviteit en een slimme ervaring biedt, vooral in kennisvragen, contentgeneratie en zoekscenario's."
},
- "llama-3.1-70b-instruct": {
- "description": "Llama 3.1 70B Instruct model, met 70B parameters, biedt uitstekende prestaties in grote tekstgeneratie- en instructietaken."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B biedt krachtigere AI-inferentiecapaciteiten, geschikt voor complexe toepassingen, ondersteunt een enorme rekenverwerking en garandeert efficiëntie en nauwkeurigheid."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B is een hoogpresterend model dat snelle tekstgeneratiecapaciteiten biedt, zeer geschikt voor toepassingen die grootschalige efficiëntie en kosteneffectiviteit vereisen."
},
- "llama-3.1-8b-instruct": {
- "description": "Llama 3.1 8B Instruct model, met 8B parameters, ondersteunt de efficiënte uitvoering van visuele instructietaken en biedt hoogwaardige tekstgeneratiecapaciteiten."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Llama 3.1 Sonar Huge Online model, met 405B parameters, ondersteunt een contextlengte van ongeveer 127.000 tokens, ontworpen voor complexe online chattoepassingen."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Llama 3.1 Sonar Large Chat model, met 70B parameters, ondersteunt een contextlengte van ongeveer 127.000 tokens, geschikt voor complexe offline chattaken."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Llama 3.1 Sonar Large Online model, met 70B parameters, ondersteunt een contextlengte van ongeveer 127.000 tokens, geschikt voor hoge capaciteit en diverse chattaken."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Llama 3.1 Sonar Small Chat model, met 8B parameters, speciaal ontworpen voor offline chat, ondersteunt een contextlengte van ongeveer 127.000 tokens."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Llama 3.1 Sonar Small Online model, met 8B parameters, ondersteunt een contextlengte van ongeveer 127.000 tokens, speciaal ontworpen voor online chat en kan efficiënt verschillende tekstinteracties verwerken."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 is ontworpen om taken te verwerken die visuele en tekstuele gegevens combineren. Het presteert uitstekend in taken zoals afbeeldingsbeschrijving en visuele vraag-en-antwoord, en overbrugt de kloof tussen taalgeneratie en visuele redeneervaardigheden."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 is het meest geavanceerde meertalige open-source grote taalmodel in de Llama-serie, dat prestaties biedt die vergelijkbaar zijn met die van het 405B-model tegen zeer lage kosten. Gebaseerd op de Transformer-structuur en verbeterd door middel van supervisie-fijnstelling (SFT) en versterkend leren met menselijke feedback (RLHF) voor gebruiksvriendelijkheid en veiligheid. De instructie-geoptimaliseerde versie is speciaal ontworpen voor meertalige dialogen en presteert beter dan veel open-source en gesloten chatmodellen op verschillende industriële benchmarks. Kennisafkapdatum is december 2023."
+ },
"llama-3.3-70b-versatile": {
"description": "Meta Llama 3.3 is een meertalige grote taalmodel (LLM) met 70B (tekstinvoer/tekstuitvoer) dat is voorgetraind en aangepast voor instructies. Het pure tekstmodel van Llama 3.3 is geoptimaliseerd voor meertalige gespreksgebruik en scoort beter dan veel beschikbare open-source en gesloten chatmodellen op gangbare industrie benchmarks."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 is ontworpen voor taken die zowel visuele als tekstuele gegevens combineren. Het presteert uitstekend in taken zoals afbeeldingsbeschrijving en visuele vraagstukken, en overbrugt de kloof tussen taalgeneratie en visuele redenering."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 is het meest geavanceerde meertalige open-source grote taalmodel uit de Llama-serie, dat een vergelijkbare prestatie biedt als het 405B model tegen zeer lage kosten. Gebaseerd op de Transformer-structuur en verbeterd in bruikbaarheid en veiligheid door middel van supervisie-fijnstelling (SFT) en versterkend leren met menselijke feedback (RLHF). De instructie-geoptimaliseerde versie is speciaal ontworpen voor meertalige gesprekken en presteert beter dan veel open-source en gesloten chatmodellen op verschillende industriële benchmarks. Kennisafkapdatum is december 2023."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Meta Llama 3.3 meertalige grote taalmodel (LLM) is een voorgetraind en instructie-aangepast generatief model van 70B (tekstinvoer/tekstuitvoer). Het Llama 3.3 instructie-aangepaste pure tekstmodel is geoptimaliseerd voor meertalige dialoogtoepassingen en presteert beter dan veel beschikbare open-source en gesloten chatmodellen op gangbare industriële benchmarks."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 is ontworpen voor taken die zowel visuele als tekstuele gegevens combineren. Het presteert uitstekend in taken zoals afbeeldingsbeschrijving en visuele vraagstukken, en overbrugt de kloof tussen taalgeneratie en visuele redenering."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "405B Llama 3.1 Turbo model biedt enorme contextondersteuning voor big data verwerking en presteert uitstekend in grootschalige AI-toepassingen."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 is een toonaangevend model van Meta, ondersteunt tot 405B parameters en kan worden toegepast in complexe gesprekken, meertalige vertalingen en data-analyse."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B biedt efficiënte gespreksondersteuning in meerdere talen."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 is ontworpen voor taken die visuele en tekstuele gegevens combineren. Het presteert uitstekend in taken zoals afbeeldingsbeschrijving en visuele vraag-en-antwoord, en overbrugt de kloof tussen taalgeneratie en visuele redenering."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 is het meest geavanceerde meertalige open-source grote taalmodel in de Llama-serie, dat prestaties biedt die vergelijkbaar zijn met die van het 405B-model tegen zeer lage kosten. Gebaseerd op de Transformer-structuur en verbeterd door middel van supervisie-fijnstelling (SFT) en versterkend leren met menselijke feedback (RLHF) voor gebruiksvriendelijkheid en veiligheid. De instructie-geoptimaliseerde versie is speciaal ontworpen voor meertalige dialogen en presteert beter dan veel open-source en gesloten chatmodellen op verschillende industriële benchmarks. Kennisafkapdatum is december 2023."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 is het meest geavanceerde meertalige open-source grote taalmodel in de Llama-serie, dat prestaties biedt die vergelijkbaar zijn met die van het 405B-model tegen zeer lage kosten. Gebaseerd op de Transformer-structuur en verbeterd door middel van supervisie-fijnstelling (SFT) en versterkend leren met menselijke feedback (RLHF) voor gebruiksvriendelijkheid en veiligheid. De instructie-geoptimaliseerde versie is speciaal ontworpen voor meertalige dialogen en presteert beter dan veel open-source en gesloten chatmodellen op verschillende industriële benchmarks. Kennisafkapdatum is december 2023."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct is het grootste en krachtigste model binnen het Llama 3.1 Instruct-model, een geavanceerd model voor conversatie-inferentie en synthetische datageneratie, dat ook kan worden gebruikt als basis voor gespecialiseerde continue pre-training of fine-tuning in specifieke domeinen. De meertalige grote taalmodellen (LLMs) die Llama 3.1 biedt, zijn een set van voorgetrainde, instructie-geoptimaliseerde generatieve modellen, waaronder 8B, 70B en 405B in grootte (tekstinvoer/uitvoer). De tekstmodellen van Llama 3.1, die zijn geoptimaliseerd voor meertalige conversatiegebruik, overtreffen veel beschikbare open-source chatmodellen in gangbare industriële benchmarktests. Llama 3.1 is ontworpen voor commercieel en onderzoeksgebruik in meerdere talen. De instructie-geoptimaliseerde tekstmodellen zijn geschikt voor assistentachtige chats, terwijl de voorgetrainde modellen zich kunnen aanpassen aan verschillende taken voor natuurlijke taalgeneratie. Het Llama 3.1-model ondersteunt ook het verbeteren van andere modellen door gebruik te maken van de output van zijn modellen, inclusief synthetische datageneratie en verfijning. Llama 3.1 is een autoregressief taalmodel dat gebruikmaakt van een geoptimaliseerde transformer-architectuur. De afgestelde versies gebruiken supervisie-finetuning (SFT) en versterkend leren met menselijke feedback (RLHF) om te voldoen aan menselijke voorkeuren voor behulpzaamheid en veiligheid."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 is een open groot taalmodel (LLM) gericht op ontwikkelaars, onderzoekers en bedrijven, ontworpen om hen te helpen bij het bouwen, experimenteren en verantwoordelijk opschalen van hun generatieve AI-ideeën. Als onderdeel van het basis systeem voor wereldwijde gemeenschapsinnovatie is het zeer geschikt voor apparaten met beperkte rekenkracht en middelen, edge-apparaten en snellere trainingstijden."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "Geavanceerd LLM, ondersteunt synthetische gegevensgeneratie, kennisdistillatie en redeneren, geschikt voor chatbots, programmeren en specifieke domeintaken."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "In staat om complexe gesprekken te ondersteunen, met uitstekende contextbegrip, redeneringsvaardigheden en tekstgeneratiecapaciteiten."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "Geavanceerd, state-of-the-art model met taalbegrip, uitstekende redeneringsvaardigheden en tekstgeneratiecapaciteiten."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "State-of-the-art visueel-taalmodel, gespecialiseerd in hoogwaardige redeneringen vanuit afbeeldingen."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "Geavanceerd, state-of-the-art klein taalmodel met taalbegrip, uitstekende redeneringsvaardigheden en tekstgeneratiecapaciteiten."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "Geavanceerd, state-of-the-art klein taalmodel met taalbegrip, uitstekende redeneringsvaardigheden en tekstgeneratiecapaciteiten."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "State-of-the-art visueel-taalmodel, gespecialiseerd in hoogwaardige redeneringen vanuit afbeeldingen."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "Geavanceerd LLM, gespecialiseerd in redeneren, wiskunde, algemene kennis en functieaanroepen."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 is een taalmodel van Microsoft AI dat uitblinkt in complexe gesprekken, meertaligheid, redenering en intelligente assistenttoepassingen."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K is een model met een superlange contextverwerkingscapaciteit, geschikt voor het genereren van zeer lange teksten, voldoet aan de behoeften van complexe generatietaken en kan tot 128.000 tokens verwerken, zeer geschikt voor onderzoek, academische en grote documentgeneratie."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Het Kimi visuele model (inclusief moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, enz.) kan de inhoud van afbeeldingen begrijpen, inclusief afbeeldingstekst, kleuren en vormen van objecten."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K biedt een gemiddelde contextverwerkingscapaciteit, kan 32.768 tokens verwerken, bijzonder geschikt voor het genereren van verschillende lange documenten en complexe gesprekken, toegepast in contentcreatie, rapportgeneratie en conversatiesystemen."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Het Kimi visuele model (inclusief moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, enz.) kan de inhoud van afbeeldingen begrijpen, inclusief afbeeldingstekst, kleuren en vormen van objecten."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K is speciaal ontworpen voor het genereren van korte teksttaken, met efficiënte verwerkingsprestaties, kan 8.192 tokens verwerken, zeer geschikt voor korte gesprekken, notities en snelle contentgeneratie."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Het Kimi visuele model (inclusief moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, enz.) kan de inhoud van afbeeldingen begrijpen, inclusief afbeeldingstekst, kleuren en vormen van objecten."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B is een upgrade van Nous Hermes 2, met de nieuwste intern ontwikkelde datasets."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B is een op maat gemaakt groot taalmodel van NVIDIA, ontworpen om de hulp te verbeteren die LLM-gebaseerde reacties bieden op gebruikersvragen."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B is een op maat gemaakt groot taalmodel van NVIDIA, ontworpen om de hulp van LLM-gegenereerde reacties op gebruikersvragen te verbeteren. Dit model presteert uitstekend in benchmarktests zoals Arena Hard, AlpacaEval 2 LC en GPT-4-Turbo MT-Bench, en staat per 1 oktober 2024 op de eerste plaats in alle drie de automatische afstemmingsbenchmarktests. Het model is getraind met RLHF (met name REINFORCE), Llama-3.1-Nemotron-70B-Reward en HelpSteer2-Preference prompts op basis van het Llama-3.1-70B-Instruct model."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "Uniek taalmodel dat ongeëvenaarde nauwkeurigheid en efficiëntie biedt."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct is een op maat gemaakt groot taalmodel van NVIDIA, ontworpen om de nuttigheid van de door LLM gegenereerde reacties te verbeteren."
+ },
"o1": {
"description": "Gefocust op geavanceerd redeneren en het oplossen van complexe problemen, inclusief wiskunde en wetenschappelijke taken. Zeer geschikt voor toepassingen die diepgaand begrip van context en agentwerkstromen vereisen."
},
- "o1-2024-12-17": {
- "description": "o1 is het nieuwe redeneer model van OpenAI, dat tekst- en afbeeldingsinvoer ondersteunt en tekstuitvoer genereert, geschikt voor complexe taken die uitgebreide algemene kennis vereisen. Dit model heeft een context van 200K en een kennisafkapdatum van oktober 2023."
- },
"o1-mini": {
"description": "o1-mini is een snel en kosteneffectief redeneermodel dat is ontworpen voor programmeer-, wiskunde- en wetenschappelijke toepassingen. Dit model heeft een context van 128K en een kennisafkapdatum van oktober 2023."
},
"o1-preview": {
"description": "o1 is het nieuwe redeneermodel van OpenAI, geschikt voor complexe taken die uitgebreide algemene kennis vereisen. Dit model heeft een context van 128K en een kennisafkapdatum van oktober 2023."
},
+ "o3-mini": {
+ "description": "o3-mini is ons nieuwste kleine inferentiemodel dat hoge intelligentie biedt met dezelfde kosten- en vertragingdoelen als o1-mini."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba is een Mamba 2-taalmodel dat zich richt op codegeneratie en krachtige ondersteuning biedt voor geavanceerde code- en inferentietaken."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini is het nieuwste model van OpenAI, gelanceerd na GPT-4 Omni, dat tekst- en afbeeldingsinvoer ondersteunt en tekstuitvoer genereert. Als hun meest geavanceerde kleine model is het veel goedkoper dan andere recente toonaangevende modellen en meer dan 60% goedkoper dan GPT-3.5 Turbo. Het behoudt de meest geavanceerde intelligentie met een aanzienlijke prijs-kwaliteitverhouding. GPT-4o mini behaalde 82% op de MMLU-test en staat momenteel hoger in chatvoorkeuren dan GPT-4."
},
- "openai/o1": {
- "description": "o1 is het nieuwe redeneer model van OpenAI, dat tekst- en afbeeldingsinvoer ondersteunt en tekstuitvoer genereert, geschikt voor complexe taken die uitgebreide algemene kennis vereisen. Dit model heeft een context van 200K en een kennisafkapdatum van oktober 2023."
- },
"openai/o1-mini": {
"description": "o1-mini is een snel en kosteneffectief redeneermodel dat is ontworpen voor programmeer-, wiskunde- en wetenschappelijke toepassingen. Dit model heeft een context van 128K en een kennisafkapdatum van oktober 2023."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K is uitgerust met een zeer grote contextverwerkingscapaciteit, in staat om tot 128K contextinformatie te verwerken. Het is bijzonder geschikt voor lange teksten die een volledige analyse en langdurige logische verbanden vereisen, en biedt een vloeiende en consistente logica met diverse ondersteuningen voor citaten in complexe tekstcommunicatie."
},
+ "qvq-72b-preview": {
+ "description": "Het QVQ-model is een experimenteel onderzoeksmodel ontwikkeld door het Qwen-team, gericht op het verbeteren van visuele redeneervaardigheden, vooral in het domein van wiskundige redenering."
+ },
"qwen-coder-plus-latest": {
"description": "Tongyi Qianwen code model."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Het Tongyi Qianwen ultra-grootschalige visuele taalmodel. In vergelijking met de verbeterde versie, verhoogt het opnieuw de visuele redeneervaardigheden en de naleving van instructies, en biedt het een hoger niveau van visuele waarneming en cognitie."
},
+ "qwen-vl-ocr-latest": {
+ "description": "Qwen OCR is een speciaal model voor tekstextractie, gericht op het extraheren van tekst uit documenten, tabellen, examenvragen, handgeschreven teksten en andere soorten afbeeldingen. Het kan verschillende talen herkennen, waaronder: Chinees, Engels, Frans, Japans, Koreaans, Duits, Russisch, Italiaans, Vietnamees en Arabisch."
+ },
"qwen-vl-plus-latest": {
"description": "De verbeterde versie van het Tongyi Qianwen grootschalige visuele taalmodel. Het verbetert aanzienlijk de detailherkenning en tekstherkenning, ondersteunt resoluties van meer dan een miljoen pixels en afbeeldingen met elke verhouding."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 is een gloednieuwe serie grote taalmodellen met sterkere begrip- en generatiecapaciteiten."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "LLM gericht op zowel Chinees als Engels, gericht op taal, programmeren, wiskunde, redeneren en meer."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "Geavanceerd LLM, ondersteunt codegeneratie, redeneren en reparatie, dekt gangbare programmeertalen."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "Krachtig middelgroot codeermodel, ondersteunt 32K contextlengte, gespecialiseerd in meertalige programmering."
+ },
"qwen2": {
"description": "Qwen2 is Alibaba's nieuwe generatie grootschalig taalmodel, ondersteunt diverse toepassingsbehoeften met uitstekende prestaties."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "Het 14B model van Tongyi Qianwen 2.5 is open source beschikbaar."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "Qwen2.5 is een open-source model van 72B schaal."
+ },
"qwen2.5-32b-instruct": {
"description": "Het 32B model van Tongyi Qianwen 2.5 is open source beschikbaar."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Het Qwen-Math model heeft krachtige capaciteiten voor het oplossen van wiskundige problemen."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "Verbeterde instructievolging, wiskunde, probleemoplossing en code, met verbeterde herkenningscapaciteiten voor verschillende formaten, directe en nauwkeurige lokalisatie van visuele elementen, ondersteuning voor lange videobestanden (maximaal 10 minuten) en seconde-niveau gebeurtenislocatie, kan tijdsvolgorde en snelheid begrijpen, en ondersteunt het bedienen van OS of mobiele agenten op basis van analyse- en lokalisatiecapaciteiten, sterke capaciteiten voor het extraheren van belangrijke informatie en JSON-formaat uitvoer, deze versie is de 72B versie, de krachtigste versie in deze serie."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "Verbeterde instructievolging, wiskunde, probleemoplossing en code, met verbeterde herkenningscapaciteiten voor verschillende formaten, directe en nauwkeurige lokalisatie van visuele elementen, ondersteuning voor lange videobestanden (maximaal 10 minuten) en seconde-niveau gebeurtenislocatie, kan tijdsvolgorde en snelheid begrijpen, en ondersteunt het bedienen van OS of mobiele agenten op basis van analyse- en lokalisatiecapaciteiten, sterke capaciteiten voor het extraheren van belangrijke informatie en JSON-formaat uitvoer, deze versie is de 72B versie, de krachtigste versie in deze serie."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 is de nieuwe generatie grootschalig taalmodel van Alibaba, dat uitstekende prestaties levert ter ondersteuning van diverse toepassingsbehoeften."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro is een zeer intelligent LLM dat is uitgebracht door Upstage, gericht op instructievolging met één GPU, met een IFEval-score van boven de 80. Momenteel ondersteunt het Engels, met een officiële versie die gepland staat voor november 2024, die de taalondersteuning en contextlengte zal uitbreiden."
},
+ "sonar": {
+ "description": "Een lichtgewicht zoekproduct op basis van contextuele zoekopdrachten, sneller en goedkoper dan Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "Een geavanceerd zoekproduct dat contextuele zoekopdrachten ondersteunt, met geavanceerde query's en vervolgacties."
+ },
+ "sonar-reasoning": {
+ "description": "Een nieuw API-product ondersteund door het DeepSeek redeneringsmodel."
+ },
"step-1-128k": {
"description": "Biedt een balans tussen prestaties en kosten, geschikt voor algemene scenario's."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "Dit model heeft krachtige video begrip capaciteiten."
},
+ "step-1o-vision-32k": {
+ "description": "Dit model heeft krachtige beeldbegripcapaciteiten. In vergelijking met de step-1v serie modellen heeft het een sterkere visuele prestatie."
+ },
"step-1v-32k": {
"description": "Ondersteunt visuele invoer, verbetert de multimodale interactie-ervaring."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Ondersteunt grootschalige contextinteracties, geschikt voor complexe gespreksscenario's."
},
+ "step-2-mini": {
+ "description": "Een razendsnel groot model gebaseerd op de nieuwe generatie zelfontwikkelde Attention-architectuur MFA, dat met zeer lage kosten vergelijkbare resultaten als step1 behaalt, terwijl het een hogere doorvoer en snellere responstijd behoudt. Het kan algemene taken verwerken en heeft speciale vaardigheden op het gebied van codering."
+ },
+ "taichu2_mm": {
+ "description": "Gecombineerd met beeldbegrip, kennisoverdracht en logische toerekening, excelleert het in het domein van vraag-en-antwoord met tekst en afbeeldingen."
+ },
"taichu_llm": {
"description": "Het Zido Tai Chu-taalmodel heeft een sterke taalbegripcapaciteit en kan tekstcreatie, kennisvragen, codeprogrammering, wiskundige berekeningen, logische redenering, sentimentanalyse, tekstsamenvattingen en meer aan. Het combineert innovatief grote data voortraining met rijke kennis uit meerdere bronnen, door algoritmische technologie continu te verfijnen en voortdurend nieuwe kennis op te nemen uit enorme tekstdata op het gebied van vocabulaire, structuur, grammatica en semantiek, waardoor de modelprestaties voortdurend evolueren. Het biedt gebruikers gemakkelijkere informatie en diensten en een meer intelligente ervaring."
},
+ "text-embedding-3-large": {
+ "description": "Het krachtigste vectorisatie model, geschikt voor Engelse en niet-Engelse taken."
+ },
+ "text-embedding-3-small": {
+ "description": "Een efficiënte en kosteneffectieve nieuwe generatie Embedding model, geschikt voor kennisretrieval, RAG-toepassingen en andere scenario's."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) biedt verbeterde rekenkracht door middel van efficiënte strategieën en modelarchitectuur."
},
+ "tts-1": {
+ "description": "Het nieuwste tekst-naar-spraak model, geoptimaliseerd voor snelheid in realtime scenario's."
+ },
+ "tts-1-hd": {
+ "description": "Het nieuwste tekst-naar-spraak model, geoptimaliseerd voor kwaliteit."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) is geschikt voor verfijnde instructietaken en biedt uitstekende taalverwerkingscapaciteiten."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet heeft de industrienormen verbeterd, met prestaties die de concurrentiemodellen en Claude 3 Opus overtreffen, en excelleert in uitgebreide evaluaties, terwijl het de snelheid en kosten van onze middelgrote modellen behoudt."
},
+ "whisper-1": {
+ "description": "Algemeen spraakherkenningsmodel, ondersteunt meertalige spraakherkenning, spraakvertaling en taalherkenning."
+ },
"wizardlm2": {
"description": "WizardLM 2 is een taalmodel van Microsoft AI dat uitblinkt in complexe gesprekken, meertaligheid, inferentie en intelligente assistentie."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Model voor complexe visuele taken, biedt hoge prestaties in beeldbegrip en analyse."
+ },
+ "yi-vision-v2": {
+ "description": "Complex visietakenmodel dat hoge prestaties biedt in begrip en analyse op basis van meerdere afbeeldingen."
}
}
diff --git a/locales/nl-NL/providers.json b/locales/nl-NL/providers.json
index d206d601da9ab..a6f1d9df0108f 100644
--- a/locales/nl-NL/providers.json
+++ b/locales/nl-NL/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure biedt een scala aan geavanceerde AI-modellen, waaronder GPT-3.5 en de nieuwste GPT-4-serie, die verschillende datatypes en complexe taken ondersteunen, met een focus op veilige, betrouwbare en duurzame AI-oplossingen."
},
+ "azureai": {
+ "description": "Azure biedt een verscheidenheid aan geavanceerde AI-modellen, waaronder GPT-3.5 en de nieuwste GPT-4-serie, die verschillende datatypes en complexe taken ondersteunt, met een focus op veilige, betrouwbare en duurzame AI-oplossingen."
+ },
"baichuan": {
"description": "Baichuan Intelligent is een bedrijf dat zich richt op de ontwikkeling van grote modellen voor kunstmatige intelligentie, wiens modellen uitblinken in Chinese taken zoals kennisencyclopedieën, lange tekstverwerking en generatieve creatie, en de mainstream modellen uit het buitenland overtreffen. Baichuan Intelligent heeft ook toonaangevende multimodale capaciteiten en presteert uitstekend in verschillende autoritatieve evaluaties. Hun modellen omvatten Baichuan 4, Baichuan 3 Turbo en Baichuan 3 Turbo 128k, die zijn geoptimaliseerd voor verschillende toepassingsscenario's en kosteneffectieve oplossingen bieden."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek is een bedrijf dat zich richt op onderzoek en toepassing van kunstmatige intelligentietechnologie, en hun nieuwste model DeepSeek-V2.5 combineert algemene dialoog- en codeverwerkingscapaciteiten, met significante verbeteringen in het afstemmen op menselijke voorkeuren, schrijfopdrachten en het volgen van instructies."
},
+ "doubao": {
+ "description": "Een door ByteDance ontwikkelde grote model. Bewezen in meer dan 50 interne zakelijke scenario's, met een dagelijks gebruik van triljoenen tokens, biedt het verschillende modaliteiten en creëert een rijke zakelijke ervaring voor bedrijven met hoogwaardige modelprestaties."
+ },
"fireworksai": {
"description": "Fireworks AI is een toonaangevende aanbieder van geavanceerde taalmodellen, met een focus op functionele aanroepen en multimodale verwerking. Hun nieuwste model Firefunction V2 is gebaseerd op Llama-3 en geoptimaliseerd voor functieaanroepen, dialogen en het volgen van instructies. Het visuele taalmodel FireLLaVA-13B ondersteunt gemengde invoer van afbeeldingen en tekst. Andere opmerkelijke modellen zijn de Llama-serie en de Mixtral-serie, die efficiënte ondersteuning bieden voor meertalig volgen van instructies en genereren."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "Een open-source organisatie die zich richt op onderzoek en ontwikkeling van tools voor grote modellen. Biedt een efficiënt en gebruiksvriendelijk open-source platform voor alle AI-ontwikkelaars, zodat de meest geavanceerde modellen en algoritmische technologieën binnen handbereik zijn."
},
+ "jina": {
+ "description": "Jina AI, opgericht in 2020, is een toonaangevend zoek-AI-bedrijf. Ons zoekplatform bevat vectormodellen, herschikkers en kleine taalmodellen, die bedrijven helpen betrouwbare en hoogwaardige generatieve AI- en multimodale zoektoepassingen te bouwen."
+ },
+ "lmstudio": {
+ "description": "LM Studio is een desktopapplicatie voor het ontwikkelen en experimenteren met LLM's op uw computer."
+ },
"minimax": {
"description": "MiniMax is een algemeen kunstmatige intelligentietechnologiebedrijf dat in 2021 is opgericht, en zich richt op co-creatie van intelligentie met gebruikers. MiniMax heeft verschillende multimodale algemene grote modellen ontwikkeld, waaronder een MoE-tekstgrootmodel met triljoenen parameters, een spraakgrootmodel en een afbeeldingsgrootmodel. Ze hebben ook toepassingen zoals Conch AI gelanceerd."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI is een platform dat API-diensten biedt voor verschillende grote taalmodellen en AI-beeldgeneratie, flexibel, betrouwbaar en kosteneffectief. Het ondersteunt de nieuwste open-source modellen zoals Llama3 en Mistral, en biedt een uitgebreide, gebruiksvriendelijke en automatisch schaalbare API-oplossing voor de ontwikkeling van generatieve AI-toepassingen, geschikt voor de snelle groei van AI-startups."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ biedt containers voor zelf-gehoste GPU-versnelde inferentie-microservices, die de implementatie van voorgetrainde en aangepaste AI-modellen in de cloud, datacenters, RTX™ AI-pc's en werkstations ondersteunen."
+ },
"ollama": {
"description": "De modellen van Ollama bestrijken een breed scala aan gebieden, waaronder codegeneratie, wiskundige berekeningen, meertalige verwerking en interactieve dialogen, en voldoen aan de diverse behoeften van bedrijfs- en lokale implementaties."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "Het Instituut voor Automatisering van de Chinese Academie van Wetenschappen en het Wuhan Instituut voor Kunstmatige Intelligentie hebben een nieuwe generatie multimodale grote modellen gelanceerd, die ondersteuning bieden voor meerdaagse vraag-en-antwoord, tekstcreatie, beeldgeneratie, 3D-begrip, signaalanalyse en andere uitgebreide vraag-en-antwoordtaken, met sterkere cognitieve, begrip en creatiecapaciteiten, wat zorgt voor een geheel nieuwe interactie-ervaring."
},
+ "tencentcloud": {
+ "description": "De atomische capaciteiten van de kennisengine (LLM Knowledge Engine Atomic Power) zijn gebaseerd op de ontwikkeling van de kennisengine en bieden een volledige keten van kennisvraag- en antwoordmogelijkheden, gericht op bedrijven en ontwikkelaars. U kunt verschillende atomische capaciteiten gebruiken om uw eigen modelservice samen te stellen, door gebruik te maken van diensten zoals documentanalyse, splitsing, embedding en meervoudige herschrijving, om een op maat gemaakte AI-oplossing voor uw bedrijf te creëren."
+ },
"togetherai": {
"description": "Together AI streeft ernaar toonaangevende prestaties te bereiken door middel van innovatieve AI-modellen, en biedt uitgebreide aanpassingsmogelijkheden, waaronder ondersteuning voor snelle schaling en intuïtieve implementatieprocessen, om aan de verschillende behoeften van bedrijven te voldoen."
},
"upstage": {
"description": "Upstage richt zich op het ontwikkelen van AI-modellen voor verschillende zakelijke behoeften, waaronder Solar LLM en document AI, met als doel het realiseren van kunstmatige algemene intelligentie (AGI). Het creëert eenvoudige dialoogagenten via de Chat API en ondersteunt functionele aanroepen, vertalingen, insluitingen en specifieke domeintoepassingen."
},
+ "vllm": {
+ "description": "vLLM is een snelle en gebruiksvriendelijke bibliotheek voor LLM-inferentie en -diensten."
+ },
+ "volcengine": {
+ "description": "Het ontwikkelingsplatform voor de grote modellenservice van ByteDance, dat een breed scala aan functies biedt, veilig is en concurrerende prijzen heeft voor modelaanroepdiensten. Het biedt ook end-to-end functionaliteiten zoals modelgegevens, fine-tuning, inferentie en evaluatie, om de ontwikkeling van uw AI-toepassingen volledig te ondersteunen."
+ },
"wenxin": {
"description": "Een enterprise-grade, alles-in-één platform voor de ontwikkeling en service van grote modellen en AI-native applicaties, dat de meest uitgebreide en gebruiksvriendelijke toolchain biedt voor de ontwikkeling van generatieve kunstmatige intelligentiemodellen en applicaties."
},
diff --git a/locales/nl-NL/setting.json b/locales/nl-NL/setting.json
index c44e43b37f366..8087fb6fdd895 100644
--- a/locales/nl-NL/setting.json
+++ b/locales/nl-NL/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Limiet voor enkele reacties inschakelen"
},
+ "enableReasoningEffort": {
+ "title": "Inschakelen van redeneringsinspanningsaanpassing"
+ },
"frequencyPenalty": {
- "desc": "Hoe hoger de waarde, hoe waarschijnlijker het is dat herhaalde woorden worden verminderd",
- "title": "Frequentieboete"
+ "desc": "Hoe hoger de waarde, hoe rijker en gevarieerder de woordkeuze; hoe lager de waarde, hoe eenvoudiger en directer de woordkeuze",
+ "title": "Woordenschat diversiteit"
},
"maxTokens": {
"desc": "Het maximale aantal tokens dat wordt gebruikt voor een enkele interactie",
@@ -212,19 +215,31 @@
"desc": "{{provider}} model",
"title": "Model"
},
+ "params": {
+ "title": "Geavanceerde parameters"
+ },
"presencePenalty": {
- "desc": "Hoe hoger de waarde, hoe waarschijnlijker het is dat het gesprek naar nieuwe onderwerpen wordt uitgebreid",
- "title": "Onderwerpnieuwheid"
+ "desc": "Hoe hoger de waarde, hoe meer de neiging om verschillende uitdrukkingen te gebruiken en herhaling van concepten te vermijden; hoe lager de waarde, hoe meer de neiging om herhalende concepten of verhalen te gebruiken, wat zorgt voor meer consistentie in de uitdrukking",
+ "title": "Uitdrukkingsdiversiteit"
+ },
+ "reasoningEffort": {
+ "desc": "Hoe hoger de waarde, hoe sterker de redeneringscapaciteit, maar dit kan de responstijd en het tokenverbruik verhogen",
+ "options": {
+ "high": "Hoog",
+ "low": "Laag",
+ "medium": "Gemiddeld"
+ },
+ "title": "Redeneringsinspanningsniveau"
},
"temperature": {
- "desc": "Hoe hoger de waarde, hoe willekeuriger de reactie",
- "title": "Willekeurigheid",
- "titleWithValue": "Willekeurigheid {{value}}"
+ "desc": "Hoe hoger de waarde, hoe creatiever en fantasierijker het antwoord; hoe lager de waarde, hoe strikter het antwoord.",
+ "title": "Creativiteit Activiteit",
+ "warning": "Een te hoge waarde voor creativiteit activiteit kan leiden tot onleesbare output."
},
"title": "Modelinstellingen",
"topP": {
- "desc": "Vergelijkbaar met willekeurigheid, maar verander dit niet samen met willekeurigheid",
- "title": "Top-P-monstername"
+ "desc": "Hoeveel mogelijkheden er in overweging worden genomen; hoe hoger de waarde, hoe meer mogelijke antwoorden worden geaccepteerd; hoe lager de waarde, hoe meer de voorkeur uitgaat naar het meest waarschijnlijke antwoord. Het wordt niet aanbevolen om dit samen met creativiteit activiteit te wijzigen.",
+ "title": "Openheid van Denken"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Algemene instellingen",
"experiment": "Experiment",
"llm": "Taalmodel",
+ "provider": "AI-dienstverlener",
"sync": "Cloudsynchronisatie",
"system-agent": "Systeemassistent",
"tts": "Spraakdienst"
diff --git a/locales/pl-PL/changelog.json b/locales/pl-PL/changelog.json
index b6a277dc37070..65b0aec8442cc 100644
--- a/locales/pl-PL/changelog.json
+++ b/locales/pl-PL/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "Zobacz wszystkie dzienniki zmian",
"description": "Na bieżąco śledź nowe funkcje i ulepszenia {{appName}}",
"pagination": {
- "older": "Zobacz wcześniejsze zmiany",
- "prev": "Poprzednia strona"
+ "next": "Następna strona",
+ "older": "Zobacz wcześniejsze zmiany"
},
"readDetails": "Przeczytaj szczegóły",
"title": "Dziennik zmian",
diff --git a/locales/pl-PL/common.json b/locales/pl-PL/common.json
index 6a52fc52bbbdb..4c0af2303651b 100644
--- a/locales/pl-PL/common.json
+++ b/locales/pl-PL/common.json
@@ -275,6 +275,7 @@
},
"temp": "Tymczasowy",
"terms": "Warunki korzystania",
+ "update": "Aktualizuj",
"updateAgent": "Zaktualizuj informacje o agencie",
"upgradeVersion": {
"action": "Aktualizuj",
@@ -285,6 +286,7 @@
"anonymousNickName": "Użytkownik Anonimowy",
"billing": "Zarządzanie rachunkami",
"cloud": "Wypróbuj {{name}}",
+ "community": "Wersja społeczności",
"data": "Przechowywanie danych",
"defaultNickname": "Użytkownik Wersji Społecznościowej",
"discord": "Wsparcie społeczności",
@@ -294,7 +296,6 @@
"help": "Centrum pomocy",
"moveGuide": "Przenieś przycisk ustawień tutaj",
"plans": "Plan abonamentu",
- "preview": "Podgląd",
"profile": "Zarządzanie kontem",
"setting": "Ustawienia aplikacji",
"usages": "Statystyki użycia"
diff --git a/locales/pl-PL/components.json b/locales/pl-PL/components.json
index 700366da029b3..c1f013f4a7302 100644
--- a/locales/pl-PL/components.json
+++ b/locales/pl-PL/components.json
@@ -76,6 +76,7 @@
"custom": "Niestandardowy model, domyślnie obsługujący zarówno wywołania funkcji, jak i rozpoznawanie wizualne. Proszę zweryfikować możliwość użycia tych funkcji w praktyce.",
"file": "Ten model obsługuje wczytywanie plików i rozpoznawanie",
"functionCall": "Ten model obsługuje wywołania funkcji (Function Call).",
+ "reasoning": "Ten model wspiera głębokie myślenie",
"tokens": "Ten model obsługuje maksymalnie {{tokens}} tokenów w pojedynczej sesji.",
"vision": "Ten model obsługuje rozpoznawanie wizualne."
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "Brak włączonych modeli, przejdź do ustawień i włącz je",
"provider": "Dostawca"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "Z powodu ograniczeń bezpieczeństwa przeglądarki, musisz skonfigurować CORS dla Ollama, aby móc go używać.",
+ "linux": {
+ "env": "Dodaj `Environment` w sekcji [Service], dodając zmienną środowiskową OLLAMA_ORIGINS:",
+ "reboot": "Przeładuj systemd i uruchom ponownie Ollama",
+ "systemd": "Wywołaj systemd, aby edytować usługę ollama:"
+ },
+ "macos": "Otwórz aplikację „Terminal” i wklej poniższe polecenie, a następnie naciśnij Enter, aby je uruchomić",
+ "reboot": "Po zakończeniu wykonania, uruchom ponownie usługę Ollama",
+ "title": "Skonfiguruj Ollama, aby zezwolić na dostęp międzydomenowy",
+ "windows": "Na Windowsie, kliknij „Panel sterowania”, aby edytować zmienne środowiskowe systemu. Utwórz nową zmienną środowiskową o nazwie „OLLAMA_ORIGINS” dla swojego konta użytkownika, ustawiając wartość na * i kliknij „OK/Zastosuj”, aby zapisać"
+ },
+ "install": {
+ "description": "Upewnij się, że uruchomiłeś Ollama. Jeśli nie masz Ollama, przejdź na oficjalną stronę <1>pobierz",
+ "docker": "Jeśli wolisz używać Dockera, Ollama również oferuje oficjalny obraz Dockera, który możesz pobrać za pomocą poniższego polecenia:",
+ "linux": {
+ "command": "Zainstaluj za pomocą poniższego polecenia:",
+ "manual": "Alternatywnie, możesz również zapoznać się z <1>podręcznikiem instalacji ręcznej dla Linuxa, aby zainstalować samodzielnie"
+ },
+ "title": "Zainstaluj i uruchom aplikację Ollama lokalnie",
+ "windowsTab": "Windows (wersja podglądowa)"
+ }
+ },
+ "Thinking": {
+ "thinking": "Głęboko myślę...",
+ "thought": "Głęboko przemyślane (czas: {{duration}} sekund)",
+ "thoughtWithDuration": "Głęboko przemyślane"
}
}
diff --git a/locales/pl-PL/discover.json b/locales/pl-PL/discover.json
index fb78a54a82220..a9071ae39d6c9 100644
--- a/locales/pl-PL/discover.json
+++ b/locales/pl-PL/discover.json
@@ -126,6 +126,10 @@
"title": "Świeżość tematu"
},
"range": "Zakres",
+ "reasoning_effort": {
+ "desc": "To ustawienie kontroluje intensywność rozumowania modelu przed wygenerowaniem odpowiedzi. Niska intensywność priorytetowo traktuje szybkość odpowiedzi i oszczędza tokeny, podczas gdy wysoka intensywność zapewnia pełniejsze rozumowanie, ale zużywa więcej tokenów i obniża szybkość odpowiedzi. Wartość domyślna to średnia, co równoważy dokładność rozumowania z szybkością odpowiedzi.",
+ "title": "Intensywność rozumowania"
+ },
"temperature": {
"desc": "To ustawienie wpływa na różnorodność odpowiedzi modelu. Niższe wartości prowadzą do bardziej przewidywalnych i typowych odpowiedzi, podczas gdy wyższe wartości zachęcają do bardziej zróżnicowanych i rzadziej spotykanych odpowiedzi. Gdy wartość wynosi 0, model zawsze daje tę samą odpowiedź na dane wejście.",
"title": "Losowość"
diff --git a/locales/pl-PL/modelProvider.json b/locales/pl-PL/modelProvider.json
index 58bf86f192962..729380fafa9bf 100644
--- a/locales/pl-PL/modelProvider.json
+++ b/locales/pl-PL/modelProvider.json
@@ -19,6 +19,24 @@
"title": "Klucz API"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Wersja API Azure, w formacie YYYY-MM-DD, sprawdź [najnowszą wersję](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "Pobierz listę",
+ "title": "Wersja API Azure"
+ },
+ "endpoint": {
+ "desc": "Znajdź punkt końcowy wnioskowania modelu Azure AI w przeglądzie projektu Azure AI",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Punkt końcowy Azure AI"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Znajdź klucz API w przeglądzie projektu Azure AI",
+ "placeholder": "Klucz Azure",
+ "title": "Klucz"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "Wprowadź AWS Access Key Id",
@@ -63,6 +81,46 @@
"title": "ID konta Cloudflare / adres API"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Proszę wpisać swój klucz API",
+ "title": "Klucz API"
+ },
+ "basicTitle": "Podstawowe informacje",
+ "configTitle": "Informacje konfiguracyjne",
+ "confirm": "Utwórz",
+ "createSuccess": "Utworzenie zakończone sukcesem",
+ "description": {
+ "placeholder": "Opis dostawcy usług (opcjonalnie)",
+ "title": "Opis dostawcy usług"
+ },
+ "id": {
+ "desc": "Unikalny identyfikator dostawcy usług, po utworzeniu nie można go zmienić",
+ "format": "Może zawierać tylko cyfry, małe litery, myślniki (-) i podkreślenia (_) ",
+ "placeholder": "Zaleca się użycie małych liter, np. openai, po utworzeniu nie można edytować",
+ "required": "Proszę wpisać identyfikator dostawcy",
+ "title": "Identyfikator dostawcy"
+ },
+ "logo": {
+ "required": "Proszę przesłać poprawne logo dostawcy",
+ "title": "Logo dostawcy"
+ },
+ "name": {
+ "placeholder": "Proszę wpisać nazwę wyświetlaną dostawcy",
+ "required": "Proszę wpisać nazwę dostawcy",
+ "title": "Nazwa dostawcy"
+ },
+ "proxyUrl": {
+ "required": "Proszę wpisać adres proxy",
+ "title": "Adres proxy"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Proszę wybrać typ SDK",
+ "title": "Format żądania"
+ },
+ "title": "Utwórz niestandardowego dostawcę AI"
+ },
"github": {
"personalAccessToken": {
"desc": "Wprowadź swój osobisty token dostępu GitHub (PAT), kliknij [tutaj](https://github.com/settings/tokens), aby go utworzyć",
@@ -77,6 +135,23 @@
"title": "Token HuggingFace"
}
},
+ "list": {
+ "title": {
+ "disabled": "Usługa nieaktywna",
+ "enabled": "Usługa aktywna"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Dodaj niestandardowego dostawcę",
+ "all": "Wszystko",
+ "list": {
+ "disabled": "Nieaktywny",
+ "enabled": "Aktywny"
+ },
+ "notFound": "Nie znaleziono wyników wyszukiwania",
+ "searchProviders": "Szukaj dostawców...",
+ "sort": "Niestandardowe sortowanie"
+ },
"ollama": {
"checker": {
"desc": "Test czy adres proxy jest poprawnie wypełniony",
@@ -94,33 +169,9 @@
"title": "Pobieranie modelu {{model}}"
},
"endpoint": {
- "desc": "Wprowadź adres rest API Ollama, jeśli lokalnie nie określono, pozostaw puste",
+ "desc": "Musi zawierać http(s)://, lokalnie, jeśli nie określono inaczej, można pozostawić puste",
"title": "Adres proxy API"
},
- "setup": {
- "cors": {
- "description": "Z powodu ograniczeń bezpieczeństwa przeglądarki, musisz skonfigurować CORS dla Ollama, aby móc go używać.",
- "linux": {
- "env": "Dodaj `Environment` w sekcji [Service], dodaj zmienną środowiskową OLLAMA_ORIGINS:",
- "reboot": "Przeładuj systemd i uruchom ponownie Ollama",
- "systemd": "Użyj systemd, aby edytować usługę ollama:"
- },
- "macos": "Otwórz aplikację „Terminal” i wklej poniższe polecenie, a następnie naciśnij Enter",
- "reboot": "Proszę ponownie uruchomić usługę Ollama po zakończeniu",
- "title": "Konfiguracja Ollama do zezwolenia na dostęp CORS",
- "windows": "Na Windowsie, kliknij „Panel sterowania”, aby edytować zmienne środowiskowe systemu. Utwórz nową zmienną środowiskową o nazwie „OLLAMA_ORIGINS” dla swojego konta użytkownika, ustaw wartość na *, a następnie kliknij „OK/Zastosuj”, aby zapisać"
- },
- "install": {
- "description": "Proszę upewnić się, że Ollama jest uruchomione, jeśli nie pobrałeś Ollama, odwiedź oficjalną stronę <1>pobierz",
- "docker": "Jeśli wolisz używać Dockera, Ollama również oferuje oficjalny obraz Dockera, który możesz pobrać za pomocą poniższego polecenia:",
- "linux": {
- "command": "Zainstaluj za pomocą poniższego polecenia:",
- "manual": "Alternatywnie, możesz zapoznać się z <1>podręcznikiem instalacji ręcznej dla Linuxa, aby zainstalować samodzielnie"
- },
- "title": "Zainstaluj i uruchom aplikację Ollama lokalnie",
- "windowsTab": "Windows (wersja podglądowa)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Anuluj pobieranie",
@@ -131,25 +182,145 @@
"title": "Pobierz określony model Ollama"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Wprowadź Access Key z platformy Baidu Qianfan",
- "placeholder": "Access Key Qianfan",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "Twój klucz oraz adres proxy będą szyfrowane za pomocą <1>AES-GCM",
+ "apiKey": {
+ "desc": "Proszę wpisać swój {{name}} klucz API",
+ "placeholder": "{{name}} klucz API",
+ "title": "Klucz API"
+ },
+ "baseURL": {
+ "desc": "Musi zawierać http(s)://",
+ "invalid": "Proszę wprowadzić prawidłowy URL",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "Adres proxy API"
+ },
+ "checker": {
+ "button": "Sprawdź",
+ "desc": "Testuj, czy klucz API i adres proxy są poprawnie wpisane",
+ "pass": "Sprawdzenie zakończone sukcesem",
+ "title": "Sprawdzenie łączności"
+ },
+ "fetchOnClient": {
+ "desc": "Tryb żądania klienta rozpocznie sesję bezpośrednio z przeglądarki, co może przyspieszyć czas odpowiedzi",
+ "title": "Użyj trybu żądania klienta"
+ },
+ "helpDoc": "Dokumentacja konfiguracyjna",
+ "waitingForMore": "Więcej modeli jest w <1>planach integracji, proszę czekać"
},
- "checker": {
- "desc": "Sprawdź, czy AccessKey / SecretAccess zostały poprawnie wprowadzone"
+ "createNew": {
+ "title": "Utwórz niestandardowy model AI"
},
- "secretKey": {
- "desc": "Wprowadź Secret Key z platformy Baidu Qianfan",
- "placeholder": "Secret Key Qianfan",
- "title": "Secret Key"
+ "item": {
+ "config": "Konfiguracja modelu",
+ "customModelCards": {
+ "addNew": "Utwórz i dodaj model {{id}}",
+ "confirmDelete": "Zaraz usuniesz ten niestandardowy model, po usunięciu nie będzie można go przywrócić, proszę działać ostrożnie."
+ },
+ "delete": {
+ "confirm": "Czy na pewno chcesz usunąć model {{displayName}}?",
+ "success": "Usunięcie zakończone sukcesem",
+ "title": "Usuń model"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Pole, które jest rzeczywiście używane w Azure OpenAI",
+ "placeholder": "Proszę wpisać nazwę wdrożenia modelu w Azure",
+ "title": "Nazwa wdrożenia modelu"
+ },
+ "deployName": {
+ "extra": "To pole będzie używane jako identyfikator modelu podczas wysyłania żądania",
+ "placeholder": "Wprowadź rzeczywistą nazwę lub identyfikator wdrożenia modelu",
+ "title": "Nazwa wdrożenia modelu"
+ },
+ "displayName": {
+ "placeholder": "Proszę wpisać nazwę wyświetlaną modelu, np. ChatGPT, GPT-4 itp.",
+ "title": "Nazwa wyświetlana modelu"
+ },
+ "files": {
+ "extra": "Obecna implementacja przesyłania plików jest jedynie rozwiązaniem hackowym, przeznaczonym do samodzielnego testowania. Pełna funkcjonalność przesyłania plików będzie dostępna w przyszłości.",
+ "title": "Wsparcie dla przesyłania plików"
+ },
+ "functionCall": {
+ "extra": "Ta konfiguracja włączy jedynie możliwość korzystania z narzędzi przez model, co pozwoli na dodanie wtyczek narzędziowych. Jednakże, czy model rzeczywiście obsługuje korzystanie z narzędzi, zależy całkowicie od samego modelu, proszę samodzielnie przetestować jego użyteczność",
+ "title": "Wsparcie dla korzystania z narzędzi"
+ },
+ "id": {
+ "extra": "Nie można zmieniać po utworzeniu, będzie używane jako identyfikator modelu podczas wywoływania AI",
+ "placeholder": "Wprowadź identyfikator modelu, na przykład gpt-4o lub claude-3.5-sonnet",
+ "title": "ID modelu"
+ },
+ "modalTitle": "Konfiguracja niestandardowego modelu",
+ "reasoning": {
+ "extra": "Ta konfiguracja włączy jedynie zdolność modelu do głębokiego myślenia, a konkretne efekty w pełni zależą od samego modelu. Proszę samodzielnie przetestować, czy model ma zdolność do głębokiego myślenia.",
+ "title": "Wsparcie dla głębokiego myślenia"
+ },
+ "tokens": {
+ "extra": "Ustaw maksymalną liczbę tokenów wspieranych przez model",
+ "title": "Maksymalne okno kontekstu",
+ "unlimited": "Bez ograniczeń"
+ },
+ "vision": {
+ "extra": "Ta konfiguracja włączy tylko możliwość przesyłania obrazów w aplikacji, czy model obsługuje rozpoznawanie zależy od samego modelu, proszę samodzielnie przetestować dostępność rozpoznawania wizualnego tego modelu.",
+ "title": "Wsparcie dla rozpoznawania wizualnego"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/obraz",
+ "inputCharts": "${{amount}}/M znaków",
+ "inputMinutes": "${{amount}}/minut",
+ "inputTokens": "Wprowadzenie ${{amount}}/M",
+ "outputTokens": "Wyjście ${{amount}}/M"
+ },
+ "releasedAt": "Wydano {{releasedAt}}"
},
- "unlock": {
- "customRegion": "Niestandardowy obszar usług",
- "description": "Wprowadź swój AccessKey / SecretKey, aby rozpocząć sesję. Aplikacja nie zapisuje twojej konfiguracji uwierzytelniania",
- "title": "Użyj niestandardowych informacji uwierzytelniających Wenxin Yiyan"
- }
+ "list": {
+ "addNew": "Dodaj model",
+ "disabled": "Nieaktywne",
+ "disabledActions": {
+ "showMore": "Pokaż więcej"
+ },
+ "empty": {
+ "desc": "Proszę utworzyć model niestandardowy lub pobrać model, aby rozpocząć korzystanie.",
+ "title": "Brak dostępnych modeli"
+ },
+ "enabled": "Aktywne",
+ "enabledActions": {
+ "disableAll": "Dezaktywuj wszystkie",
+ "enableAll": "Aktywuj wszystkie",
+ "sort": "Sortowanie modeli niestandardowych"
+ },
+ "enabledEmpty": "Brak aktywnych modeli, aktywuj ulubione modele z poniższej listy~",
+ "fetcher": {
+ "clear": "Wyczyść pobrane modele",
+ "fetch": "Pobierz listę modeli",
+ "fetching": "Pobieranie listy modeli...",
+ "latestTime": "Ostatnia aktualizacja: {{time}}",
+ "noLatestTime": "Lista nie została jeszcze pobrana"
+ },
+ "resetAll": {
+ "conform": "Czy na pewno chcesz zresetować wszystkie zmiany w bieżącym modelu? Po zresetowaniu lista modeli wróci do stanu domyślnego",
+ "success": "Resetowanie zakończone sukcesem",
+ "title": "Zresetuj wszystkie zmiany"
+ },
+ "search": "Szukaj modeli...",
+ "searchResult": "Znaleziono {{count}} modeli",
+ "title": "Lista modeli",
+ "total": "Łącznie dostępnych modeli: {{count}}"
+ },
+ "searchNotFound": "Nie znaleziono wyników wyszukiwania"
+ },
+ "sortModal": {
+ "success": "Aktualizacja sortowania zakończona sukcesem",
+ "title": "Niestandardowe sortowanie",
+ "update": "Aktualizuj"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "Zaraz usuniesz tego dostawcę AI, po usunięciu nie będzie można go przywrócić, czy na pewno chcesz usunąć?",
+ "deleteSuccess": "Usunięcie zakończone sukcesem",
+ "tooltip": "Aktualizuj podstawowe ustawienia dostawcy",
+ "updateSuccess": "Aktualizacja zakończona sukcesem"
},
"zeroone": {
"title": "01.AI Zero Jeden Wszystko"
diff --git a/locales/pl-PL/models.json b/locales/pl-PL/models.json
index 3dff164759b16..9f2939319b3ba 100644
--- a/locales/pl-PL/models.json
+++ b/locales/pl-PL/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K kładzie nacisk na bezpieczeństwo semantyczne i odpowiedzialność, zaprojektowany specjalnie dla aplikacji o wysokich wymaganiach dotyczących bezpieczeństwa treści, zapewniając dokładność i stabilność doświadczeń użytkowników."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 wykorzystuje wyszukiwanie drzew do budowy łańcucha myślenia i wprowadza mechanizm refleksji, wykorzystując uczenie przez wzmocnienie, model posiada zdolność do samorefleksji i korekty błędów."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro to zaawansowany model przetwarzania języka naturalnego wydany przez firmę 360, charakteryzujący się doskonałymi zdolnościami generowania i rozumienia tekstu, szczególnie w obszarze generowania i tworzenia treści, zdolny do obsługi skomplikowanych zadań związanych z konwersją językową i odgrywaniem ról."
},
+ "360zhinao2-o1": {
+ "description": "Model 360zhinao2-o1 wykorzystuje wyszukiwanie drzewne do budowy łańcucha myślowego i wprowadza mechanizm refleksji, wykorzystując uczenie przez wzmocnienie do treningu, co pozwala modelowi na samorefleksję i korekcję błędów."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra to najsilniejsza wersja w serii modeli Spark, która, oprócz ulepszonego łącza wyszukiwania w sieci, zwiększa zdolność rozumienia i podsumowywania treści tekstowych. Jest to kompleksowe rozwiązanie mające na celu zwiększenie wydajności biurowej i dokładne odpowiadanie na potrzeby, stanowiące inteligentny produkt wiodący w branży."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "Model o najlepszych możliwościach w kraju, przewyższający zagraniczne modele w zadaniach związanych z wiedzą encyklopedyczną, długimi tekstami i twórczością w języku chińskim. Posiada również wiodące w branży możliwości multimodalne, osiągając doskonałe wyniki w wielu autorytatywnych testach."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Model destylacyjny DeepSeek-R1 oparty na Qwen2.5-Math-1.5B, optymalizujący wydajność wnioskowania dzięki uczeniu przez wzmocnienie i danym z zimnego startu, otwarty model ustanawiający nowe standardy w wielu zadaniach."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Model destylacyjny DeepSeek-R1 oparty na Qwen2.5-14B, optymalizujący wydajność wnioskowania dzięki uczeniu przez wzmocnienie i danym z zimnego startu, otwarty model ustanawiający nowe standardy w wielu zadaniach."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "Seria DeepSeek-R1 optymalizuje wydajność wnioskowania dzięki uczeniu przez wzmocnienie i danym z zimnego startu, otwarty model ustanawiający nowe standardy w wielu zadaniach, przewyższający poziom OpenAI-o1-mini."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "Model destylacyjny DeepSeek-R1 oparty na Qwen2.5-Math-7B, optymalizujący wydajność wnioskowania dzięki uczeniu przez wzmocnienie i danym z zimnego startu, otwarty model ustanawiający nowe standardy w wielu zadaniach."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite cechuje się ekstremalną szybkością reakcji i lepszym stosunkiem jakości do ceny, oferując klientom elastyczność w różnych scenariuszach. Obsługuje wnioskowanie i dostosowywanie z kontekstem 128k."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Flagowy model ultra dużego języka opracowany przez Baidu, w porównaniu do ERNIE 3.5, oferujący kompleksową aktualizację możliwości modelu, szeroko stosowany w złożonych scenariuszach w różnych dziedzinach; wspiera automatyczne połączenie z wtyczką wyszukiwania Baidu, zapewniając aktualność informacji."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Flagowy model językowy opracowany przez Baidu, o dużej skali, wykazujący doskonałe wyniki w złożonych zadaniach w różnych dziedzinach; wspiera automatyczne połączenie z wtyczką wyszukiwania Baidu, zapewniając aktualność informacji w odpowiedziach. W porównaniu do ERNIE 4.0, wykazuje lepsze osiągi."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "Opracowany przez Baidu flagowy, ultra-duży model językowy, który wykazuje doskonałe ogólne rezultaty i jest szeroko stosowany w złożonych zadaniach w różnych dziedzinach; obsługuje automatyczne łączenie z wtyczką wyszukiwania Baidu, zapewniając aktualność informacji odpowiadających na pytania. W porównaniu do ERNIE 4.0 wykazuje lepszą wydajność."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct to jeden z najnowszych modeli dużych języków wydanych przez Alibaba Cloud. Model 7B ma znacząco poprawione zdolności w zakresie kodowania i matematyki. Oferuje również wsparcie dla wielu języków, obejmując ponad 29 języków, w tym chiński i angielski. Model ten wykazuje znaczną poprawę w zakresie przestrzegania instrukcji, rozumienia danych strukturalnych oraz generowania strukturalnych wyników (szczególnie JSON)."
},
+ "MiniMax-Text-01": {
+ "description": "W serii modeli MiniMax-01 wprowadziliśmy odważne innowacje: po raz pierwszy na dużą skalę zrealizowano mechanizm liniowej uwagi, tradycyjna architektura Transformera nie jest już jedynym wyborem. Liczba parametrów tego modelu wynosi aż 456 miliardów, z aktywacją wynoszącą 45,9 miliarda. Ogólna wydajność modelu dorównuje najlepszym modelom zagranicznym, jednocześnie efektywnie przetwarzając kontekst o długości do 4 milionów tokenów, co stanowi 32 razy więcej niż GPT-4o i 20 razy więcej niż Claude-3.5-Sonnet."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO to wysoce elastyczna fuzja wielu modeli, mająca na celu zapewnienie doskonałego doświadczenia twórczego."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 pokazuje wyjątkowe wyniki w różnych zadaniach językowych i wizualnych, w tym zrozumieniu dokumentów i wykresów, zrozumieniu tekstu w scenach, OCR, rozwiązywaniu problemów naukowych i matematycznych."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 pokazuje wyjątkowe wyniki w różnych zadaniach językowych i wizualnych, w tym zrozumieniu dokumentów i wykresów, zrozumieniu tekstu w scenach, OCR, rozwiązywaniu problemów naukowych i matematycznych."
- },
"Phi-3-medium-128k-instruct": {
"description": "Ten sam model Phi-3-medium, ale z większym rozmiarem kontekstu do RAG lub kilku strzałowego wywoływania."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat to otwarta wersja modelu pretrenowanego z serii GLM-4, wydana przez Zhipu AI. Model ten wykazuje doskonałe wyniki w zakresie semantyki, matematyki, wnioskowania, kodu i wiedzy. Oprócz wsparcia dla wieloetapowych rozmów, GLM-4-9B-Chat oferuje również zaawansowane funkcje, takie jak przeglądanie stron internetowych, wykonywanie kodu, wywoływanie niestandardowych narzędzi (Function Call) oraz wnioskowanie z długich tekstów. Model obsługuje 26 języków, w tym chiński, angielski, japoński, koreański i niemiecki. W wielu testach benchmarkowych, takich jak AlignBench-v2, MT-Bench, MMLU i C-Eval, GLM-4-9B-Chat wykazuje doskonałą wydajność. Model obsługuje maksymalną długość kontekstu 128K, co czyni go odpowiednim do badań akademickich i zastosowań komercyjnych."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 to model wnioskowania napędzany uczeniem ze wzmocnieniem (RL), który rozwiązuje problemy z powtarzalnością i czytelnością modeli. Przed RL, DeepSeek-R1 wprowadził dane do zimnego startu, co dodatkowo zoptymalizowało wydajność wnioskowania. W zadaniach matematycznych, kodowych i wnioskowania, osiąga wyniki porównywalne z OpenAI-o1, a dzięki starannie zaprojektowanym metodom treningowym poprawia ogólne wyniki."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 to model językowy z 6710 miliardami parametrów, oparty na architekturze mieszanych ekspertów (MoE), wykorzystujący wielogłowicową potencjalną uwagę (MLA) oraz strategię równoważenia obciążenia bez dodatkowych strat, co optymalizuje wydajność wnioskowania i treningu. Dzięki wstępnemu treningowi na 14,8 bilionach wysokiej jakości tokenów oraz nadzorowanemu dostrajaniu i uczeniu ze wzmocnieniem, DeepSeek-V3 przewyższa inne modele open source, zbliżając się do wiodących modeli zamkniętych."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma to jedna z lekkich, nowoczesnych otwartych serii modeli opracowanych przez Google. Jest to duży model językowy z jedynie dekoderem, wspierający język angielski, oferujący otwarte wagi, pretrenowane warianty oraz warianty dostosowane do instrukcji. Model Gemma nadaje się do różnych zadań generowania tekstu, w tym pytania-odpowiedzi, streszczenia i wnioskowania. Model 9B został przeszkolony na 8 bilionach tokenów. Jego stosunkowo mała skala umożliwia wdrożenie w środowiskach o ograniczonych zasobach, takich jak laptopy, komputery stacjonarne lub własna infrastruktura chmurowa, co umożliwia większej liczbie osób dostęp do nowoczesnych modeli AI i wspiera innowacje."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 to rodzina dużych modeli językowych opracowanych przez Meta, obejmująca pretrenowane i dostosowane do instrukcji warianty o rozmiarach parametrów 8B, 70B i 405B. Model 8B dostosowany do instrukcji został zoptymalizowany do scenariuszy rozmów wielojęzycznych, osiągając doskonałe wyniki w wielu branżowych testach benchmarkowych. Trening modelu wykorzystał ponad 150 bilionów tokenów danych publicznych oraz zastosował techniki takie jak nadzorowane dostrajanie i uczenie przez wzmacnianie z ludzkim feedbackiem, aby zwiększyć użyteczność i bezpieczeństwo modelu. Llama 3.1 wspiera generowanie tekstu i kodu, a data graniczna wiedzy to grudzień 2023 roku."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview to innowacyjny model przetwarzania języka naturalnego, który efektywnie radzi sobie z złożonymi zadaniami generowania dialogów i rozumienia kontekstu."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview to model badawczy opracowany przez zespół Qwen, skoncentrowany na zdolnościach wnioskowania wizualnego, który ma unikalne zalety w zrozumieniu złożonych scenariuszy i rozwiązywaniu wizualnie związanych problemów matematycznych."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview to najnowszy eksperymentalny model badawczy Qwen, skoncentrowany na zwiększeniu zdolności wnioskowania AI. Poprzez eksplorację złożonych mechanizmów, takich jak mieszanie języków i wnioskowanie rekurencyjne, główne zalety obejmują silne zdolności analizy wnioskowania, matematyki i programowania. Jednocześnie występują problemy z przełączaniem języków, cyklami wnioskowania, kwestiami bezpieczeństwa oraz różnicami w innych zdolnościach."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct to najnowsza wersja serii dużych modeli językowych specyficznych dla kodu wydana przez Alibaba Cloud. Model ten, oparty na Qwen2.5, został przeszkolony na 55 bilionach tokenów, znacznie poprawiając zdolności generowania kodu, wnioskowania i naprawy. Wzmacnia on nie tylko zdolności kodowania, ale także utrzymuje przewagę w zakresie matematyki i ogólnych umiejętności. Model ten stanowi bardziej kompleksową podstawę dla rzeczywistych zastosowań, takich jak inteligentne agenty kodowe."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math koncentruje się na rozwiązywaniu problemów w dziedzinie matematyki, oferując profesjonalne odpowiedzi na trudne pytania."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 to najnowsza seria modeli Qwen, obsługująca kontekst 128k. W porównaniu do obecnie najlepszych modeli open source, Qwen2-72B znacznie przewyższa w zakresie rozumienia języka naturalnego, wiedzy, kodowania, matematyki i wielu języków."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 to najnowsza seria modeli Qwen, która przewyższa najlepsze modele open source o podobnej skali, a nawet większe. Qwen2 7B osiągnęła znaczną przewagę w wielu testach, szczególnie w zakresie kodowania i rozumienia języka chińskiego."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B to potężny model językowo-wizualny, wspierający przetwarzanie multimodalne obrazów i tekstu, zdolny do precyzyjnego rozpoznawania treści obrazów i generowania odpowiednich opisów lub odpowiedzi."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct to model językowy z 14 miliardami parametrów, o doskonałej wydajności, optymalizujący scenariusze w języku chińskim i wielojęzyczne, wspierający inteligentne odpowiedzi, generowanie treści i inne zastosowania."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct to model językowy z 7 miliardami parametrów, wspierający wywołania funkcji i bezproblemową interakcję z systemami zewnętrznymi, znacznie zwiększając elastyczność i skalowalność. Optymalizuje scenariusze w języku chińskim i wielojęzyczne, wspierając inteligentne odpowiedzi, generowanie treści i inne zastosowania."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct to model instrukcji programowania oparty na dużych wstępnych treningach, posiadający silne zdolności rozumienia i generowania kodu, zdolny do efektywnego przetwarzania różnych zadań programistycznych, szczególnie odpowiedni do inteligentnego pisania kodu, generowania skryptów automatycznych i rozwiązywania problemów programistycznych."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct to duży model językowy zaprojektowany specjalnie do generowania kodu, rozumienia kodu i efektywnych scenariuszy rozwoju, wykorzystujący wiodącą w branży skalę 32B parametrów, zdolny do zaspokojenia różnorodnych potrzeb programistycznych."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "Model TeleMM to model wielomodalny opracowany przez China Telecom, który potrafi przetwarzać różne rodzaje wejść, takie jak tekst i obrazy, wspierając funkcje rozumienia obrazów oraz analizy wykresów, oferując użytkownikom usługi rozumienia międzymodalnego. Model ten potrafi prowadzić interakcje wielomodalne z użytkownikami, dokładnie rozumiejąc wprowadzone treści, odpowiadając na pytania, wspierając twórczość oraz efektywnie dostarczając informacji i inspiracji w różnych modalnościach. Wykazuje doskonałe wyniki w zadaniach wielomodalnych, takich jak precyzyjne postrzeganie i rozumowanie logiczne."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large to największy w branży otwarty model Transformer MoE, z 389 miliardami parametrów ogółem i 52 miliardami aktywowanych parametrów."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct to model dużego języka z serii Qwen2, dostosowany do instrukcji, o rozmiarze parametrów wynoszącym 72B. Model ten oparty jest na architekturze Transformer, wykorzystując funkcję aktywacji SwiGLU, przesunięcia QKV w uwadze oraz grupowe zapytania uwagi. Może obsługiwać duże wejścia. Model ten wykazuje doskonałe wyniki w wielu testach benchmarkowych dotyczących rozumienia języka, generowania, zdolności wielojęzycznych, kodowania, matematyki i wnioskowania, przewyższając większość modeli open-source i wykazując konkurencyjność z modelami własnościowymi w niektórych zadaniach."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct to jeden z najnowszych modeli dużych języków wydanych przez Alibaba Cloud. Model 72B ma znacząco poprawione zdolności w zakresie kodowania i matematyki. Oferuje również wsparcie dla wielu języków, obejmując ponad 29 języków, w tym chiński i angielski. Model ten wykazuje znaczną poprawę w zakresie przestrzegania instrukcji, rozumienia danych strukturalnych oraz generowania strukturalnych wyników (szczególnie JSON)."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Optymalizowany do scenariuszy dialogowych w języku chińskim, oferujący płynne i zgodne z chińskimi zwyczajami generowanie dialogów."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Open source model wywołań funkcji od Fireworks, oferujący doskonałe możliwości wykonania poleceń i otwarte, konfigurowalne cechy."
+ "abab7-chat-preview": {
+ "description": "W porównaniu do modeli z serii abab6.5, zdolności w zakresie długich tekstów, matematyki i pisania uległy znacznej poprawie."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Firefunction-v2, najnowszy model firmy Fireworks, to wydajny model wywołań funkcji, opracowany na bazie Llama-3, zoptymalizowany do wywołań funkcji, dialogów i śledzenia poleceń."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 to zaawansowany model językowy, który został zoptymalizowany dzięki uczeniu przez wzmocnienie i danym z zimnego startu, oferując doskonałe możliwości wnioskowania, matematyki i programowania."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b to model językowy wizualny, który może jednocześnie przyjmować obrazy i tekst, przeszkolony na wysokiej jakości danych, idealny do zadań multimodalnych."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Potężny model językowy Mixture-of-Experts (MoE) oferowany przez Deepseek, z całkowitą liczbą parametrów wynoszącą 671 miliardów, aktywującym 37 miliardów parametrów na każdy token."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Model Llama 3 70B Instruct, zaprojektowany do wielojęzycznych dialogów i rozumienia języka naturalnego, przewyższa większość konkurencyjnych modeli."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Model Llama 3 70B Instruct (wersja HF), zgodny z wynikami oficjalnej implementacji, idealny do wysokiej jakości zadań śledzenia poleceń."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Model Llama 3 8B Instruct, zoptymalizowany do dialogów i zadań wielojęzycznych, oferuje doskonałe i efektywne osiągi."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Model wnioskowania wizualnego z 11B parametrów od Meta. Model zoptymalizowany do rozpoznawania wizualnego, wnioskowania obrazów, opisywania obrazów oraz odpowiadania na ogólne pytania dotyczące obrazów. Model potrafi rozumieć dane wizualne, takie jak wykresy i grafiki, a dzięki generowaniu tekstowych opisów szczegółów obrazów, łączy wizję z językiem."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Model instruktażowy Llama 3.2 1B to lekki model wielojęzyczny zaprezentowany przez Meta. Zaprojektowany, aby poprawić wydajność, oferując znaczące usprawnienia w opóźnieniu i kosztach w porównaniu do większych modeli. Przykładowe przypadki użycia tego modelu obejmują wyszukiwanie i podsumowywanie."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Model instruktażowy Llama 3.2 3B to lekki model wielojęzyczny zaprezentowany przez Meta. Zaprojektowany, aby poprawić wydajność, oferując znaczące usprawnienia w opóźnieniu i kosztach w porównaniu do większych modeli. Przykładowe przypadki użycia tego modelu obejmują zapytania i przepisanie sugestii oraz pomoc w pisaniu."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Model wnioskowania wizualnego z 90B parametrów od Meta. Model zoptymalizowany do rozpoznawania wizualnego, wnioskowania obrazów, opisywania obrazów oraz odpowiadania na ogólne pytania dotyczące obrazów. Model potrafi rozumieć dane wizualne, takie jak wykresy i grafiki, a dzięki generowaniu tekstowych opisów szczegółów obrazów, łączy wizję z językiem."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct to zaktualizowana wersja Llama 3.1 70B z grudnia. Model ten został ulepszony w oparciu o Llama 3.1 70B (wydany w lipcu 2024), wzmacniając możliwości wywoływania narzędzi, wsparcie dla tekstów w wielu językach, a także umiejętności matematyczne i programistyczne. Model osiągnął wiodący w branży poziom w zakresie wnioskowania, matematyki i przestrzegania instrukcji, oferując wydajność porównywalną z 3.1 405B, jednocześnie zapewniając znaczące korzyści w zakresie szybkości i kosztów."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "Model z 24 miliardami parametrów, oferujący zaawansowane możliwości porównywalne z większymi modelami."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Model Mixtral MoE 8x22B Instruct, z dużą liczbą parametrów i architekturą wielu ekspertów, kompleksowo wspierający efektywne przetwarzanie złożonych zadań."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Model Mixtral MoE 8x7B Instruct, architektura wielu ekspertów, oferująca efektywne śledzenie i wykonanie poleceń."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Model Mixtral MoE 8x7B Instruct (wersja HF), wydajność zgodna z oficjalną implementacją, idealny do różnych scenariuszy efektywnych zadań."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "Model MythoMax L2 13B, łączący nowatorskie techniki łączenia, doskonały w narracji i odgrywaniu ról."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "Model QwQ to eksperymentalny model badawczy opracowany przez zespół Qwen, skoncentrowany na zwiększeniu zdolności wnioskowania AI."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "Wersja 72B modelu Qwen-VL to najnowszy owoc iteracji Alibaba, reprezentujący innowacje z ostatniego roku."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 to seria modeli językowych opracowana przez zespół Qwen na chmurze Alibaba, która zawiera jedynie dekodery. Modele te występują w różnych rozmiarach, w tym 0.5B, 1.5B, 3B, 7B, 14B, 32B i 72B, i oferują dwie wersje: bazową (base) i instruktażową (instruct)."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct to najnowsza wersja serii dużych modeli językowych specyficznych dla kodu wydana przez Alibaba Cloud. Model ten, oparty na Qwen2.5, został przeszkolony na 55 bilionach tokenów, znacznie poprawiając zdolności generowania kodu, wnioskowania i naprawy. Wzmacnia on nie tylko zdolności kodowania, ale także utrzymuje przewagę w zakresie matematyki i ogólnych umiejętności. Model ten stanowi bardziej kompleksową podstawę dla rzeczywistych zastosowań, takich jak inteligentne agenty kodowe."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "Model StarCoder 15.5B, wspierający zaawansowane zadania programistyczne, z wzmocnionymi możliwościami wielojęzycznymi, idealny do złożonego generowania i rozumienia kodu."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "Model StarCoder 7B, przeszkolony w ponad 80 językach programowania, oferujący doskonałe możliwości uzupełniania kodu i rozumienia kontekstu."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Model Yi-Large, oferujący doskonałe możliwości przetwarzania wielojęzycznego, nadający się do różnych zadań generowania i rozumienia języka."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus to najpotężniejszy model Anthropic do obsługi wysoce złożonych zadań. Wyróżnia się doskonałymi osiągami, inteligencją, płynnością i zdolnością rozumienia."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku to najszybszy model nowej generacji od Anthropic. W porównaniu do Claude 3 Haiku, Claude 3.5 Haiku wykazuje poprawę w różnych umiejętnościach i przewyższa największy model poprzedniej generacji, Claude 3 Opus, w wielu testach inteligencji."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet oferuje możliwości przewyższające Opus oraz szybsze tempo niż Sonnet, zachowując tę samą cenę. Sonnet szczególnie dobrze radzi sobie z programowaniem, nauką o danych, przetwarzaniem wizualnym i zadaniami agenta."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet zapewnia idealną równowagę między inteligencją a szybkością dla obciążeń roboczych w przedsiębiorstwach. Oferuje maksymalną użyteczność przy niższej cenie, jest niezawodny i odpowiedni do dużych wdrożeń."
},
- "code-raccoon-v1": {
- "description": "Code Raccoon to inteligentny asystent do rozwoju oprogramowania oparty na dużym modelu językowym SenseTime, obejmujący analizę wymagań oprogramowania, projektowanie architektury, pisanie kodu, testowanie oprogramowania i inne etapy, zaspokajający różnorodne potrzeby użytkowników w zakresie pisania kodu i nauki programowania. Code Raccoon wspiera ponad 90 popularnych języków programowania, takich jak Python, Java, JavaScript, C++, Go, SQL oraz popularne IDE, takie jak VS Code i IntelliJ IDEA. W praktyce Code Raccoon może pomóc programistom zwiększyć wydajność programowania o ponad 50%."
- },
"codegeex-4": {
"description": "CodeGeeX-4 to potężny asystent programowania AI, obsługujący inteligentne pytania i odpowiedzi oraz uzupełnianie kodu w różnych językach programowania, zwiększając wydajność programistów."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ to model językowy o wysokiej wydajności, zaprojektowany z myślą o rzeczywistych scenariuszach biznesowych i złożonych zastosowaniach."
},
+ "dall-e-2": {
+ "description": "Druga generacja modelu DALL·E, obsługująca bardziej realistyczne i dokładne generowanie obrazów, o rozdzielczości czterokrotnie większej niż pierwsza generacja."
+ },
+ "dall-e-3": {
+ "description": "Najnowocześniejszy model DALL·E, wydany w listopadzie 2023 roku. Obsługuje bardziej realistyczne i dokładne generowanie obrazów, z lepszą zdolnością do oddawania szczegółów."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct oferuje wysoką niezawodność w przetwarzaniu poleceń, wspierając różne branże."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 to potężny, ekonomiczny model językowy typu mieszany ekspert (MoE). Został wstępnie przeszkolony na wysokiej jakości korpusie danych liczącym 8,1 biliona tokenów, a jego zdolności zostały dodatkowo poprawione dzięki nadzorowanemu dostrajaniu (SFT) i uczeniu przez wzmacnianie (RL). W porównaniu do DeepSeek 67B, DeepSeek-V2 osiąga lepszą wydajność, oszczędzając 42,5% kosztów szkolenia, redukując 93,3% pamięci podręcznej KV i zwiększając maksymalną przepustowość generacji do 5,76 razy. Model obsługuje długość kontekstu 128k i osiąga doskonałe wyniki w standardowych testach benchmarkowych oraz w otwartych ocenach generacji."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 to model wnioskowania napędzany uczeniem przez wzmacnianie (RL), który rozwiązuje problemy z powtarzalnością i czytelnością modelu. Przed RL, DeepSeek-R1 wprowadził dane z zimnego startu, co dodatkowo zoptymalizowało wydajność wnioskowania. W zadaniach matematycznych, kodowania i wnioskowania osiąga wyniki porównywalne z OpenAI-o1, a dzięki starannie zaprojektowanym metodom treningowym poprawia ogólne efekty."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "Model destylacyjny DeepSeek-R1, optymalizujący wydajność wnioskowania dzięki uczeniu przez wzmocnienie i danym z zimnego startu, otwarty model ustanawiający nowe standardy w wielu zadaniach."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B to model destylacyjny oparty na Llama-3.1-8B. Model ten został dostosowany przy użyciu próbek wygenerowanych przez DeepSeek-R1, wykazując doskonałe zdolności wnioskowania. Osiągnął dobre wyniki w wielu testach referencyjnych, w tym 89,1% dokładności w MATH-500, 50,4% wskaźnika zdawalności w AIME 2024 oraz 1205 punktów w CodeForces, demonstrując silne zdolności matematyczne i programistyczne jako model o skali 8B."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Model destylacyjny DeepSeek-R1, optymalizujący wydajność wnioskowania dzięki uczeniu przez wzmocnienie i danym z zimnego startu, otwarty model ustanawiający nowe standardy w wielu zadaniach."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Model destylacyjny DeepSeek-R1, optymalizujący wydajność wnioskowania dzięki uczeniu przez wzmocnienie i danym z zimnego startu, otwarty model ustanawiający nowe standardy w wielu zadaniach."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B to model uzyskany przez destylację Qwen2.5-32B. Model ten został dostosowany przy użyciu 800 000 starannie wybranych próbek wygenerowanych przez DeepSeek-R1, wykazując doskonałe osiągi w wielu dziedzinach, takich jak matematyka, programowanie i wnioskowanie. Osiągnął znakomite wyniki w wielu testach referencyjnych, w tym 94,3% dokładności w MATH-500, co pokazuje jego silne zdolności wnioskowania matematycznego."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B to model uzyskany przez destylację Qwen2.5-Math-7B. Model ten został dostosowany przy użyciu 800 000 starannie wybranych próbek wygenerowanych przez DeepSeek-R1, wykazując doskonałe zdolności wnioskowania. Osiągnął znakomite wyniki w wielu testach referencyjnych, w tym 92,8% dokładności w MATH-500, 55,5% wskaźnika zdawalności w AIME 2024 oraz 1189 punktów w CodeForces, demonstrując silne zdolności matematyczne i programistyczne jako model o skali 7B."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 łączy doskonałe cechy wcześniejszych wersji, wzmacniając zdolności ogólne i kodowania."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 to model językowy z 6710 miliardami parametrów, oparty na mieszanych ekspertach (MoE), wykorzystujący wielogłowicową potencjalną uwagę (MLA) oraz architekturę DeepSeekMoE, łączącą strategię równoważenia obciążenia bez dodatkowych strat, co optymalizuje wydajność wnioskowania i treningu. Dzięki wstępnemu treningowi na 14,8 bilionach wysokiej jakości tokenów oraz nadzorowanemu dostrajaniu i uczeniu przez wzmacnianie, DeepSeek-V3 przewyższa inne modele open source, zbliżając się do wiodących modeli zamkniętych."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B to zaawansowany model przeszkolony do złożonych dialogów."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "Najnowocześniejszy, wydajny LLM, specjalizujący się w wnioskowaniu, matematyce i programowaniu."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 to model wizualno-językowy oparty na DeepSeekMoE-27B, wykorzystujący architekturę MoE z rzadką aktywacją, osiągający doskonałe wyniki przy aktywacji jedynie 4,5 miliarda parametrów. Model ten wyróżnia się w wielu zadaniach, takich jak wizualne pytania i odpowiedzi, optyczne rozpoznawanie znaków, zrozumienie dokumentów/tabel/wykresów oraz lokalizacja wizualna."
+ },
"deepseek-chat": {
"description": "Nowy otwarty model łączący zdolności ogólne i kodowe, który nie tylko zachowuje ogólne zdolności dialogowe oryginalnego modelu czatu i potężne zdolności przetwarzania kodu modelu Coder, ale także lepiej dostosowuje się do ludzkich preferencji. Ponadto, DeepSeek-V2.5 osiągnął znaczne poprawy w zadaniach pisarskich, przestrzeganiu instrukcji i innych obszarach."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 to otwarty model kodowy Mixture-of-Experts, który doskonale radzi sobie z zadaniami kodowymi, porównywalny z GPT4-Turbo."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 to model wnioskowania napędzany uczeniem przez wzmacnianie (RL), który rozwiązuje problemy z powtarzalnością i czytelnością modelu. Przed RL, DeepSeek-R1 wprowadził dane z zimnego startu, co dodatkowo zoptymalizowało wydajność wnioskowania. W zadaniach matematycznych, kodowania i wnioskowania osiąga wyniki porównywalne z OpenAI-o1, a dzięki starannie zaprojektowanym metodom treningowym poprawia ogólne efekty."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 — większy i inteligentniejszy model w zestawie DeepSeek — został destylowany do architektury Llama 70B. Na podstawie testów referencyjnych i ocen ręcznych, model ten jest bardziej inteligentny niż oryginalna Llama 70B, szczególnie w zadaniach wymagających precyzji matematycznej i faktograficznej."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "Modele z serii DeepSeek-R1-Distill są dostosowywane do modeli open source, takich jak Qwen i Llama, poprzez technologię destylacji wiedzy, na podstawie próbek generowanych przez DeepSeek-R1."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "Modele z serii DeepSeek-R1-Distill są dostosowywane do modeli open source, takich jak Qwen i Llama, poprzez technologię destylacji wiedzy, na podstawie próbek generowanych przez DeepSeek-R1."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "Modele z serii DeepSeek-R1-Distill są dostosowywane do modeli open source, takich jak Qwen i Llama, poprzez technologię destylacji wiedzy, na podstawie próbek generowanych przez DeepSeek-R1."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "Modele z serii DeepSeek-R1-Distill są dostosowywane do modeli open source, takich jak Qwen i Llama, poprzez technologię destylacji wiedzy, na podstawie próbek generowanych przez DeepSeek-R1."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "Modele z serii DeepSeek-R1-Distill są dostosowywane do modeli open source, takich jak Qwen i Llama, poprzez technologię destylacji wiedzy, na podstawie próbek generowanych przez DeepSeek-R1."
+ },
+ "deepseek-reasoner": {
+ "description": "Model inferency wprowadzony przez DeepSeek. Przed wygenerowaniem ostatecznej odpowiedzi, model najpierw przedstawia fragment łańcucha myślowego, aby zwiększyć dokładność końcowej odpowiedzi."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 to wydajny model językowy Mixture-of-Experts, odpowiedni do ekonomicznych potrzeb przetwarzania."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B to model kodowy zaprojektowany przez DeepSeek, oferujący potężne możliwości generowania kodu."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 to model MoE opracowany przez Hangzhou DeepSeek AI Technology Research Co., Ltd., który osiągnął znakomite wyniki w wielu testach, zajmując pierwsze miejsce wśród modeli open-source na głównych listach. W porównaniu do modelu V2.5, prędkość generowania wzrosła trzykrotnie, co zapewnia użytkownikom szybsze i płynniejsze doświadczenia."
+ },
"deepseek/deepseek-chat": {
"description": "Nowy, otwarty model łączący zdolności ogólne i kodowe, który nie tylko zachowuje ogólne zdolności dialogowe oryginalnego modelu Chat, ale także potężne zdolności przetwarzania kodu modelu Coder, lepiej dostosowując się do ludzkich preferencji. Ponadto, DeepSeek-V2.5 osiągnął znaczne poprawy w zadaniach pisarskich, przestrzeganiu instrukcji i wielu innych obszarach."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 znacznie poprawił zdolności wnioskowania modelu przy minimalnej ilości oznaczonych danych. Przed wygenerowaniem ostatecznej odpowiedzi, model najpierw wygeneruje fragment myślenia, aby zwiększyć dokładność końcowej odpowiedzi."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 znacznie poprawił zdolności wnioskowania modelu przy minimalnej ilości oznaczonych danych. Przed wygenerowaniem ostatecznej odpowiedzi, model najpierw wygeneruje fragment myślenia, aby zwiększyć dokładność końcowej odpowiedzi."
+ },
"emohaa": {
"description": "Emohaa to model psychologiczny, posiadający profesjonalne umiejętności doradcze, pomagający użytkownikom zrozumieć problemy emocjonalne."
},
+ "ernie-3.5-128k": {
+ "description": "Flagowy model językowy opracowany przez Baidu, obejmujący ogromne zbiory danych w języku chińskim i angielskim, charakteryzujący się silnymi zdolnościami ogólnymi, spełniającym wymagania większości zastosowań w dialogach, generowaniu treści i aplikacjach wtyczek; wspiera automatyczne połączenie z wtyczką wyszukiwania Baidu, zapewniając aktualność informacji."
+ },
+ "ernie-3.5-8k": {
+ "description": "Flagowy model językowy opracowany przez Baidu, obejmujący ogromne zbiory danych w języku chińskim i angielskim, charakteryzujący się silnymi zdolnościami ogólnymi, spełniającym wymagania większości zastosowań w dialogach, generowaniu treści i aplikacjach wtyczek; wspiera automatyczne połączenie z wtyczką wyszukiwania Baidu, zapewniając aktualność informacji."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "Flagowy model językowy opracowany przez Baidu, obejmujący ogromne zbiory danych w języku chińskim i angielskim, charakteryzujący się silnymi zdolnościami ogólnymi, spełniającym wymagania większości zastosowań w dialogach, generowaniu treści i aplikacjach wtyczek; wspiera automatyczne połączenie z wtyczką wyszukiwania Baidu, zapewniając aktualność informacji."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "Flagowy model językowy Baidu o ultra dużej skali, w porównaniu do ERNIE 3.5, oferujący kompleksową aktualizację zdolności modelu, szeroko stosowany w złożonych zadaniach w różnych dziedzinach; wspiera automatyczne połączenie z wtyczką wyszukiwania Baidu, zapewniając aktualność informacji."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "Flagowy model językowy Baidu o ultra dużej skali, w porównaniu do ERNIE 3.5, oferujący kompleksową aktualizację zdolności modelu, szeroko stosowany w złożonych zadaniach w różnych dziedzinach; wspiera automatyczne połączenie z wtyczką wyszukiwania Baidu, zapewniając aktualność informacji."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "Flagowy model językowy Baidu o ultra dużej skali, charakteryzujący się doskonałymi wynikami ogólnymi, szeroko stosowany w złożonych zadaniach w różnych dziedzinach; wspiera automatyczne połączenie z wtyczką wyszukiwania Baidu, zapewniając aktualność informacji. W porównaniu do ERNIE 4.0, oferuje lepsze wyniki wydajności."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "Flagowy model językowy Baidu o ultra dużej skali, charakteryzujący się doskonałymi wynikami ogólnymi, szeroko stosowany w złożonych zadaniach w różnych dziedzinach; wspiera automatyczne połączenie z wtyczką wyszukiwania Baidu, zapewniając aktualność informacji. W porównaniu do ERNIE 4.0, oferuje lepsze wyniki wydajności."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "Flagowy model językowy Baidu o ultra dużej skali, charakteryzujący się doskonałymi wynikami ogólnymi, szeroko stosowany w złożonych zadaniach w różnych dziedzinach; wspiera automatyczne połączenie z wtyczką wyszukiwania Baidu, zapewniając aktualność informacji. W porównaniu do ERNIE 4.0, oferuje lepsze wyniki wydajności."
+ },
+ "ernie-char-8k": {
+ "description": "Model językowy opracowany przez Baidu, skoncentrowany na specyficznych scenariuszach, odpowiedni do zastosowań w grach NPC, dialogach obsługi klienta, odgrywaniu ról w dialogach, charakteryzujący się wyraźnym i spójnym stylem postaci, silniejszą zdolnością do podążania za instrukcjami oraz lepszą wydajnością wnioskowania."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Model językowy opracowany przez Baidu, skoncentrowany na specyficznych scenariuszach, odpowiedni do zastosowań w grach NPC, dialogach obsługi klienta, odgrywaniu ról w dialogach, charakteryzujący się wyraźnym i spójnym stylem postaci, silniejszą zdolnością do podążania za instrukcjami oraz lepszą wydajnością wnioskowania."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite to lekki model językowy opracowany przez Baidu, łączący doskonałe wyniki modelu z wydajnością wnioskowania, odpowiedni do użycia na kartach przyspieszających AI o niskiej mocy obliczeniowej."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Lekki model językowy opracowany przez Baidu, łączący doskonałe wyniki modelu z wydajnością wnioskowania, oferujący lepsze wyniki niż ERNIE Lite, odpowiedni do użycia na kartach przyspieszających AI o niskiej mocy obliczeniowej."
+ },
+ "ernie-novel-8k": {
+ "description": "Ogólny model językowy opracowany przez Baidu, który wykazuje wyraźne przewagi w zakresie kontynuacji powieści, może być również stosowany w scenariuszach krótkich dramatów i filmów."
+ },
+ "ernie-speed-128k": {
+ "description": "Najnowszy model językowy o wysokiej wydajności opracowany przez Baidu w 2024 roku, charakteryzujący się doskonałymi zdolnościami ogólnymi, odpowiedni jako model bazowy do dalszego dostosowania, lepiej radzący sobie z problemami w specyficznych scenariuszach, a także oferujący doskonałą wydajność wnioskowania."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "Najnowszy model językowy o wysokiej wydajności opracowany przez Baidu w 2024 roku, charakteryzujący się doskonałymi zdolnościami ogólnymi, oferujący lepsze wyniki niż ERNIE Speed, odpowiedni jako model bazowy do dalszego dostosowania, lepiej radzący sobie z problemami w specyficznych scenariuszach, a także oferujący doskonałą wydajność wnioskowania."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny to model językowy o ultra wysokiej wydajności opracowany przez Baidu, charakteryzujący się najniższymi kosztami wdrożenia i dostosowania w serii modeli Wenxin."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Tuning) oferuje stabilną i dostosowywalną wydajność, co czyni go idealnym wyborem dla rozwiązań złożonych zadań."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro obsługuje do 2 milionów tokenów, co czyni go idealnym wyborem dla średniej wielkości modeli multimodalnych, odpowiednim do wszechstronnej obsługi złożonych zadań."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash oferuje funkcje i ulepszenia nowej generacji, w tym doskonałą prędkość, natywne korzystanie z narzędzi, generowanie multimodalne oraz okno kontekstowe o długości 1M tokenów."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash oferuje funkcje i ulepszenia nowej generacji, w tym doskonałą prędkość, natywne korzystanie z narzędzi, generowanie multimodalne oraz okno kontekstowe o długości 1M tokenów."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp to najnowszy eksperymentalny model AI o wielu modalnościach od Google, który oferuje cechy nowej generacji, doskonałą prędkość, natywne wywołania narzędzi oraz generowanie multimodalne."
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "Model Gemini 2.0 Flash, zoptymalizowany pod kątem efektywności kosztowej i niskiej latencji."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Exp to najnowszy eksperymentalny model AI multimodalnego Google, posiadający cechy nowej generacji, doskonałą prędkość, natywne wywołania narzędzi oraz generację multimodalną."
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Exp to najnowszy eksperymentalny model AI multimodalnego Google, oferujący funkcje nowej generacji, doskonałą prędkość, natywne wywołania narzędzi oraz generację multimodalną."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 to najnowszy eksperymentalny model AI o wielu modalnościach od Google, który charakteryzuje się szybkim przetwarzaniem i obsługuje wejścia tekstowe, obrazowe oraz wideo, co umożliwia wydajne rozszerzanie go na różne zadania."
- },
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 to najnowszy eksperymentalny model AI Google, który charakteryzuje się szybkim przetwarzaniem i obsługuje wejścia tekstowe, obrazowe i wideo, co czyni go wydajnym narzędziem do rozwiązywania różnych zadań."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental to najnowszy eksperymentalny model AI o wielu modalnościach od Google, który w porównaniu do wcześniejszych wersji oferuje pewne poprawy jakości, szczególnie w zakresie wiedzy o świecie, kodu i długiego kontekstu."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 to najnowszy eksperymentalny model AI Google w multimodalności, z zauważalnym poprawieniem jakości w porównaniu do wcześniejszych wersji."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus ma zdolność rozumienia treści wideo oraz wielu obrazów, odpowiedni do zadań multimodalnych."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview posiada silne zdolności do złożonego wnioskowania, wyróżniając się w dziedzinach takich jak wnioskowanie logiczne, matematyka i programowanie."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash oferuje funkcje i ulepszenia nowej generacji, w tym doskonałą prędkość, natywne korzystanie z narzędzi, generowanie multimodalne oraz okno kontekstowe o długości 1M tokenów."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental to najnowszy eksperymentalny model AI o wielu modalnościach od Google, który w porównaniu do wcześniejszych wersji oferuje pewne poprawy jakości, szczególnie w zakresie wiedzy o świecie, kodu i długiego kontekstu."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash oferuje zoptymalizowane możliwości przetwarzania multimodalnego, odpowiednie do różnych złożonych scenariuszy zadań."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro łączy najnowsze technologie optymalizacji, oferując bardziej efektywne przetwarzanie danych multimodalnych."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 to wydajny model wydany przez Google, obejmujący różnorodne scenariusze zastosowań, od małych aplikacji po złożone przetwarzanie danych."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 kontynuuje ideę lekkiego i wydajnego projektowania."
},
"google/gemma-2-2b-it": {
"description": "Lekki model dostosowywania instrukcji od Google."
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 to wydajny model wydany przez Google, obejmujący różnorodne scenariusze zastosowań, od małych aplikacji po złożone przetwarzanie danych."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 to lekka seria modeli tekstowych open source od Google."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo, odpowiedni do różnych zadań generowania i rozumienia tekstu, obecnie wskazuje na gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, model generacji tekstu o dużej pojemności, odpowiedni do złożonych zadań."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo, odpowiedni do różnych zadań generowania i rozumienia tekstu, obecnie wskazuje na gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o to dynamiczny model, aktualizowany w czasie rzeczywistym, aby być zawsze na bieżąco z najnowszą wersją. Łączy potężne zdolności rozumienia i generowania języka, idealny do zastosowań w dużej skali, w tym obsłudze klienta, edukacji i wsparciu technicznym."
},
+ "gpt-4o-audio-preview": {
+ "description": "Model audio GPT-4o, obsługujący wejście i wyjście audio."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini to najnowszy model OpenAI, wprowadzony po GPT-4 Omni, obsługujący wejścia tekstowe i wizualne oraz generujący tekst. Jako ich najnowocześniejszy model w małej skali, jest znacznie tańszy niż inne niedawno wprowadzone modele, a jego cena jest o ponad 60% niższa niż GPT-3.5 Turbo. Utrzymuje najnowocześniejszą inteligencję, jednocześnie oferując znaczną wartość za pieniądze. GPT-4o mini uzyskał wynik 82% w teście MMLU i obecnie zajmuje wyższą pozycję w preferencjach czatu niż GPT-4."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "Wersja na żywo GPT-4o-mini, obsługująca wejście i wyjście audio oraz tekstowe w czasie rzeczywistym."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "Wersja na żywo GPT-4o, obsługująca wejście i wyjście audio oraz tekstowe w czasie rzeczywistym."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "Wersja na żywo GPT-4o, obsługująca wejście i wyjście audio oraz tekstowe w czasie rzeczywistym."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "Wersja na żywo GPT-4o, obsługująca wejście i wyjście audio oraz tekstowe w czasie rzeczywistym."
+ },
"grok-2-1212": {
"description": "Model ten poprawił dokładność, przestrzeganie instrukcji oraz zdolności wielojęzyczne."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "Najnowocześniejszy model FunctionCall w architekturze MOE Hunyuan, przeszkolony na wysokiej jakości danych FunctionCall, z oknem kontekstowym o długości 32K, osiągający wiodące wyniki w wielu wymiarach oceny."
},
+ "hunyuan-large": {
+ "description": "Model Hunyuan-large ma całkowitą liczbę parametrów wynoszącą około 389B, z aktywowanymi parametrami wynoszącymi około 52B, co czyni go obecnie największym i najlepiej działającym modelem MoE w architekturze Transformer w branży."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Specjalizuje się w zadaniach związanych z długimi tekstami, takich jak streszczenia dokumentów i pytania i odpowiedzi dotyczące dokumentów, a także ma zdolność do obsługi ogólnych zadań generowania tekstu. Wykazuje doskonałe wyniki w analizie i generowaniu długich tekstów, skutecznie radząc sobie z złożonymi i szczegółowymi wymaganiami dotyczącymi przetwarzania długich treści."
+ },
"hunyuan-lite": {
"description": "Zaktualizowana do struktury MOE, z oknem kontekstowym o długości 256k, prowadzi w wielu zestawach testowych w NLP, kodowaniu, matematyce i innych dziedzinach w porównaniu do wielu modeli open source."
},
+ "hunyuan-lite-vision": {
+ "description": "Najnowocześniejszy model multimodalny 7B Hunyuan, z oknem kontekstowym 32K, wspierający multimodalne dialogi w języku chińskim i angielskim, rozpoznawanie obiektów w obrazach, zrozumienie dokumentów i tabel, multimodalną matematykę itp., z wynikami w wielu wymiarach lepszymi niż modele konkurencyjne 7B."
+ },
"hunyuan-pro": {
"description": "Model długiego tekstu MOE-32K o skali bilionów parametrów. Osiąga absolutnie wiodący poziom w różnych benchmarkach, obsługując złożone instrukcje i wnioskowanie, posiadając zaawansowane umiejętności matematyczne, wspierając wywołania funkcji, z optymalizacjami w obszarach takich jak tłumaczenia wielojęzyczne, prawo finansowe i medyczne."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Zastosowano lepszą strategię routingu, jednocześnie łagodząc problemy z równoważeniem obciążenia i zbieżnością ekspertów. W przypadku długich tekstów wskaźnik 'znalezienia igły w stogu siana' osiąga 99,9%. MOE-256K dokonuje dalszych przełomów w długości i efektywności, znacznie rozszerzając możliwą długość wejścia."
},
+ "hunyuan-standard-vision": {
+ "description": "Najnowocześniejszy model multimodalny Hunyuan, wspierający odpowiedzi w wielu językach, z równoważnymi zdolnościami w języku chińskim i angielskim."
+ },
"hunyuan-turbo": {
"description": "Hunyuan to nowa generacja dużego modelu językowego w wersji próbnej, wykorzystująca nową strukturę modelu mieszanych ekspertów (MoE), która w porównaniu do hunyuan-pro charakteryzuje się szybszą efektywnością wnioskowania i lepszymi wynikami."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Stała wersja hunyuan-turbo z dnia 20 listopada 2024 roku, będąca pomiędzy hunyuan-turbo a hunyuan-turbo-latest."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "Optymalizacja tej wersji: skalowanie danych instrukcji, znaczne zwiększenie ogólnej zdolności generalizacji modelu; znaczne zwiększenie zdolności w zakresie matematyki, kodowania i rozumowania logicznego; optymalizacja zdolności związanych z rozumieniem tekstu i słów; optymalizacja jakości generowania treści w tworzeniu tekstów."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "Ogólna optymalizacja doświadczeń, w tym zrozumienie NLP, tworzenie tekstów, rozmowy, pytania i odpowiedzi, tłumaczenia, obszary tematyczne itp.; zwiększenie humanizacji, optymalizacja inteligencji emocjonalnej modelu; poprawa zdolności modelu do aktywnego wyjaśniania w przypadku niejasnych intencji; poprawa zdolności do rozwiązywania problemów związanych z analizą słów; poprawa jakości i interaktywności twórczości; poprawa doświadczeń w wielokrotnych interakcjach."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "Nowa generacja flagowego modelu językowo-wizualnego Hunyuan, wykorzystująca nową strukturę modelu mieszanych ekspertów (MoE), z pełnym zwiększeniem zdolności w zakresie podstawowego rozpoznawania, tworzenia treści, pytań i odpowiedzi oraz analizy i rozumowania w porównaniu do poprzedniej generacji modeli."
+ },
"hunyuan-vision": {
"description": "Najnowocześniejszy model multimodalny Hunyuan, wspierający generowanie treści tekstowych na podstawie obrazów i tekstu."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "Nasza najnowsza seria modeli, charakteryzująca się doskonałymi osiągami wnioskowania, obsługująca długość kontekstu do 1M oraz lepsze możliwości śledzenia instrukcji i wywoływania narzędzi."
},
+ "internlm3-latest": {
+ "description": "Nasza najnowsza seria modeli, charakteryzująca się doskonałą wydajnością wnioskowania, prowadzi wśród modeli open-source o podobnej skali. Domyślnie wskazuje na naszą najnowszą wersję modelu InternLM3."
+ },
+ "jina-deepsearch-v1": {
+ "description": "Głębokie wyszukiwanie łączy wyszukiwanie w sieci, czytanie i wnioskowanie, umożliwiając kompleksowe badania. Możesz to traktować jako agenta, który przyjmuje Twoje zadania badawcze - przeprowadza szerokie poszukiwania i wielokrotne iteracje, zanim poda odpowiedź. Proces ten obejmuje ciągłe badania, wnioskowanie i rozwiązywanie problemów z różnych perspektyw. To zasadniczo różni się od standardowych dużych modeli, które generują odpowiedzi bezpośrednio z wstępnie wytrenowanych danych oraz od tradycyjnych systemów RAG, które polegają na jednorazowym powierzchownym wyszukiwaniu."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM to eksperymentalny model językowy, specyficzny dla zadań, przeszkolony zgodnie z zasadami nauki o uczeniu się, który może przestrzegać systemowych instrukcji w scenariuszach nauczania i uczenia się, pełniąc rolę eksperta mentora."
},
"lite": {
"description": "Spark Lite to lekki model językowy o dużej skali, charakteryzujący się niezwykle niskim opóźnieniem i wysoką wydajnością przetwarzania, całkowicie darmowy i otwarty, wspierający funkcje wyszukiwania w czasie rzeczywistym. Jego cechy szybkiej reakcji sprawiają, że doskonale sprawdza się w zastosowaniach inferencyjnych na urządzeniach o niskiej mocy obliczeniowej oraz w dostosowywaniu modeli, oferując użytkownikom znakomity stosunek kosztów do korzyści oraz inteligentne doświadczenie, szczególnie w kontekście pytań i odpowiedzi, generowania treści oraz wyszukiwania."
},
- "llama-3.1-70b-instruct": {
- "description": "Model Llama 3.1 70B Instruct, z 70B parametrami, oferujący doskonałe osiągi w dużych zadaniach generowania tekstu i poleceń."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B oferuje potężne możliwości wnioskowania AI, odpowiednie do złożonych zastosowań, wspierające ogromne przetwarzanie obliczeniowe przy zachowaniu efektywności i dokładności."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B to model o wysokiej wydajności, oferujący szybkie możliwości generowania tekstu, idealny do zastosowań wymagających dużej efektywności i opłacalności."
},
- "llama-3.1-8b-instruct": {
- "description": "Model Llama 3.1 8B Instruct, z 8B parametrami, wspierający efektywne wykonanie zadań wskazujących, oferujący wysoką jakość generowania tekstu."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Model Llama 3.1 Sonar Huge Online, z 405B parametrami, obsługujący kontekst o długości około 127,000 tokenów, zaprojektowany do złożonych aplikacji czatu online."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Model Llama 3.1 Sonar Large Chat, z 70B parametrami, obsługujący kontekst o długości około 127,000 tokenów, idealny do złożonych zadań czatu offline."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Model Llama 3.1 Sonar Large Online, z 70B parametrami, obsługujący kontekst o długości około 127,000 tokenów, idealny do zadań czatu o dużej pojemności i różnorodności."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Model Llama 3.1 Sonar Small Chat, z 8B parametrami, zaprojektowany do czatów offline, obsługujący kontekst o długości około 127,000 tokenów."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Model Llama 3.1 Sonar Small Online, z 8B parametrami, obsługujący kontekst o długości około 127,000 tokenów, zaprojektowany do czatów online, efektywnie przetwarzający różne interakcje tekstowe."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 jest zaprojektowana do obsługi zadań łączących dane wizualne i tekstowe. Wykazuje doskonałe wyniki w zadaniach takich jak opisywanie obrazów i wizualne pytania i odpowiedzi, przekraczając przepaść między generowaniem języka a wnioskowaniem wizualnym."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 to najnowocześniejszy wielojęzyczny, otwarty model językowy z serii Llama, który oferuje wydajność porównywalną z modelem 405B przy bardzo niskich kosztach. Opiera się na strukturze Transformer i poprawia użyteczność oraz bezpieczeństwo dzięki nadzorowanemu dostrajaniu (SFT) i uczeniu ze wzmocnieniem na podstawie ludzkich opinii (RLHF). Jego wersja dostosowana do instrukcji jest zoptymalizowana do wielojęzycznych rozmów i w wielu branżowych benchmarkach przewyższa wiele otwartych i zamkniętych modeli czatu. Data graniczna wiedzy to grudzień 2023."
+ },
"llama-3.3-70b-versatile": {
"description": "Meta Llama 3.3 to wielojęzyczny model językowy (LLM) 70B, pretrenowany i dostosowany do poleceń. Model Llama 3.3, dostosowany do poleceń, jest zoptymalizowany do zastosowań w dialogach wielojęzycznych i przewyższa wiele dostępnych modeli czatu, zarówno open source, jak i zamkniętych, w popularnych branżowych benchmarkach."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 zaprojektowana do przetwarzania zadań łączących dane wizualne i tekstowe. Doskonała w zadaniach takich jak opisywanie obrazów i wizualne pytania odpowiedzi, przekracza granice między generowaniem języka a wnioskowaniem wizualnym."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 to najnowocześniejszy wielojęzyczny model językowy open-source z serii Llama, oferujący wydajność porównywalną z modelem 405B przy bardzo niskich kosztach. Oparty na strukturze Transformer, poprawiony dzięki nadzorowanemu dostrajaniu (SFT) oraz uczeniu się z ludzkiego feedbacku (RLHF), co zwiększa użyteczność i bezpieczeństwo. Jego wersja dostosowana do instrukcji jest zoptymalizowana do wielojęzycznych rozmów, osiągając lepsze wyniki w wielu branżowych benchmarkach niż wiele modeli czatu open-source i zamkniętych. Data graniczna wiedzy to grudzień 2023 roku."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Meta Llama 3.3 to wielojęzyczny model językowy (LLM) o skali 70B (wejście/wyjście tekstowe), będący modelem generacyjnym wstępnie wytrenowanym i dostosowanym do instrukcji. Model Llama 3.3 dostosowany do instrukcji jest zoptymalizowany pod kątem zastosowań w dialogach wielojęzycznych i przewyższa wiele dostępnych modeli open-source i zamkniętych w popularnych testach branżowych."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 zaprojektowana do przetwarzania zadań łączących dane wizualne i tekstowe. Doskonała w zadaniach takich jak opisywanie obrazów i wizualne pytania odpowiedzi, przekracza granice między generowaniem języka a wnioskowaniem wizualnym."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "Model Llama 3.1 Turbo 405B oferuje ogromną pojemność kontekstową dla przetwarzania dużych danych, wyróżniając się w zastosowaniach sztucznej inteligencji o dużej skali."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 to wiodący model wydany przez Meta, wspierający do 405B parametrów, mogący być stosowany w złożonych rozmowach, tłumaczeniach wielojęzycznych i analizie danych."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B oferuje efektywne wsparcie dialogowe w wielu językach."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 jest zaprojektowana do przetwarzania zadań łączących dane wizualne i tekstowe. Wykazuje doskonałe wyniki w zadaniach takich jak opisywanie obrazów i wizualne pytania i odpowiedzi, przekraczając granice między generowaniem języka a wnioskowaniem wizualnym."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 to najnowocześniejszy wielojęzyczny, otwarty model językowy z serii Llama, który oferuje wydajność porównywalną z modelem 405B przy bardzo niskich kosztach. Opiera się na strukturze Transformer i poprawia użyteczność oraz bezpieczeństwo dzięki nadzorowanemu dostrajaniu (SFT) i uczeniu ze wzmocnieniem na podstawie ludzkich opinii (RLHF). Jego wersja dostosowana do instrukcji jest zoptymalizowana do wielojęzycznych rozmów i w wielu branżowych benchmarkach przewyższa wiele otwartych i zamkniętych modeli czatu. Data graniczna wiedzy to grudzień 2023."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 to najnowocześniejszy wielojęzyczny, otwarty model językowy z serii Llama, który oferuje wydajność porównywalną z modelem 405B przy bardzo niskich kosztach. Opiera się na strukturze Transformer i poprawia użyteczność oraz bezpieczeństwo dzięki nadzorowanemu dostrajaniu (SFT) i uczeniu ze wzmocnieniem na podstawie ludzkich opinii (RLHF). Jego wersja dostosowana do instrukcji jest zoptymalizowana do wielojęzycznych rozmów i w wielu branżowych benchmarkach przewyższa wiele otwartych i zamkniętych modeli czatu. Data graniczna wiedzy to grudzień 2023."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct to największy i najpotężniejszy model w rodzinie modeli Llama 3.1 Instruct. Jest to wysoko zaawansowany model do dialogów, wnioskowania i generowania danych, który może być również używany jako podstawa do specjalistycznego, ciągłego wstępnego szkolenia lub dostosowywania w określonych dziedzinach. Llama 3.1 oferuje wielojęzyczne duże modele językowe (LLM), które są zestawem wstępnie wytrenowanych, dostosowanych do instrukcji modeli generacyjnych, obejmujących rozmiary 8B, 70B i 405B (wejście/wyjście tekstowe). Modele tekstowe Llama 3.1 dostosowane do instrukcji (8B, 70B, 405B) zostały zoptymalizowane do zastosowań w wielojęzycznych dialogach i przewyższają wiele dostępnych modeli czatu open source w powszechnych testach branżowych. Llama 3.1 jest zaprojektowana do użytku komercyjnego i badawczego w wielu językach. Modele tekstowe dostosowane do instrukcji nadają się do czatu w stylu asystenta, podczas gdy modele wstępnie wytrenowane mogą być dostosowane do różnych zadań generowania języka naturalnego. Modele Llama 3.1 wspierają również wykorzystanie ich wyjść do poprawy innych modeli, w tym generowania danych syntetycznych i udoskonalania. Llama 3.1 jest modelem językowym autoregresywnym opartym na zoptymalizowanej architekturze transformatora. Dostosowane wersje wykorzystują nadzorowane dostosowywanie (SFT) oraz uczenie się ze wzmocnieniem z ludzkim feedbackiem (RLHF), aby odpowiadać ludzkim preferencjom dotyczącym pomocności i bezpieczeństwa."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 to otwarty duży model językowy (LLM) skierowany do deweloperów, badaczy i przedsiębiorstw, mający na celu pomoc w budowaniu, eksperymentowaniu i odpowiedzialnym rozwijaniu ich pomysłów na generatywną sztuczną inteligencję. Jako część podstawowego systemu innowacji globalnej społeczności, jest idealny dla urządzeń o ograniczonej mocy obliczeniowej i zasobach, a także dla szybszego czasu szkolenia."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "Zaawansowany LLM, wspierający generowanie danych syntetycznych, destylację wiedzy i wnioskowanie, odpowiedni do chatbotów, programowania i zadań w określonych dziedzinach."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "Umożliwia złożone rozmowy, posiadając doskonałe zrozumienie kontekstu, zdolności wnioskowania i generowania tekstu."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "Zaawansowany, nowoczesny model, posiadający zrozumienie języka, doskonałe zdolności wnioskowania i generowania tekstu."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "Nowoczesny model wizualno-językowy, specjalizujący się w wysokiej jakości wnioskowaniu z obrazów."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "Zaawansowany, nowoczesny mały model językowy, posiadający zrozumienie języka, doskonałe zdolności wnioskowania i generowania tekstu."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "Zaawansowany, nowoczesny mały model językowy, posiadający zrozumienie języka, doskonałe zdolności wnioskowania i generowania tekstu."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "Nowoczesny model wizualno-językowy, specjalizujący się w wysokiej jakości wnioskowaniu z obrazów."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "Zaawansowany LLM, specjalizujący się w wnioskowaniu, matematyce, zdrowym rozsądku i wywoływaniu funkcji."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 to model językowy oferowany przez Microsoft AI, który wyróżnia się w złożonych rozmowach, wielojęzyczności, wnioskowaniu i jako inteligentny asystent."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K to model o zdolności przetwarzania kontekstu o ultra-długiej długości, odpowiedni do generowania bardzo długich tekstów, spełniający wymagania złożonych zadań generacyjnych, zdolny do przetwarzania treści do 128 000 tokenów, idealny do zastosowań w badaniach, akademickich i generowaniu dużych dokumentów."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Model wizualny Kimi (w tym moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview itp.) potrafi rozumieć treść obrazów, w tym teksty na obrazach, kolory obrazów i kształty obiektów."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K oferuje zdolność przetwarzania kontekstu o średniej długości, zdolną do przetwarzania 32 768 tokenów, szczególnie odpowiednią do generowania różnych długich dokumentów i złożonych dialogów, stosowaną w tworzeniu treści, generowaniu raportów i systemach dialogowych."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Model wizualny Kimi (w tym moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview itp.) potrafi rozumieć treść obrazów, w tym teksty na obrazach, kolory obrazów i kształty obiektów."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K zaprojektowany do generowania krótkich tekstów, charakteryzuje się wydajnością przetwarzania, zdolny do przetwarzania 8 192 tokenów, idealny do krótkich dialogów, notatek i szybkiego generowania treści."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Model wizualny Kimi (w tym moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview itp.) potrafi rozumieć treść obrazów, w tym teksty na obrazach, kolory obrazów i kształty obiektów."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B to ulepszona wersja Nous Hermes 2, zawierająca najnowsze wewnętrznie opracowane zbiory danych."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B to duży model językowy stworzony przez NVIDIA, zaprojektowany w celu zwiększenia użyteczności odpowiedzi generowanych przez LLM dla zapytań użytkowników."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B to dostosowany przez NVIDIA duży model językowy, mający na celu zwiększenie użyteczności odpowiedzi generowanych przez LLM w odpowiedzi na zapytania użytkowników. Model ten osiągnął doskonałe wyniki w testach benchmarkowych, takich jak Arena Hard, AlpacaEval 2 LC i GPT-4-Turbo MT-Bench, zajmując pierwsze miejsce we wszystkich trzech automatycznych testach do 1 października 2024 roku. Model został przeszkolony przy użyciu RLHF (szczególnie REINFORCE), Llama-3.1-Nemotron-70B-Reward i HelpSteer2-Preference na bazie modelu Llama-3.1-70B-Instruct."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "Unikalny model językowy, oferujący niezrównaną dokładność i wydajność."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct to dostosowany przez NVIDIA duży model językowy, zaprojektowany w celu zwiększenia użyteczności odpowiedzi generowanych przez LLM."
+ },
"o1": {
"description": "Skupia się na zaawansowanym wnioskowaniu i rozwiązywaniu złożonych problemów, w tym zadań matematycznych i naukowych. Doskonale nadaje się do aplikacji wymagających głębokiego zrozumienia kontekstu i zarządzania procesami."
},
- "o1-2024-12-17": {
- "description": "o1 to nowy model wnioskowania OpenAI, który obsługuje wejścia tekstowe i graficzne oraz generuje tekst, idealny do złożonych zadań wymagających szerokiej wiedzy ogólnej. Model ma kontekst 200K i datę graniczną wiedzy do października 2023 roku."
- },
"o1-mini": {
"description": "o1-mini to szybki i ekonomiczny model wnioskowania zaprojektowany z myślą o programowaniu, matematyce i zastosowaniach naukowych. Model ten ma kontekst 128K i datę graniczną wiedzy z października 2023 roku."
},
"o1-preview": {
"description": "o1 to nowy model wnioskowania OpenAI, odpowiedni do złożonych zadań wymagających szerokiej wiedzy ogólnej. Model ten ma kontekst 128K i datę graniczną wiedzy z października 2023 roku."
},
+ "o3-mini": {
+ "description": "o3-mini to nasz najnowszy mały model wnioskowania, który oferuje wysoką inteligencję przy tych samych kosztach i celach opóźnienia co o1-mini."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba to model językowy Mamba 2 skoncentrowany na generowaniu kodu, oferujący silne wsparcie dla zaawansowanych zadań kodowania i wnioskowania."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini to najnowszy model OpenAI, wydany po GPT-4 Omni, obsługujący wejścia tekstowe i wizualne. Jako ich najnowocześniejszy mały model, jest znacznie tańszy od innych niedawnych modeli czołowych i kosztuje o ponad 60% mniej niż GPT-3.5 Turbo. Utrzymuje najnowocześniejszą inteligencję, oferując jednocześnie znaczną wartość za pieniądze. GPT-4o mini uzyskał wynik 82% w teście MMLU i obecnie zajmuje wyższą pozycję w preferencjach czatu niż GPT-4."
},
- "openai/o1": {
- "description": "o1 to nowy model wnioskowania OpenAI, który obsługuje wejścia tekstowe i graficzne oraz generuje tekst, idealny do złożonych zadań wymagających szerokiej wiedzy ogólnej. Model ma kontekst 200K i datę graniczną wiedzy do października 2023 roku."
- },
"openai/o1-mini": {
"description": "o1-mini to szybki i ekonomiczny model wnioskowania zaprojektowany z myślą o programowaniu, matematyce i zastosowaniach naukowych. Model ten ma kontekst 128K i datę graniczną wiedzy z października 2023 roku."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K jest wyposażony w wyjątkową zdolność przetwarzania kontekstu, mogąc obsługiwać do 128K informacji kontekstowych, co czyni go idealnym do analizy całościowej i długoterminowego przetwarzania logicznych powiązań w długich treściach, zapewniając płynność i spójność logiczną oraz różnorodne wsparcie cytatów w złożonej komunikacji tekstowej."
},
+ "qvq-72b-preview": {
+ "description": "Model QVQ jest eksperymentalnym modelem badawczym opracowanym przez zespół Qwen, skoncentrowanym na zwiększeniu zdolności w zakresie rozumowania wizualnego, szczególnie w dziedzinie rozumowania matematycznego."
+ },
"qwen-coder-plus-latest": {
"description": "Model kodowania Qwen, oparty na ogólnym zrozumieniu."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Model wizualno-językowy Qwen o ultra dużej skali. W porównaniu do wersji rozszerzonej, ponownie zwiększa zdolności wnioskowania wizualnego i przestrzegania instrukcji, oferując wyższy poziom percepcji wizualnej i poznawczej."
},
+ "qwen-vl-ocr-latest": {
+ "description": "Model OCR Tongyi Qianwen to specjalistyczny model do ekstrakcji tekstu, skoncentrowany na zdolności do wydobywania tekstu z obrazów dokumentów, tabel, zadań testowych i pisma ręcznego. Potrafi rozpoznawać wiele języków, w tym: chiński, angielski, francuski, japoński, koreański, niemiecki, rosyjski, włoski, wietnamski i arabski."
+ },
"qwen-vl-plus-latest": {
"description": "Wersja rozszerzona modelu wizualno-językowego Qwen. Znacząco poprawia zdolność rozpoznawania szczegółów i tekstu, obsługuje obrazy o rozdzielczości przekraczającej milion pikseli oraz dowolnych proporcjach."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 to nowa seria dużych modeli językowych, charakteryzująca się silniejszymi zdolnościami rozumienia i generowania."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "LLM skierowany na język chiński i angielski, skoncentrowany na języku, programowaniu, matematyce, wnioskowaniu i innych dziedzinach."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "Zaawansowany LLM, wspierający generowanie kodu, wnioskowanie i naprawę, obejmujący główne języki programowania."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "Potężny średniej wielkości model kodu, wspierający długość kontekstu 32K, specjalizujący się w programowaniu wielojęzycznym."
+ },
"qwen2": {
"description": "Qwen2 to nowa generacja dużego modelu językowego Alibaba, wspierająca różnorodne potrzeby aplikacyjne dzięki doskonałej wydajności."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "Model Qwen 2.5 o skali 14B, udostępniony na zasadzie open source."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "Model o skali 72B, udostępniony przez Tongyi Qianwen 2.5."
+ },
"qwen2.5-32b-instruct": {
"description": "Model Qwen 2.5 o skali 32B, udostępniony na zasadzie open source."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Model Qwen-Math, który ma silne zdolności rozwiązywania problemów matematycznych."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "Zwiększona zdolność do podążania za instrukcjami, matematyki, rozwiązywania problemów i kodowania, poprawiona zdolność do rozpoznawania obiektów, wsparcie dla różnych formatów do precyzyjnego lokalizowania elementów wizualnych, zdolność do rozumienia długich plików wideo (do 10 minut) oraz lokalizowania momentów zdarzeń w czasie rzeczywistym, zdolność do rozumienia kolejności czasowej i szybkości, wsparcie dla operacji na systemach OS lub Mobile, silna zdolność do ekstrakcji kluczowych informacji i generowania wyjścia w formacie JSON. Ta wersja to wersja 72B, najsilniejsza w tej serii."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "Zwiększona zdolność do podążania za instrukcjami, matematyki, rozwiązywania problemów i kodowania, poprawiona zdolność do rozpoznawania obiektów, wsparcie dla różnych formatów do precyzyjnego lokalizowania elementów wizualnych, zdolność do rozumienia długich plików wideo (do 10 minut) oraz lokalizowania momentów zdarzeń w czasie rzeczywistym, zdolność do rozumienia kolejności czasowej i szybkości, wsparcie dla operacji na systemach OS lub Mobile, silna zdolność do ekstrakcji kluczowych informacji i generowania wyjścia w formacie JSON. Ta wersja to wersja 72B, najsilniejsza w tej serii."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 to nowa generacja dużego modelu językowego Alibaba, który wspiera różnorodne potrzeby aplikacyjne dzięki doskonałej wydajności."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro to model LLM o wysokiej inteligencji wydany przez Upstage, koncentrujący się na zdolności do przestrzegania instrukcji na pojedynczym GPU, osiągając wynik IFEval powyżej 80. Obecnie wspiera język angielski, a wersja oficjalna planowana jest na listopad 2024, z rozszerzeniem wsparcia językowego i długości kontekstu."
},
+ "sonar": {
+ "description": "Lekki produkt wyszukiwania oparty na kontekście, szybszy i tańszy niż Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "Zaawansowany produkt wyszukiwania wspierający kontekst wyszukiwania, oferujący zaawansowane zapytania i śledzenie."
+ },
+ "sonar-reasoning": {
+ "description": "Nowy produkt API wspierany przez model wnioskowania DeepSeek."
+ },
"step-1-128k": {
"description": "Równoważy wydajność i koszty, odpowiedni do ogólnych scenariuszy."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "Ten model ma potężne zdolności rozumienia wideo."
},
+ "step-1o-vision-32k": {
+ "description": "Ten model ma potężne zdolności rozumienia obrazów. W porównaniu do modeli z serii step-1v, oferuje lepsze osiągi wizualne."
+ },
"step-1v-32k": {
"description": "Obsługuje wejścia wizualne, wzmacniając doświadczenie interakcji multimodalnych."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Obsługuje interakcje z dużą ilością kontekstu, idealny do złożonych scenariuszy dialogowych."
},
+ "step-2-mini": {
+ "description": "Model oparty na nowej generacji własnej architektury Attention MFA, osiągający podobne wyniki jak step1 przy bardzo niskich kosztach, jednocześnie zapewniając wyższą przepustowość i szybszy czas reakcji. Potrafi obsługiwać ogólne zadania, a w zakresie umiejętności kodowania ma szczególne zdolności."
+ },
+ "taichu2_mm": {
+ "description": "Łączy zdolności rozumienia obrazów, transferu wiedzy, logicznego wnioskowania i wyróżnia się w dziedzinie pytań i odpowiedzi związanych z obrazem i tekstem."
+ },
"taichu_llm": {
"description": "Model językowy TaiChu charakteryzuje się wyjątkową zdolnością rozumienia języka oraz umiejętnościami w zakresie tworzenia tekstów, odpowiadania na pytania, programowania, obliczeń matematycznych, wnioskowania logicznego, analizy emocji i streszczenia tekstu. Innowacyjnie łączy wstępne uczenie się na dużych zbiorach danych z bogatą wiedzą z wielu źródeł, stale doskonaląc technologię algorytmiczną i nieustannie przyswajając nową wiedzę z zakresu słownictwa, struktury, gramatyki i semantyki z ogromnych zbiorów danych tekstowych, co prowadzi do ciągłej ewolucji modelu. Umożliwia użytkownikom łatwiejszy dostęp do informacji i usług oraz bardziej inteligentne doświadczenia."
},
+ "text-embedding-3-large": {
+ "description": "Najpotężniejszy model wektoryzacji, odpowiedni do zadań w języku angielskim i innych językach."
+ },
+ "text-embedding-3-small": {
+ "description": "Nowej generacji model Embedding, efektywny i ekonomiczny, odpowiedni do wyszukiwania wiedzy, aplikacji RAG i innych scenariuszy."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) oferuje zwiększoną moc obliczeniową dzięki efektywnym strategiom i architekturze modelu."
},
+ "tts-1": {
+ "description": "Najnowocześniejszy model tekstu na mowę, zoptymalizowany pod kątem szybkości w scenariuszach w czasie rzeczywistym."
+ },
+ "tts-1-hd": {
+ "description": "Najnowocześniejszy model tekstu na mowę, zoptymalizowany pod kątem jakości."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) jest przeznaczony do precyzyjnych zadań poleceniowych, oferując doskonałe możliwości przetwarzania języka."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet podnosi standardy branżowe, przewyższając modele konkurencji oraz Claude 3 Opus, osiągając doskonałe wyniki w szerokim zakresie ocen, przy zachowaniu prędkości i kosztów naszych modeli średniego poziomu."
},
+ "whisper-1": {
+ "description": "Uniwersalny model rozpoznawania mowy, obsługujący rozpoznawanie mowy w wielu językach, tłumaczenie mowy i rozpoznawanie języków."
+ },
"wizardlm2": {
"description": "WizardLM 2 to model językowy dostarczany przez Microsoft AI, który wyróżnia się w złożonych dialogach, wielojęzyczności, wnioskowaniu i inteligentnych asystentach."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Model do złożonych zadań wizualnych, oferujący wysoką wydajność w zakresie rozumienia i analizy obrazów."
+ },
+ "yi-vision-v2": {
+ "description": "Model do złożonych zadań wizualnych, oferujący wysokowydajną zdolność rozumienia i analizy na podstawie wielu obrazów."
}
}
diff --git a/locales/pl-PL/providers.json b/locales/pl-PL/providers.json
index 719a12c3cb8b7..8199be3199b07 100644
--- a/locales/pl-PL/providers.json
+++ b/locales/pl-PL/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure oferuje różnorodne zaawansowane modele AI, w tym GPT-3.5 i najnowszą serię GPT-4, wspierające różne typy danych i złożone zadania, koncentrując się na bezpiecznych, niezawodnych i zrównoważonych rozwiązaniach AI."
},
+ "azureai": {
+ "description": "Azure oferuje wiele zaawansowanych modeli AI, w tym GPT-3.5 i najnowszą serię GPT-4, wspierając różne typy danych i złożone zadania, dążąc do bezpiecznych, niezawodnych i zrównoważonych rozwiązań AI."
+ },
"baichuan": {
"description": "Baichuan Intelligent to firma skoncentrowana na badaniach nad dużymi modelami sztucznej inteligencji, której modele osiągają doskonałe wyniki w krajowych zadaniach związanych z encyklopedią wiedzy, przetwarzaniem długich tekstów i generowaniem treści w języku chińskim, przewyższając zagraniczne modele mainstreamowe. Baichuan Intelligent dysponuje również wiodącymi w branży zdolnościami multimodalnymi, osiągając doskonałe wyniki w wielu autorytatywnych ocenach. Jej modele obejmują Baichuan 4, Baichuan 3 Turbo i Baichuan 3 Turbo 128k, zoptymalizowane pod kątem różnych scenariuszy zastosowań, oferując opłacalne rozwiązania."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek to firma skoncentrowana na badaniach i zastosowaniach technologii sztucznej inteligencji, której najnowszy model DeepSeek-V2.5 łączy zdolności do prowadzenia ogólnych rozmów i przetwarzania kodu, osiągając znaczące postępy w zakresie dostosowywania do preferencji ludzkich, zadań pisarskich i przestrzegania instrukcji."
},
+ "doubao": {
+ "description": "Model dużych rozmiarów opracowany przez ByteDance. Potwierdzony w ponad 50 scenariuszach biznesowych wewnątrz firmy, z codziennym użyciem bilionów tokenów, co pozwala na ciągłe doskonalenie. Oferuje różnorodne możliwości modalne, tworząc bogate doświadczenia biznesowe dla przedsiębiorstw dzięki wysokiej jakości modelom."
+ },
"fireworksai": {
"description": "Fireworks AI to wiodący dostawca zaawansowanych modeli językowych, skoncentrowany na wywołaniach funkcji i przetwarzaniu multimodalnym. Jego najnowszy model Firefunction V2 oparty na Llama-3, zoptymalizowany do wywołań funkcji, dialogów i przestrzegania instrukcji. Model wizualny FireLLaVA-13B wspiera mieszane wejścia obrazów i tekstu. Inne znaczące modele to seria Llama i seria Mixtral, oferujące efektywne wsparcie dla wielojęzycznego przestrzegania instrukcji i generacji."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "Organizacja open source poświęcona badaniom i rozwojowi narzędzi dla dużych modeli. Oferuje wszystkim deweloperom AI wydajną i łatwą w użyciu platformę open source, umożliwiającą dostęp do najnowocześniejszych technologii modeli i algorytmów."
},
+ "jina": {
+ "description": "Jina AI została założona w 2020 roku i jest wiodącą firmą zajmującą się AI w zakresie wyszukiwania. Nasza platforma bazowa do wyszukiwania zawiera modele wektorowe, przetasowujące i małe modele językowe, które pomagają firmom budować niezawodne i wysokiej jakości aplikacje wyszukiwania generatywnego i multimodalnego."
+ },
+ "lmstudio": {
+ "description": "LM Studio to aplikacja desktopowa do rozwijania i eksperymentowania z LLM-ami na Twoim komputerze."
+ },
"minimax": {
"description": "MiniMax to firma technologiczna zajmująca się ogólną sztuczną inteligencją, założona w 2021 roku, dążąca do współtworzenia inteligencji z użytkownikami. MiniMax opracowało różne modele dużych modeli o różnych modalnościach, w tym model tekstowy MoE z bilionem parametrów, model głosowy oraz model obrazowy. Wprowadziło również aplikacje takie jak Conch AI."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI to platforma oferująca API do różnych dużych modeli językowych i generacji obrazów AI, elastyczna, niezawodna i opłacalna. Wspiera najnowsze modele open-source, takie jak Llama3, Mistral, i oferuje kompleksowe, przyjazne dla użytkownika oraz automatycznie skalowalne rozwiązania API dla rozwoju aplikacji generatywnej AI, odpowiednie dla szybkiego rozwoju startupów AI."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ oferuje kontenery do samodzielnego hostowania przyspieszonych przez GPU mikroserwisów wnioskowania, wspierając wdrażanie w chmurze, centrach danych, komputerach osobistych RTX™ AI i stacjach roboczych wstępnie wytrenowanych i dostosowanych modeli AI."
+ },
"ollama": {
"description": "Modele oferowane przez Ollama obejmują szeroki zakres zastosowań, w tym generowanie kodu, obliczenia matematyczne, przetwarzanie wielojęzyczne i interakcje konwersacyjne, wspierając różnorodne potrzeby wdrożeń na poziomie przedsiębiorstw i lokalnych."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "Nowa generacja multimodalnych dużych modeli opracowana przez Instytut Automatyki Chińskiej Akademii Nauk i Wuhan Institute of Artificial Intelligence wspiera wielorundowe pytania i odpowiedzi, tworzenie tekstów, generowanie obrazów, zrozumienie 3D, analizę sygnałów i inne kompleksowe zadania pytaniowe, posiadając silniejsze zdolności poznawcze, rozumienia i tworzenia, oferując nową interaktywną doświadczenie."
},
+ "tencentcloud": {
+ "description": "Atomowe możliwości silnika wiedzy (LLM Knowledge Engine Atomic Power) oparte na silniku wiedzy, oferujące pełne możliwości zadawania pytań i odpowiedzi, skierowane do przedsiębiorstw i deweloperów, zapewniające elastyczność w budowaniu i rozwijaniu aplikacji modelowych. Możesz tworzyć własne usługi modelowe za pomocą różnych atomowych możliwości, korzystając z usług takich jak analiza dokumentów, podział, embedding, wielokrotne przeredagowywanie i inne, aby dostosować AI do specyficznych potrzeb Twojej firmy."
+ },
"togetherai": {
"description": "Together AI dąży do osiągnięcia wiodącej wydajności poprzez innowacyjne modele AI, oferując szerokie możliwości dostosowywania, w tym wsparcie dla szybkiej ekspansji i intuicyjnych procesów wdrożeniowych, aby zaspokoić różnorodne potrzeby przedsiębiorstw."
},
"upstage": {
"description": "Upstage koncentruje się na opracowywaniu modeli AI dla różnych potrzeb biznesowych, w tym Solar LLM i dokumentów AI, mając na celu osiągnięcie sztucznej ogólnej inteligencji (AGI). Umożliwia tworzenie prostych agentów konwersacyjnych za pomocą Chat API oraz wspiera wywołania funkcji, tłumaczenia, osadzenia i zastosowania w określonych dziedzinach."
},
+ "vllm": {
+ "description": "vLLM to szybka i łatwa w użyciu biblioteka do wnioskowania i usług LLM."
+ },
+ "volcengine": {
+ "description": "Platforma deweloperska usług dużych modeli wprowadzona przez ByteDance, oferująca bogate w funkcje, bezpieczne i konkurencyjne cenowo usługi wywoływania modeli. Oferuje również dane modelowe, dostosowywanie, wnioskowanie, ocenę i inne funkcje end-to-end, zapewniając kompleksowe wsparcie dla rozwoju aplikacji AI."
+ },
"wenxin": {
"description": "Platforma do rozwoju i usług aplikacji AI oraz dużych modeli w skali przedsiębiorstwa, oferująca najbardziej kompleksowy i łatwy w użyciu zestaw narzędzi do rozwoju modeli sztucznej inteligencji generatywnej oraz całego procesu tworzenia aplikacji."
},
diff --git a/locales/pl-PL/setting.json b/locales/pl-PL/setting.json
index 0a2e25083f954..ec76e38d9ba16 100644
--- a/locales/pl-PL/setting.json
+++ b/locales/pl-PL/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Włącz limit jednorazowej odpowiedzi"
},
+ "enableReasoningEffort": {
+ "title": "Włącz dostosowanie intensywności rozumowania"
+ },
"frequencyPenalty": {
- "desc": "Im większa wartość, tym większe prawdopodobieństwo zmniejszenia powtarzających się słów",
- "title": "Kara za częstość"
+ "desc": "Im większa wartość, tym bardziej zróżnicowane i bogate słownictwo; im mniejsza wartość, tym prostsze i bardziej bezpośrednie słownictwo",
+ "title": "Różnorodność słownictwa"
},
"maxTokens": {
"desc": "Maksymalna liczba tokenów używanych w pojedynczej interakcji",
@@ -212,19 +215,31 @@
"desc": "{{provider}} model",
"title": "Model"
},
+ "params": {
+ "title": "Zaawansowane parametry"
+ },
"presencePenalty": {
- "desc": "Im większa wartość, tym większe prawdopodobieństwo rozszerzenia się na nowe tematy",
- "title": "Świeżość tematu"
+ "desc": "Im większa wartość, tym większa tendencja do różnorodnych wyrażeń, unikanie powtórzeń; im mniejsza wartość, tym większa tendencja do używania powtarzających się koncepcji lub narracji, co prowadzi do większej spójności",
+ "title": "Różnorodność wyrażeń"
+ },
+ "reasoningEffort": {
+ "desc": "Im wyższa wartość, tym silniejsza zdolność rozumowania, ale może to zwiększyć czas odpowiedzi i zużycie tokenów",
+ "options": {
+ "high": "Wysoki",
+ "low": "Niski",
+ "medium": "Średni"
+ },
+ "title": "Intensywność rozumowania"
},
"temperature": {
- "desc": "Im większa wartość, tym odpowiedzi są bardziej losowe",
- "title": "Losowość",
- "titleWithValue": "Losowość {{value}}"
+ "desc": "Im większa wartość, tym bardziej kreatywne i wyobrażeniowe będą odpowiedzi; im mniejsza wartość, tym bardziej rygorystyczne odpowiedzi",
+ "title": "Aktywność kreatywna",
+ "warning": "Zbyt wysoka wartość aktywności kreatywnej może prowadzić do nieczytelnych wyników"
},
"title": "Ustawienia modelu",
"topP": {
- "desc": "Podobne do losowości, ale nie należy zmieniać razem z losowością",
- "title": "Najlepsze P"
+ "desc": "Ile możliwości należy rozważyć, im większa wartość, tym więcej możliwych odpowiedzi; im mniejsza wartość, tym większa tendencja do wyboru najbardziej prawdopodobnej odpowiedzi. Nie zaleca się jednoczesnej zmiany z aktywnością kreatywną",
+ "title": "Otwartość myślenia"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Ustawienia ogólne",
"experiment": "Eksperyment",
"llm": "Model językowy",
+ "provider": "Dostawca usług AI",
"sync": "Synchronizacja w chmurze",
"system-agent": "System Agent",
"tts": "Usługa głosowa"
diff --git a/locales/pt-BR/changelog.json b/locales/pt-BR/changelog.json
index 50b34c1dd02c9..384c7eb3159c8 100644
--- a/locales/pt-BR/changelog.json
+++ b/locales/pt-BR/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "Veja todos os registros de alterações",
"description": "Acompanhe as novas funcionalidades e melhorias do {{appName}}",
"pagination": {
- "older": "Ver alterações anteriores",
- "prev": "Página anterior"
+ "next": "Próxima página",
+ "older": "Ver alterações anteriores"
},
"readDetails": "Leia os detalhes",
"title": "Registro de Atualizações",
diff --git a/locales/pt-BR/common.json b/locales/pt-BR/common.json
index d704d8af05aa0..3328ad36b2eae 100644
--- a/locales/pt-BR/common.json
+++ b/locales/pt-BR/common.json
@@ -275,6 +275,7 @@
},
"temp": "Temporário",
"terms": "Termos de Serviço",
+ "update": "Atualizar",
"updateAgent": "Atualizar informações do assistente",
"upgradeVersion": {
"action": "Atualizar",
@@ -285,6 +286,7 @@
"anonymousNickName": "Usuário Anônimo",
"billing": "Gerenciamento de faturas",
"cloud": "Experimente {{name}}",
+ "community": "Versão Comunitária",
"data": "Armazenamento de dados",
"defaultNickname": "Usuário da Comunidade",
"discord": "Suporte da Comunidade",
@@ -294,7 +296,6 @@
"help": "Central de Ajuda",
"moveGuide": "O botão de configurações foi movido para cá",
"plans": "Planos de Assinatura",
- "preview": "Versão de visualização",
"profile": "Gerenciamento de Conta",
"setting": "Configurações do Aplicativo",
"usages": "Estatísticas de Uso"
diff --git a/locales/pt-BR/components.json b/locales/pt-BR/components.json
index 53f87d52c3ab9..2768dddaf8e31 100644
--- a/locales/pt-BR/components.json
+++ b/locales/pt-BR/components.json
@@ -76,6 +76,7 @@
"custom": "Modelo personalizado, por padrão, suporta chamadas de função e reconhecimento visual. Por favor, verifique a disponibilidade dessas capacidades de acordo com a situação real.",
"file": "Este modelo suporta leitura e reconhecimento de arquivos enviados.",
"functionCall": "Este modelo suporta chamadas de função.",
+ "reasoning": "Este modelo suporta pensamento profundo",
"tokens": "Este modelo suporta no máximo {{tokens}} tokens por sessão.",
"vision": "Este modelo suporta reconhecimento visual."
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "Nenhum modelo habilitado. Por favor, vá para as configurações e habilite um.",
"provider": "Fornecedor"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "Devido a restrições de segurança do navegador, você precisa configurar o CORS para o Ollama antes de usá-lo normalmente.",
+ "linux": {
+ "env": "Adicione `Environment` na seção [Service] e adicione a variável de ambiente OLLAMA_ORIGINS:",
+ "reboot": "Recarregue o systemd e reinicie o Ollama",
+ "systemd": "Chame o systemd para editar o serviço ollama:"
+ },
+ "macos": "Abra o aplicativo 'Terminal', cole o seguinte comando e pressione Enter para executar",
+ "reboot": "Reinicie o serviço Ollama após a conclusão da execução",
+ "title": "Configurar o Ollama para permitir acesso CORS",
+ "windows": "No Windows, clique em 'Painel de Controle' e entre na edição das variáveis de ambiente do sistema. Crie uma nova variável de ambiente chamada 'OLLAMA_ORIGINS' para sua conta de usuário, com o valor * e clique em 'OK/Aplicar' para salvar."
+ },
+ "install": {
+ "description": "Por favor, confirme que você já ativou o Ollama. Se não tiver baixado o Ollama, visite o site oficial <1>para baixar",
+ "docker": "Se você preferir usar o Docker, o Ollama também oferece uma imagem oficial do Docker, que você pode puxar com o seguinte comando:",
+ "linux": {
+ "command": "Instale com o seguinte comando:",
+ "manual": "Ou, você também pode consultar o <1>Guia de Instalação Manual do Linux para instalar por conta própria."
+ },
+ "title": "Instalar e iniciar o aplicativo Ollama localmente",
+ "windowsTab": "Windows (versão de pré-visualização)"
+ }
+ },
+ "Thinking": {
+ "thinking": "Pensando profundamente...",
+ "thought": "Já pensei profundamente (tempo gasto {{duration}} segundos)",
+ "thoughtWithDuration": "Já pensei profundamente"
}
}
diff --git a/locales/pt-BR/discover.json b/locales/pt-BR/discover.json
index 4b88bc81c6b29..6f2c6de8923fc 100644
--- a/locales/pt-BR/discover.json
+++ b/locales/pt-BR/discover.json
@@ -126,6 +126,10 @@
"title": "Novidade do Tópico"
},
"range": "Faixa",
+ "reasoning_effort": {
+ "desc": "Esta configuração é usada para controlar a intensidade de raciocínio do modelo antes de gerar uma resposta. Intensidade baixa prioriza a velocidade de resposta e economiza Tokens, enquanto intensidade alta oferece um raciocínio mais completo, mas consome mais Tokens e reduz a velocidade de resposta. O valor padrão é médio, equilibrando a precisão do raciocínio com a velocidade de resposta.",
+ "title": "Intensidade de Raciocínio"
+ },
"temperature": {
"desc": "Esta configuração afeta a diversidade das respostas do modelo. Valores mais baixos resultam em respostas mais previsíveis e típicas, enquanto valores mais altos incentivam respostas mais variadas e incomuns. Quando o valor é 0, o modelo sempre dá a mesma resposta para uma entrada dada.",
"title": "Aleatoriedade"
diff --git a/locales/pt-BR/modelProvider.json b/locales/pt-BR/modelProvider.json
index de62bc6036c71..d3212261f7d1b 100644
--- a/locales/pt-BR/modelProvider.json
+++ b/locales/pt-BR/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API Key"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Versão da API do Azure, seguindo o formato AAAA-MM-DD. Consulte a [versão mais recente](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "Obter lista",
+ "title": "Versão da API do Azure"
+ },
+ "endpoint": {
+ "desc": "Encontre o ponto de extremidade de inferência do modelo do Azure AI na visão geral do projeto Azure AI",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Ponto de extremidade do Azure AI"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Encontre a chave da API na visão geral do projeto Azure AI",
+ "placeholder": "Chave do Azure",
+ "title": "Chave"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "Insira o AWS Access Key Id",
@@ -63,6 +81,46 @@
"title": "ID da conta do Cloudflare / Endereço da API"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Por favor, insira sua API Key",
+ "title": "API Key"
+ },
+ "basicTitle": "Informações Básicas",
+ "configTitle": "Informações de Configuração",
+ "confirm": "Criar Novo",
+ "createSuccess": "Criação bem-sucedida",
+ "description": {
+ "placeholder": "Descrição do provedor (opcional)",
+ "title": "Descrição do Provedor"
+ },
+ "id": {
+ "desc": "Identificador único do provedor de serviços, não pode ser modificado após a criação",
+ "format": "Só pode conter números, letras minúsculas, hífens (-) e sublinhados (_) ",
+ "placeholder": "Sugestão: tudo em minúsculas, por exemplo, openai, não poderá ser modificado após a criação",
+ "required": "Por favor, insira o ID do provedor",
+ "title": "ID do Provedor"
+ },
+ "logo": {
+ "required": "Por favor, envie um logo correto do provedor",
+ "title": "Logo do Provedor"
+ },
+ "name": {
+ "placeholder": "Por favor, insira o nome de exibição do provedor",
+ "required": "Por favor, insira o nome do provedor",
+ "title": "Nome do Provedor"
+ },
+ "proxyUrl": {
+ "required": "Por favor, insira o endereço do proxy",
+ "title": "Endereço do Proxy"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Por favor, selecione o tipo de SDK",
+ "title": "Formato da Requisição"
+ },
+ "title": "Criar Provedor de AI Personalizado"
+ },
"github": {
"personalAccessToken": {
"desc": "Insira seu PAT do Github, clique [aqui](https://github.com/settings/tokens) para criar",
@@ -77,6 +135,23 @@
"title": "Token do HuggingFace"
}
},
+ "list": {
+ "title": {
+ "disabled": "Fornecedor não habilitado",
+ "enabled": "Fornecedor habilitado"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Adicionar Provedor Personalizado",
+ "all": "Todos",
+ "list": {
+ "disabled": "Desativado",
+ "enabled": "Ativado"
+ },
+ "notFound": "Nenhum resultado encontrado",
+ "searchProviders": "Pesquisar Provedores...",
+ "sort": "Ordenação Personalizada"
+ },
"ollama": {
"checker": {
"desc": "Teste se o endereço do proxy está corretamente preenchido",
@@ -94,33 +169,9 @@
"title": "Baixando o modelo {{model}} "
},
"endpoint": {
- "desc": "Insira o endereço do proxy de interface da Ollama, se não foi especificado localmente, pode deixar em branco",
+ "desc": "Deve incluir http(s)://, pode deixar em branco se não houver especificação local adicional",
"title": "Endereço do Proxy de Interface"
},
- "setup": {
- "cors": {
- "description": "Devido às restrições de segurança do navegador, você precisa configurar o Ollama para permitir o acesso entre domínios.",
- "linux": {
- "env": "Sob a seção [Service], adicione `Environment` e inclua a variável de ambiente OLLAMA_ORIGINS:",
- "reboot": "Recarregue o systemd e reinicie o Ollama.",
- "systemd": "Chame o systemd para editar o serviço ollama:"
- },
- "macos": "Abra o aplicativo 'Terminal', cole o comando abaixo e pressione Enter para executar:",
- "reboot": "Após a conclusão, reinicie o serviço Ollama.",
- "title": "Configurar o Ollama para permitir acesso entre domínios",
- "windows": "No Windows, acesse o 'Painel de Controle' e edite as variáveis de ambiente do sistema. Crie uma nova variável de ambiente chamada 'OLLAMA_ORIGINS' para sua conta de usuário, com o valor '*', e clique em 'OK/Aplicar' para salvar."
- },
- "install": {
- "description": "Certifique-se de que você ativou o Ollama. Se ainda não o fez, baixe o Ollama no site oficial <1>aqui.",
- "docker": "Se preferir usar o Docker, o Ollama também oferece uma imagem oficial. Você pode puxá-la com o comando:",
- "linux": {
- "command": "Instale com o comando a seguir:",
- "manual": "Ou, se preferir, consulte o <1>Guia de Instalação Manual do Linux para instalar manualmente."
- },
- "title": "Instale e inicie o aplicativo Ollama localmente",
- "windowsTab": "Windows (Versão de Visualização)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Cancel Download",
@@ -131,25 +182,145 @@
"title": "Download specified Ollama model"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Insira a Access Key da plataforma Qianfan do Baidu",
- "placeholder": "Access Key Qianfan",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "Sua chave e o endereço do proxy serão criptografados usando o algoritmo de criptografia <1>AES-GCM",
+ "apiKey": {
+ "desc": "Por favor, insira sua {{name}} API Key",
+ "placeholder": "{{name}} API Key",
+ "title": "API Key"
+ },
+ "baseURL": {
+ "desc": "Deve incluir http(s)://",
+ "invalid": "Por favor, insira uma URL válida",
+ "placeholder": "https://seu-endereco-proxy.com/v1",
+ "title": "Endereço do Proxy API"
+ },
+ "checker": {
+ "button": "Verificar",
+ "desc": "Teste se a API Key e o endereço do proxy estão preenchidos corretamente",
+ "pass": "Verificação bem-sucedida",
+ "title": "Verificação de Conectividade"
+ },
+ "fetchOnClient": {
+ "desc": "O modo de requisição do cliente iniciará a requisição de sessão diretamente do navegador, podendo aumentar a velocidade de resposta",
+ "title": "Usar Modo de Requisição do Cliente"
+ },
+ "helpDoc": "Tutorial de Configuração",
+ "waitingForMore": "Mais modelos estão <1>planejados para integração, fique atento"
},
- "checker": {
- "desc": "Teste se a AccessKey / SecretAccess está preenchida corretamente"
+ "createNew": {
+ "title": "Criar Modelo de AI Personalizado"
},
- "secretKey": {
- "desc": "Insira a Secret Key da plataforma Qianfan do Baidu",
- "placeholder": "Secret Key Qianfan",
- "title": "Secret Key"
+ "item": {
+ "config": "Configurar Modelo",
+ "customModelCards": {
+ "addNew": "Criar e adicionar modelo {{id}}",
+ "confirmDelete": "Você está prestes a excluir este modelo personalizado, após a exclusão não poderá ser recuperado, por favor, proceda com cautela."
+ },
+ "delete": {
+ "confirm": "Confirmar exclusão do modelo {{displayName}}?",
+ "success": "Exclusão bem-sucedida",
+ "title": "Excluir Modelo"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Campo solicitado na Azure OpenAI",
+ "placeholder": "Por favor, insira o nome de implantação do modelo na Azure",
+ "title": "Nome de Implantação do Modelo"
+ },
+ "deployName": {
+ "extra": "Este campo será usado como ID do modelo ao enviar a solicitação",
+ "placeholder": "Insira o nome ou ID real do modelo implantado",
+ "title": "Nome da implantação do modelo"
+ },
+ "displayName": {
+ "placeholder": "Por favor, insira o nome de exibição do modelo, por exemplo, ChatGPT, GPT-4, etc.",
+ "title": "Nome de Exibição do Modelo"
+ },
+ "files": {
+ "extra": "A implementação atual de upload de arquivos é apenas uma solução temporária, limitada a tentativas pessoais. A capacidade completa de upload de arquivos será implementada posteriormente.",
+ "title": "Suporte a Upload de Arquivos"
+ },
+ "functionCall": {
+ "extra": "Esta configuração ativará apenas a capacidade do modelo de usar ferramentas, permitindo assim a adição de plugins do tipo ferramenta. No entanto, se o uso real das ferramentas é suportado depende inteiramente do modelo em si, teste a usabilidade por conta própria.",
+ "title": "Suporte ao uso de ferramentas"
+ },
+ "id": {
+ "extra": "Não pode ser modificado após a criação, será usado como ID do modelo ao chamar a IA",
+ "placeholder": "Insira o ID do modelo, por exemplo, gpt-4o ou claude-3.5-sonnet",
+ "title": "ID do Modelo"
+ },
+ "modalTitle": "Configuração do Modelo Personalizado",
+ "reasoning": {
+ "extra": "Esta configuração ativará apenas a capacidade de pensamento profundo do modelo, e o efeito específico depende totalmente do próprio modelo. Por favor, teste se este modelo possui a capacidade de pensamento profundo utilizável.",
+ "title": "Suporte a Pensamento Profundo"
+ },
+ "tokens": {
+ "extra": "Configurar o número máximo de tokens suportados pelo modelo",
+ "title": "Janela de contexto máxima",
+ "unlimited": "Ilimitado"
+ },
+ "vision": {
+ "extra": "Esta configuração apenas habilitará a configuração de upload de imagens no aplicativo, se o reconhecimento for suportado depende do modelo em si, teste a capacidade de reconhecimento visual desse modelo.",
+ "title": "Suporte a Reconhecimento Visual"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/imagem",
+ "inputCharts": "${{amount}}/M caracteres",
+ "inputMinutes": "${{amount}}/minuto",
+ "inputTokens": "Entrada ${{amount}}/M",
+ "outputTokens": "Saída ${{amount}}/M"
+ },
+ "releasedAt": "Lançado em {{releasedAt}}"
},
- "unlock": {
- "customRegion": "Região de serviço personalizada",
- "description": "Insira sua AccessKey / SecretKey para iniciar a sessão. O aplicativo não registrará suas configurações de autenticação",
- "title": "Usar informações de autenticação personalizadas do Wenxin Yiyan"
- }
+ "list": {
+ "addNew": "Adicionar Modelo",
+ "disabled": "Não habilitado",
+ "disabledActions": {
+ "showMore": "Mostrar tudo"
+ },
+ "empty": {
+ "desc": "Por favor, crie um modelo personalizado ou importe um modelo para começar a usar.",
+ "title": "Nenhum modelo disponível"
+ },
+ "enabled": "Habilitado",
+ "enabledActions": {
+ "disableAll": "Desabilitar todos",
+ "enableAll": "Habilitar todos",
+ "sort": "Ordenar modelos personalizados"
+ },
+ "enabledEmpty": "Nenhum modelo habilitado no momento, por favor habilite os modelos desejados na lista abaixo~",
+ "fetcher": {
+ "clear": "Limpar modelos obtidos",
+ "fetch": "Obter lista de modelos",
+ "fetching": "Obtendo lista de modelos...",
+ "latestTime": "Última atualização: {{time}}",
+ "noLatestTime": "Lista ainda não obtida"
+ },
+ "resetAll": {
+ "conform": "Você tem certeza de que deseja redefinir todas as modificações do modelo atual? Após a redefinição, a lista de modelos atuais voltará ao estado padrão",
+ "success": "Redefinição bem-sucedida",
+ "title": "Redefinir todas as modificações"
+ },
+ "search": "Pesquisar modelos...",
+ "searchResult": "Encontrados {{count}} modelos",
+ "title": "Lista de Modelos",
+ "total": "Um total de {{count}} modelos disponíveis"
+ },
+ "searchNotFound": "Nenhum resultado encontrado"
+ },
+ "sortModal": {
+ "success": "Ordenação atualizada com sucesso",
+ "title": "Ordenação Personalizada",
+ "update": "Atualizar"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "Você está prestes a excluir este provedor de AI, após a exclusão não poderá ser recuperado, deseja confirmar a exclusão?",
+ "deleteSuccess": "Exclusão bem-sucedida",
+ "tooltip": "Atualizar configurações básicas do provedor",
+ "updateSuccess": "Atualização bem-sucedida"
},
"zeroone": {
"title": "01.AI Zero e Um"
diff --git a/locales/pt-BR/models.json b/locales/pt-BR/models.json
index 5c1d72b355cac..cd059712aa4f7 100644
--- a/locales/pt-BR/models.json
+++ b/locales/pt-BR/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K enfatiza segurança semântica e responsabilidade, projetado especificamente para cenários de aplicação com altas exigências de segurança de conteúdo, garantindo precisão e robustez na experiência do usuário."
},
+ "360gpt2-o1": {
+ "description": "O 360gpt2-o1 utiliza busca em árvore para construir cadeias de pensamento e introduz um mecanismo de reflexão, sendo treinado com aprendizado por reforço, o modelo possui a capacidade de auto-reflexão e correção de erros."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro é um modelo avançado de processamento de linguagem natural lançado pela 360, com excelente capacidade de geração e compreensão de texto, destacando-se especialmente na geração e criação de conteúdo, capaz de lidar com tarefas complexas de conversão de linguagem e interpretação de papéis."
},
+ "360zhinao2-o1": {
+ "description": "O 360zhinao2-o1 utiliza busca em árvore para construir cadeias de pensamento e introduz um mecanismo de reflexão, utilizando aprendizado por reforço para treinar, permitindo que o modelo tenha a capacidade de auto-reflexão e correção de erros."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra é a versão mais poderosa da série de grandes modelos Xinghuo, que, ao atualizar a conexão de busca online, melhora a capacidade de compreensão e resumo de conteúdo textual. É uma solução abrangente para aumentar a produtividade no trabalho e responder com precisão às demandas, sendo um produto inteligente líder na indústria."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "Modelo com a melhor capacidade do país, superando modelos estrangeiros em tarefas em chinês como enciclopédia, textos longos e criação de conteúdo. Também possui capacidades multimodais líderes da indústria, com excelente desempenho em várias avaliações de referência."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Modelo de destilação DeepSeek-R1 baseado no Qwen2.5-Math-1.5B, otimizado para desempenho de inferência através de aprendizado por reforço e dados de inicialização fria, modelo de código aberto que redefine os padrões de múltiplas tarefas."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Modelo de destilação DeepSeek-R1 baseado no Qwen2.5-14B, otimizado para desempenho de inferência através de aprendizado por reforço e dados de inicialização fria, modelo de código aberto que redefine os padrões de múltiplas tarefas."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "A série DeepSeek-R1 otimiza o desempenho de inferência através de aprendizado por reforço e dados de inicialização fria, modelo de código aberto que redefine os padrões de múltiplas tarefas, superando o nível do OpenAI-o1-mini."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "Modelo de destilação DeepSeek-R1 baseado no Qwen2.5-Math-7B, otimizado para desempenho de inferência através de aprendizado por reforço e dados de inicialização fria, modelo de código aberto que redefine os padrões de múltiplas tarefas."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite possui uma velocidade de resposta excepcional e uma melhor relação custo-benefício, oferecendo opções mais flexíveis para diferentes cenários dos clientes. Suporta raciocínio e ajuste fino em janelas de contexto de 128k."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Modelo de linguagem ultra grande escala desenvolvido pela Baidu, que em comparação com o ERNIE 3.5, apresenta uma atualização completa nas capacidades do modelo, amplamente aplicável em cenários de tarefas complexas em diversas áreas; suporta integração automática com o plugin de busca da Baidu, garantindo a atualidade das informações de perguntas e respostas."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Modelo de linguagem de grande escala desenvolvido pela Baidu, com desempenho excepcional em uma ampla gama de cenários de tarefas complexas; suporta integração automática com plugins de busca da Baidu, garantindo a atualidade das informações de perguntas e respostas. Em comparação com o ERNIE 4.0, apresenta desempenho superior."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "Modelo de linguagem de última geração desenvolvido pela Baidu, com desempenho excepcional em uma ampla gama de cenários de tarefas complexas; suporta integração automática com plugins de busca da Baidu, garantindo a relevância da informação nas respostas. Supera o desempenho do ERNIE 4.0."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct é um dos mais recentes modelos de linguagem de grande escala lançados pela Alibaba Cloud. Este modelo de 7B apresenta melhorias significativas em áreas como codificação e matemática. O modelo também oferece suporte multilíngue, abrangendo mais de 29 idiomas, incluindo chinês e inglês. O modelo teve melhorias significativas em seguir instruções, entender dados estruturados e gerar saídas estruturadas (especialmente JSON)."
},
+ "MiniMax-Text-01": {
+ "description": "Na série de modelos MiniMax-01, fizemos inovações ousadas: pela primeira vez, implementamos em larga escala um mecanismo de atenção linear, tornando a arquitetura Transformer tradicional não mais a única opção. Este modelo possui um total de 456 bilhões de parâmetros, com 45,9 bilhões ativados em uma única vez. O desempenho geral do modelo é comparável aos melhores modelos internacionais, enquanto lida eficientemente com contextos de até 4 milhões de tokens, 32 vezes mais que o GPT-4o e 20 vezes mais que o Claude-3.5-Sonnet."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO é uma fusão de múltiplos modelos altamente flexível, projetada para oferecer uma experiência criativa excepcional."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 demonstrou desempenho excepcional em diversas tarefas de linguagem visual, incluindo compreensão de documentos e gráficos, compreensão de texto em cena, OCR, e resolução de problemas científicos e matemáticos."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 demonstrou desempenho excepcional em diversas tarefas de linguagem visual, incluindo compreensão de documentos e gráficos, compreensão de texto em cena, OCR, e resolução de problemas científicos e matemáticos."
- },
"Phi-3-medium-128k-instruct": {
"description": "Mesmo modelo Phi-3-medium, mas com um tamanho de contexto maior para RAG ou prompting de poucos exemplos."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat é a versão de código aberto da série de modelos pré-treinados GLM-4 lançada pela Zhipu AI. Este modelo se destaca em semântica, matemática, raciocínio, código e conhecimento. Além de suportar diálogos de múltiplas rodadas, o GLM-4-9B-Chat também possui recursos avançados como navegação na web, execução de código, chamadas de ferramentas personalizadas (Function Call) e raciocínio de longo texto. O modelo suporta 26 idiomas, incluindo chinês, inglês, japonês, coreano e alemão. Em vários benchmarks, o GLM-4-9B-Chat demonstrou desempenho excepcional, como AlignBench-v2, MT-Bench, MMLU e C-Eval. O modelo suporta um comprimento de contexto máximo de 128K, adequado para pesquisa acadêmica e aplicações comerciais."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 é um modelo de inferência impulsionado por aprendizado por reforço (RL), que resolve problemas de repetitividade e legibilidade no modelo. Antes do RL, o DeepSeek-R1 introduziu dados de inicialização a frio, otimizando ainda mais o desempenho de inferência. Ele se compara ao OpenAI-o1 em tarefas matemáticas, de código e de inferência, e melhora o desempenho geral por meio de métodos de treinamento cuidadosamente projetados."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 é um modelo de linguagem com 671 bilhões de parâmetros, utilizando uma arquitetura de especialistas mistos (MoE) com atenção potencial de múltiplas cabeças (MLA) e uma estratégia de balanceamento de carga sem perda auxiliar, otimizando a eficiência de inferência e treinamento. Pré-treinado em 14,8 trilhões de tokens de alta qualidade, e ajustado por supervisão e aprendizado por reforço, o DeepSeek-V3 supera outros modelos de código aberto, aproximando-se de modelos fechados líderes."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma é uma das séries de modelos abertos mais avançadas e leves desenvolvidas pelo Google. É um modelo de linguagem em larga escala apenas de decodificação, que suporta inglês, oferecendo pesos abertos, variantes pré-treinadas e variantes de ajuste fino para instruções. O modelo Gemma é adequado para várias tarefas de geração de texto, incluindo perguntas e respostas, resumos e raciocínio. Este modelo de 9B foi treinado com 80 trilhões de tokens. Seu tamanho relativamente pequeno permite que seja implantado em ambientes com recursos limitados, como laptops, desktops ou sua própria infraestrutura em nuvem, permitindo que mais pessoas acessem modelos de IA de ponta e promovam inovações."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 é uma família de modelos de linguagem em larga escala multilíngue desenvolvida pela Meta, incluindo variantes pré-treinadas e de ajuste fino para instruções com tamanhos de parâmetros de 8B, 70B e 405B. Este modelo de 8B foi otimizado para cenários de diálogo multilíngue e se destacou em vários benchmarks da indústria. O treinamento do modelo utilizou mais de 150 trilhões de tokens de dados públicos e empregou técnicas como ajuste fino supervisionado e aprendizado por reforço com feedback humano para melhorar a utilidade e segurança do modelo. Llama 3.1 suporta geração de texto e geração de código, com data de corte de conhecimento em dezembro de 2023."
},
+ "QwQ-32B-Preview": {
+ "description": "O QwQ-32B-Preview é um modelo de processamento de linguagem natural inovador, capaz de lidar eficientemente com tarefas complexas de geração de diálogos e compreensão de contexto."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview é um modelo de pesquisa desenvolvido pela equipe Qwen, focado em capacidades de raciocínio visual, apresentando vantagens únicas na compreensão de cenários complexos e na resolução de problemas matemáticos relacionados à visão."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview é o mais recente modelo de pesquisa experimental da Qwen, focado em melhorar a capacidade de raciocínio da IA. Ao explorar mecanismos complexos como mistura de linguagem e raciocínio recursivo, suas principais vantagens incluem forte capacidade de análise de raciocínio, habilidades matemáticas e de programação. Ao mesmo tempo, existem questões de troca de linguagem, ciclos de raciocínio, considerações de segurança e diferenças em outras capacidades."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct é a versão mais recente da série de modelos de linguagem de grande escala específicos para código lançada pela Alibaba Cloud. Este modelo, baseado no Qwen2.5, foi treinado com 55 trilhões de tokens, melhorando significativamente a capacidade de geração, raciocínio e correção de código. Ele não apenas aprimora a capacidade de codificação, mas também mantém as vantagens em matemática e habilidades gerais. O modelo fornece uma base mais abrangente para aplicações práticas, como agentes de código."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math foca na resolução de problemas na área de matemática, oferecendo respostas especializadas para questões de alta dificuldade."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 é a mais recente série do modelo Qwen, suportando 128k de contexto. Em comparação com os melhores modelos de código aberto atuais, o Qwen2-72B supera significativamente os modelos líderes em várias capacidades, incluindo compreensão de linguagem natural, conhecimento, código, matemática e multilinguismo."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 é a mais recente série do modelo Qwen, capaz de superar modelos de código aberto de tamanho equivalente e até mesmo modelos de maior escala. O Qwen2 7B obteve vantagens significativas em várias avaliações, especialmente em compreensão de código e chinês."
},
+ "Qwen2-VL-72B": {
+ "description": "O Qwen2-VL-72B é um poderoso modelo de linguagem visual, que suporta processamento multimodal de imagens e texto, capaz de reconhecer com precisão o conteúdo das imagens e gerar descrições ou respostas relacionadas."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct é um grande modelo de linguagem com 14 bilhões de parâmetros, com desempenho excelente, otimizado para cenários em chinês e multilíngues, suportando aplicações como perguntas e respostas inteligentes e geração de conteúdo."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct é um grande modelo de linguagem com 7 bilhões de parâmetros, que suporta chamadas de função e interação sem costura com sistemas externos, aumentando significativamente a flexibilidade e escalabilidade. Otimizado para cenários em chinês e multilíngues, suporta aplicações como perguntas e respostas inteligentes e geração de conteúdo."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "O Qwen2.5-Coder-14B-Instruct é um modelo de instrução de programação baseado em pré-treinamento em larga escala, com forte capacidade de compreensão e geração de código, capaz de lidar eficientemente com diversas tarefas de programação, especialmente adequado para escrita inteligente de código, geração de scripts automatizados e resolução de problemas de programação."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct é um grande modelo de linguagem projetado para geração de código, compreensão de código e cenários de desenvolvimento eficiente, com uma escala de 32 bilhões de parâmetros, atendendo a diversas necessidades de programação."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "O modelo TeleMM é um modelo de compreensão multimodal desenvolvido de forma independente pela China Telecom, capaz de lidar com entradas de múltiplas modalidades, como texto e imagem, suportando funções como compreensão de imagem e análise de gráficos, oferecendo serviços de compreensão multimodal aos usuários. O modelo pode interagir com os usuários de forma multimodal, compreendendo com precisão o conteúdo de entrada, respondendo perguntas, auxiliando na criação e fornecendo informações e suporte de inspiração multimodal de forma eficiente. O modelo se destaca em tarefas multimodais, como percepção de granularidade fina e raciocínio lógico."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large é o maior modelo MoE de arquitetura Transformer open source da indústria, com um total de 389 bilhões de parâmetros e 52 bilhões de parâmetros ativados."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct é um modelo de linguagem de grande escala com ajuste fino para instruções na série Qwen2, com um tamanho de parâmetro de 72B. Este modelo é baseado na arquitetura Transformer, utilizando funções de ativação SwiGLU, viés de atenção QKV e atenção de consulta em grupo. Ele é capaz de lidar com entradas em larga escala. O modelo se destaca em compreensão de linguagem, geração, capacidade multilíngue, codificação, matemática e raciocínio em vários benchmarks, superando a maioria dos modelos de código aberto e demonstrando competitividade comparável a modelos proprietários em algumas tarefas."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct é um dos mais recentes modelos de linguagem de grande escala lançados pela Alibaba Cloud. Este modelo de 72B apresenta melhorias significativas em áreas como codificação e matemática. O modelo também oferece suporte multilíngue, abrangendo mais de 29 idiomas, incluindo chinês e inglês. O modelo teve melhorias significativas em seguir instruções, entender dados estruturados e gerar saídas estruturadas (especialmente JSON)."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Otimizado para cenários de diálogo de personagens em chinês, oferecendo capacidade de geração de diálogos fluentes e que respeitam os hábitos de expressão em chinês."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Modelo de chamada de função de código aberto da Fireworks, oferecendo excelente capacidade de execução de instruções e características personalizáveis."
+ "abab7-chat-preview": {
+ "description": "Em comparação com a série de modelos abab6.5, houve uma melhoria significativa nas habilidades em textos longos, matemática, escrita, entre outros."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "O Firefunction-v2 da Fireworks é um modelo de chamada de função de alto desempenho, desenvolvido com base no Llama-3 e otimizado para cenários como chamadas de função, diálogos e seguimento de instruções."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 é um modelo de linguagem grande de última geração, otimizado com aprendizado por reforço e dados de inicialização a frio, apresentando desempenho excepcional em raciocínio, matemática e programação."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b é um modelo de linguagem visual que pode receber entradas de imagem e texto simultaneamente, treinado com dados de alta qualidade, adequado para tarefas multimodais."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Modelo de linguagem poderoso da Deepseek, baseado em Mixture-of-Experts (MoE), com um total de 671B de parâmetros, ativando 37B de parâmetros por token."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "O modelo Llama 3 70B Instruct é otimizado para diálogos multilíngues e compreensão de linguagem natural, superando a maioria dos modelos concorrentes."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "O modelo Llama 3 70B Instruct (versão HF) mantém consistência com os resultados da implementação oficial, adequado para tarefas de seguimento de instruções de alta qualidade."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "O modelo Llama 3 8B Instruct é otimizado para diálogos e tarefas multilíngues, apresentando desempenho excepcional e eficiência."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Modelo de raciocínio visual de 11B parâmetros da Meta, otimizado para reconhecimento visual, raciocínio visual, descrição de imagens e resposta a perguntas gerais sobre imagens. Este modelo é capaz de entender dados visuais, como gráficos e diagramas, e preencher a lacuna entre visão e linguagem gerando descrições textuais dos detalhes das imagens."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "O modelo de instrução Llama 3.2 1B é um modelo multilíngue leve lançado pela Meta. Este modelo visa aumentar a eficiência, oferecendo melhorias significativas em latência e custo em comparação com modelos maiores. Exemplos de uso incluem recuperação e resumo."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "O modelo de instrução Llama 3.2 3B é um modelo multilíngue leve lançado pela Meta. Este modelo visa aumentar a eficiência, oferecendo melhorias significativas em latência e custo em comparação com modelos maiores. Exemplos de uso incluem consultas, reescrita de prompts e auxílio na redação."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Modelo de raciocínio visual de 90B parâmetros da Meta, otimizado para reconhecimento visual, raciocínio visual, descrição de imagens e resposta a perguntas gerais sobre imagens. Este modelo é capaz de entender dados visuais, como gráficos e diagramas, e preencher a lacuna entre visão e linguagem gerando descrições textuais dos detalhes das imagens."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct é a versão atualizada de dezembro do Llama 3.1 70B. Este modelo foi aprimorado com base no Llama 3.1 70B (lançado em julho de 2024), melhorando a chamada de ferramentas, suporte a textos multilíngues, habilidades matemáticas e de programação. O modelo alcançou níveis de liderança da indústria em raciocínio, matemática e seguimento de instruções, e é capaz de oferecer desempenho semelhante ao 3.1 405B, ao mesmo tempo em que apresenta vantagens significativas em velocidade e custo."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "Modelo com 24B de parâmetros, com capacidades de ponta comparáveis a modelos maiores."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "O modelo Mixtral MoE 8x22B Instruct, com parâmetros em grande escala e arquitetura de múltiplos especialistas, suporta o processamento eficiente de tarefas complexas."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "O modelo Mixtral MoE 8x7B Instruct, com uma arquitetura de múltiplos especialistas, oferece seguimento e execução de instruções de forma eficiente."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "O modelo Mixtral MoE 8x7B Instruct (versão HF) apresenta desempenho consistente com a implementação oficial, adequado para uma variedade de cenários de tarefas eficientes."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "O modelo MythoMax L2 13B combina novas técnicas de fusão, sendo especializado em narrativas e interpretação de personagens."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "O modelo QwQ é um modelo de pesquisa experimental desenvolvido pela equipe Qwen, focado em aprimorar a capacidade de raciocínio da IA."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "A versão 72B do modelo Qwen-VL é o resultado da mais recente iteração da Alibaba, representando quase um ano de inovações."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 é uma série de modelos de linguagem com apenas decodificadores, desenvolvida pela equipe Qwen da Alibaba Cloud. Estes modelos têm tamanhos variados, incluindo 0.5B, 1.5B, 3B, 7B, 14B, 32B e 72B, com variantes base (base) e de instrução (instruct)."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct é a versão mais recente da série de modelos de linguagem de grande escala específicos para código lançada pela Alibaba Cloud. Este modelo, baseado no Qwen2.5, foi treinado com 55 trilhões de tokens, melhorando significativamente a capacidade de geração, raciocínio e correção de código. Ele não apenas aprimora a capacidade de codificação, mas também mantém as vantagens em matemática e habilidades gerais. O modelo fornece uma base mais abrangente para aplicações práticas, como agentes de código."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "O modelo StarCoder 15.5B suporta tarefas de programação avançadas, com capacidade multilíngue aprimorada, adequado para geração e compreensão de código complexos."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "O modelo StarCoder 7B é treinado para mais de 80 linguagens de programação, apresentando excelente capacidade de preenchimento de código e compreensão de contexto."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "O modelo Yi-Large oferece excelente capacidade de processamento multilíngue, adequado para diversas tarefas de geração e compreensão de linguagem."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus é o modelo mais poderoso da Anthropic para lidar com tarefas altamente complexas. Ele se destaca em desempenho, inteligência, fluência e compreensão."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku é o modelo de próxima geração mais rápido da Anthropic. Em comparação com Claude 3 Haiku, Claude 3.5 Haiku apresenta melhorias em várias habilidades e supera o maior modelo da geração anterior, Claude 3 Opus, em muitos testes de inteligência."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet oferece capacidades que vão além do Opus e uma velocidade superior ao Sonnet, mantendo o mesmo preço do Sonnet. O Sonnet é especialmente habilidoso em programação, ciência de dados, processamento visual e tarefas de agente."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet oferece um equilíbrio ideal entre inteligência e velocidade para cargas de trabalho empresariais. Ele fornece máxima utilidade a um custo mais baixo, sendo confiável e adequado para implantação em larga escala."
},
- "code-raccoon-v1": {
- "description": "O Código Raccoon é um assistente de desenvolvimento inteligente baseado no grande modelo de linguagem da SenseTime, cobrindo análise de requisitos de software, design de arquitetura, escrita de código, testes de software e outros aspectos, atendendo a diversas necessidades de escrita de código e aprendizado de programação. O Código Raccoon suporta mais de 90 linguagens de programação populares, como Python, Java, JavaScript, C++, Go, SQL, e IDEs populares como VS Code e IntelliJ IDEA. Na prática, o Código Raccoon pode ajudar os desenvolvedores a aumentar a eficiência da programação em mais de 50%."
- },
"codegeex-4": {
"description": "O CodeGeeX-4 é um poderoso assistente de programação AI, suportando perguntas e respostas inteligentes e autocompletar em várias linguagens de programação, aumentando a eficiência do desenvolvimento."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ é um modelo de linguagem de grande porte de alto desempenho, projetado para cenários empresariais reais e aplicações complexas."
},
+ "dall-e-2": {
+ "description": "O segundo modelo DALL·E, suporta geração de imagens mais realistas e precisas, com resolução quatro vezes maior que a da primeira geração."
+ },
+ "dall-e-3": {
+ "description": "O mais recente modelo DALL·E, lançado em novembro de 2023. Suporta geração de imagens mais realistas e precisas, com maior capacidade de detalhamento."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct oferece capacidade de processamento de instruções altamente confiável, suportando aplicações em diversos setores."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 é um poderoso e econômico modelo de linguagem de especialistas mistos (MoE). Ele foi pré-treinado em um corpus de alta qualidade de 81 trilhões de tokens e aprimorado por meio de ajuste fino supervisionado (SFT) e aprendizado por reforço (RL). Em comparação com o DeepSeek 67B, o DeepSeek-V2 não só apresenta desempenho superior, mas também economiza 42,5% nos custos de treinamento, reduz 93,3% do cache KV e aumenta a taxa de geração máxima em 5,76 vezes. Este modelo suporta um comprimento de contexto de 128k e se destaca em benchmarks padrão e avaliações de geração aberta."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 é um modelo de inferência impulsionado por aprendizado por reforço (RL), que resolve problemas de repetitividade e legibilidade no modelo. Antes do RL, o DeepSeek-R1 introduziu dados de inicialização a frio, otimizando ainda mais o desempenho da inferência. Ele apresenta desempenho comparável ao OpenAI-o1 em tarefas matemáticas, de código e de inferência, e melhora o resultado geral por meio de métodos de treinamento cuidadosamente projetados."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "Modelo de destilação DeepSeek-R1, otimizado para desempenho de inferência através de aprendizado por reforço e dados de inicialização fria, modelo de código aberto que redefine os padrões de múltiplas tarefas."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B é um modelo de destilação desenvolvido com base no Llama-3.1-8B. Este modelo foi ajustado com amostras geradas pelo DeepSeek-R1, demonstrando excelente capacidade de inferência. Apresentou bom desempenho em vários testes de referência, alcançando uma precisão de 89,1% no MATH-500, uma taxa de aprovação de 50,4% no AIME 2024 e uma pontuação de 1205 no CodeForces, demonstrando forte capacidade matemática e de programação para um modelo de 8B."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Modelo de destilação DeepSeek-R1, otimizado para desempenho de inferência através de aprendizado por reforço e dados de inicialização fria, modelo de código aberto que redefine os padrões de múltiplas tarefas."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Modelo de destilação DeepSeek-R1, otimizado para desempenho de inferência através de aprendizado por reforço e dados de inicialização fria, modelo de código aberto que redefine os padrões de múltiplas tarefas."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B é um modelo obtido através da destilação do Qwen2.5-32B. Este modelo foi ajustado com 800 mil amostras selecionadas geradas pelo DeepSeek-R1, demonstrando desempenho excepcional em várias áreas, como matemática, programação e raciocínio. Obteve resultados notáveis em vários testes de referência, alcançando uma precisão de 94,3% no MATH-500, demonstrando forte capacidade de raciocínio matemático."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B é um modelo obtido através da destilação do Qwen2.5-Math-7B. Este modelo foi ajustado com 800 mil amostras selecionadas geradas pelo DeepSeek-R1, demonstrando excelente capacidade de inferência. Apresentou desempenho notável em vários testes de referência, alcançando uma precisão de 92,8% no MATH-500, uma taxa de aprovação de 55,5% no AIME 2024 e uma pontuação de 1189 no CodeForces, demonstrando forte capacidade matemática e de programação para um modelo de 7B."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 combina as excelentes características das versões anteriores, aprimorando a capacidade geral e de codificação."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 é um modelo de linguagem de especialistas mistos (MoE) com 671 bilhões de parâmetros, utilizando atenção latente de múltiplas cabeças (MLA) e a arquitetura DeepSeekMoE, combinando uma estratégia de balanceamento de carga sem perda auxiliar para otimizar a eficiência de inferência e treinamento. Após ser pré-treinado em 14,8 trilhões de tokens de alta qualidade e passar por ajuste fino supervisionado e aprendizado por reforço, o DeepSeek-V3 supera outros modelos de código aberto em desempenho, aproximando-se de modelos fechados líderes."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B é um modelo avançado treinado para diálogos de alta complexidade."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "LLM avançado e eficiente, especializado em raciocínio, matemática e programação."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 é um modelo de linguagem visual baseado no DeepSeekMoE-27B, desenvolvido como um especialista misto (MoE), utilizando uma arquitetura de MoE com ativação esparsa, alcançando desempenho excepcional com apenas 4,5 bilhões de parâmetros ativados. Este modelo se destaca em várias tarefas, incluindo perguntas visuais, reconhecimento óptico de caracteres, compreensão de documentos/tabelas/gráficos e localização visual."
+ },
"deepseek-chat": {
"description": "Um novo modelo de código aberto que combina capacidades gerais e de codificação, não apenas preservando a capacidade de diálogo geral do modelo Chat original e a poderosa capacidade de processamento de código do modelo Coder, mas também alinhando-se melhor às preferências humanas. Além disso, o DeepSeek-V2.5 também alcançou melhorias significativas em várias áreas, como tarefas de escrita e seguimento de instruções."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 é um modelo de código de especialistas abertos, destacando-se em tarefas de codificação, comparável ao GPT4-Turbo."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 é um modelo de inferência impulsionado por aprendizado por reforço (RL), que resolve problemas de repetitividade e legibilidade no modelo. Antes do RL, o DeepSeek-R1 introduziu dados de inicialização a frio, otimizando ainda mais o desempenho da inferência. Ele apresenta desempenho comparável ao OpenAI-o1 em tarefas matemáticas, de código e de inferência, e melhora o resultado geral por meio de métodos de treinamento cuidadosamente projetados."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 — um modelo maior e mais inteligente dentro do pacote DeepSeek — foi destilado para a arquitetura Llama 70B. Com base em testes de referência e avaliações humanas, este modelo é mais inteligente que o Llama 70B original, destacando-se especialmente em tarefas que exigem precisão matemática e factual."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "O modelo da série DeepSeek-R1-Distill é obtido através da técnica de destilação de conhecimento, ajustando amostras geradas pelo DeepSeek-R1 em modelos de código aberto como Qwen e Llama."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "O modelo da série DeepSeek-R1-Distill é obtido através da técnica de destilação de conhecimento, ajustando amostras geradas pelo DeepSeek-R1 em modelos de código aberto como Qwen e Llama."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "O modelo da série DeepSeek-R1-Distill é obtido através da técnica de destilação de conhecimento, ajustando amostras geradas pelo DeepSeek-R1 em modelos de código aberto como Qwen e Llama."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "O modelo da série DeepSeek-R1-Distill é obtido através da técnica de destilação de conhecimento, ajustando amostras geradas pelo DeepSeek-R1 em modelos de código aberto como Qwen e Llama."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "O modelo da série DeepSeek-R1-Distill é obtido através da técnica de destilação de conhecimento, ajustando amostras geradas pelo DeepSeek-R1 em modelos de código aberto como Qwen e Llama."
+ },
+ "deepseek-reasoner": {
+ "description": "Modelo de raciocínio lançado pela DeepSeek. Antes de fornecer a resposta final, o modelo gera uma cadeia de pensamento para aumentar a precisão da resposta final."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 é um modelo de linguagem eficiente Mixture-of-Experts, adequado para demandas de processamento econômico."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B é o modelo de código projetado do DeepSeek, oferecendo forte capacidade de geração de código."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 é um modelo MoE desenvolvido pela Hangzhou DeepSeek Artificial Intelligence Technology Research Co., Ltd., com desempenho destacado em várias avaliações, ocupando o primeiro lugar entre os modelos de código aberto nas principais listas. Em comparação com o modelo V2.5, a velocidade de geração do V3 foi aumentada em 3 vezes, proporcionando uma experiência de uso mais rápida e fluida."
+ },
"deepseek/deepseek-chat": {
"description": "Um novo modelo de código aberto que integra capacidades gerais e de codificação, não apenas preservando a capacidade de diálogo geral do modelo Chat original e a poderosa capacidade de processamento de código do modelo Coder, mas também alinhando-se melhor às preferências humanas. Além disso, o DeepSeek-V2.5 também alcançou melhorias significativas em várias áreas, como tarefas de escrita e seguimento de instruções."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 melhorou significativamente a capacidade de raciocínio do modelo com muito poucos dados rotulados. Antes de fornecer a resposta final, o modelo gera uma cadeia de pensamento para aumentar a precisão da resposta final."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 melhorou significativamente a capacidade de raciocínio do modelo com muito poucos dados rotulados. Antes de fornecer a resposta final, o modelo gera uma cadeia de pensamento para aumentar a precisão da resposta final."
+ },
"emohaa": {
"description": "O Emohaa é um modelo psicológico com capacidade de consultoria profissional, ajudando os usuários a entender questões emocionais."
},
+ "ernie-3.5-128k": {
+ "description": "Modelo de linguagem de grande escala de nível flagship desenvolvido pela Baidu, cobrindo uma vasta quantidade de dados em chinês e inglês, com forte capacidade geral, capaz de atender à maioria das demandas de diálogo, geração criativa e aplicações de plugins; suporta integração automática com plugins de busca da Baidu, garantindo a atualidade das informações de perguntas e respostas."
+ },
+ "ernie-3.5-8k": {
+ "description": "Modelo de linguagem de grande escala de nível flagship desenvolvido pela Baidu, cobrindo uma vasta quantidade de dados em chinês e inglês, com forte capacidade geral, capaz de atender à maioria das demandas de diálogo, geração criativa e aplicações de plugins; suporta integração automática com plugins de busca da Baidu, garantindo a atualidade das informações de perguntas e respostas."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "Modelo de linguagem de grande escala de nível flagship desenvolvido pela Baidu, cobrindo uma vasta quantidade de dados em chinês e inglês, com forte capacidade geral, capaz de atender à maioria das demandas de diálogo, geração criativa e aplicações de plugins; suporta integração automática com plugins de busca da Baidu, garantindo a atualidade das informações de perguntas e respostas."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "Modelo de linguagem de grande escala de nível flagship desenvolvido pela Baidu, com capacidade de modelo amplamente aprimorada em comparação com o ERNIE 3.5, amplamente aplicável a cenários de tarefas complexas em várias áreas; suporta integração automática com plugins de busca da Baidu, garantindo a atualidade das informações de perguntas e respostas."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "Modelo de linguagem de grande escala de nível flagship desenvolvido pela Baidu, com capacidade de modelo amplamente aprimorada em comparação com o ERNIE 3.5, amplamente aplicável a cenários de tarefas complexas em várias áreas; suporta integração automática com plugins de busca da Baidu, garantindo a atualidade das informações de perguntas e respostas."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "Modelo de linguagem de grande escala de nível flagship desenvolvido pela Baidu, com desempenho geral excepcional, amplamente aplicável a cenários de tarefas complexas em várias áreas; suporta integração automática com plugins de busca da Baidu, garantindo a atualidade das informações de perguntas e respostas. Em comparação com o ERNIE 4.0, apresenta desempenho superior."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "Modelo de linguagem de grande escala de nível flagship desenvolvido pela Baidu, com desempenho geral excepcional, amplamente aplicável a cenários de tarefas complexas em várias áreas; suporta integração automática com plugins de busca da Baidu, garantindo a atualidade das informações de perguntas e respostas. Em comparação com o ERNIE 4.0, apresenta desempenho superior."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "Modelo de linguagem de grande escala de nível flagship desenvolvido pela Baidu, com desempenho geral excepcional, amplamente aplicável a cenários de tarefas complexas em várias áreas; suporta integração automática com plugins de busca da Baidu, garantindo a atualidade das informações de perguntas e respostas. Em comparação com o ERNIE 4.0, apresenta desempenho superior."
+ },
+ "ernie-char-8k": {
+ "description": "Modelo de linguagem de grande escala vertical desenvolvido pela Baidu, adequado para aplicações como NPCs de jogos, diálogos de atendimento ao cliente e interpretação de personagens, com estilo de personagem mais distinto e consistente, capacidade de seguir instruções mais forte e desempenho de inferência superior."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Modelo de linguagem de grande escala vertical desenvolvido pela Baidu, adequado para aplicações como NPCs de jogos, diálogos de atendimento ao cliente e interpretação de personagens, com estilo de personagem mais distinto e consistente, capacidade de seguir instruções mais forte e desempenho de inferência superior."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite é um modelo de linguagem de grande escala leve desenvolvido pela Baidu, equilibrando excelente desempenho do modelo e eficiência de inferência, adequado para uso em placas de aceleração de IA de baixa potência."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Modelo de linguagem de grande escala leve desenvolvido pela Baidu, equilibrando excelente desempenho do modelo e eficiência de inferência, com desempenho superior ao ERNIE Lite, adequado para uso em placas de aceleração de IA de baixa potência."
+ },
+ "ernie-novel-8k": {
+ "description": "Modelo de linguagem de grande escala geral desenvolvido pela Baidu, com vantagens notáveis na capacidade de continuar histórias, também aplicável em cenários como peças curtas e filmes."
+ },
+ "ernie-speed-128k": {
+ "description": "Modelo de linguagem de alto desempenho desenvolvido pela Baidu, lançado em 2024, com excelente capacidade geral, adequado para ser usado como modelo base para ajuste fino, lidando melhor com problemas de cenários específicos, enquanto apresenta excelente desempenho de inferência."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "Modelo de linguagem de alto desempenho desenvolvido pela Baidu, lançado em 2024, com excelente capacidade geral, desempenho superior ao ERNIE Speed, adequado para ser usado como modelo base para ajuste fino, lidando melhor com problemas de cenários específicos, enquanto apresenta excelente desempenho de inferência."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny é um modelo de linguagem de grande escala de alto desempenho desenvolvido pela Baidu, com os menores custos de implantação e ajuste entre os modelos da série Wenxin."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Ajuste) oferece desempenho estável e ajustável, sendo a escolha ideal para soluções de tarefas complexas."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro suporta até 2 milhões de tokens, sendo a escolha ideal para modelos multimodais de médio porte, adequados para suporte multifacetado em tarefas complexas."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash oferece funcionalidades e melhorias de próxima geração, incluindo velocidade excepcional, uso nativo de ferramentas, geração multimodal e uma janela de contexto de 1M tokens."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash oferece funcionalidades e melhorias de próxima geração, incluindo velocidade excepcional, uso nativo de ferramentas, geração multimodal e uma janela de contexto de 1M tokens."
+ },
"gemini-2.0-flash-exp": {
"description": "O Gemini 2.0 Flash Exp é o mais recente modelo de IA multiodal experimental do Google, com recursos de próxima geração, velocidade extraordinária, chamadas nativas de ferramentas e geração multimodal."
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "Um modelo Gemini 2.0 Flash otimizado para custo-benefício e baixa latência."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "O Gemini 2.0 Flash Exp é o mais recente modelo experimental de IA multimodal do Google, com características de próxima geração, velocidade excepcional, chamadas nativas de ferramentas e geração multimodal."
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Exp é o mais recente modelo de IA multimodal experimental do Google, com recursos de próxima geração, velocidade excepcional, chamadas nativas de ferramentas e geração multimodal."
},
- "gemini-exp-1114": {
- "description": "O Gemini Exp 1114 é o mais recente modelo experimental de IA multimodal da Google, com capacidade de processamento rápido, suportando entradas de texto, imagem e vídeo, adequado para a expansão eficiente de várias tarefas."
- },
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 é o mais recente modelo experimental de IA multimodal do Google, com capacidade de processamento rápido, suportando entradas de texto, imagem e vídeo, adequado para a escalabilidade eficiente em várias tarefas."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental é o mais recente modelo de IA multimodal experimental do Google, apresentando melhorias de qualidade em comparação com versões anteriores, especialmente em conhecimento mundial, código e contextos longos."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 é o mais recente modelo experimental multimodal da Google, apresentando uma melhoria na qualidade em comparação com versões anteriores."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "O GLM-4V-Plus possui a capacidade de entender conteúdo de vídeo e múltiplas imagens, adequado para tarefas multimodais."
},
+ "glm-zero-preview": {
+ "description": "O GLM-Zero-Preview possui uma poderosa capacidade de raciocínio complexo, destacando-se em áreas como raciocínio lógico, matemática e programação."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash oferece funcionalidades e melhorias de próxima geração, incluindo velocidade excepcional, uso nativo de ferramentas, geração multimodal e uma janela de contexto de 1M tokens."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental é o mais recente modelo de IA multimodal experimental do Google, apresentando melhorias de qualidade em comparação com versões anteriores, especialmente em conhecimento mundial, código e contextos longos."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash oferece capacidades de processamento multimodal otimizadas, adequadas para uma variedade de cenários de tarefas complexas."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro combina as mais recentes tecnologias de otimização, proporcionando uma capacidade de processamento de dados multimodais mais eficiente."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 é um modelo eficiente lançado pelo Google, abrangendo uma variedade de cenários de aplicação, desde pequenos aplicativos até processamento de dados complexos."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 continua a filosofia de design leve e eficiente."
},
"google/gemma-2-2b-it": {
"description": "Modelo leve de ajuste de instruções do Google."
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 é um modelo eficiente lançado pelo Google, abrangendo uma variedade de cenários de aplicação, desde pequenos aplicativos até processamento de dados complexos."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 é uma série de modelos de texto de código aberto leve da Google."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "O GPT 3.5 Turbo é adequado para uma variedade de tarefas de geração e compreensão de texto, atualmente apontando para gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, um modelo de geração de texto de alta capacidade, adequado para tarefas complexas."
+ },
"gpt-3.5-turbo-instruct": {
"description": "O GPT 3.5 Turbo é adequado para uma variedade de tarefas de geração e compreensão de texto, atualmente apontando para gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o é um modelo dinâmico, atualizado em tempo real para manter a versão mais atualizada. Combina uma poderosa compreensão e capacidade de geração de linguagem, adequado para cenários de aplicação em larga escala, incluindo atendimento ao cliente, educação e suporte técnico."
},
+ "gpt-4o-audio-preview": {
+ "description": "Modelo de áudio GPT-4o, suporta entrada e saída de áudio."
+ },
"gpt-4o-mini": {
"description": "O GPT-4o mini é o mais recente modelo lançado pela OpenAI após o GPT-4 Omni, suportando entrada de texto e imagem e gerando texto como saída. Como seu modelo compacto mais avançado, ele é muito mais acessível do que outros modelos de ponta recentes, custando mais de 60% menos que o GPT-3.5 Turbo. Ele mantém uma inteligência de ponta, ao mesmo tempo que oferece um custo-benefício significativo. O GPT-4o mini obteve uma pontuação de 82% no teste MMLU e atualmente está classificado acima do GPT-4 em preferências de chat."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "Versão em tempo real do GPT-4o-mini, suporta entrada e saída de áudio e texto em tempo real."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "Versão em tempo real do GPT-4o, suporta entrada e saída de áudio e texto em tempo real."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "Versão em tempo real do GPT-4o, suporta entrada e saída de áudio e texto em tempo real."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "Versão em tempo real do GPT-4o, suporta entrada e saída de áudio e texto em tempo real."
+ },
"grok-2-1212": {
"description": "Este modelo apresenta melhorias em precisão, conformidade com instruções e capacidade multilíngue."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "O mais recente modelo FunctionCall da arquitetura MOE Hunyuan, treinado com dados de alta qualidade de FunctionCall, com uma janela de contexto de 32K, liderando em várias métricas de avaliação."
},
+ "hunyuan-large": {
+ "description": "O modelo Hunyuan-large possui um total de aproximadamente 389B de parâmetros, com cerca de 52B de parâmetros ativados, sendo o modelo MoE de código aberto com a maior escala de parâmetros e melhor desempenho na arquitetura Transformer atualmente disponível no mercado."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Especializado em tarefas de texto longo, como resumo de documentos e perguntas e respostas de documentos, também possui a capacidade de lidar com tarefas gerais de geração de texto. Apresenta desempenho excepcional na análise e geração de textos longos, conseguindo atender efetivamente às demandas complexas e detalhadas de processamento de conteúdo longo."
+ },
"hunyuan-lite": {
"description": "Atualizado para uma estrutura MOE, com uma janela de contexto de 256k, liderando em várias avaliações em NLP, código, matemática e setores diversos em comparação com muitos modelos de código aberto."
},
+ "hunyuan-lite-vision": {
+ "description": "Modelo multimodal mais recente de 7B da Hunyuan, com janela de contexto de 32K, suporta diálogos multimodais em cenários em chinês e português, reconhecimento de objetos em imagens, compreensão de documentos e tabelas, matemática multimodal, entre outros, superando modelos concorrentes de 7B em várias métricas de avaliação."
+ },
"hunyuan-pro": {
"description": "Modelo de texto longo MOE-32K com trilhões de parâmetros. Alcança níveis de liderança absoluta em vários benchmarks, com capacidades complexas de instrução e raciocínio, habilidades matemáticas complexas, suporte a chamadas de função, otimizado para áreas como tradução multilíngue, finanças, direito e saúde."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Adota uma estratégia de roteamento superior, ao mesmo tempo que mitiga problemas de balanceamento de carga e convergência de especialistas. Em termos de textos longos, o índice de precisão atinge 99,9%. O MOE-256K rompe ainda mais em comprimento e desempenho, expandindo significativamente o comprimento de entrada permitido."
},
+ "hunyuan-standard-vision": {
+ "description": "Modelo multimodal mais recente da Hunyuan, suporta respostas em múltiplas línguas, com habilidades equilibradas em chinês e português."
+ },
"hunyuan-turbo": {
"description": "Versão de pré-visualização do novo modelo de linguagem de próxima geração Hunyuan, utilizando uma nova estrutura de modelo de especialistas mistos (MoE), com eficiência de inferência mais rápida e desempenho superior em comparação ao Hunyuan-Pro."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Versão fixa do hunyuan-turbo de 20 de novembro de 2024, uma versão intermediária entre hunyuan-turbo e hunyuan-turbo-latest."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "Esta versão otimiza: escalonamento de instruções de dados, aumentando significativamente a capacidade de generalização do modelo; melhoria substancial nas habilidades matemáticas, de codificação e de raciocínio lógico; otimização das capacidades de compreensão de texto e palavras; melhoria na qualidade da geração de conteúdo de criação de texto."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "Otimização da experiência geral, incluindo compreensão de NLP, criação de texto, conversas informais, perguntas e respostas de conhecimento, tradução, entre outros; aumento da humanização, otimização da inteligência emocional do modelo; melhoria na capacidade do modelo de esclarecer ativamente em casos de intenção ambígua; aprimoramento na capacidade de lidar com questões de análise de palavras; melhoria na qualidade e interatividade da criação; aprimoramento da experiência em múltiplas interações."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "Novo modelo de linguagem visual de próxima geração da Hunyuan, adotando uma nova estrutura de modelo de especialistas mistos (MoE), com melhorias abrangentes em relação ao modelo anterior nas capacidades de reconhecimento básico, criação de conteúdo, perguntas e respostas de conhecimento, e análise e raciocínio relacionados à compreensão de texto e imagem."
+ },
"hunyuan-vision": {
"description": "O mais recente modelo multimodal Hunyuan, que suporta a entrada de imagens e texto para gerar conteúdo textual."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "Nossa mais recente série de modelos, com desempenho de raciocínio excepcional, suportando um comprimento de contexto de 1M e capacidades aprimoradas de seguimento de instruções e chamadas de ferramentas."
},
+ "internlm3-latest": {
+ "description": "Nossa mais recente série de modelos, com desempenho de inferência excepcional, liderando entre modelos de código aberto de mesma escala. Aponta por padrão para nossa mais recente série de modelos InternLM3."
+ },
+ "jina-deepsearch-v1": {
+ "description": "A busca profunda combina pesquisa na web, leitura e raciocínio para realizar investigações abrangentes. Você pode vê-la como um agente que aceita suas tarefas de pesquisa - ela realizará uma busca extensa e passará por várias iterações antes de fornecer uma resposta. Esse processo envolve pesquisa contínua, raciocínio e resolução de problemas sob diferentes ângulos. Isso é fundamentalmente diferente de gerar respostas diretamente a partir de dados pré-treinados de grandes modelos padrão e de sistemas RAG tradicionais que dependem de buscas superficiais únicas."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM é um modelo de linguagem experimental e específico para tarefas, treinado para atender aos princípios da ciência da aprendizagem, podendo seguir instruções sistemáticas em cenários de ensino e aprendizagem, atuando como um mentor especialista, entre outros."
},
"lite": {
"description": "Spark Lite é um modelo de linguagem grande leve, com latência extremamente baixa e alta eficiência de processamento, totalmente gratuito e aberto, suportando funcionalidades de busca online em tempo real. Sua característica de resposta rápida o torna excelente para aplicações de inferência em dispositivos de baixo poder computacional e ajuste fino de modelos, proporcionando aos usuários uma excelente relação custo-benefício e experiência inteligente, especialmente em cenários de perguntas e respostas, geração de conteúdo e busca."
},
- "llama-3.1-70b-instruct": {
- "description": "O modelo Llama 3.1 70B Instruct possui 70B de parâmetros, capaz de oferecer desempenho excepcional em tarefas de geração de texto e instrução em larga escala."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B oferece capacidade de raciocínio AI mais poderosa, adequada para aplicações complexas, suportando um processamento computacional extenso e garantindo eficiência e precisão."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B é um modelo de alto desempenho, oferecendo capacidade de geração de texto rápida, ideal para cenários de aplicação que exigem eficiência em larga escala e custo-benefício."
},
- "llama-3.1-8b-instruct": {
- "description": "O modelo Llama 3.1 8B Instruct possui 8B de parâmetros, suportando a execução eficiente de tarefas de instrução, oferecendo excelente capacidade de geração de texto."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "O modelo Llama 3.1 Sonar Huge Online possui 405B de parâmetros, suportando um comprimento de contexto de aproximadamente 127.000 tokens, projetado para aplicações de chat online complexas."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "O modelo Llama 3.1 Sonar Large Chat possui 70B de parâmetros, suportando um comprimento de contexto de aproximadamente 127.000 tokens, adequado para tarefas de chat offline complexas."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "O modelo Llama 3.1 Sonar Large Online possui 70B de parâmetros, suportando um comprimento de contexto de aproximadamente 127.000 tokens, adequado para tarefas de chat de alta capacidade e diversidade."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "O modelo Llama 3.1 Sonar Small Chat possui 8B de parâmetros, projetado para chats offline, suportando um comprimento de contexto de aproximadamente 127.000 tokens."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "O modelo Llama 3.1 Sonar Small Online possui 8B de parâmetros, suportando um comprimento de contexto de aproximadamente 127.000 tokens, projetado para chats online, capaz de processar eficientemente diversas interações textuais."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 é projetado para lidar com tarefas que combinam dados visuais e textuais. Ele se destaca em tarefas como descrição de imagens e perguntas visuais, superando a lacuna entre geração de linguagem e raciocínio visual."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 é o modelo de linguagem de código aberto multilíngue mais avançado da série Llama, oferecendo desempenho comparável ao modelo 405B a um custo extremamente baixo. Baseado na estrutura Transformer, e aprimorado por meio de ajuste fino supervisionado (SFT) e aprendizado por reforço com feedback humano (RLHF) para aumentar a utilidade e a segurança. Sua versão ajustada para instruções é otimizada para diálogos multilíngues, superando muitos modelos de chat de código aberto e fechado em vários benchmarks da indústria. A data limite de conhecimento é dezembro de 2023."
+ },
"llama-3.3-70b-versatile": {
"description": "O modelo de linguagem multilíngue Meta Llama 3.3 (LLM) é um modelo gerador pré-treinado e ajustado para instruções, com 70B (entrada/saída de texto). O modelo de texto puro ajustado para instruções do Llama 3.3 é otimizado para casos de uso de diálogo multilíngue e supera muitos modelos de chat open source e fechados disponíveis em benchmarks comuns da indústria."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 é projetado para lidar com tarefas que combinam dados visuais e textuais. Ele se destaca em tarefas como descrição de imagens e perguntas visuais, superando a lacuna entre geração de linguagem e raciocínio visual."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 é o modelo de linguagem de código aberto multilíngue mais avançado da série Llama, oferecendo uma experiência de desempenho comparável ao modelo de 405B a um custo extremamente baixo. Baseado na estrutura Transformer e aprimorado por meio de ajuste fino supervisionado (SFT) e aprendizado por reforço com feedback humano (RLHF) para aumentar a utilidade e segurança. Sua versão ajustada para instruções é otimizada para diálogos multilíngues, superando muitos modelos de chat de código aberto e fechado em vários benchmarks da indústria. Data limite de conhecimento é dezembro de 2023."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "O Meta Llama 3.3 é um modelo de linguagem de grande escala multilíngue (LLM) com 70B (entrada/saída de texto) que é um modelo gerado por pré-treinamento e ajuste de instruções. O modelo de texto puro ajustado por instruções do Llama 3.3 foi otimizado para casos de uso de diálogo multilíngue e supera muitos modelos de chat de código aberto e fechados disponíveis em benchmarks de indústria comuns."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 é projetado para lidar com tarefas que combinam dados visuais e textuais. Ele se destaca em tarefas como descrição de imagens e perguntas visuais, superando a lacuna entre geração de linguagem e raciocínio visual."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "O modelo Llama 3.1 Turbo 405B oferece suporte a um contexto de capacidade extremamente grande para processamento de grandes volumes de dados, destacando-se em aplicações de inteligência artificial em larga escala."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 é o modelo líder lançado pela Meta, suportando até 405B de parâmetros, aplicável em diálogos complexos, tradução multilíngue e análise de dados."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B oferece suporte a diálogos multilíngues de forma eficiente."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 é projetado para lidar com tarefas que combinam dados visuais e textuais. Ele se destaca em tarefas como descrição de imagens e perguntas visuais, superando a lacuna entre geração de linguagem e raciocínio visual."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 é o modelo de linguagem de código aberto multilíngue mais avançado da série Llama, oferecendo desempenho comparável ao modelo 405B a um custo extremamente baixo. Baseado na estrutura Transformer, e aprimorado por meio de ajuste fino supervisionado (SFT) e aprendizado por reforço com feedback humano (RLHF) para aumentar a utilidade e a segurança. Sua versão ajustada para instruções é otimizada para diálogos multilíngues, superando muitos modelos de chat de código aberto e fechado em vários benchmarks da indústria. A data limite de conhecimento é dezembro de 2023."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 é o modelo de linguagem de código aberto multilíngue mais avançado da série Llama, oferecendo desempenho comparável ao modelo 405B a um custo extremamente baixo. Baseado na estrutura Transformer, e aprimorado por meio de ajuste fino supervisionado (SFT) e aprendizado por reforço com feedback humano (RLHF) para aumentar a utilidade e a segurança. Sua versão ajustada para instruções é otimizada para diálogos multilíngues, superando muitos modelos de chat de código aberto e fechado em vários benchmarks da indústria. A data limite de conhecimento é dezembro de 2023."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct é o maior e mais poderoso modelo da série Llama 3.1 Instruct, sendo um modelo altamente avançado para raciocínio conversacional e geração de dados sintéticos, que também pode ser usado como base para pré-treinamento ou ajuste fino em domínios específicos. Os modelos de linguagem de grande escala (LLMs) multilíngues oferecidos pelo Llama 3.1 são um conjunto de modelos geradores pré-treinados e ajustados por instruções, incluindo tamanhos de 8B, 70B e 405B (entrada/saída de texto). Os modelos de texto ajustados por instruções do Llama 3.1 (8B, 70B, 405B) são otimizados para casos de uso de diálogo multilíngue e superaram muitos modelos de chat de código aberto disponíveis em benchmarks comuns da indústria. O Llama 3.1 é projetado para uso comercial e de pesquisa em várias línguas. Os modelos de texto ajustados por instruções são adequados para chats semelhantes a assistentes, enquanto os modelos pré-treinados podem se adaptar a várias tarefas de geração de linguagem natural. O modelo Llama 3.1 também suporta a utilização de sua saída para melhorar outros modelos, incluindo geração de dados sintéticos e refinamento. O Llama 3.1 é um modelo de linguagem autoregressivo que utiliza uma arquitetura de transformador otimizada. As versões ajustadas utilizam ajuste fino supervisionado (SFT) e aprendizado por reforço com feedback humano (RLHF) para alinhar-se às preferências humanas em relação à utilidade e segurança."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 é um modelo de linguagem de grande escala (LLM) aberto voltado para desenvolvedores, pesquisadores e empresas, projetado para ajudá-los a construir, experimentar e expandir suas ideias de IA geradora de forma responsável. Como parte de um sistema de base para inovação da comunidade global, é ideal para dispositivos de borda com capacidade de computação e recursos limitados, além de tempos de treinamento mais rápidos."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "LLM avançado, suporta geração de dados sintéticos, destilação de conhecimento e raciocínio, adequado para chatbots, programação e tarefas de domínio específico."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "Capacita diálogos complexos, com excelente compreensão de contexto, capacidade de raciocínio e geração de texto."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "Modelo de ponta avançado, com compreensão de linguagem, excelente capacidade de raciocínio e geração de texto."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "Modelo de visão-linguagem de ponta, especializado em raciocínio de alta qualidade a partir de imagens."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "Modelo de linguagem de ponta avançado e compacto, com compreensão de linguagem, excelente capacidade de raciocínio e geração de texto."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "Modelo de linguagem de ponta avançado e compacto, com compreensão de linguagem, excelente capacidade de raciocínio e geração de texto."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "Modelo de visão-linguagem de ponta, especializado em raciocínio de alta qualidade a partir de imagens."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "Modelo LLM avançado, especializado em raciocínio, matemática, conhecimento geral e chamadas de função."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 é um modelo de linguagem fornecido pela Microsoft AI, que se destaca em diálogos complexos, multilíngue, raciocínio e assistentes inteligentes."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K é um modelo com capacidade de processamento de contexto ultra longo, adequado para gerar textos muito longos, atendendo a demandas complexas de geração, capaz de lidar com até 128.000 tokens, ideal para pesquisa, acadêmicos e geração de documentos extensos."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "O modelo visual Kimi (incluindo moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, etc.) é capaz de entender o conteúdo das imagens, incluindo texto, cores e formas dos objetos."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K oferece capacidade de processamento de contexto de comprimento médio, capaz de lidar com 32.768 tokens, especialmente adequado para gerar vários documentos longos e diálogos complexos, aplicável em criação de conteúdo, geração de relatórios e sistemas de diálogo."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "O modelo visual Kimi (incluindo moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, etc.) é capaz de entender o conteúdo das imagens, incluindo texto, cores e formas dos objetos."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K é projetado para tarefas de geração de texto curto, com desempenho de processamento eficiente, capaz de lidar com 8.192 tokens, ideal para diálogos curtos, anotações e geração rápida de conteúdo."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "O modelo visual Kimi (incluindo moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, etc.) é capaz de entender o conteúdo das imagens, incluindo texto, cores e formas dos objetos."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B é uma versão aprimorada do Nous Hermes 2, contendo os conjuntos de dados mais recentes desenvolvidos internamente."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B é um grande modelo de linguagem personalizado pela NVIDIA, visando aumentar a utilidade das respostas geradas pelo LLM para as consultas dos usuários."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B é um modelo de linguagem em larga escala personalizado pela NVIDIA, projetado para aumentar a utilidade das respostas geradas pelo LLM em relação às consultas dos usuários. Este modelo se destacou em benchmarks como Arena Hard, AlpacaEval 2 LC e GPT-4-Turbo MT-Bench, ocupando o primeiro lugar em todos os três benchmarks de alinhamento automático até 1º de outubro de 2024. O modelo foi treinado usando RLHF (especialmente REINFORCE), Llama-3.1-Nemotron-70B-Reward e HelpSteer2-Preference prompts, com base no modelo Llama-3.1-70B-Instruct."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "Modelo de linguagem único, oferecendo precisão e eficiência incomparáveis."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct é um modelo de linguagem de grande porte personalizado pela NVIDIA, projetado para melhorar a utilidade das respostas geradas pelo LLM."
+ },
"o1": {
"description": "Focado em raciocínio avançado e resolução de problemas complexos, incluindo tarefas matemáticas e científicas. Muito adequado para aplicativos que exigem compreensão profunda do contexto e gerenciamento de fluxos de trabalho."
},
- "o1-2024-12-17": {
- "description": "o1 é o novo modelo de raciocínio da OpenAI, que suporta entrada de texto e imagem e gera texto como saída, adequado para tarefas complexas que exigem amplo conhecimento geral. Este modelo possui um contexto de 200K e uma data limite de conhecimento em outubro de 2023."
- },
"o1-mini": {
"description": "o1-mini é um modelo de raciocínio rápido e econômico, projetado para cenários de programação, matemática e ciências. Este modelo possui um contexto de 128K e uma data limite de conhecimento em outubro de 2023."
},
"o1-preview": {
"description": "o1 é o novo modelo de raciocínio da OpenAI, adequado para tarefas complexas que exigem amplo conhecimento geral. Este modelo possui um contexto de 128K e uma data limite de conhecimento em outubro de 2023."
},
+ "o3-mini": {
+ "description": "o3-mini é nosso mais recente modelo de inferência em miniatura, oferecendo alta inteligência com os mesmos custos e metas de latência que o o1-mini."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba é um modelo de linguagem Mamba 2 focado em geração de código, oferecendo forte suporte para tarefas avançadas de codificação e raciocínio."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini é o mais recente modelo da OpenAI, lançado após o GPT-4 Omni, que suporta entrada de texto e imagem e saída de texto. Como seu modelo compacto mais avançado, é muito mais barato do que outros modelos de ponta recentes e custa mais de 60% menos que o GPT-3.5 Turbo. Ele mantém inteligência de ponta, ao mesmo tempo que oferece uma relação custo-benefício significativa. O GPT-4o mini obteve uma pontuação de 82% no teste MMLU e atualmente está classificado acima do GPT-4 em preferências de chat."
},
- "openai/o1": {
- "description": "o1 é o novo modelo de raciocínio da OpenAI, que suporta entrada de texto e imagem e gera texto como saída, adequado para tarefas complexas que exigem amplo conhecimento geral. Este modelo possui um contexto de 200K e uma data limite de conhecimento em outubro de 2023."
- },
"openai/o1-mini": {
"description": "o1-mini é um modelo de raciocínio rápido e econômico, projetado para cenários de programação, matemática e ciências. Este modelo possui um contexto de 128K e uma data limite de conhecimento em outubro de 2023."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K possui uma capacidade de processamento de contexto extremamente grande, capaz de lidar com até 128K de informações contextuais, especialmente adequado para análise completa e processamento de associações lógicas de longo prazo em conteúdos longos, podendo oferecer lógica fluida e consistente e suporte a diversas citações em comunicações textuais complexas."
},
+ "qvq-72b-preview": {
+ "description": "O modelo QVQ é um modelo de pesquisa experimental desenvolvido pela equipe Qwen, focado em melhorar a capacidade de raciocínio visual, especialmente na área de raciocínio matemático."
+ },
"qwen-coder-plus-latest": {
"description": "Modelo de código Qwen."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Modelo de linguagem visual em escala ultra grande Qwen. Em comparação com a versão aprimorada, melhora ainda mais a capacidade de raciocínio visual e de seguir instruções, oferecendo um nível mais alto de percepção e cognição visual."
},
+ "qwen-vl-ocr-latest": {
+ "description": "O OCR Qwen é um modelo especializado em extração de texto, focado na capacidade de extrair texto de imagens de documentos, tabelas, questões de exames, escrita manual, entre outros. Ele pode reconhecer vários idiomas, atualmente suportando: chinês, inglês, francês, japonês, coreano, alemão, russo, italiano, vietnamita e árabe."
+ },
"qwen-vl-plus-latest": {
"description": "Versão aprimorada do modelo de linguagem visual em larga escala Qwen. Aumenta significativamente a capacidade de reconhecimento de detalhes e de texto, suportando resolução de mais de um milhão de pixels e imagens de qualquer proporção."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 é uma nova série de grandes modelos de linguagem, com capacidades de compreensão e geração mais robustas."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "LLM voltado para chinês e inglês, focado em linguagem, programação, matemática, raciocínio e outras áreas."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "LLM avançado, suporta geração de código, raciocínio e correção, abrangendo linguagens de programação populares."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "Modelo de código de médio porte poderoso, suporta comprimento de contexto de 32K, especializado em programação multilíngue."
+ },
"qwen2": {
"description": "Qwen2 é a nova geração de modelo de linguagem em larga escala da Alibaba, oferecendo desempenho excepcional para atender a diversas necessidades de aplicação."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "Modelo de 14B parâmetros do Qwen 2.5, disponível como código aberto."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "Modelo de 72B de código aberto do Qwen2.5."
+ },
"qwen2.5-32b-instruct": {
"description": "Modelo de 32B parâmetros do Qwen 2.5, disponível como código aberto."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "O modelo Qwen-Math possui uma forte capacidade de resolução de problemas matemáticos."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "Aprimoramento geral em seguimento de instruções, matemática, resolução de problemas e código, com capacidade de reconhecimento de objetos aprimorada, suporte a formatos diversos para localização precisa de elementos visuais, compreensão de arquivos de vídeo longos (até 10 minutos) e localização de eventos em segundos, capaz de entender a sequência e a velocidade do tempo, suportando controle de agentes em OS ou Mobile com forte capacidade de extração de informações e saída em formato Json. Esta versão é a de 72B, a mais poderosa da série."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "Aprimoramento geral em seguimento de instruções, matemática, resolução de problemas e código, com capacidade de reconhecimento de objetos aprimorada, suporte a formatos diversos para localização precisa de elementos visuais, compreensão de arquivos de vídeo longos (até 10 minutos) e localização de eventos em segundos, capaz de entender a sequência e a velocidade do tempo, suportando controle de agentes em OS ou Mobile com forte capacidade de extração de informações e saída em formato Json. Esta versão é a de 72B, a mais poderosa da série."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 é a nova geração de modelo de linguagem em larga escala da Alibaba, oferecendo desempenho excepcional para atender a diversas necessidades de aplicação."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro é um LLM de alta inteligência lançado pela Upstage, focado na capacidade de seguir instruções em um único GPU, com pontuação IFEval acima de 80. Atualmente suporta inglês, com uma versão oficial planejada para lançamento em novembro de 2024, que expandirá o suporte a idiomas e comprimento de contexto."
},
+ "sonar": {
+ "description": "Produto de busca leve baseado em contexto de busca, mais rápido e mais barato que o Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "Produto de busca avançada que suporta contexto de busca, consultas avançadas e acompanhamento."
+ },
+ "sonar-reasoning": {
+ "description": "Novo produto API suportado pelo modelo de raciocínio da DeepSeek."
+ },
"step-1-128k": {
"description": "Equilibra desempenho e custo, adequado para cenários gerais."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "Este modelo possui uma poderosa capacidade de compreensão de vídeo."
},
+ "step-1o-vision-32k": {
+ "description": "Este modelo possui uma poderosa capacidade de compreensão de imagens. Em comparação com a série de modelos step-1v, apresenta um desempenho visual superior."
+ },
"step-1v-32k": {
"description": "Suporta entradas visuais, aprimorando a experiência de interação multimodal."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Suporta interações de contexto em larga escala, adequado para cenários de diálogo complexos."
},
+ "step-2-mini": {
+ "description": "Um modelo de grande escala de alta velocidade baseado na nova arquitetura de atenção auto-desenvolvida MFA, alcançando resultados semelhantes ao step1 com um custo muito baixo, enquanto mantém uma maior taxa de transferência e um tempo de resposta mais rápido. Capaz de lidar com tarefas gerais, possui especialização em habilidades de codificação."
+ },
+ "taichu2_mm": {
+ "description": "Integra capacidades de compreensão de imagem, transferência de conhecimento e atribuição lógica, destacando-se no campo de perguntas e respostas baseadas em texto e imagem."
+ },
"taichu_llm": {
"description": "O modelo de linguagem Taichu possui uma forte capacidade de compreensão de linguagem, além de habilidades em criação de texto, perguntas e respostas, programação de código, cálculos matemáticos, raciocínio lógico, análise de sentimentos e resumo de texto. Inova ao combinar pré-treinamento com grandes dados e conhecimento rico de múltiplas fontes, aprimorando continuamente a tecnologia de algoritmos e absorvendo novos conhecimentos de vocabulário, estrutura, gramática e semântica de grandes volumes de dados textuais, proporcionando aos usuários informações e serviços mais convenientes e uma experiência mais inteligente."
},
+ "text-embedding-3-large": {
+ "description": "O modelo de vetorização mais poderoso, adequado para tarefas em inglês e não inglês."
+ },
+ "text-embedding-3-small": {
+ "description": "Modelo de Embedding de nova geração, eficiente e econômico, adequado para recuperação de conhecimento, aplicações RAG e outros cenários."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) oferece capacidade de computação aprimorada através de estratégias e arquiteturas de modelo eficientes."
},
+ "tts-1": {
+ "description": "O mais recente modelo de texto para fala, otimizado para velocidade em cenários em tempo real."
+ },
+ "tts-1-hd": {
+ "description": "O mais recente modelo de texto para fala, otimizado para qualidade."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) é adequado para tarefas de instrução refinadas, oferecendo excelente capacidade de processamento de linguagem."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet eleva o padrão da indústria, superando modelos concorrentes e Claude 3 Opus, apresentando um desempenho excepcional em uma ampla gama de avaliações, enquanto mantém a velocidade e o custo de nossos modelos de nível médio."
},
+ "whisper-1": {
+ "description": "Modelo de reconhecimento de voz universal, suporta reconhecimento de voz multilíngue, tradução de voz e identificação de idiomas."
+ },
"wizardlm2": {
"description": "WizardLM 2 é um modelo de linguagem fornecido pela Microsoft AI, destacando-se em diálogos complexos, multilíngue, raciocínio e assistentes inteligentes."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Modelo para tarefas visuais complexas, oferecendo alta performance em compreensão e análise de imagens."
+ },
+ "yi-vision-v2": {
+ "description": "Modelo para tarefas visuais complexas, oferecendo alta performance em compreensão e análise baseadas em múltiplas imagens."
}
}
diff --git a/locales/pt-BR/providers.json b/locales/pt-BR/providers.json
index db2cfcb4b2129..cc46d71555fb3 100644
--- a/locales/pt-BR/providers.json
+++ b/locales/pt-BR/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure oferece uma variedade de modelos avançados de IA, incluindo GPT-3.5 e a mais recente série GPT-4, suportando diversos tipos de dados e tarefas complexas, com foco em soluções de IA seguras, confiáveis e sustentáveis."
},
+ "azureai": {
+ "description": "A Azure oferece uma variedade de modelos de IA avançados, incluindo o GPT-3.5 e a mais recente série GPT-4, suportando diversos tipos de dados e tarefas complexas, comprometendo-se com soluções de IA seguras, confiáveis e sustentáveis."
+ },
"baichuan": {
"description": "Baichuan Intelligent é uma empresa focada no desenvolvimento de grandes modelos de inteligência artificial, cujos modelos se destacam em tarefas em chinês, como enciclopédias de conhecimento, processamento de textos longos e criação de conteúdo, superando modelos mainstream estrangeiros. A Baichuan Intelligent também possui capacidades multimodais líderes do setor, destacando-se em várias avaliações de autoridade. Seus modelos incluem Baichuan 4, Baichuan 3 Turbo e Baichuan 3 Turbo 128k, otimizados para diferentes cenários de aplicação, oferecendo soluções com alta relação custo-benefício."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "A DeepSeek é uma empresa focada em pesquisa e aplicação de tecnologia de inteligência artificial, cujo modelo mais recente, DeepSeek-V2.5, combina capacidades de diálogo geral e processamento de código, alcançando melhorias significativas em alinhamento com preferências humanas, tarefas de escrita e seguimento de instruções."
},
+ "doubao": {
+ "description": "Um grande modelo desenvolvido internamente pela ByteDance. Validado através da prática em mais de 50 cenários de negócios dentro da ByteDance, com um uso diário de trilhões de tokens, continuamente aprimorado, oferece diversas capacidades multimodais, criando uma rica experiência de negócios para as empresas com resultados de modelo de alta qualidade."
+ },
"fireworksai": {
"description": "Fireworks AI é um fornecedor líder de serviços de modelos de linguagem avançados, focando em chamadas de função e processamento multimodal. Seu modelo mais recente, Firefunction V2, baseado em Llama-3, é otimizado para chamadas de função, diálogos e seguimento de instruções. O modelo de linguagem visual FireLLaVA-13B suporta entradas mistas de imagem e texto. Outros modelos notáveis incluem a série Llama e a série Mixtral, oferecendo suporte eficiente para seguimento e geração de instruções multilíngues."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "Uma organização de código aberto dedicada à pesquisa e desenvolvimento de ferramentas para grandes modelos. Oferece uma plataforma de código aberto eficiente e fácil de usar para todos os desenvolvedores de IA, tornando as tecnologias e algoritmos de ponta acessíveis."
},
+ "jina": {
+ "description": "A Jina AI foi fundada em 2020 e é uma empresa líder em IA de busca. Nossa plataforma de busca base contém modelos vetoriais, reordenadores e pequenos modelos de linguagem, ajudando empresas a construir aplicações de busca generativa e multimodal confiáveis e de alta qualidade."
+ },
+ "lmstudio": {
+ "description": "LM Studio é um aplicativo de desktop para desenvolver e experimentar LLMs em seu computador."
+ },
"minimax": {
"description": "MiniMax é uma empresa de tecnologia de inteligência artificial geral fundada em 2021, dedicada a co-criar inteligência com os usuários. A MiniMax desenvolveu internamente diferentes modelos gerais de grande escala, incluindo um modelo de texto MoE com trilhões de parâmetros, um modelo de voz e um modelo de imagem. Também lançou aplicações como Conch AI."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI é uma plataforma que oferece uma variedade de modelos de linguagem de grande escala e serviços de geração de imagens de IA, sendo flexível, confiável e econômica. Suporta os mais recentes modelos de código aberto, como Llama3 e Mistral, e fornece soluções de API abrangentes, amigáveis ao usuário e escaláveis para o desenvolvimento de aplicações de IA, adequadas para o rápido crescimento de startups de IA."
},
+ "nvidia": {
+ "description": "O NVIDIA NIM™ fornece contêineres para inferência de microserviços acelerados por GPU autogerenciados, suportando a implantação de modelos de IA pré-treinados e personalizados na nuvem, em data centers, em PCs RTX™ AI e estações de trabalho."
+ },
"ollama": {
"description": "Os modelos oferecidos pela Ollama abrangem amplamente áreas como geração de código, operações matemáticas, processamento multilíngue e interações de diálogo, atendendo a diversas necessidades de implantação em nível empresarial e local."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "O Instituto de Automação da Academia Chinesa de Ciências e o Instituto de Pesquisa em Inteligência Artificial de Wuhan lançaram uma nova geração de grandes modelos multimodais, suportando tarefas abrangentes de perguntas e respostas, criação de texto, geração de imagens, compreensão 3D, análise de sinais, entre outras, com capacidades cognitivas, de compreensão e criação mais fortes, proporcionando uma nova experiência interativa."
},
+ "tencentcloud": {
+ "description": "A capacidade atômica do mecanismo de conhecimento (LLM Knowledge Engine Atomic Power) é uma capacidade completa de perguntas e respostas baseada no desenvolvimento do mecanismo de conhecimento, voltada para empresas e desenvolvedores, oferecendo a capacidade de montar e desenvolver aplicações de modelo de forma flexível. Você pode montar seu serviço de modelo exclusivo usando várias capacidades atômicas, chamando serviços de análise de documentos, divisão, embedding, reescrita em várias rodadas, entre outros, para personalizar negócios de IA exclusivos para sua empresa."
+ },
"togetherai": {
"description": "A Together AI se dedica a alcançar desempenho de ponta por meio de modelos de IA inovadores, oferecendo amplas capacidades de personalização, incluindo suporte para escalabilidade rápida e processos de implantação intuitivos, atendendo a diversas necessidades empresariais."
},
"upstage": {
"description": "Upstage se concentra no desenvolvimento de modelos de IA para diversas necessidades comerciais, incluindo Solar LLM e Document AI, visando alcançar uma inteligência geral artificial (AGI) que funcione. Crie agentes de diálogo simples por meio da API de Chat e suporte chamadas de função, tradução, incorporação e aplicações em domínios específicos."
},
+ "vllm": {
+ "description": "vLLM é uma biblioteca rápida e fácil de usar para inferência e serviços de LLM."
+ },
+ "volcengine": {
+ "description": "A plataforma de desenvolvimento de serviços de grandes modelos lançada pela ByteDance, que oferece serviços de chamada de modelos ricos em funcionalidades, seguros e com preços competitivos, além de fornecer dados de modelos, ajuste fino, inferência, avaliação e outras funcionalidades de ponta a ponta, garantindo de forma abrangente a implementação do seu desenvolvimento de aplicações de IA."
+ },
"wenxin": {
"description": "Plataforma de desenvolvimento e serviços de aplicativos nativos de IA e modelos de grande escala, voltada para empresas, que oferece a mais completa e fácil ferramenta de cadeia de ferramentas para o desenvolvimento de modelos de inteligência artificial generativa e aplicativos."
},
diff --git a/locales/pt-BR/setting.json b/locales/pt-BR/setting.json
index 8a18207517a41..d98ad43dffa98 100644
--- a/locales/pt-BR/setting.json
+++ b/locales/pt-BR/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Ativar limite de resposta única"
},
+ "enableReasoningEffort": {
+ "title": "Ativar ajuste de intensidade de raciocínio"
+ },
"frequencyPenalty": {
- "desc": "Quanto maior o valor, maior a probabilidade de reduzir palavras repetidas",
- "title": "Penalidade de frequência"
+ "desc": "Quanto maior o valor, mais rica e variada será a escolha de palavras; quanto menor o valor, mais simples e direta será a escolha de palavras.",
+ "title": "Riqueza do Vocabulário"
},
"maxTokens": {
"desc": "Número máximo de tokens a serem usados em uma interação única",
@@ -212,19 +215,31 @@
"desc": "{{provider}} modelo",
"title": "Modelo"
},
+ "params": {
+ "title": "Parâmetros Avançados"
+ },
"presencePenalty": {
- "desc": "Quanto maior o valor, maior a probabilidade de expandir para novos tópicos",
- "title": "Penalidade de novidade do tópico"
+ "desc": "Quanto maior o valor, mais inclinado a diferentes formas de expressão, evitando repetições de conceitos; quanto menor o valor, mais inclinado a usar conceitos ou narrativas repetidas, resultando em uma expressão mais consistente.",
+ "title": "Diversidade de Expressão"
+ },
+ "reasoningEffort": {
+ "desc": "Quanto maior o valor, mais forte será a capacidade de raciocínio, mas isso pode aumentar o tempo de resposta e o consumo de tokens",
+ "options": {
+ "high": "Alto",
+ "low": "Baixo",
+ "medium": "Médio"
+ },
+ "title": "Intensidade de raciocínio"
},
"temperature": {
- "desc": "Quanto maior o valor, mais aleatória será a resposta",
- "title": "Aleatoriedade",
- "titleWithValue": "Aleatoriedade {{value}}"
+ "desc": "Quanto maior o valor, mais criativas e imaginativas serão as respostas; quanto menor o valor, mais rigorosas serão as respostas",
+ "title": "Atividade Criativa",
+ "warning": "Valor de atividade criativa muito alto pode resultar em saídas confusas"
},
- "title": "Configurações do modelo",
+ "title": "Configurações do Modelo",
"topP": {
- "desc": "Semelhante à aleatoriedade, mas não deve ser alterado junto com a aleatoriedade",
- "title": "Amostragem principal"
+ "desc": "Quantas possibilidades considerar; quanto maior o valor, mais respostas possíveis serão aceitas; quanto menor o valor, mais se tende a escolher a resposta mais provável. Não é recomendado alterar junto com a atividade criativa",
+ "title": "Abertura Mental"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Configurações Comuns",
"experiment": "Experimento",
"llm": "Modelo de Linguagem",
+ "provider": "Fornecedor de IA",
"sync": "Sincronização na nuvem",
"system-agent": "Assistente do Sistema",
"tts": "Serviço de Voz"
diff --git a/locales/ru-RU/changelog.json b/locales/ru-RU/changelog.json
index 42a31753681ac..da64b0a69a0fc 100644
--- a/locales/ru-RU/changelog.json
+++ b/locales/ru-RU/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "Просмотреть все журналы изменений",
"description": "Постоянно следите за новыми функциями и улучшениями {{appName}}",
"pagination": {
- "older": "Посмотреть историю изменений",
- "prev": "Предыдущая страница"
+ "next": "Следующая страница",
+ "older": "Посмотреть историю изменений"
},
"readDetails": "Читать детали",
"title": "Журнал изменений",
diff --git a/locales/ru-RU/common.json b/locales/ru-RU/common.json
index 67d128277da2d..f5499b41ba244 100644
--- a/locales/ru-RU/common.json
+++ b/locales/ru-RU/common.json
@@ -275,6 +275,7 @@
},
"temp": "Временный",
"terms": "Условия использования",
+ "update": "Обновить",
"updateAgent": "Обновить информацию об агенте",
"upgradeVersion": {
"action": "обновить",
@@ -285,6 +286,7 @@
"anonymousNickName": "Анонимный пользователь",
"billing": "Управление счетами",
"cloud": "Опыт {{name}}",
+ "community": "Сообщество",
"data": "Хранилище данных",
"defaultNickname": "Пользователь сообщества",
"discord": "Поддержка сообщества",
@@ -294,7 +296,6 @@
"help": "Центр помощи",
"moveGuide": "Кнопка настроек перемещена сюда",
"plans": "Планы подписки",
- "preview": "Предпросмотр",
"profile": "Управление аккаунтом",
"setting": "Настройки приложения",
"usages": "Статистика использования"
diff --git a/locales/ru-RU/components.json b/locales/ru-RU/components.json
index fc522a24a2fb0..b785e24fdd980 100644
--- a/locales/ru-RU/components.json
+++ b/locales/ru-RU/components.json
@@ -76,6 +76,7 @@
"custom": "Пользовательская модель по умолчанию поддерживает как вызов функций, так и распознавание изображений. Пожалуйста, проверьте доступность указанных возможностей в вашем случае",
"file": "Эта модель поддерживает загрузку и распознавание файлов",
"functionCall": "Эта модель поддерживает вызов функций",
+ "reasoning": "Эта модель поддерживает глубокое мышление",
"tokens": "Эта модель поддерживает до {{tokens}} токенов в одной сессии",
"vision": "Эта модель поддерживает распознавание изображений"
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "Нет активированных моделей. Пожалуйста, перейдите в настройки и включите модель",
"provider": "Поставщик"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "Из-за ограничений безопасности браузера вам необходимо настроить кросс-доменные запросы для корректного использования Ollama.",
+ "linux": {
+ "env": "Добавьте `Environment` в раздел [Service] и добавьте переменную окружения OLLAMA_ORIGINS:",
+ "reboot": "Перезагрузите systemd и перезапустите Ollama",
+ "systemd": "Вызовите systemd для редактирования службы ollama:"
+ },
+ "macos": "Откройте приложение «Терминал», вставьте следующую команду и нажмите Enter для выполнения",
+ "reboot": "Пожалуйста, перезапустите службу Ollama после завершения выполнения",
+ "title": "Настройка Ollama для разрешения кросс-доменных запросов",
+ "windows": "На Windows нажмите «Панель управления», перейдите к редактированию системных переменных окружения. Создайте новую переменную окружения с именем «OLLAMA_ORIGINS» для вашей учетной записи пользователя, значение - * , нажмите «OK/Применить» для сохранения"
+ },
+ "install": {
+ "description": "Пожалуйста, убедитесь, что вы запустили Ollama. Если вы еще не скачали Ollama, перейдите на официальный сайт <1>для загрузки",
+ "docker": "Если вы предпочитаете использовать Docker, Ollama также предоставляет официальный образ Docker, который вы можете загрузить с помощью следующей команды:",
+ "linux": {
+ "command": "Установите с помощью следующей команды:",
+ "manual": "Или вы можете обратиться к <1>руководству по ручной установке для Linux для самостоятельной установки"
+ },
+ "title": "Установите и запустите приложение Ollama локально",
+ "windowsTab": "Windows (предварительная версия)"
+ }
+ },
+ "Thinking": {
+ "thinking": "Глубокое размышление...",
+ "thought": "Глубоко обдумано (время: {{duration}} секунд)",
+ "thoughtWithDuration": "Глубоко обдумано"
}
}
diff --git a/locales/ru-RU/discover.json b/locales/ru-RU/discover.json
index 1259a910fb9aa..9553af416a675 100644
--- a/locales/ru-RU/discover.json
+++ b/locales/ru-RU/discover.json
@@ -126,6 +126,10 @@
"title": "Свежесть темы"
},
"range": "Диапазон",
+ "reasoning_effort": {
+ "desc": "Эта настройка используется для управления интенсивностью размышлений модели перед генерацией ответа. Низкая интенсивность приоритизирует скорость ответа и экономит токены, высокая интенсивность обеспечивает более полное размышление, но потребляет больше токенов и снижает скорость ответа. Значение по умолчанию - среднее, что обеспечивает баланс между точностью размышлений и скоростью ответа.",
+ "title": "Интенсивность размышлений"
+ },
"temperature": {
"desc": "Эта настройка влияет на разнообразие ответов модели. Более низкие значения приводят к более предсказуемым и типичным ответам, в то время как более высокие значения поощряют более разнообразные и необычные ответы. Когда значение установлено на 0, модель всегда дает один и тот же ответ на данный ввод.",
"title": "Случайность"
diff --git a/locales/ru-RU/modelProvider.json b/locales/ru-RU/modelProvider.json
index 60b7ab31cb907..5da480dda8bac 100644
--- a/locales/ru-RU/modelProvider.json
+++ b/locales/ru-RU/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API Key"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Версия API Azure, формат YYYY-MM-DD, смотрите [последнюю версию](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "Получить список",
+ "title": "Версия API Azure"
+ },
+ "endpoint": {
+ "desc": "Найдите конечную точку вывода модели Azure AI в обзоре проекта Azure AI",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Конечная точка Azure AI"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Найдите API-ключ в обзоре проекта Azure AI",
+ "placeholder": "Ключ Azure",
+ "title": "Ключ"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "Введите ваш AWS Access Key ID",
@@ -63,6 +81,46 @@
"title": "ID аккаунта Cloudflare / адрес API"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Пожалуйста, введите ваш API Key",
+ "title": "API Key"
+ },
+ "basicTitle": "Основная информация",
+ "configTitle": "Конфигурационная информация",
+ "confirm": "Создать",
+ "createSuccess": "Создание успешно",
+ "description": {
+ "placeholder": "Описание провайдера (необязательно)",
+ "title": "Описание провайдера"
+ },
+ "id": {
+ "desc": "Уникальный идентификатор для поставщика услуг, который нельзя изменить после создания",
+ "format": "Может содержать только цифры, строчные буквы, дефисы (-) и подчеркивания (_) ",
+ "placeholder": "Рекомендуется использовать строчные буквы, например, openai, после создания изменить нельзя",
+ "required": "Пожалуйста, введите ID провайдера",
+ "title": "ID провайдера"
+ },
+ "logo": {
+ "required": "Пожалуйста, загрузите правильный логотип провайдера",
+ "title": "Логотип провайдера"
+ },
+ "name": {
+ "placeholder": "Пожалуйста, введите отображаемое имя провайдера",
+ "required": "Пожалуйста, введите имя провайдера",
+ "title": "Имя провайдера"
+ },
+ "proxyUrl": {
+ "required": "Пожалуйста, введите адрес прокси",
+ "title": "Адрес прокси"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Пожалуйста, выберите тип SDK",
+ "title": "Формат запроса"
+ },
+ "title": "Создание пользовательского AI провайдера"
+ },
"github": {
"personalAccessToken": {
"desc": "Введите ваш персональный токен доступа GitHub (PAT), нажмите [здесь](https://github.com/settings/tokens), чтобы создать его",
@@ -77,6 +135,23 @@
"title": "Токен HuggingFace"
}
},
+ "list": {
+ "title": {
+ "disabled": "Поставщик не активирован",
+ "enabled": "Поставщик активирован"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Добавить пользовательского провайдера",
+ "all": "Все",
+ "list": {
+ "disabled": "Не активирован",
+ "enabled": "Активирован"
+ },
+ "notFound": "Результаты поиска не найдены",
+ "searchProviders": "Поиск провайдеров...",
+ "sort": "Пользовательская сортировка"
+ },
"ollama": {
"checker": {
"desc": "Проверить правильность адреса прокси",
@@ -94,33 +169,9 @@
"title": "Загрузка модели {{model}} "
},
"endpoint": {
- "desc": "Введите адрес прокси-интерфейса Ollama, если локально не указано иное, можете оставить пустым",
+ "desc": "Должен содержать http(s)://, если локально не указано иное, можно оставить пустым",
"title": "Адрес прокси-интерфейса"
},
- "setup": {
- "cors": {
- "description": "Из-за ограничений безопасности браузера вам необходимо настроить кросс-доменные запросы для правильной работы Ollama.",
- "linux": {
- "env": "Добавьте переменную среды OLLAMA_ORIGINS в разделе [Service],",
- "reboot": "Перезагрузите systemd и перезапустите Ollama.",
- "systemd": "Вызовите редактирование службы ollama в systemd:"
- },
- "macos": "Откройте приложение \"Терминал\", вставьте и выполните следующую команду, затем нажмите Enter.",
- "reboot": "Пожалуйста, перезагрузите службу Ollama после завершения выполнения команды.",
- "title": "Настройка разрешений на кросс-доменный доступ для Ollama",
- "windows": "На Windows откройте \"Панель управления\", зайдите в настройки системных переменных. Создайте новую переменную среды для вашей учетной записи с именем \"OLLAMA_ORIGINS\" и значением * , затем нажмите \"OK/Применить\" для сохранения."
- },
- "install": {
- "description": "Пожалуйста, убедитесь, что вы установили Ollama. Если вы еще не скачали Ollama, перейдите на официальный сайт <1> для загрузки",
- "docker": "Если вы предпочитаете использовать Docker, Ollama также предоставляет официальное образ Docker. Вы можете загрузить его с помощью следующей команды:",
- "linux": {
- "command": "Установите с помощью следующей команды:",
- "manual": "Или вы можете установить его вручную, следуя <1>руководству по установке на Linux."
- },
- "title": "Установка и запуск приложения Ollama локально",
- "windowsTab": "Windows (превью)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Cancel Download",
@@ -131,25 +182,145 @@
"title": "Download specified Ollama model"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Введите Access Key платформы Baidu Qianfan",
- "placeholder": "Access Key Qianfan",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "Ваши ключи и адрес прокси будут зашифрованы с использованием <1>AES-GCM",
+ "apiKey": {
+ "desc": "Пожалуйста, введите ваш {{name}} API Key",
+ "placeholder": "{{name}} API Key",
+ "title": "API Key"
+ },
+ "baseURL": {
+ "desc": "Должен содержать http(s)://",
+ "invalid": "Пожалуйста, введите действительный URL",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "API адрес прокси"
+ },
+ "checker": {
+ "button": "Проверить",
+ "desc": "Проверьте, правильно ли заполнены API Key и адрес прокси",
+ "pass": "Проверка пройдена",
+ "title": "Проверка соединения"
+ },
+ "fetchOnClient": {
+ "desc": "Клиентский режим запросов будет инициировать сессии напрямую из браузера, что может ускорить время отклика",
+ "title": "Использовать клиентский режим запросов"
+ },
+ "helpDoc": "Документация по настройке",
+ "waitingForMore": "Больше моделей находится в <1>планировании подключения, ожидайте с нетерпением"
},
- "checker": {
- "desc": "Проверьте, правильно ли заполнены AccessKey / SecretAccess"
+ "createNew": {
+ "title": "Создание пользовательской AI модели"
},
- "secretKey": {
- "desc": "Введите Secret Key платформы Baidu Qianfan",
- "placeholder": "Secret Key Qianfan",
- "title": "Secret Key"
+ "item": {
+ "config": "Настроить модель",
+ "customModelCards": {
+ "addNew": "Создать и добавить модель {{id}}",
+ "confirmDelete": "Вы собираетесь удалить эту пользовательскую модель, после удаления восстановить ее будет невозможно, будьте осторожны."
+ },
+ "delete": {
+ "confirm": "Подтвердите удаление модели {{displayName}}?",
+ "success": "Удаление успешно",
+ "title": "Удалить модель"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Поле, запрашиваемое в Azure OpenAI",
+ "placeholder": "Пожалуйста, введите имя развертывания модели в Azure",
+ "title": "Имя развертывания модели"
+ },
+ "deployName": {
+ "extra": "Это поле будет использоваться как идентификатор модели при отправке запроса",
+ "placeholder": "Введите фактическое имя или id развертывания модели",
+ "title": "Имя развертывания модели"
+ },
+ "displayName": {
+ "placeholder": "Пожалуйста, введите отображаемое имя модели, например, ChatGPT, GPT-4 и т.д.",
+ "title": "Отображаемое имя модели"
+ },
+ "files": {
+ "extra": "Текущая реализация загрузки файлов является лишь хакерским решением, предназначенным только для самостоятельного тестирования. Полные возможности загрузки файлов ожидайте в будущем.",
+ "title": "Поддержка загрузки файлов"
+ },
+ "functionCall": {
+ "extra": "Эта настройка позволит модели использовать инструменты, что даст возможность добавлять плагины инструментов. Однако возможность фактического использования инструментов полностью зависит от самой модели, пожалуйста, протестируйте их работоспособность самостоятельно",
+ "title": "Поддержка использования инструментов"
+ },
+ "id": {
+ "extra": "После создания изменить нельзя, будет использоваться как идентификатор модели при вызове AI",
+ "placeholder": "Введите идентификатор модели, например, gpt-4o или claude-3.5-sonnet",
+ "title": "ID модели"
+ },
+ "modalTitle": "Настройка пользовательской модели",
+ "reasoning": {
+ "extra": "Эта настройка активирует возможность глубокого мышления модели, конкретный эффект полностью зависит от самой модели, пожалуйста, протестируйте, обладает ли модель доступной способностью к глубокому мышлению",
+ "title": "Поддержка глубокого мышления"
+ },
+ "tokens": {
+ "extra": "Установите максимальное количество токенов, поддерживаемое моделью",
+ "title": "Максимальное окно контекста",
+ "unlimited": "Без ограничений"
+ },
+ "vision": {
+ "extra": "Эта настройка только активирует возможность загрузки изображений в приложении, поддержка распознавания полностью зависит от самой модели, пожалуйста, протестируйте доступность визуального распознавания этой модели.",
+ "title": "Поддержка визуального распознавания"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/изображение",
+ "inputCharts": "${{amount}}/M символов",
+ "inputMinutes": "${{amount}}/минуты",
+ "inputTokens": "Ввод ${{amount}}/М",
+ "outputTokens": "Вывод ${{amount}}/М"
+ },
+ "releasedAt": "Выпущено {{releasedAt}}"
},
- "unlock": {
- "customRegion": "Пользовательский регион сервиса",
- "description": "Введите ваш AccessKey / SecretKey, чтобы начать сессию. Приложение не будет сохранять ваши данные аутентификации",
- "title": "Использовать пользовательскую информацию аутентификации Wenxin Yiyan"
- }
+ "list": {
+ "addNew": "Добавить модель",
+ "disabled": "Не активирован",
+ "disabledActions": {
+ "showMore": "Показать все"
+ },
+ "empty": {
+ "desc": "Пожалуйста, создайте пользовательскую модель или загрузите модель, чтобы начать использовать.",
+ "title": "Нет доступных моделей"
+ },
+ "enabled": "Активирован",
+ "enabledActions": {
+ "disableAll": "Отключить все",
+ "enableAll": "Включить все",
+ "sort": "Сортировка моделей по индивидуальному порядку"
+ },
+ "enabledEmpty": "Нет активированных моделей, пожалуйста, активируйте понравившиеся модели из списка ниже~",
+ "fetcher": {
+ "clear": "Очистить полученные модели",
+ "fetch": "Получить список моделей",
+ "fetching": "Получение списка моделей...",
+ "latestTime": "Последнее обновление: {{time}}",
+ "noLatestTime": "Список еще не получен"
+ },
+ "resetAll": {
+ "conform": "Вы уверены, что хотите сбросить все изменения текущей модели? После сброса список текущих моделей вернется к состоянию по умолчанию",
+ "success": "Сброс выполнен успешно",
+ "title": "Сбросить все изменения"
+ },
+ "search": "Поиск моделей...",
+ "searchResult": "Найдено {{count}} моделей",
+ "title": "Список моделей",
+ "total": "Всего доступно {{count}} моделей"
+ },
+ "searchNotFound": "Результаты поиска не найдены"
+ },
+ "sortModal": {
+ "success": "Сортировка обновлена успешно",
+ "title": "Пользовательская сортировка",
+ "update": "Обновить"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "Вы собираетесь удалить этого AI провайдера, после удаления его будет невозможно восстановить, подтвердите, хотите ли вы удалить?",
+ "deleteSuccess": "Удаление успешно",
+ "tooltip": "Обновить базовую конфигурацию провайдера",
+ "updateSuccess": "Обновление успешно"
},
"zeroone": {
"title": "01.AI Цифровая Вселенная"
diff --git a/locales/ru-RU/models.json b/locales/ru-RU/models.json
index 0165ace335e0b..466f2da9424a4 100644
--- a/locales/ru-RU/models.json
+++ b/locales/ru-RU/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K акцентирует внимание на семантической безопасности и ответственности, специально разработан для приложений с высокими требованиями к безопасности контента, обеспечивая точность и надежность пользовательского опыта."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 использует дерево поиска для построения цепочек размышлений и вводит механизм рефлексии, обучаясь с помощью усиленного обучения, модель обладает способностью к саморефлексии и исправлению ошибок."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro — это продвинутая модель обработки естественного языка, выпущенная компанией 360, обладающая выдающимися способностями к генерации и пониманию текста, особенно в области генерации и творчества, способная обрабатывать сложные языковые преобразования и ролевые задачи."
},
+ "360zhinao2-o1": {
+ "description": "Модель 360zhinao2-o1 использует дерево поиска для построения цепочки размышлений и включает механизм рефлексии, обучаясь с помощью усиленного обучения, что позволяет модели самостоятельно рефлексировать и исправлять ошибки."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra — это самая мощная версия в серии больших моделей Xinghuo, которая, обновив сетевые поисковые связи, улучшает понимание и обобщение текстового контента. Это всестороннее решение для повышения производительности в офисе и точного реагирования на запросы, являющееся ведущим интеллектуальным продуктом в отрасли."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "Модель обладает лучшими в стране возможностями, превосходя зарубежные модели в задачах на китайском языке, таких как энциклопедические знания, длинные тексты и генерация контента. Также обладает передовыми мультимодальными возможностями и демонстрирует отличные результаты в нескольких авторитетных оценочных тестах."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Модель DeepSeek-R1, основанная на Qwen2.5-Math-1.5B, оптимизирует производительность вывода с помощью усиленного обучения и данных холодного старта, обновляя стандарт многозадачности в открытых моделях."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Модель DeepSeek-R1, основанная на Qwen2.5-14B, оптимизирует производительность вывода с помощью усиленного обучения и данных холодного старта, обновляя стандарт многозадачности в открытых моделях."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "Серия DeepSeek-R1 оптимизирует производительность вывода с помощью усиленного обучения и данных холодного старта, обновляя стандарт многозадачности в открытых моделях, превосходя уровень OpenAI-o1-mini."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "Модель DeepSeek-R1, основанная на Qwen2.5-Math-7B, оптимизирует производительность вывода с помощью усиленного обучения и данных холодного старта, обновляя стандарт многозадачности в открытых моделях."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite обеспечивает выдающуюся скорость отклика и лучшее соотношение цены и качества, предлагая клиентам больше гибкости в различных сценариях. Поддерживает вывод и настройку с 128k контекстным окном."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Флагманская сверхкрупномасштабная языковая модель, разработанная Baidu, которая по сравнению с ERNIE 3.5 обеспечивает полное обновление возможностей модели и широко применяется в сложных задачах в различных областях; поддерживает автоматическую интеграцию с плагином поиска Baidu, обеспечивая актуальность информации в ответах."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Флагманская модель Baidu, разработанная самостоятельно, с огромным масштабом, демонстрирует отличные результаты и широко применяется в сложных задачах в различных областях; поддерживает автоматическую интеграцию с плагином поиска Baidu, обеспечивая актуальность информации в ответах. По сравнению с ERNIE 4.0, демонстрирует лучшие результаты."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "Флагманская 超大型 языковая модель, разработанная Baidu, демонстрирует отличные результаты и хорошо подходит для сложных задач в различных областях; поддерживает автоматическую интеграцию с плагином поиска Baidu, обеспечивая своевременность ответов. По сравнению с ERNIE 4.0 имеет лучшие показатели производительности."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct — это одна из последних языковых моделей, выпущенных Alibaba Cloud. Эта 7B модель значительно улучшила способности в области кодирования и математики. Модель также поддерживает множество языков, охватывающих более 29 языков, включая китайский и английский. Она значительно улучшила выполнение инструкций, понимание структурированных данных и генерацию структурированных выходных данных (особенно JSON)."
},
+ "MiniMax-Text-01": {
+ "description": "В серии моделей MiniMax-01 мы сделали смелые инновации: впервые в крупномасштабном масштабе реализован линейный механизм внимания, традиционная архитектура Transformer больше не является единственным выбором. Объем параметров этой модели достигает 456 миллиардов, из которых 45,9 миллиарда активируются за один раз. Комплексная производительность модели сопоставима с ведущими зарубежными моделями, при этом она может эффективно обрабатывать контекст длиной до 4 миллионов токенов, что в 32 раза больше, чем у GPT-4o, и в 20 раз больше, чем у Claude-3.5-Sonnet."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO — это высокоадаптивная многомодельная комбинация, предназначенная для предоставления выдающегося творческого опыта."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 демонстрирует превосходные результаты в различных визуально-языковых задачах, включая понимание документов и графиков, понимание текстов сцены, OCR, решение научных и математических задач."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 демонстрирует превосходные результаты в различных визуально-языковых задачах, включая понимание документов и графиков, понимание текстов сцены, OCR, решение научных и математических задач."
- },
"Phi-3-medium-128k-instruct": {
"description": "Та же модель Phi-3-medium, но с большим размером контекста для RAG или нескольких подсказок."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat — это открытая версия предобученной модели из серии GLM-4, выпущенная Zhizhu AI. Эта модель показывает отличные результаты в семантике, математике, выводах, коде и знаниях. Кроме поддержки многократных диалогов, GLM-4-9B-Chat также обладает продвинутыми функциями, такими как веб-браузинг, выполнение кода, вызов пользовательских инструментов (Function Call) и вывод длинных текстов. Модель поддерживает 26 языков, включая китайский, английский, японский, корейский и немецкий. В нескольких бенчмарках GLM-4-9B-Chat демонстрирует отличные результаты, такие как AlignBench-v2, MT-Bench, MMLU и C-Eval. Эта модель поддерживает максимальную длину контекста 128K и подходит для академических исследований и коммерческих приложений."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 — это модель вывода, управляемая обучением с подкреплением (RL), которая решает проблемы повторяемости и читаемости в модели. Перед RL DeepSeek-R1 вводит данные холодного старта, что дополнительно оптимизирует производительность вывода. Она показывает сопоставимые результаты с OpenAI-o1 в математических, кодовых и задачах вывода и улучшает общую эффективность благодаря тщательно продуманным методам обучения."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 — это языковая модель с 6710 миллиардами параметров, использующая архитектуру смешанных экспертов (MoE) и многофункциональное внимание (MLA), в сочетании с стратегией балансировки нагрузки без вспомогательных потерь, оптимизирующая эффективность вывода и обучения. После предобучения на 14.8 триллионах высококачественных токенов и последующей контролируемой донастройки и обучения с подкреплением, DeepSeek-V3 превосходит другие открытые модели и приближается к ведущим закрытым моделям."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma — это одна из легковесных, передовых открытых моделей, разработанных Google. Это крупная языковая модель с только декодером, поддерживающая английский язык, предлагающая открытые веса, предобученные варианты и варианты с дообучением на инструкциях. Модель Gemma подходит для различных задач генерации текста, включая вопросы и ответы, резюме и выводы. Эта 9B модель была обучена на 8 триллионах токенов. Ее относительно небольшой размер позволяет развертывать ее в условиях ограниченных ресурсов, таких как ноутбуки, настольные компьютеры или ваша собственная облачная инфраструктура, что позволяет большему количеству людей получить доступ к передовым моделям ИИ и способствовать инновациям."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 — это семейство многоязычных крупных языковых моделей, разработанных Meta, включая предобученные и дообученные на инструкциях варианты с параметрами 8B, 70B и 405B. Эта 8B модель с дообучением на инструкциях оптимизирована для многоязычных диалоговых сценариев и показывает отличные результаты в нескольких отраслевых бенчмарках. Обучение модели использовало более 150 триллионов токенов открытых данных и применяло такие технологии, как контролируемое дообучение и обучение с подкреплением на основе человеческой обратной связи для повышения полезности и безопасности модели. Llama 3.1 поддерживает генерацию текста и кода, с датой окончания знаний в декабре 2023 года."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview — это инновационная модель обработки естественного языка, способная эффективно обрабатывать сложные задачи генерации диалогов и понимания контекста."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview — это исследовательская модель, разработанная командой Qwen, сосредоточенная на способностях визуального вывода, обладающая уникальными преимуществами в понимании сложных сцен и решении визуально связанных математических задач."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview — это последняя экспериментальная исследовательская модель Qwen, сосредоточенная на повышении возможностей вывода ИИ. Исследуя сложные механизмы, такие как смешение языков и рекурсивные выводы, основные преимущества включают мощные аналитические способности, математические и программные навыки. В то же время существуют проблемы с переключением языков, циклом вывода, соображениями безопасности и различиями в других способностях."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct — это последняя версия серии языковых моделей, специфичных для кода, выпущенная Alibaba Cloud. Эта модель значительно улучшила способности генерации кода, вывода и исправления на основе Qwen2.5, обучаясь на 5.5 триллионах токенов. Она не только усилила кодирование, но и сохранила преимущества в математике и общих способностях. Модель предоставляет более полную основу для практических приложений, таких как интеллектуальные агенты кода."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math сосредоточен на решении математических задач, предоставляя профессиональные ответы на сложные вопросы."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 — это последняя серия моделей Qwen, поддерживающая контекст до 128k. По сравнению с текущими лучшими открытыми моделями, Qwen2-72B значительно превосходит ведущие модели по многим аспектам, включая понимание естественного языка, знания, код, математику и многоязычность."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 — это последняя серия моделей Qwen, способная превосходить лучшие открытые модели сопоставимого размера и даже более крупные модели. Qwen2 7B демонстрирует значительные преимущества в нескольких тестах, особенно в понимании кода и китайского языка."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B — это мощная модель визуального языка, поддерживающая многомодальную обработку изображений и текста, способная точно распознавать содержимое изображений и генерировать соответствующие описания или ответы."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct — это языковая модель с 14 миллиардами параметров, с отличными показателями производительности, оптимизированная для китайского и многоязычного контекста, поддерживает интеллектуальные ответы, генерацию контента и другие приложения."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct — это языковая модель с 7 миллиардами параметров, поддерживающая вызовы функций и бесшовное взаимодействие с внешними системами, что значительно повышает гибкость и масштабируемость. Оптимизирована для китайского и многоязычного контекста, поддерживает интеллектуальные ответы, генерацию контента и другие приложения."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct — это модель программирования на основе масштабного предварительного обучения, обладающая мощными способностями к пониманию и генерации кода, способная эффективно решать различные задачи программирования, особенно подходит для интеллектуального написания кода, автоматизации скриптов и ответов на программные вопросы."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct — это крупная языковая модель, специально разработанная для генерации кода, понимания кода и эффективных сценариев разработки, с передовым масштабом параметров 32B, способная удовлетворить разнообразные потребности программирования."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "Модель TeleMM — это многомодальная модель, разработанная China Telecom, способная обрабатывать текстовые, графические и другие виды входных данных, поддерживающая функции понимания изображений, анализа графиков и т.д., предоставляя пользователям услуги понимания на разных модальностях. Модель может взаимодействовать с пользователями в многомодальном формате, точно понимая входной контент, отвечая на вопросы, помогая в творчестве и эффективно предоставляя многомодальную информацию и поддержку вдохновения. Она показывает отличные результаты в задачах многомодального восприятия и логического вывода."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large — это крупнейшая в отрасли открытая модель Transformer архитектуры MoE с общим количеством параметров 389 миллиардов и 52 миллиарда активных параметров."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct — это языковая модель с дообучением на инструкциях в серии Qwen2, с параметрами 72B. Эта модель основана на архитектуре Transformer и использует такие технологии, как активационная функция SwiGLU, смещение внимания QKV и групповой запрос внимания. Она может обрабатывать большие объемы входных данных. Эта модель показывает отличные результаты в понимании языка, генерации, многоязычных способностях, кодировании, математике и выводах в различных бенчмарках, превосходя большинство открытых моделей и демонстрируя конкурентоспособность с проприетарными моделями в некоторых задачах."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct — это одна из последних языковых моделей, выпущенных Alibaba Cloud. Эта 72B модель значительно улучшила способности в области кодирования и математики. Модель также поддерживает множество языков, охватывающих более 29 языков, включая китайский и английский. Она значительно улучшила выполнение инструкций, понимание структурированных данных и генерацию структурированных выходных данных (особенно JSON)."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Оптимизирован для диалогов на китайском языке, обеспечивая плавную генерацию диалогов, соответствующую китайским языковым привычкам."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Открытая модель вызова функций от Fireworks, обеспечивающая выдающиеся возможности выполнения команд и открытые настраиваемые функции."
+ "abab7-chat-preview": {
+ "description": "По сравнению с серией моделей abab6.5, значительно улучшены способности в области длинных текстов, математики и написания."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Firefunction-v2 от компании Fireworks — это высокопроизводительная модель вызова функций, разработанная на основе Llama-3 и оптимизированная для вызова функций, диалогов и выполнения команд."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 — это передовая большая языковая модель, оптимизированная с помощью обучения с подкреплением и холодных стартовых данных, обладающая выдающимися показателями вывода, математики и программирования."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b — это визуальная языковая модель, способная одновременно обрабатывать изображения и текстовые вводы, обученная на высококачественных данных, подходящая для мультимодальных задач."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Мощная языковая модель Mixture-of-Experts (MoE) от Deepseek с общим количеством параметров 671B, активирующая 37B параметров на каждый токен."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Модель Llama 3 70B для команд, специально оптимизированная для многоязычных диалогов и понимания естественного языка, превосходит большинство конкурентных моделей."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Модель Llama 3 70B для команд (HF версия), результаты которой совпадают с официальной реализацией, подходит для высококачественных задач выполнения команд."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Модель Llama 3 8B для команд, оптимизированная для диалогов и многоязычных задач, демонстрирует выдающиеся и эффективные результаты."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Модель Meta с 11B параметрами, оптимизированная для вывода изображений. Эта модель предназначена для визуального распознавания, вывода изображений, описания изображений и ответа на общие вопросы о изображениях. Эта модель способна понимать визуальные данные, такие как графики и диаграммы, и преодолевать разрыв между визуальным и языковым пониманием, генерируя текстовые описания деталей изображений."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Модель Llama 3.2 1B для инструкций - это компактная многоязычная модель, запущенная Meta. Эта модель предназначена для повышения эффективности и обеспечивает значительное улучшение в задержке и стоимости по сравнению с более крупными моделями. Примеры использования модели включают извлечение информации и резюме."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Модель Llama 3.2 3B для инструкций - это компактная многоязычная модель, запущенная Meta. Эта модель предназначена для повышения эффективности и обеспечивает значительное улучшение в задержке и стоимости по сравнению с более крупными моделями. Примеры использования модели включают запросы, переоформление подсказок и помощь в написании."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Модель Meta с 90B параметрами, оптимизированная для вывода изображений. Эта модель предназначена для визуального распознавания, вывода изображений, описания изображений и ответа на общие вопросы о изображениях. Эта модель способна понимать визуальные данные, такие как графики и диаграммы, и преодолевать разрыв между визуальным и языковым пониманием, генерируя текстовые описания деталей изображений."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct — это обновленная версия Llama 3.1 70B от декабря. Эта модель улучшена на основе Llama 3.1 70B (выпущенной в июле 2024 года), с усиленной поддержкой вызовов инструментов, многоязычного текста, математических и программных возможностей. Модель достигла ведущих в отрасли показателей в области вывода, математики и соблюдения инструкций, обеспечивая производительность, сопоставимую с 3.1 405B, при этом обладая значительными преимуществами по скорости и стоимости."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "Модель с 24B параметрами, обладающая передовыми возможностями, сопоставимыми с более крупными моделями."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Mixtral MoE 8x22B для команд, с большим количеством параметров и архитектурой с несколькими экспертами, всесторонне поддерживает эффективную обработку сложных задач."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Mixtral MoE 8x7B для команд, архитектура с несколькими экспертами обеспечивает эффективное выполнение и следование командам."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Mixtral MoE 8x7B для команд (HF версия), производительность которой совпадает с официальной реализацией, подходит для множества эффективных задач."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "Модель MythoMax L2 13B, использующая новые технологии объединения, хорошо подходит для повествования и ролевых игр."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "Модель QwQ — это экспериментальная исследовательская модель, разработанная командой Qwen, сосредоточенная на улучшении возможностей вывода ИИ."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "72B версия модели Qwen-VL — это результат последней итерации Alibaba, представляющий собой инновации почти за год."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 - это серия языковых моделей, содержащая только декодеры, разработанная командой Qwen от Alibaba Cloud. Эти модели предлагаются в различных размерах: 0.5B, 1.5B, 3B, 7B, 14B, 32B и 72B, с вариантами базовой и инструкционной версии."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct — это последняя версия серии языковых моделей, специфичных для кода, выпущенная Alibaba Cloud. Эта модель значительно улучшила способности генерации кода, вывода и исправления на основе Qwen2.5, обучаясь на 5.5 триллионах токенов. Она не только усилила кодирование, но и сохранила преимущества в математике и общих способностях. Модель предоставляет более полную основу для практических приложений, таких как интеллектуальные агенты кода."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "Модель StarCoder 15.5B, поддерживающая сложные задачи программирования, с улучшенными многоязычными возможностями, подходит для генерации и понимания сложного кода."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "Модель StarCoder 7B, обученная на более чем 80 языках программирования, обладает выдающимися способностями к заполнению кода и пониманию контекста."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Модель Yi-Large, обладающая выдающимися возможностями обработки нескольких языков, подходит для различных задач генерации и понимания языка."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus — это самая мощная модель от Anthropic для обработки высококомплексных задач. Она демонстрирует выдающиеся результаты по производительности, интеллекту, плавности и пониманию."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku — это самая быстрая модель следующего поколения от Anthropic. По сравнению с Claude 3 Haiku, Claude 3.5 Haiku продемонстрировала улучшения во всех навыках и превзошла предыдущую крупнейшую модель Claude 3 Opus во многих интеллектуальных бенчмарках."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet предлагает возможности, превосходящие Opus, и скорость, превышающую Sonnet, при этом сохраняя ту же цену. Sonnet особенно хорошо справляется с программированием, наукой о данных, визуальной обработкой и агентскими задачами."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet обеспечивает идеальный баланс между интеллектом и скоростью для корпоративных рабочих нагрузок. Он предлагает максимальную полезность по более низкой цене, надежен и подходит для масштабного развертывания."
},
- "code-raccoon-v1": {
- "description": "Кодовый енот — это программный интеллектуальный помощник на основе языковой модели SenseTime, охватывающий такие этапы, как анализ требований к программному обеспечению, проектирование архитектуры, написание кода, тестирование программного обеспечения и т. д., удовлетворяющий различные потребности пользователей в написании кода и обучении программированию. Кодовый енот поддерживает более 90 популярных языков программирования, таких как Python, Java, JavaScript, C++, Go, SQL, а также популярные IDE, такие как VS Code и IntelliJ IDEA. В реальных приложениях кодовый енот может помочь разработчикам повысить эффективность программирования более чем на 50%."
- },
"codegeex-4": {
"description": "CodeGeeX-4 — это мощный AI помощник по программированию, поддерживающий интеллектуальные ответы и автозаполнение кода на различных языках программирования, повышая эффективность разработки."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ — это высокопроизводительная большая языковая модель, специально разработанная для реальных бизнес-сценариев и сложных приложений."
},
+ "dall-e-2": {
+ "description": "Вторая генерация модели DALL·E, поддерживающая более реалистичную и точную генерацию изображений с разрешением в 4 раза выше, чем у первой генерации."
+ },
+ "dall-e-3": {
+ "description": "Последняя модель DALL·E, выпущенная в ноябре 2023 года. Поддерживает более реалистичную и точную генерацию изображений с более сильной детализацией."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct предлагает высокую надежность в обработке команд, поддерживая приложения в различных отраслях."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 — это мощная и экономически эффективная языковая модель с гибридными экспертами (MoE). Она была предварительно обучена на высококачественном корпусе из 8.1 триллиона токенов и дополнительно улучшена с помощью контролируемой дообучения (SFT) и обучения с подкреплением (RL). По сравнению с DeepSeek 67B, DeepSeek-V2 обеспечивает более высокую производительность, экономя 42.5% затрат на обучение, снижая использование KV-кэша на 93.3% и увеличивая максимальную пропускную способность генерации в 5.76 раз. Эта модель поддерживает длину контекста до 128k и показывает отличные результаты в стандартных бенчмарках и оценках открытой генерации."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 — это модель вывода, управляемая методом обучения с подкреплением (RL), которая решает проблемы повторяемости и читаемости модели. Перед применением RL DeepSeek-R1 вводит данные холодного старта, что дополнительно оптимизирует производительность вывода. Она показывает сопоставимые результаты с OpenAI-o1 в математических, кодовых и задачах вывода, а также улучшает общую эффективность благодаря тщательно разработанным методам обучения."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "Модель DeepSeek-R1, дистиллированная с помощью усиленного обучения и данных холодного старта, оптимизирует производительность вывода, обновляя стандарт многозадачности в открытых моделях."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B — это дистиллированная модель, основанная на Llama-3.1-8B. Эта модель была дообучена на образцах, сгенерированных DeepSeek-R1, и демонстрирует отличные способности вывода. Она показала хорошие результаты в нескольких бенчмарках, включая 89.1% точности на MATH-500, 50.4% проходной уровень на AIME 2024 и 1205 баллов на CodeForces, демонстрируя сильные математические и программные способности для модели объемом 8B."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Модель DeepSeek-R1, дистиллированная с помощью усиленного обучения и данных холодного старта, оптимизирует производительность вывода, обновляя стандарт многозадачности в открытых моделях."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Модель DeepSeek-R1, дистиллированная с помощью усиленного обучения и данных холодного старта, оптимизирует производительность вывода, обновляя стандарт многозадачности в открытых моделях."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B — это модель, полученная с помощью дистилляции на основе Qwen2.5-32B. Эта модель была дообучена на 800000 отобранных образцах, сгенерированных DeepSeek-R1, и демонстрирует выдающуюся производительность в таких областях, как математика, программирование и логика. Она показала отличные результаты в нескольких бенчмарках, включая AIME 2024, MATH-500 и GPQA Diamond, достигнув 94.3% точности на MATH-500, демонстрируя мощные способности математического вывода."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B — это модель, полученная с помощью дистилляции на основе Qwen2.5-Math-7B. Эта модель была дообучена на 800000 отобранных образцах, сгенерированных DeepSeek-R1, и демонстрирует отличные способности вывода. Она показала выдающиеся результаты в нескольких бенчмарках, включая 92.8% точности на MATH-500, 55.5% проходной уровень на AIME 2024 и 1189 баллов на CodeForces, демонстрируя сильные математические и программные способности для модели объемом 7B."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 объединяет отличительные черты предыдущих версий, улучшая общие и кодировочные способности."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 — это языковая модель смешанных экспертов (MoE) с 6710 миллиардами параметров, использующая многоголовое потенциальное внимание (MLA) и архитектуру DeepSeekMoE, в сочетании с стратегией балансировки нагрузки без вспомогательных потерь, оптимизирующей эффективность вывода и обучения. После предобучения на 14,8 триллионах высококачественных токенов и последующей супервизионной донастройки и обучения с подкреплением, DeepSeek-V3 превосходит другие открытые модели и приближается к ведущим закрытым моделям."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B — это передовая модель, обученная для высококомплексных диалогов."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "Современная эффективная LLM, специализирующаяся на рассуждениях, математике и программировании."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 — это модель визуального языка, разработанная на основе DeepSeekMoE-27B, использующая архитектуру MoE с разреженной активацией, которая демонстрирует выдающуюся производительность при активации всего 4,5 миллиарда параметров. Эта модель показывает отличные результаты в таких задачах, как визуальные вопросы и ответы, оптическое распознавание символов, понимание документов/таблиц/графиков и визуальная локализация."
+ },
"deepseek-chat": {
"description": "Новая открытая модель, объединяющая общие и кодовые возможности, не только сохраняет общие диалоговые способности оригинальной модели Chat и мощные возможности обработки кода модели Coder, но и лучше согласуется с человеческими предпочтениями. Кроме того, DeepSeek-V2.5 значительно улучшила производительность в таких задачах, как написание текстов и следование инструкциям."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 — это открытая смешанная экспертная модель кода, показывающая отличные результаты в задачах кода, сопоставимая с GPT4-Turbo."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 — это модель вывода, управляемая методом обучения с подкреплением (RL), которая решает проблемы повторяемости и читаемости модели. Перед применением RL DeepSeek-R1 вводит данные холодного старта, что дополнительно оптимизирует производительность вывода. Она показывает сопоставимые результаты с OpenAI-o1 в математических, кодовых и задачах вывода, а также улучшает общую эффективность благодаря тщательно разработанным методам обучения."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 — более крупная и умная модель в наборе DeepSeek, была дистиллирована в архитектуру Llama 70B. На основе бенчмарков и ручной оценки эта модель более умная, особенно в задачах, требующих математической и фактической точности."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "Модели серии DeepSeek-R1-Distill были получены с помощью технологии дистилляции знаний, донастраивая образцы, сгенерированные DeepSeek-R1, на открытых моделях, таких как Qwen и Llama."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "Модели серии DeepSeek-R1-Distill были получены с помощью технологии дистилляции знаний, донастраивая образцы, сгенерированные DeepSeek-R1, на открытых моделях, таких как Qwen и Llama."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "Модели серии DeepSeek-R1-Distill были получены с помощью технологии дистилляции знаний, донастраивая образцы, сгенерированные DeepSeek-R1, на открытых моделях, таких как Qwen и Llama."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "Модели серии DeepSeek-R1-Distill были получены с помощью технологии дистилляции знаний, донастраивая образцы, сгенерированные DeepSeek-R1, на открытых моделях, таких как Qwen и Llama."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "Модели серии DeepSeek-R1-Distill были получены с помощью технологии дистилляции знаний, донастраивая образцы, сгенерированные DeepSeek-R1, на открытых моделях, таких как Qwen и Llama."
+ },
+ "deepseek-reasoner": {
+ "description": "Модель вывода, представленная DeepSeek. Перед тем как выдать окончательный ответ, модель сначала выводит цепочку размышлений, чтобы повысить точность окончательного ответа."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 — это эффективная языковая модель Mixture-of-Experts, подходящая для экономически эффективных потребностей обработки."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B — это модель кода DeepSeek, обеспечивающая мощные возможности генерации кода."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 — это модель MoE, разработанная компанией Hangzhou DeepSeek AI Technology Research Co., Ltd., которая показывает выдающиеся результаты в нескольких тестах и занимает первое место среди открытых моделей в основных рейтингах. V3 по сравнению с моделью V2.5 увеличила скорость генерации в 3 раза, обеспечивая пользователям более быстрое и плавное использование."
+ },
"deepseek/deepseek-chat": {
"description": "Новая открытая модель, объединяющая общие и кодовые возможности, не только сохраняет общие диалоговые способности оригинальной модели Chat и мощные возможности обработки кода модели Coder, но и лучше соответствует человеческим предпочтениям. Кроме того, DeepSeek-V2.5 значительно улучшила свои результаты в задачах написания, следования инструкциям и других областях."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 значительно улучшила способности модели к рассуждению при наличии лишь очень ограниченных размеченных данных. Перед тем как предоставить окончательный ответ, модель сначала выводит цепочку размышлений, чтобы повысить точность окончательного ответа."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 значительно улучшила способности модели к рассуждению при наличии лишь очень ограниченных размеченных данных. Перед тем как предоставить окончательный ответ, модель сначала выводит цепочку размышлений, чтобы повысить точность окончательного ответа."
+ },
"emohaa": {
"description": "Emohaa — это психологическая модель, обладающая профессиональными консультационными способностями, помогающая пользователям понимать эмоциональные проблемы."
},
+ "ernie-3.5-128k": {
+ "description": "Флагманская большая языковая модель, разработанная Baidu, охватывающая огромные объемы китайских и английских текстов, обладающая мощными универсальными способностями, способная удовлетворить требования большинства сценариев диалогов, генерации контента и применения плагинов; поддерживает автоматическое подключение к плагину поиска Baidu, обеспечивая актуальность информации."
+ },
+ "ernie-3.5-8k": {
+ "description": "Флагманская большая языковая модель, разработанная Baidu, охватывающая огромные объемы китайских и английских текстов, обладающая мощными универсальными способностями, способная удовлетворить требования большинства сценариев диалогов, генерации контента и применения плагинов; поддерживает автоматическое подключение к плагину поиска Baidu, обеспечивая актуальность информации."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "Флагманская большая языковая модель, разработанная Baidu, охватывающая огромные объемы китайских и английских текстов, обладающая мощными универсальными способностями, способная удовлетворить требования большинства сценариев диалогов, генерации контента и применения плагинов; поддерживает автоматическое подключение к плагину поиска Baidu, обеспечивая актуальность информации."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "Флагманская сверхбольшая языковая модель, разработанная Baidu, по сравнению с ERNIE 3.5 демонстрирует полное обновление возможностей модели, широко применима в сложных задачах различных областей; поддерживает автоматическое подключение к плагину поиска Baidu, обеспечивая актуальность информации."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "Флагманская сверхбольшая языковая модель, разработанная Baidu, по сравнению с ERNIE 3.5 демонстрирует полное обновление возможностей модели, широко применима в сложных задачах различных областей; поддерживает автоматическое подключение к плагину поиска Baidu, обеспечивая актуальность информации."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "Флагманская сверхбольшая языковая модель, разработанная Baidu, демонстрирует отличные результаты в комплексных задачах, широко применима в различных областях; поддерживает автоматическое подключение к плагину поиска Baidu, обеспечивая актуальность информации. По сравнению с ERNIE 4.0, она показывает лучшие результаты."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "Флагманская сверхбольшая языковая модель, разработанная Baidu, демонстрирует отличные результаты в комплексных задачах, широко применима в различных областях; поддерживает автоматическое подключение к плагину поиска Baidu, обеспечивая актуальность информации. По сравнению с ERNIE 4.0, она показывает лучшие результаты."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "Флагманская сверхбольшая языковая модель, разработанная Baidu, демонстрирует отличные результаты в комплексных задачах, широко применима в различных областях; поддерживает автоматическое подключение к плагину поиска Baidu, обеспечивая актуальность информации. По сравнению с ERNIE 4.0, она показывает лучшие результаты."
+ },
+ "ernie-char-8k": {
+ "description": "Специализированная большая языковая модель, разработанная Baidu, подходящая для применения в игровых NPC, диалогах службы поддержки, ролевых играх и других сценариях, с более ярким и последовательным стилем персонажей, более высокой способностью следовать инструкциям и лучшей производительностью вывода."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Специализированная большая языковая модель, разработанная Baidu, подходящая для применения в игровых NPC, диалогах службы поддержки, ролевых играх и других сценариях, с более ярким и последовательным стилем персонажей, более высокой способностью следовать инструкциям и лучшей производительностью вывода."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite — это легковесная большая языковая модель, разработанная Baidu, которая сочетает в себе отличные результаты модели и производительность вывода, подходит для использования на AI-ускорителях с низкой вычислительной мощностью."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Легковесная большая языковая модель, разработанная Baidu, которая сочетает в себе отличные результаты модели и производительность вывода, превосходя ERNIE Lite, подходит для использования на AI-ускорителях с низкой вычислительной мощностью."
+ },
+ "ernie-novel-8k": {
+ "description": "Универсальная большая языковая модель, разработанная Baidu, обладающая явными преимуществами в способности продолжать написание романов, также может использоваться в сценариях коротких пьес и фильмов."
+ },
+ "ernie-speed-128k": {
+ "description": "Новая высокопроизводительная большая языковая модель, разработанная Baidu в 2024 году, обладающая выдающимися универсальными способностями, подходит для использования в качестве базовой модели для тонкой настройки, лучше справляясь с проблемами конкретных сценариев, при этом обладая отличной производительностью вывода."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "Новая высокопроизводительная большая языковая модель, разработанная Baidu в 2024 году, обладающая выдающимися универсальными способностями, превосходя ERNIE Speed, подходит для использования в качестве базовой модели для тонкой настройки, лучше справляясь с проблемами конкретных сценариев, при этом обладая отличной производительностью вывода."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny — это сверхвысокопроизводительная большая языковая модель, стоимость развертывания и тонкой настройки которой является самой низкой среди моделей серии Wenxin."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Тюнинг) предлагает стабильную и настраиваемую производительность, что делает её идеальным выбором для решения сложных задач."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro поддерживает до 2 миллионов токенов и является идеальным выбором для средних многомодальных моделей, обеспечивая многостороннюю поддержку для сложных задач."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash предлагает функции следующего поколения и улучшения, включая выдающуюся скорость, использование встроенных инструментов, многомодальную генерацию и контекстное окно на 1M токенов."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash предлагает функции следующего поколения и улучшения, включая выдающуюся скорость, использование встроенных инструментов, многомодальную генерацию и контекстное окно на 1M токенов."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp — это новейшая экспериментальная мультимодальная AI-модель от Google, обладающая функциями следующего поколения, выдающейся скоростью, нативными инструментами и мультимодальной генерацией."
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "Модель Gemini 2.0 Flash, оптимизированная для экономической эффективности и низкой задержки."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Exp — это последняя экспериментальная многомодальная AI модель от Google, обладающая следующими поколениями характеристик, выдающейся скоростью, нативным вызовом инструментов и многомодальной генерацией."
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Exp — это последняя экспериментальная многомодальная AI-модель от Google, обладающая следующими поколениями функций, выдающейся скоростью, нативными вызовами инструментов и многомодальной генерацией."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 — это новейшая экспериментальная многомодальная ИИ модель от Google, обладающая высокой скоростью обработки и поддерживающая текстовые, изображенческие и видеовходы, что позволяет эффективно расширять применение для различных задач."
- },
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 — это последняя экспериментальная многомодальная AI модель от Google, обладающая высокой скоростью обработки, поддерживающая текстовые, графические и видеовходы, что делает её эффективной для масштабирования различных задач."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental — это последняя экспериментальная многомодальная AI модель от Google, которая демонстрирует определенное улучшение качества по сравнению с предыдущими версиями, особенно в области мировых знаний, кода и длинного контекста."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 — это новейшая экспериментальная многомодальная AI модель от Google, которая продемонстрировала определенное улучшение качества по сравнению с предыдущими версиями."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus обладает способностью понимать видео-контент и множество изображений, подходит для мультимодальных задач."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview обладает мощными способностями к сложному выводу, демонстрируя отличные результаты в области логического вывода, математики и программирования."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash предлагает функции следующего поколения и улучшения, включая выдающуюся скорость, использование встроенных инструментов, многомодальную генерацию и контекстное окно на 1M токенов."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental — это последняя экспериментальная многомодальная AI модель от Google, которая демонстрирует определенное улучшение качества по сравнению с предыдущими версиями, особенно в области мировых знаний, кода и длинного контекста."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash предлагает оптимизированные возможности многомодальной обработки, подходящие для различных сложных задач."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro сочетает в себе новейшие технологии оптимизации, обеспечивая более эффективную обработку многомодальных данных."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 — это эффективная модель, представленная Google, охватывающая широкий спектр приложений от небольших до сложных задач обработки данных."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 продолжает концепцию легковесного и эффективного дизайна."
},
"google/gemma-2-2b-it": {
"description": "Легковесная модель настройки инструкций от Google."
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 — это эффективная модель, представленная Google, охватывающая широкий спектр приложений от небольших до сложных задач обработки данных."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 — это легковесная серия текстовых моделей с открытым исходным кодом от Google."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo подходит для различных задач генерации и понимания текста, в настоящее время ссылается на gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, высокоемкий текстовый генеративный модель, подходящая для сложных задач."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo подходит для различных задач генерации и понимания текста, в настоящее время ссылается на gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o — это динамическая модель, которая обновляется в реальном времени для поддержания актуальной версии. Она сочетает в себе мощное понимание языка и генерацию текста, подходя для широкого спектра приложений, включая обслуживание клиентов, образование и техническую поддержку."
},
+ "gpt-4o-audio-preview": {
+ "description": "Модель GPT-4o Audio, поддерживающая аудиовход и аудиовыход."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini — это последняя модель, выпущенная OpenAI после GPT-4 Omni, поддерживающая ввод изображений и текстов с выводом текста. Как их самый продвинутый компактный модель, она значительно дешевле других недавних передовых моделей и более чем на 60% дешевле GPT-3.5 Turbo. Она сохраняет передовой уровень интеллекта при значительном соотношении цена-качество. GPT-4o mini набрала 82% на тесте MMLU и в настоящее время занимает более высокое место в предпочтениях чата по сравнению с GPT-4."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "Реальная версия GPT-4o-mini, поддерживающая аудио и текстовый ввод и вывод в реальном времени."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "Реальная версия GPT-4o, поддерживающая аудио и текстовый ввод и вывод в реальном времени."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "Реальная версия GPT-4o, поддерживающая аудио и текстовый ввод и вывод в реальном времени."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "Реальная версия GPT-4o, поддерживающая аудио и текстовый ввод и вывод в реальном времени."
+ },
"grok-2-1212": {
"description": "Модель улучшена в точности, соблюдении инструкций и многоязычных возможностях."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "Последняя модель Hunyuan с архитектурой MOE FunctionCall, обученная на высококачественных данных FunctionCall, с контекстным окном до 32K, занимает лидирующие позиции по множеству оценочных показателей."
},
+ "hunyuan-large": {
+ "description": "Модель Hunyuan-large имеет общее количество параметров около 389B, активных параметров около 52B, что делает её самой крупной и эффективной открытой моделью MoE с архитектурой Transformer в отрасли."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Специализируется на обработке длинных текстовых задач, таких как резюме документов и вопросы и ответы по документам, а также обладает способностью обрабатывать общие задачи генерации текста. Отлично справляется с анализом и генерацией длинных текстов, эффективно справляясь с требованиями к обработке сложного и детального длинного контента."
+ },
"hunyuan-lite": {
"description": "Обновленная версия с MOE-структурой, контекстное окно составляет 256k, она опережает множество открытых моделей в оценках по NLP, коду, математике и другим областям."
},
+ "hunyuan-lite-vision": {
+ "description": "Последняя многомодальная модель Hunyuan с 7B параметрами, окно контекста 32K, поддерживает многомодальный диалог на китайском и английском языках, распознавание объектов на изображениях, понимание документов и таблиц, многомодальную математику и т. д., по многим измерениям превосходит модели конкурентов с 7B параметрами."
+ },
"hunyuan-pro": {
"description": "Модель длинного текста с параметрами уровня триллиона MOE-32K. Она достигает абсолютного лидерства на различных бенчмарках, обладает сложными инструкциями и выводом, имеет сложные математические способности и поддерживает вызовы функций, с акцентом на оптимизацию в области многоязычного перевода, финансов, права и медицины."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Использует более оптимальную стратегию маршрутизации, одновременно смягчая проблемы с балансировкой нагрузки и сходимостью экспертов. В области длинных текстов показатель «найти иголку в стоге сена» достигает 99,9%. MOE-256K делает дальнейший прорыв в длине и качестве, значительно расширяя допустимую длину ввода."
},
+ "hunyuan-standard-vision": {
+ "description": "Последняя многомодальная модель Hunyuan, поддерживающая многоязычные ответы, с сбалансированными способностями на китайском и английском языках."
+ },
"hunyuan-turbo": {
"description": "Предварительная версия нового поколения языковой модели Hunyuan, использующая совершенно новую структуру смешанной экспертной модели (MoE), которая обеспечивает более быструю эффективность вывода и более сильные результаты по сравнению с hunyuan-pro."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Фиксированная версия hunyuan-turbo от 20 ноября 2024 года, промежуточная между hunyuan-turbo и hunyuan-turbo-latest."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "Оптимизация этой версии: масштабирование данных и инструкций, значительное повышение общей обобщающей способности модели; значительное улучшение математических, кодовых и логических способностей; оптимизация понимания текста и связанных с ним способностей понимания слов; оптимизация качества генерации контента при создании текста."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "Оптимизация общего опыта, включая понимание NLP, создание текста, общение, вопросы и ответы на знания, перевод, области и т. д.; повышение человечности, оптимизация эмоционального интеллекта модели; улучшение способности модели активно прояснять неясные намерения; повышение способности обработки вопросов, связанных с анализом слов; улучшение качества и интерактивности творчества; улучшение многократного взаимодействия."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "Флагманская модель нового поколения Hunyuan в области визуального языка, использующая совершенно новую структуру смешанной экспертной модели (MoE), с полным улучшением способностей в области базового распознавания, создания контента, вопросов и ответов на знания, анализа и вывода по сравнению с предыдущей моделью."
+ },
"hunyuan-vision": {
"description": "Последняя многомодальная модель Hunyuan, поддерживающая ввод изображений и текста для генерации текстового контента."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "Наша последняя серия моделей с выдающимися показателями вывода, поддерживающая длину контекста до 1M и обладающая улучшенными возможностями следования инструкциям и вызова инструментов."
},
+ "internlm3-latest": {
+ "description": "Наша последняя серия моделей с выдающейся производительностью вывода, лидирующая среди моделей открытого кода того же уровня. По умолчанию указывает на нашу последнюю выпущенную серию моделей InternLM3."
+ },
+ "jina-deepsearch-v1": {
+ "description": "Глубокий поиск сочетает в себе сетевой поиск, чтение и рассуждение, позволяя проводить всесторонние исследования. Вы можете рассматривать его как агента, который принимает ваши исследовательские задачи — он проводит обширный поиск и проходит через множество итераций, прежде чем предоставить ответ. Этот процесс включает в себя постоянные исследования, рассуждения и решение проблем с разных точек зрения. Это принципиально отличается от стандартных больших моделей, которые генерируют ответы непосредственно из предобученных данных, и от традиционных систем RAG, полагающихся на одноразовый поверхностный поиск."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM — это экспериментальная языковая модель, ориентированная на конкретные задачи, обученная в соответствии с принципами науки о обучении, которая может следовать системным инструкциям в учебных и образовательных сценариях, выступая в роли эксперта-наставника и т.д."
},
"lite": {
"description": "Spark Lite — это легковесная большая языковая модель с крайне низкой задержкой и высокой эффективностью обработки, полностью бесплатная и открытая, поддерживающая функции онлайн-поиска в реальном времени. Ее быстрая реакция делает ее отличным выбором для применения в устройствах с низкой вычислительной мощностью и для тонкой настройки моделей, обеспечивая пользователям отличное соотношение цены и качества, особенно в сценариях вопросов и ответов, генерации контента и поиска."
},
- "llama-3.1-70b-instruct": {
- "description": "Модель Llama 3.1 70B для команд, обладающая 70B параметрами, обеспечивает выдающуюся производительность в задачах генерации текста и выполнения команд."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B предлагает более мощные возможности ИИ вывода, подходит для сложных приложений, поддерживает огромное количество вычислительных процессов и гарантирует эффективность и точность."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B — это высокоэффективная модель, обеспечивающая быструю генерацию текста, идеально подходящая для приложений, требующих масштабной эффективности и экономичности."
},
- "llama-3.1-8b-instruct": {
- "description": "Модель Llama 3.1 8B для команд, обладающая 8B параметрами, обеспечивает эффективное выполнение задач с указаниями и предлагает высококачественные возможности генерации текста."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Модель Llama 3.1 Sonar Huge Online, обладающая 405B параметрами, поддерживает контекст длиной около 127,000 токенов, предназначена для сложных онлайн-чат-приложений."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Модель Llama 3.1 Sonar Large Chat, обладающая 70B параметрами, поддерживает контекст длиной около 127,000 токенов, подходит для сложных оффлайн-чатов."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Модель Llama 3.1 Sonar Large Online, обладающая 70B параметрами, поддерживает контекст длиной около 127,000 токенов, подходит для задач с высокой нагрузкой и разнообразными чатами."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Модель Llama 3.1 Sonar Small Chat, обладающая 8B параметрами, специально разработана для оффлайн-чатов и поддерживает контекст длиной около 127,000 токенов."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Модель Llama 3.1 Sonar Small Online, обладающая 8B параметрами, поддерживает контекст длиной около 127,000 токенов, специально разработана для онлайн-чатов и эффективно обрабатывает различные текстовые взаимодействия."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 предназначена для обработки задач, сочетающих визуальные и текстовые данные. Она демонстрирует отличные результаты в задачах описания изображений и визуального вопросно-ответного взаимодействия, преодолевая разрыв между генерацией языка и визуальным выводом."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 — это самая современная многоязычная открытая языковая модель из серии Llama, которая позволяет получить производительность, сопоставимую с 405B моделями, по очень низкой цене. Основана на структуре Transformer и улучшена с помощью контролируемой донастройки (SFT) и обучения с подкреплением на основе человеческой обратной связи (RLHF) для повышения полезности и безопасности. Ее версия с оптимизацией под инструкции специально разработана для многоязычных диалогов и показывает лучшие результаты по сравнению с множеством открытых и закрытых моделей чата на различных отраслевых бенчмарках. Дата окончания знаний — декабрь 2023 года."
+ },
"llama-3.3-70b-versatile": {
"description": "Многоязычная большая языковая модель Meta Llama 3.3 (LLM) — это предобученная и откорректированная модель генерации на 70B (текстовый ввод/текстовый вывод). Откорректированная на чистом тексте модель Llama 3.3 оптимизирована для многоязычных диалоговых задач и превосходит многие доступные открытые и закрытые модели чата по общим промышленным стандартам."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 предназначена для выполнения задач, объединяющих визуальные и текстовые данные. Она отлично справляется с задачами по описанию изображений и визуальному вопросу-ответу, преодолевая разрыв между генерацией языка и визуальным пониманием."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 — это самая современная многоязычная открытая языковая модель серии Llama, позволяющая получить производительность, сопоставимую с 405B моделью, по очень низкой цене. Основана на структуре Transformer и улучшена с помощью контролируемой донастройки (SFT) и обучения с подкреплением на основе человеческой обратной связи (RLHF) для повышения полезности и безопасности. Ее версия с оптимизацией под инструкции специально разработана для многоязычного диалога и показывает лучшие результаты по сравнению с многими открытыми и закрытыми чат-моделями на нескольких отраслевых бенчмарках. Дата окончания знаний — декабрь 2023 года."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Многоязычная большая языковая модель Meta Llama 3.3 (LLM) — это предобученная и настроенная на инструкции генеративная модель объемом 70B (входной/выходной текст). Модель Llama 3.3, настроенная на инструкции, оптимизирована для многоязычных диалоговых случаев и превосходит многие доступные открытые и закрытые модели чата по общим отраслевым бенчмаркам."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 предназначена для выполнения задач, объединяющих визуальные и текстовые данные. Она отлично справляется с задачами по описанию изображений и визуальному вопросу-ответу, преодолевая разрыв между генерацией языка и визуальным пониманием."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "Модель Llama 3.1 Turbo 405B предлагает огромную поддержку контекста для обработки больших данных и демонстрирует выдающиеся результаты в масштабных приложениях искусственного интеллекта."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 — это передовая модель, представленная Meta, поддерживающая до 405B параметров, применимая в сложных диалогах, многоязычном переводе и анализе данных."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B предлагает эффективную поддержку диалогов на нескольких языках."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 предназначена для обработки задач, сочетающих визуальные и текстовые данные. Она демонстрирует отличные результаты в задачах описания изображений и визуального вопросно-ответного взаимодействия, преодолевая разрыв между генерацией языка и визуальным выводом."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 — это самая современная многоязычная открытая языковая модель из серии Llama, которая позволяет получить производительность, сопоставимую с 405B моделями, по очень низкой цене. Основана на структуре Transformer и улучшена с помощью контролируемой донастройки (SFT) и обучения с подкреплением на основе человеческой обратной связи (RLHF) для повышения полезности и безопасности. Ее версия с оптимизацией под инструкции специально разработана для многоязычных диалогов и показывает лучшие результаты по сравнению с множеством открытых и закрытых моделей чата на различных отраслевых бенчмарках. Дата окончания знаний — декабрь 2023 года."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 — это самая современная многоязычная открытая языковая модель из серии Llama, которая позволяет получить производительность, сопоставимую с 405B моделями, по очень низкой цене. Основана на структуре Transformer и улучшена с помощью контролируемой донастройки (SFT) и обучения с подкреплением на основе человеческой обратной связи (RLHF) для повышения полезности и безопасности. Ее версия с оптимизацией под инструкции специально разработана для многоязычных диалогов и показывает лучшие результаты по сравнению с множеством открытых и закрытых моделей чата на различных отраслевых бенчмарках. Дата окончания знаний — декабрь 2023 года."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct — это самая большая и мощная модель в линейке Llama 3.1 Instruct, представляющая собой высокоразвёрнутую модель для диалогового вывода и генерации синтетических данных, также может использоваться в качестве основы для специализированного предобучения или дообучения в определённых областях. Многоязычные большие языковые модели (LLMs), предлагаемые Llama 3.1, представляют собой набор предобученных генеративных моделей с настройкой на инструкции, включая размеры 8B, 70B и 405B (вход/выход текста). Модели текста с настройкой на инструкции Llama 3.1 (8B, 70B, 405B) оптимизированы для многоязычных диалоговых случаев и превосходят многие доступные открытые модели чата в общепринятых отраслевых бенчмарках. Llama 3.1 предназначена для коммерческого и исследовательского использования на нескольких языках. Модели текста с настройкой на инструкции подходят для диалогов, похожих на помощников, в то время как предобученные модели могут адаптироваться к различным задачам генерации естественного языка. Модели Llama 3.1 также поддерживают использование их вывода для улучшения других моделей, включая генерацию синтетических данных и уточнение. Llama 3.1 является саморегрессионной языковой моделью, использующей оптимизированную архитектуру трансформеров. Настроенные версии используют контролируемое дообучение (SFT) и обучение с подкреплением с человеческой обратной связью (RLHF), чтобы соответствовать предпочтениям людей в отношении полезности и безопасности."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 — это открытая большая языковая модель (LLM), ориентированная на разработчиков, исследователей и предприятия, предназначенная для помощи в создании, экспериментировании и ответственном масштабировании их идей по генеративному ИИ. В качестве части базовой системы для инноваций глобального сообщества она идеально подходит для устройств с ограниченными вычислительными мощностями и ресурсами, а также для более быстрого времени обучения."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "Современная LLM, поддерживающая генерацию синтетических данных, дистилляцию знаний и рассуждения, подходит для чат-ботов, программирования и специализированных задач."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "Обеспечивает сложные диалоги, обладая выдающимся пониманием контекста, способностями к рассуждению и генерации текста."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "Современная передовая модель, обладающая пониманием языка, выдающимися способностями к рассуждению и генерации текста."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "Современная визуально-языковая модель, специализирующаяся на высококачественном рассуждении на основе изображений."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "Современная передовая компактная языковая модель, обладающая пониманием языка, выдающимися способностями к рассуждению и генерации текста."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "Современная передовая компактная языковая модель, обладающая пониманием языка, выдающимися способностями к рассуждению и генерации текста."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "Современная визуально-языковая модель, специализирующаяся на высококачественном рассуждении на основе изображений."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "Современная LLM, специализирующаяся на рассуждениях, математике, здравом смысле и вызовах функций."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 — это языковая модель от Microsoft AI, которая особенно хорошо справляется с сложными диалогами, многоязычностью, выводами и интеллектуальными помощниками."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K — это модель с возможностями обработки сверхдлинного контекста, подходящая для генерации очень длинных текстов, удовлетворяющая требованиям сложных задач генерации, способная обрабатывать до 128 000 токенов, идеально подходящая для научных исследований, академических и крупных документальных приложений."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Модель визуализации Kimi (включая moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview и др.) может понимать содержимое изображений, включая текст на изображениях, цвета изображений и формы объектов."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K предлагает возможности обработки контекста средней длины, способная обрабатывать 32 768 токенов, особенно подходит для генерации различных длинных документов и сложных диалогов, применяется в создании контента, генерации отчетов и диалоговых систем."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Модель визуализации Kimi (включая moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview и др.) может понимать содержимое изображений, включая текст на изображениях, цвета изображений и формы объектов."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K специально разработан для генерации коротких текстов, обладая высокой производительностью обработки, способный обрабатывать 8 192 токена, идеально подходит для кратких диалогов, стенографирования и быстрой генерации контента."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Модель визуализации Kimi (включая moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview и др.) может понимать содержимое изображений, включая текст на изображениях, цвета изображений и формы объектов."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B — это обновленная версия Nous Hermes 2, содержащая последние внутренние разработанные наборы данных."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B - это специализированная языковая модель от NVIDIA, предназначенная для повышения степени полезности ответов, генерируемых LLM, к пользовательским запросам."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B — это крупная языковая модель, созданная NVIDIA, предназначенная для повышения полезности ответов, генерируемых LLM, на запросы пользователей. Эта модель показала отличные результаты в таких бенчмарках, как Arena Hard, AlpacaEval 2 LC и GPT-4-Turbo MT-Bench, и на 1 октября 2024 года занимает первое место во всех трех автоматических тестах на согласование. Модель обучалась с использованием RLHF (в частности, REINFORCE), Llama-3.1-Nemotron-70B-Reward и HelpSteer2-Preference на основе модели Llama-3.1-70B-Instruct."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "Уникальная языковая модель, обеспечивающая непревзойденную точность и эффективность."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B — это крупная языковая модель, разработанная NVIDIA, предназначенная для повышения полезности ответов, генерируемых LLM."
+ },
"o1": {
"description": "Сосредоточена на высокоуровневом выводе и решении сложных задач, включая математические и научные задачи. Идеально подходит для приложений, требующих глубокого понимания контекста и управления рабочими процессами."
},
- "o1-2024-12-17": {
- "description": "o1 — это новая модель вывода от OpenAI, поддерживающая ввод изображений и текста с выводом текста, предназначенная для сложных задач, требующих обширных общих знаний. Эта модель имеет контекст 200K и срок знания до октября 2023 года."
- },
"o1-mini": {
"description": "o1-mini — это быстрое и экономичное модель вывода, разработанная для программирования, математики и научных приложений. Модель имеет контекст 128K и срок знания до октября 2023 года."
},
"o1-preview": {
"description": "o1 — это новая модель вывода от OpenAI, подходящая для сложных задач, требующих обширных общих знаний. Модель имеет контекст 128K и срок знания до октября 2023 года."
},
+ "o3-mini": {
+ "description": "o3-mini — это наша последняя компактная модель вывода, обеспечивающая высокий уровень интеллекта при тех же затратах и задержках, что и o1-mini."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba — это языковая модель Mamba 2, сосредоточенная на генерации кода, обеспечивающая мощную поддержку для сложных задач по коду и выводу."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini — это последняя модель от OpenAI, выпущенная после GPT-4 Omni, поддерживающая ввод изображений и текста с выводом текста. Как их самый продвинутый компактный модель, она значительно дешевле других недавних передовых моделей и более чем на 60% дешевле GPT-3.5 Turbo. Она сохраняет передовой уровень интеллекта при значительном соотношении цена-качество. GPT-4o mini набрала 82% в тесте MMLU и в настоящее время занимает более высокое место по предпочтениям в чате, чем GPT-4."
},
- "openai/o1": {
- "description": "o1 — это новая модель вывода от OpenAI, поддерживающая ввод изображений и текста с выводом текста, предназначенная для сложных задач, требующих обширных общих знаний. Эта модель имеет контекст 200K и срок знания до октября 2023 года."
- },
"openai/o1-mini": {
"description": "o1-mini — это быстрое и экономичное модель вывода, разработанная для программирования, математики и научных приложений. Модель имеет контекст 128K и срок знания до октября 2023 года."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K оснащен огромной способностью обработки контекста, способной обрабатывать до 128K контекстной информации, что делает его особенно подходящим для анализа длинных текстов и обработки долгосрочных логических связей, обеспечивая плавную и последовательную логику и разнообразную поддержку ссылок в сложных текстовых коммуникациях."
},
+ "qvq-72b-preview": {
+ "description": "Модель QVQ, разработанная командой Qwen, является экспериментальной исследовательской моделью, сосредоточенной на повышении визуальных способностей рассуждения, особенно в области математического рассуждения."
+ },
"qwen-coder-plus-latest": {
"description": "Модель кода Tongyi Qianwen."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Супер масштабная визуально-языковая модель Tongyi Qianwen. По сравнению с улучшенной версией, еще больше повышает способности визуального вывода и соблюдения инструкций, обеспечивая более высокий уровень визуального восприятия и когнитивных способностей."
},
+ "qwen-vl-ocr-latest": {
+ "description": "OCR Qwen — это специализированная модель для извлечения текста, сосредоточенная на способности извлекать текст из изображений различных типов, таких как документы, таблицы, тесты и рукописный текст. Она может распознавать множество языков, включая: китайский, английский, французский, японский, корейский, немецкий, русский, итальянский, вьетнамский и арабский."
+ },
"qwen-vl-plus-latest": {
"description": "Улучшенная версия масштабной визуально-языковой модели Tongyi Qianwen. Значительно повышает способность распознавания деталей и текста, поддерживает разрешение более миллиона пикселей и изображения с произвольным соотношением сторон."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 — это новая серия крупных языковых моделей с более сильными возможностями понимания и генерации."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "LLM, ориентированная на китайский и английский языки, охватывающая области языка, программирования, математики, рассуждений и др."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "Современная LLM, поддерживающая генерацию кода, рассуждения и исправления, охватывающая основные языки программирования."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "Мощная средняя модель кода, поддерживающая контекст длиной 32K, специализирующаяся на многоязычном программировании."
+ },
"qwen2": {
"description": "Qwen2 — это новое поколение крупномасштабной языковой модели от Alibaba, обеспечивающее отличные результаты для разнообразных приложений."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "Модель Tongyi Qwen 2.5 с открытым исходным кодом объемом 14B."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "Модель Qwen2.5 с открытым исходным кодом объемом 72B."
+ },
"qwen2.5-32b-instruct": {
"description": "Модель Tongyi Qwen 2.5 с открытым исходным кодом объемом 32B."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Модель Qwen-Math с мощными способностями решения математических задач."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "Улучшение следования инструкциям, математики, решения задач и кода, улучшение способности распознавания объектов, поддержка точного позиционирования визуальных элементов в различных форматах, поддержка понимания длинных видеофайлов (максимум 10 минут) и локализация событий на уровне секунд, способность понимать последовательность времени и скорость, поддержка управления агентами ОС или мобильными устройствами на основе аналитических и позиционных возможностей, высокая способность извлечения ключевой информации и вывода в формате Json. Эта версия является 72B, самой мощной в серии."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "Улучшение следования инструкциям, математики, решения задач и кода, улучшение способности распознавания объектов, поддержка точного позиционирования визуальных элементов в различных форматах, поддержка понимания длинных видеофайлов (максимум 10 минут) и локализация событий на уровне секунд, способность понимать последовательность времени и скорость, поддержка управления агентами ОС или мобильными устройствами на основе аналитических и позиционных возможностей, высокая способность извлечения ключевой информации и вывода в формате Json. Эта версия является 72B, самой мощной в серии."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 — это новое поколение масштабной языковой модели от Alibaba, обеспечивающее отличные результаты для разнообразных потребностей приложений."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro — это высокоинтеллектуальная LLM, выпущенная Upstage, сосредоточенная на способности следовать инструкциям на одном GPU, с оценкой IFEval выше 80. В настоящее время поддерживает английский язык, официальная версия запланирована на ноябрь 2024 года, с расширением языковой поддержки и длины контекста."
},
+ "sonar": {
+ "description": "Легковесный продукт поиска на основе контекста, быстрее и дешевле, чем Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "Расширенный продукт поиска, поддерживающий контекст поиска, сложные запросы и последующие действия."
+ },
+ "sonar-reasoning": {
+ "description": "Новый API продукт, поддерживаемый моделью вывода DeepSeek."
+ },
"step-1-128k": {
"description": "Балансирует производительность и стоимость, подходит для общих сценариев."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "Эта модель обладает мощными возможностями понимания видео."
},
+ "step-1o-vision-32k": {
+ "description": "Эта модель обладает мощными способностями к пониманию изображений. По сравнению с серией моделей step-1v, она имеет более высокую визуальную производительность."
+ },
"step-1v-32k": {
"description": "Поддерживает визуальный ввод, улучшая мультимодальный опыт взаимодействия."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Поддерживает масштабные взаимодействия контекста, подходит для сложных диалоговых сценариев."
},
+ "step-2-mini": {
+ "description": "Супербыстрая большая модель на основе новой самодельной архитектуры внимания MFA, достигающая аналогичных результатов, как step1, при очень низких затратах, одновременно обеспечивая более высокую пропускную способность и более быстрое время отклика. Способна обрабатывать общие задачи и обладает особыми навыками в кодировании."
+ },
+ "taichu2_mm": {
+ "description": "Объединяет способности понимания изображений, переноса знаний, логической атрибуции и демонстрирует выдающиеся результаты в области вопросов и ответов на основе текста и изображений."
+ },
"taichu_llm": {
"description": "Модель языка TaiChu обладает выдающимися способностями к пониманию языка, а также к созданию текстов, ответам на вопросы, программированию, математическим вычислениям, логическому выводу, анализу эмоций и резюмированию текстов. Инновационно сочетает предобучение на больших данных с богатством многопоточных знаний, постоянно совершенствуя алгоритмические технологии и поглощая новые знания о словах, структуре, грамматике и семантике из огромных объемов текстовых данных, обеспечивая пользователям более удобную информацию и услуги, а также более интеллектуальный опыт."
},
+ "text-embedding-3-large": {
+ "description": "Самая мощная модель векторизации, подходящая для английских и неанглийских задач."
+ },
+ "text-embedding-3-small": {
+ "description": "Эффективная и экономичная новая генерация модели Embedding, подходящая для поиска знаний, приложений RAG и других сценариев."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) обеспечивает повышенные вычислительные возможности благодаря эффективным стратегиям и архитектуре модели."
},
+ "tts-1": {
+ "description": "Последняя модель преобразования текста в речь, оптимизированная для скорости в реальных сценариях."
+ },
+ "tts-1-hd": {
+ "description": "Последняя модель преобразования текста в речь, оптимизированная для качества."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) подходит для детализированных командных задач, обеспечивая отличные возможности обработки языка."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet устанавливает новые отраслевые стандарты, превосходя модели конкурентов и Claude 3 Opus, демонстрируя отличные результаты в широком спектре оценок, при этом обладая скоростью и стоимостью наших моделей среднего уровня."
},
+ "whisper-1": {
+ "description": "Универсальная модель распознавания речи, поддерживающая многоязычное распознавание речи, перевод речи и распознавание языка."
+ },
"wizardlm2": {
"description": "WizardLM 2 — это языковая модель, предоставляемая Microsoft AI, которая особенно хорошо проявляет себя в сложных диалогах, многоязычных задачах, выводе и интеллектуальных помощниках."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Модель для сложных визуальных задач, обеспечивающая высокую производительность в понимании и анализе изображений."
+ },
+ "yi-vision-v2": {
+ "description": "Модель для сложных визуальных задач, обеспечивающая высокопроизводительное понимание и анализ на основе нескольких изображений."
}
}
diff --git a/locales/ru-RU/providers.json b/locales/ru-RU/providers.json
index fd99f05b94b42..8f24493f8d636 100644
--- a/locales/ru-RU/providers.json
+++ b/locales/ru-RU/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure предлагает множество передовых AI-моделей, включая GPT-3.5 и новейшую серию GPT-4, поддерживающих различные типы данных и сложные задачи, с акцентом на безопасность, надежность и устойчивые AI-решения."
},
+ "azureai": {
+ "description": "Azure предлагает множество современных AI моделей, включая GPT-3.5 и последнюю серию GPT-4, поддерживающих различные типы данных и сложные задачи, нацеленных на безопасные, надежные и устойчивые AI решения."
+ },
"baichuan": {
"description": "Baichuan Intelligent — это компания, сосредоточенная на разработке больших моделей искусственного интеллекта, чьи модели показывают выдающиеся результаты в области китайских задач, таких как знаниевые энциклопедии, обработка длинных текстов и генерация контента, превосходя зарубежные модели. Baichuan Intelligent также обладает передовыми мультимодальными возможностями и показала отличные результаты в нескольких авторитетных оценках. Их модели включают Baichuan 4, Baichuan 3 Turbo и Baichuan 3 Turbo 128k, оптимизированные для различных сценариев применения, предлагая высокоэффективные решения."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek — это компания, сосредоточенная на исследованиях и применении технологий искусственного интеллекта, ее последняя модель DeepSeek-V2.5 объединяет возможности общего диалога и обработки кода, достигнув значительных улучшений в области согласования с человеческими предпочтениями, написания текстов и выполнения инструкций."
},
+ "doubao": {
+ "description": "Модель большого размера, разработанная ByteDance. Проверенная на более чем 50 внутренних бизнес-сценариях, с ежедневным использованием триллионов токенов, она продолжает совершенствоваться, предлагая множество модальных возможностей и создавая богатый бизнес-опыт для компаний с помощью качественных моделей."
+ },
"fireworksai": {
"description": "Fireworks AI — это ведущий поставщик высококлассных языковых моделей, сосредоточенный на вызовах функций и мультимодальной обработке. Их последняя модель Firefunction V2 основана на Llama-3 и оптимизирована для вызовов функций, диалогов и выполнения инструкций. Модель визуального языка FireLLaVA-13B поддерживает смешанный ввод изображений и текста. Другие заметные модели включают серию Llama и серию Mixtral, предлагая эффективную поддержку многоязычных инструкций и генерации."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "Открытая организация, занимающаяся исследованием и разработкой инструментов для больших моделей. Предоставляет всем разработчикам ИИ эффективную и удобную открытую платформу, позволяя получить доступ к самым современным технологиям больших моделей и алгоритмов."
},
+ "jina": {
+ "description": "Jina AI была основана в 2020 году и является ведущей компанией в области поискового AI. Наша платформа поискового базиса включает векторные модели, реорганизаторы и небольшие языковые модели, которые помогают предприятиям создавать надежные и высококачественные генеративные AI и мультимодальные поисковые приложения."
+ },
+ "lmstudio": {
+ "description": "LM Studio — это настольное приложение для разработки и экспериментов с LLM на вашем компьютере."
+ },
"minimax": {
"description": "MiniMax — это компания по разработке универсального искусственного интеллекта, основанная в 2021 году, стремящаяся к совместному созданию интеллекта с пользователями. MiniMax самостоятельно разработала универсальные большие модели различных модальностей, включая текстовые модели с триллионом параметров, модели речи и модели изображений. Также были запущены приложения, такие как Conch AI."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI — это платформа, предлагающая API-сервисы для различных больших языковых моделей и генерации изображений AI, гибкая, надежная и экономически эффективная. Она поддерживает новейшие открытые модели, такие как Llama3, Mistral и предоставляет комплексные, удобные для пользователя и автоматически масштабируемые API-решения для разработки генеративных AI-приложений, подходящие для быстрого роста AI-стартапов."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ предоставляет контейнеры для самообслуживания GPU-ускоренного вывода микросервисов, поддерживающих развертывание предобученных и пользовательских AI моделей в облаке, центрах обработки данных, на персональных компьютерах RTX™ AI и рабочих станциях."
+ },
"ollama": {
"description": "Модели, предлагаемые Ollama, охватывают широкий спектр областей, включая генерацию кода, математические вычисления, многоязыковую обработку и диалоговое взаимодействие, поддерживая разнообразные потребности в развертывании на уровне предприятий и локализации."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "Новая генерация мультимодальных больших моделей, разработанная Институтом автоматизации Китайской академии наук и Институтом искусственного интеллекта Уханя, поддерживает многораундные вопросы и ответы, создание текстов, генерацию изображений, 3D-понимание, анализ сигналов и другие комплексные задачи, обладая более сильными когнитивными, понимательными и творческими способностями, предлагая новый опыт взаимодействия."
},
+ "tencentcloud": {
+ "description": "Атомные возможности движка знаний (LLM Knowledge Engine Atomic Power) основаны на разработке движка знаний и представляют собой полную цепочку возможностей для вопросов и ответов, ориентированную на предприятия и разработчиков. Вы можете создать собственный сервис модели, используя различные атомные возможности, комбинируя такие услуги, как анализ документов, разбиение, встраивание, многократное переписывание и другие, чтобы настроить уникальный AI-бизнес для вашей компании."
+ },
"togetherai": {
"description": "Together AI стремится достичь передовых результатов с помощью инновационных AI-моделей, предлагая широкий спектр возможностей для настройки, включая поддержку быстрого масштабирования и интуитивно понятные процессы развертывания, чтобы удовлетворить различные потребности бизнеса."
},
"upstage": {
"description": "Upstage сосредоточен на разработке AI-моделей для различных бизнес-потребностей, включая Solar LLM и документальный AI, с целью достижения искусственного общего интеллекта (AGI). Создавайте простые диалоговые агенты через Chat API и поддерживайте вызовы функций, переводы, встраивания и приложения в конкретных областях."
},
+ "vllm": {
+ "description": "vLLM — это быстрая и простая в использовании библиотека для вывода и обслуживания LLM."
+ },
+ "volcengine": {
+ "description": "Платформа разработки сервисов больших моделей, запущенная ByteDance, предлагает функционально богатые, безопасные и конкурентоспособные по цене услуги вызова моделей, а также предоставляет полные функции от данных моделей, тонкой настройки, вывода до оценки, обеспечивая всестороннюю поддержку разработки ваших AI приложений."
+ },
"wenxin": {
"description": "Корпоративная платформа для разработки и обслуживания крупных моделей и нативных приложений ИИ, предлагающая самый полный и удобный инструментарий для разработки генеративных моделей искусственного интеллекта и полного процесса разработки приложений."
},
diff --git a/locales/ru-RU/setting.json b/locales/ru-RU/setting.json
index 9d92f97bc079a..1ec5967d46e81 100644
--- a/locales/ru-RU/setting.json
+++ b/locales/ru-RU/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Включить ограничение максимального количества токенов"
},
+ "enableReasoningEffort": {
+ "title": "Включить настройку интенсивности вывода"
+ },
"frequencyPenalty": {
- "desc": "Чем выше значение, тем меньше вероятность повторения слов",
- "title": "Штраф за повторение"
+ "desc": "Чем больше значение, тем разнообразнее и богаче словарный запас; чем меньше значение, тем проще и понятнее слова",
+ "title": "Разнообразие словарного запаса"
},
"maxTokens": {
"desc": "Максимальное количество токенов для одного взаимодействия",
@@ -212,19 +215,31 @@
"desc": "{{provider}} модель",
"title": "Модель"
},
+ "params": {
+ "title": "Расширенные параметры"
+ },
"presencePenalty": {
- "desc": "Чем выше значение, тем больше вероятность перехода на новые темы",
- "title": "Штраф за однообразие"
+ "desc": "Чем больше значение, тем больше склонность к различным выражениям, избегая повторения концепций; чем меньше значение, тем больше склонность к использованию повторяющихся концепций или нарративов, выражение становится более последовательным",
+ "title": "Разнообразие выражений"
+ },
+ "reasoningEffort": {
+ "desc": "Чем больше значение, тем сильнее способность вывода, но это может увеличить время отклика и потребление токенов",
+ "options": {
+ "high": "Высокий",
+ "low": "Низкий",
+ "medium": "Средний"
+ },
+ "title": "Интенсивность вывода"
},
"temperature": {
- "desc": "Чем выше значение, тем более непредсказуемым будет ответ",
- "title": "Непредсказуемость",
- "titleWithValue": "Непредсказуемость {{value}}"
+ "desc": "Чем больше значение, тем более креативными и воображаемыми будут ответы; чем меньше значение, тем более строгими будут ответы",
+ "title": "Креативность",
+ "warning": "Слишком высокое значение креативности может привести к искажению вывода"
},
"title": "Настройки модели",
"topP": {
- "desc": "Похоже на непредсказуемость, но не изменяется вместе с параметром непредсказуемости",
- "title": "Верхний процент P"
+ "desc": "Сколько возможностей учитывать, чем больше значение, тем больше возможных ответов принимается; чем меньше значение, тем больше склонность к выбору наиболее вероятного ответа. Не рекомендуется изменять вместе с креативностью",
+ "title": "Открытость мышления"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Общие настройки",
"experiment": "Эксперимент",
"llm": "Языковая модель",
+ "provider": "Поставщик ИИ услуг",
"sync": "Синхронизация с облаком",
"system-agent": "Системный агент",
"tts": "Голосовые услуги"
diff --git a/locales/tr-TR/changelog.json b/locales/tr-TR/changelog.json
index 8ca7fee5b03f0..a4f3b74479176 100644
--- a/locales/tr-TR/changelog.json
+++ b/locales/tr-TR/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "Tüm güncelleme günlüklerini görüntüle",
"description": "{{appName}}'in yeni özelliklerini ve iyileştirmelerini sürekli takip edin",
"pagination": {
- "older": "Geçmiş değişiklikleri görüntüle",
- "prev": "Önceki sayfa"
+ "next": "Sonraki Sayfa",
+ "older": "Geçmiş değişiklikleri görüntüle"
},
"readDetails": "Detayları okuyun",
"title": "Güncelleme Günlüğü",
diff --git a/locales/tr-TR/common.json b/locales/tr-TR/common.json
index 4e25f113b807c..99de1c3fc6730 100644
--- a/locales/tr-TR/common.json
+++ b/locales/tr-TR/common.json
@@ -275,6 +275,7 @@
},
"temp": "Geçici",
"terms": "Hizmet Koşulları",
+ "update": "Güncelle",
"updateAgent": "Asistan Bilgilerini Güncelle",
"upgradeVersion": {
"action": "Güncelle",
@@ -285,6 +286,7 @@
"anonymousNickName": "Anonim Kullanıcı",
"billing": "Fatura Yönetimi",
"cloud": "{{name}}'i Deneyin",
+ "community": "Topluluk Sürümü",
"data": "Veri Depolama",
"defaultNickname": "Topluluk Kullanıcısı",
"discord": "Topluluk Destek",
@@ -294,7 +296,6 @@
"help": "Yardım Merkezi",
"moveGuide": "Ayarlar düğmesini buraya taşıyın",
"plans": "Planlar",
- "preview": "Önizleme",
"profile": "Hesap Yönetimi",
"setting": "Uygulama Ayarları",
"usages": "Kullanım İstatistikleri"
diff --git a/locales/tr-TR/components.json b/locales/tr-TR/components.json
index 0ad79d37cc9a0..d27c4e92eb7cb 100644
--- a/locales/tr-TR/components.json
+++ b/locales/tr-TR/components.json
@@ -76,6 +76,7 @@
"custom": "Özel model, varsayılan olarak hem fonksiyon çağrısını hem de görüntü tanımayı destekler, yukarıdaki yeteneklerin kullanılabilirliğini doğrulamak için lütfen gerçek durumu kontrol edin",
"file": "Bu model dosya yükleme ve tanımayı destekler",
"functionCall": "Bu model fonksiyon çağrısını destekler",
+ "reasoning": "Bu model derin düşünmeyi destekler",
"tokens": "Bu model tek bir oturumda en fazla {{tokens}} Token destekler",
"vision": "Bu model görüntü tanımıyı destekler"
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "Etkinleştirilmiş model bulunmamaktadır, lütfen ayarlara giderek açın",
"provider": "Sağlayıcı"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "Tarayıcı güvenlik kısıtlamaları nedeniyle, Ollama'yı düzgün bir şekilde kullanabilmek için çapraz alan yapılandırması yapmanız gerekmektedir.",
+ "linux": {
+ "env": "[Service] bölümüne `Environment` ekleyin ve OLLAMA_ORIGINS ortam değişkenini ekleyin:",
+ "reboot": "systemd'yi yeniden yükleyin ve Ollama'yı yeniden başlatın",
+ "systemd": "ollama hizmetini düzenlemek için systemd'yi çağırın:"
+ },
+ "macos": "Lütfen 'Terminal' uygulamasını açın, aşağıdaki komutu yapıştırın ve çalıştırmak için Enter tuşuna basın",
+ "reboot": "İşlem tamamlandıktan sonra Ollama hizmetini yeniden başlatın",
+ "title": "Ollama'nın çapraz alan erişimine izin vermek için yapılandırma",
+ "windows": "Windows'ta, 'Denetim Masası'na tıklayın ve sistem ortam değişkenlerini düzenleyin. Kullanıcı hesabınız için 'OLLAMA_ORIGINS' adında bir ortam değişkeni oluşturun, değeri * olarak ayarlayın ve 'Tamam/Uygula'ya tıklayarak kaydedin."
+ },
+ "install": {
+ "description": "Lütfen Ollama'nın açık olduğundan emin olun, eğer Ollama'yı indirmediyseniz, lütfen resmi web sitesinden <1>indirin",
+ "docker": "Eğer Docker kullanmayı tercih ediyorsanız, Ollama'nın resmi Docker imajı da mevcuttur, aşağıdaki komutla çekebilirsiniz:",
+ "linux": {
+ "command": "Aşağıdaki komutla kurulum yapın:",
+ "manual": "Alternatif olarak, <1>Linux Manuel Kurulum Kılavuzu'na başvurarak kendiniz de kurulum yapabilirsiniz."
+ },
+ "title": "Ollama uygulamasını yerel olarak kurun ve başlatın",
+ "windowsTab": "Windows (önizleme sürümü)"
+ }
+ },
+ "Thinking": {
+ "thinking": "Derin düşünme aşamasında...",
+ "thought": "Derinlemesine düşündüm (geçen süre {{duration}} saniye)",
+ "thoughtWithDuration": "Derinlemesine düşündüm"
}
}
diff --git a/locales/tr-TR/discover.json b/locales/tr-TR/discover.json
index 0896540a6138b..d75758a6c4e27 100644
--- a/locales/tr-TR/discover.json
+++ b/locales/tr-TR/discover.json
@@ -126,6 +126,10 @@
"title": "Konu Tazeliği"
},
"range": "Aralık",
+ "reasoning_effort": {
+ "desc": "Bu ayar, modelin yanıt üretmeden önceki akıl yürütme gücünü kontrol etmek için kullanılır. Düşük güç, yanıt hızını önceliklendirir ve Token tasarrufu sağlar; yüksek güç ise daha kapsamlı bir akıl yürütme sunar, ancak daha fazla Token tüketir ve yanıt hızını düşürür. Varsayılan değer orta seviyedir, akıl yürütme doğruluğu ile yanıt hızı arasında bir denge sağlar.",
+ "title": "Akıl Yürütme Gücü"
+ },
"temperature": {
"desc": "Bu ayar, modelin yanıtlarının çeşitliliğini etkiler. Daha düşük değerler daha öngörülebilir ve tipik yanıtlar verirken, daha yüksek değerler daha çeşitli ve nadir yanıtları teşvik eder. Değer 0 olarak ayarlandığında, model belirli bir girdi için her zaman aynı yanıtı verir.",
"title": "Rastgelelik"
diff --git a/locales/tr-TR/modelProvider.json b/locales/tr-TR/modelProvider.json
index e8b78d143ca33..9cecf54a9e732 100644
--- a/locales/tr-TR/modelProvider.json
+++ b/locales/tr-TR/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API Key"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Azure API sürümü, YYYY-AA-GG formatına uymaktadır, [en son sürümü](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions) kontrol edin",
+ "fetch": "Listeyi al",
+ "title": "Azure API Sürümü"
+ },
+ "endpoint": {
+ "desc": "Azure AI proje özetinden Azure AI model çıkarım uç noktasını bulun",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Azure AI Uç Noktası"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Azure AI proje özetinden API anahtarını bulun",
+ "placeholder": "Azure Anahtarı",
+ "title": "Anahtar"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "AWS Access Key Id girin",
@@ -63,6 +81,46 @@
"title": "Cloudflare Hesap ID / API Adresi"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Lütfen API Anahtarınızı girin",
+ "title": "API Anahtarı"
+ },
+ "basicTitle": "Temel Bilgiler",
+ "configTitle": "Yapılandırma Bilgileri",
+ "confirm": "Yeni Oluştur",
+ "createSuccess": "Başarıyla oluşturuldu",
+ "description": {
+ "placeholder": "Hizmet sağlayıcı tanımı (isteğe bağlı)",
+ "title": "Hizmet Sağlayıcı Tanımı"
+ },
+ "id": {
+ "desc": "Hizmet sağlayıcının benzersiz kimliği, oluşturulduktan sonra değiştirilemez",
+ "format": "Sadece rakamlar, küçük harfler, tire (-) ve alt çizgi (_) içerebilir",
+ "placeholder": "Küçük harflerle yazılması önerilir, örneğin openai, oluşturduktan sonra değiştirilemez",
+ "required": "Lütfen hizmet sağlayıcı ID'sini girin",
+ "title": "Hizmet Sağlayıcı ID"
+ },
+ "logo": {
+ "required": "Lütfen geçerli bir hizmet sağlayıcı logosu yükleyin",
+ "title": "Hizmet Sağlayıcı Logosu"
+ },
+ "name": {
+ "placeholder": "Lütfen hizmet sağlayıcının gösterim adını girin",
+ "required": "Lütfen hizmet sağlayıcı adını girin",
+ "title": "Hizmet Sağlayıcı Adı"
+ },
+ "proxyUrl": {
+ "required": "Lütfen proxy adresini girin",
+ "title": "Proxy Adresi"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Lütfen SDK türünü seçin",
+ "title": "İstek Formatı"
+ },
+ "title": "Özel AI Hizmet Sağlayıcısı Oluştur"
+ },
"github": {
"personalAccessToken": {
"desc": "Github PAT'nizi girin, [buraya](https://github.com/settings/tokens) tıklayarak oluşturun",
@@ -77,6 +135,23 @@
"title": "HuggingFace Token"
}
},
+ "list": {
+ "title": {
+ "disabled": "Hizmet sağlayıcı devre dışı",
+ "enabled": "Hizmet sağlayıcı etkin"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Özel Hizmet Sağlayıcı Ekle",
+ "all": "Tümü",
+ "list": {
+ "disabled": "Devre Dışı",
+ "enabled": "Aktif"
+ },
+ "notFound": "Arama sonuçları bulunamadı",
+ "searchProviders": "Hizmet sağlayıcıları ara...",
+ "sort": "Özel Sıralama"
+ },
"ollama": {
"checker": {
"desc": "Proxy adresinin doğru girilip girilmediğini test edin",
@@ -94,33 +169,9 @@
"title": "正在下载模型 {{model}} "
},
"endpoint": {
- "desc": "Ollama arayüz proxy adresini girin, yerel olarak belirtilmemişse boş bırakılabilir",
+ "desc": "http(s):// içermelidir, yerel olarak belirtilmemişse boş bırakılabilir",
"title": "Arayüz Proxy Adresi"
},
- "setup": {
- "cors": {
- "description": "Ollama'nın normal şekilde çalışabilmesi için, tarayıcı güvenlik kısıtlamaları nedeniyle Ollama'nın çapraz kaynak isteklerine izin verilmesi gerekmektedir.",
- "linux": {
- "env": "[Service] bölümüne `Environment` ekleyerek OLLAMA_ORIGINS ortam değişkenini ekleyin:",
- "reboot": "systemd'yi yeniden yükleyin ve Ollama'yı yeniden başlatın",
- "systemd": "systemd'yi çağırarak ollama servisini düzenleyin:"
- },
- "macos": "Lütfen 'Terminal' uygulamasını açın ve aşağıdaki komutu yapıştırıp Enter tuşuna basın",
- "reboot": "Komut tamamlandıktan sonra Ollama servisini yeniden başlatın",
- "title": "Ollama'nın çapraz kaynak erişimine izin vermek için yapılandırma",
- "windows": "Windows'ta, 'Control Panel'ı tıklayarak sistem ortam değişkenlerini düzenleyin. Kullanıcı hesabınıza * değerinde 'OLLAMA_ORIGINS' adında bir ortam değişkeni oluşturun ve 'OK/Apply' düğmesine tıklayarak kaydedin"
- },
- "install": {
- "description": "Ollama'nın açık olduğundan emin olun. Ollama'yı indirmediyseniz, lütfen resmi web sitesine giderek <1>indirin.",
- "docker": "Docker kullanmayı tercih ediyorsanız, Ollama resmi Docker görüntüsünü aşağıdaki komutla çekebilirsiniz:",
- "linux": {
- "command": "Aşağıdaki komutları kullanarak yükleyin:",
- "manual": "Ya da, <1>Linux için el ile kurulum kılavuzuna bakarak kendiniz kurabilirsiniz"
- },
- "title": "Yerel olarak Ollama uygulamasını kurun ve başlatın",
- "windowsTab": "Windows (Önizleme)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "取消下载",
@@ -131,25 +182,145 @@
"title": "下载指定的 Ollama 模型"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Baidu Qianfan platform için Access Key girin",
- "placeholder": "Qianfan Access Key",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "Anahtarınız ve proxy adresi gibi bilgiler <1>AES-GCM şifreleme algoritması ile şifrelenecektir",
+ "apiKey": {
+ "desc": "{{name}} API Anahtarınızı girin",
+ "placeholder": "{{name}} API Anahtarı",
+ "title": "API Anahtarı"
+ },
+ "baseURL": {
+ "desc": "http(s):// içermelidir",
+ "invalid": "Lütfen geçerli bir URL girin",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "API Proxy Adresi"
+ },
+ "checker": {
+ "button": "Kontrol Et",
+ "desc": "API Anahtarı ve proxy adresinin doğru girilip girilmediğini test edin",
+ "pass": "Kontrol başarılı",
+ "title": "Bağlantı Kontrolü"
+ },
+ "fetchOnClient": {
+ "desc": "İstemci istek modu, tarayıcıdan doğrudan oturum isteği başlatır, yanıt hızını artırabilir",
+ "title": "İstemci İstek Modunu Kullan"
+ },
+ "helpDoc": "Yapılandırma Kılavuzu",
+ "waitingForMore": "Daha fazla model <1>planlanıyor, lütfen bekleyin"
},
- "checker": {
- "desc": "AccessKey / SecretAccess'ın doğru girilip girilmediğini test et"
+ "createNew": {
+ "title": "Özel AI Modeli Oluştur"
+ },
+ "item": {
+ "config": "Modeli Yapılandır",
+ "customModelCards": {
+ "addNew": "{{id}} modelini oluştur ve ekle",
+ "confirmDelete": "Bu özel modeli silmek üzeresiniz, silindikten sonra geri alınamaz, lütfen dikkatli olun."
+ },
+ "delete": {
+ "confirm": "{{displayName}} modelini silmek istediğinize emin misiniz?",
+ "success": "Silme işlemi başarılı",
+ "title": "Modeli Sil"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Azure OpenAI'de gerçek istek için alan",
+ "placeholder": "Lütfen Azure'daki model dağıtım adını girin",
+ "title": "Model Dağıtım Adı"
+ },
+ "deployName": {
+ "extra": "Bu alan, isteği gönderirken model kimliği olarak kullanılacaktır",
+ "placeholder": "Modelin gerçek dağıtım adını veya kimliğini girin",
+ "title": "Model Dağıtım Adı"
+ },
+ "displayName": {
+ "placeholder": "Lütfen modelin gösterim adını girin, örneğin ChatGPT, GPT-4 vb.",
+ "title": "Model Gösterim Adı"
+ },
+ "files": {
+ "extra": "Mevcut dosya yükleme uygulaması yalnızca bir Hack çözümüdür, yalnızca denemek için geçerlidir. Tam dosya yükleme yeteneği için lütfen sonraki uygulamayı bekleyin.",
+ "title": "Dosya Yüklemeyi Destekle"
+ },
+ "functionCall": {
+ "extra": "Bu yapılandırma, modelin araçları kullanma yeteneğini açacak ve böylece modele araç sınıfı eklentileri eklenebilecektir. Ancak, gerçek araç kullanımı tamamen modele bağlıdır, kullanılabilirliğini kendiniz test etmelisiniz.",
+ "title": "Araç kullanımını destekle"
+ },
+ "id": {
+ "extra": "Oluşturulduktan sonra değiştirilemez, AI çağrıldığında model kimliği olarak kullanılacaktır",
+ "placeholder": "Model kimliğini girin, örneğin gpt-4o veya claude-3.5-sonnet",
+ "title": "Model ID"
+ },
+ "modalTitle": "Özel Model Yapılandırması",
+ "reasoning": {
+ "extra": "Bu yapılandırma yalnızca modelin derin düşünme yeteneğini açacaktır, belirli etkiler tamamen modelin kendisine bağlıdır, lütfen bu modelin kullanılabilir derin düşünme yeteneğine sahip olup olmadığını kendiniz test edin",
+ "title": "Derin düşünmeyi destekler"
+ },
+ "tokens": {
+ "extra": "Modelin desteklediği maksimum Token sayısını ayarlayın",
+ "title": "Maksimum bağlam penceresi",
+ "unlimited": "Sınırsız"
+ },
+ "vision": {
+ "extra": "Bu yapılandırma yalnızca uygulamadaki resim yükleme yapılandırmasını açacaktır, tanıma desteği tamamen modele bağlıdır, lütfen bu modelin görsel tanıma yeteneğini test edin.",
+ "title": "Görsel Tanımayı Destekle"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/Resim",
+ "inputCharts": "${{amount}}/M Karakter",
+ "inputMinutes": "${{amount}}/Dakika",
+ "inputTokens": "Girdi ${{amount}}/M",
+ "outputTokens": "Çıktı ${{amount}}/M"
+ },
+ "releasedAt": "Yayınlanma tarihi: {{releasedAt}}"
},
- "secretKey": {
- "desc": "Baidu Qianfan platform için Secret Key girin",
- "placeholder": "Qianfan Secret Key",
- "title": "Secret Key"
+ "list": {
+ "addNew": "Model Ekle",
+ "disabled": "Devre dışı",
+ "disabledActions": {
+ "showMore": "Hepsini Göster"
+ },
+ "empty": {
+ "desc": "Lütfen özel bir model oluşturun veya kullanmaya başlamadan önce bir model çekin",
+ "title": "Kullanılabilir model yok"
+ },
+ "enabled": "Etkin",
+ "enabledActions": {
+ "disableAll": "Hepsini devre dışı bırak",
+ "enableAll": "Hepsini etkinleştir",
+ "sort": "Özel model sıralaması"
+ },
+ "enabledEmpty": "Etkin model yok, lütfen aşağıdaki listeden beğendiğiniz modeli etkinleştirin~",
+ "fetcher": {
+ "clear": "Alınan modelleri temizle",
+ "fetch": "Model listesini al",
+ "fetching": "Model listesi alınıyor...",
+ "latestTime": "Son güncelleme zamanı: {{time}}",
+ "noLatestTime": "Henüz liste alınmadı"
+ },
+ "resetAll": {
+ "conform": "Mevcut modelin tüm değişikliklerini sıfırlamak istediğinize emin misiniz? Sıfırladıktan sonra mevcut model listesi varsayılan duruma dönecektir.",
+ "success": "Sıfırlama başarılı",
+ "title": "Tüm değişiklikleri sıfırla"
+ },
+ "search": "Model ara...",
+ "searchResult": "{{count}} model bulundu",
+ "title": "Model Listesi",
+ "total": "Toplam {{count}} adet model mevcut"
},
- "unlock": {
- "customRegion": "Özel hizmet bölgesi",
- "description": "Oturuma başlamak için AccessKey / SecretKey'nizi girin. Uygulama kimlik doğrulama yapılandırmanızı kaydetmeyecek",
- "title": "Özel Wenxin Yiyan kimlik doğrulama bilgilerini kullan"
- }
+ "searchNotFound": "Arama sonuçları bulunamadı"
+ },
+ "sortModal": {
+ "success": "Sıralama güncellemesi başarılı",
+ "title": "Özel Sıralama",
+ "update": "Güncelle"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "Bu AI hizmet sağlayıcısını silmek üzeresiniz, silindikten sonra geri alınamaz, silmek istediğinize emin misiniz?",
+ "deleteSuccess": "Silme işlemi başarılı",
+ "tooltip": "Hizmet sağlayıcının temel yapılandırmasını güncelle",
+ "updateSuccess": "Güncelleme başarılı"
},
"zeroone": {
"title": "01.AI Sıfır Bir"
diff --git a/locales/tr-TR/models.json b/locales/tr-TR/models.json
index fddbc868d5e49..d792f2ba79f12 100644
--- a/locales/tr-TR/models.json
+++ b/locales/tr-TR/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K, anlam güvenliği ve sorumluluk odaklılığı vurgular, içerik güvenliği konusunda yüksek gereksinimlere sahip uygulama senaryoları için tasarlanmıştır, kullanıcı deneyiminin doğruluğunu ve sağlamlığını garanti eder."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1, düşünce zincirini ağaç arama ile inşa eder ve yansıtma mekanizmasını entegre eder, pekiştirme öğrenimi ile eğitilir, model kendini yansıtma ve hata düzeltme yeteneğine sahiptir."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro, 360 şirketi tarafından sunulan yüksek düzeyde doğal dil işleme modelidir, mükemmel metin oluşturma ve anlama yeteneğine sahiptir, özellikle oluşturma ve yaratma alanında olağanüstü performans gösterir, karmaşık dil dönüşümleri ve rol canlandırma görevlerini işleyebilir."
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1, düşünce zincirini oluşturmak için ağaç araması kullanır ve yansıtma mekanizmasını entegre eder, pekiştirme öğrenimi ile eğitilir, model kendini yansıtma ve hata düzeltme yeteneğine sahiptir."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra, Xinghuo büyük model serisinin en güçlü versiyonudur, çevrimiçi arama bağlantısını yükseltirken, metin içeriğini anlama ve özetleme yeteneğini artırır. Ofis verimliliğini artırmak ve taleplere doğru yanıt vermek için kapsamlı bir çözüm sunar, sektördeki akıllı ürünlerin öncüsüdür."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "Model yetenekleri ülke içinde birinci, bilgi ansiklopedisi, uzun metinler, yaratıcı üretim gibi Çince görevlerde uluslararası ana akım modelleri aşmaktadır. Ayrıca, sektörde lider çok modlu yeteneklere sahip olup, birçok yetkili değerlendirme ölçütünde mükemmel performans sergilemektedir."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Qwen2.5-Math-1.5B temel alınarak oluşturulmuş DeepSeek-R1 damıtma modeli, pekiştirme öğrenimi ve soğuk başlatma verileri ile çıkarım performansını optimize eder, açık kaynak model çoklu görev standartlarını yeniler."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Qwen2.5-14B temel alınarak oluşturulmuş DeepSeek-R1 damıtma modeli, pekiştirme öğrenimi ve soğuk başlatma verileri ile çıkarım performansını optimize eder, açık kaynak model çoklu görev standartlarını yeniler."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1 serisi, pekiştirme öğrenimi ve soğuk başlatma verileri ile çıkarım performansını optimize eder, açık kaynak model çoklu görev standartlarını yeniler, OpenAI-o1-mini seviyesini aşar."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "Qwen2.5-Math-7B temel alınarak oluşturulmuş DeepSeek-R1 damıtma modeli, pekiştirme öğrenimi ve soğuk başlatma verileri ile çıkarım performansını optimize eder, açık kaynak model çoklu görev standartlarını yeniler."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite, mükemmel yanıt hızı ve daha iyi maliyet Performansı ile müşterilere farklı senaryolar için daha esnek seçenekler sunar. 128k bağlam penceresi çıkarım ve ince ayar destekler."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Baidu'nun kendi geliştirdiği amiral gemisi ultra büyük ölçekli dil modeli, ERNIE 3.5'e kıyasla model yeteneklerinde kapsamlı bir yükseltme gerçekleştirmiştir, çeşitli alanlardaki karmaşık görev senaryolarında geniş bir şekilde uygulanabilir; Baidu arama eklentisi ile otomatik entegrasyonu destekler, yanıt bilgilerini güncel tutar."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Baidu'nun kendi geliştirdiği amiral gemisi ultra büyük ölçekli dil modeli, genel performansı mükemmel olup, çeşitli alanlardaki karmaşık görev senaryolarında geniş bir şekilde uygulanabilir; Baidu arama eklentisi ile otomatik entegrasyonu destekler, soru-cevap bilgilerini güncel tutar. ERNIE 4.0'a kıyasla performans açısından daha üstündür."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "Baidu tarafından geliştirilen, geniş ölçekli büyük dil modeli, genel performansı mükemmeldir ve her alanda karmaşık görev sahneleri için geniş bir şekilde kullanılabilir; Baidu arama eklentisi ile otomatik entegrasyonu destekler, yanıt bilgi güncellemelerinin zamanlamasını güvence altına alır. ERNIE 4.0'a kıyasla, performans olarak daha üstündür."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct, Alibaba Cloud tarafından yayınlanan en son büyük dil modeli serilerinden biridir. Bu 7B modeli, kodlama ve matematik gibi alanlarda önemli ölçüde geliştirilmiş yeteneklere sahiptir. Model ayrıca, Çince, İngilizce gibi 29'dan fazla dili kapsayan çok dilli destek sunmaktadır. Model, talimat takibi, yapılandırılmış verileri anlama ve yapılandırılmış çıktı (özellikle JSON) üretme konularında önemli iyileştirmeler göstermektedir."
},
+ "MiniMax-Text-01": {
+ "description": "MiniMax-01 serisi modellerinde cesur yenilikler yaptık: ilk kez büyük ölçekli lineer dikkat mekanizmasını gerçekleştirdik, geleneksel Transformer mimarisi artık tek seçenek değil. Bu modelin parametre sayısı 456 milyara kadar çıkmakta, tek bir aktivasyonda 45.9 milyar. Modelin genel performansı, yurtdışındaki en iyi modellerle karşılaştırılabilirken, dünya genelinde 4 milyon token uzunluğundaki bağlamı verimli bir şekilde işleyebilir, bu da GPT-4o'nun 32 katı, Claude-3.5-Sonnet'in 20 katıdır."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO, olağanüstü yaratıcı deneyimler sunmak için tasarlanmış son derece esnek bir çoklu model birleşimidir."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2, belgelere ve grafiklere anlama, sahne metni anlama, OCR, bilimsel ve matematik soruları çözme gibi çeşitli görsel dil görevlerinde mükemmel performans sergilemiştir."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2, belgelere ve grafiklere anlama, sahne metni anlama, OCR, bilimsel ve matematik soruları çözme gibi çeşitli görsel dil görevlerinde mükemmel performans sergilemiştir."
- },
"Phi-3-medium-128k-instruct": {
"description": "Aynı Phi-3-medium modeli, ancak RAG veya az sayıda örnek isteme için daha büyük bir bağlam boyutuna sahiptir."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat, Zhipu AI tarafından sunulan GLM-4 serisi önceden eğitilmiş modellerin açık kaynak versiyonudur. Bu model, anlam, matematik, akıl yürütme, kod ve bilgi gibi birçok alanda mükemmel performans sergilemektedir. Çoklu diyalogları desteklemenin yanı sıra, GLM-4-9B-Chat, web tarayıcı, kod yürütme, özelleştirilmiş araç çağrısı (Function Call) ve uzun metin akıl yürütme gibi gelişmiş özelliklere de sahiptir. Model, Çince, İngilizce, Japonca, Korece ve Almanca gibi 26 dili desteklemektedir. GLM-4-9B-Chat, AlignBench-v2, MT-Bench, MMLU ve C-Eval gibi birçok standart testte mükemmel performans sergilemiştir. Bu model, maksimum 128K bağlam uzunluğunu desteklemekte olup, akademik araştırmalar ve ticari uygulamalar için uygundur."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1, modeldeki tekrarlılık ve okunabilirlik sorunlarını çözen bir güçlendirilmiş öğrenme (RL) destekli çıkarım modelidir. RL'den önce, DeepSeek-R1 soğuk başlangıç verileri tanıtarak çıkarım performansını daha da optimize etmiştir. Matematik, kod ve çıkarım görevlerinde OpenAI-o1 ile benzer performans göstermektedir ve özenle tasarlanmış eğitim yöntemleri ile genel etkisini artırmıştır."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3, 6710 milyar parametreye sahip bir karma uzman (MoE) dil modelidir ve çok başlı potansiyel dikkat (MLA) ve DeepSeekMoE mimarisini kullanarak, yardımcı kayıplar olmadan yük dengeleme stratejileri ile çıkarım ve eğitim verimliliğini optimize etmektedir. 14.8 trilyon yüksek kaliteli token üzerinde önceden eğitilmiş ve denetimli ince ayar ve güçlendirilmiş öğrenme ile, DeepSeek-V3 performans açısından diğer açık kaynak modelleri geride bırakmakta ve lider kapalı kaynak modellere yaklaşmaktadır."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma, Google tarafından geliştirilen hafif, en son açık model serilerinden biridir. Bu, yalnızca kodlayıcıdan oluşan büyük bir dil modelidir ve İngilizceyi desteklemekte, açık ağırlıklar, önceden eğitilmiş varyantlar ve talimat ince ayar varyantları sunmaktadır. Gemma modeli, soru yanıtlama, özetleme ve akıl yürütme gibi çeşitli metin üretim görevleri için uygundur. Bu 9B modeli, 8 trilyon token ile eğitilmiştir. Göreceli olarak küçük boyutu, onu dizüstü bilgisayarlar, masaüstü bilgisayarlar veya kendi bulut altyapınız gibi kaynak sınırlı ortamlarda dağıtılabilir hale getirir ve daha fazla kişinin en son AI modellerine erişimini sağlar ve yeniliği teşvik eder."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1, Meta tarafından geliştirilen çok dilli büyük dil modeli ailesidir ve 8B, 70B ve 405B olmak üzere üç parametre ölçeği ile önceden eğitilmiş ve talimat ince ayar varyantları içermektedir. Bu 8B talimat ince ayar modeli, çok dilli diyalog senaryoları için optimize edilmiştir ve birçok endüstri standart testinde mükemmel performans sergilemektedir. Model, 15 trilyon token'dan fazla açık veriler kullanılarak eğitilmiş ve modelin faydasını ve güvenliğini artırmak için denetimli ince ayar ve insan geri bildirimi pekiştirmeli öğrenme gibi teknikler kullanılmıştır. Llama 3.1, metin üretimi ve kod üretimini desteklemekte olup, bilgi kesim tarihi 2023 Aralık'tır."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview, karmaşık diyalog oluşturma ve bağlam anlama görevlerini etkili bir şekilde işleyebilen yenilikçi bir doğal dil işleme modelidir."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview, Qwen ekibi tarafından geliştirilen ve görsel çıkarım yeteneklerine odaklanan bir araştırma modelidir. Karmaşık sahne anlayışı ve görsel ile ilgili matematiksel sorunları çözme konusundaki benzersiz avantajları ile dikkat çekmektedir."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview, Qwen'in en son deneysel araştırma modelidir ve AI akıl yürütme yeteneklerini artırmaya odaklanmaktadır. Dil karışımı, özyinelemeli akıl yürütme gibi karmaşık mekanizmaları keşfederek, güçlü akıl yürütme analizi, matematik ve programlama yetenekleri gibi ana avantajlar sunmaktadır. Bununla birlikte, dil geçiş sorunları, akıl yürütme döngüleri, güvenlik endişeleri ve diğer yetenek farklılıkları gibi zorluklar da bulunmaktadır."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct, Alibaba Cloud tarafından yayınlanan kod odaklı büyük dil modeli serisinin en son versiyonudur. Bu model, Qwen2.5 temelinde, 5.5 trilyon token ile eğitilerek kod üretimi, akıl yürütme ve düzeltme yeteneklerini önemli ölçüde artırmıştır. Hem kodlama yeteneklerini geliştirmiş hem de matematik ve genel yetenek avantajlarını korumuştur. Model, kod akıllı ajanları gibi pratik uygulamalar için daha kapsamlı bir temel sunmaktadır."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math, matematik alanındaki sorunları çözmeye odaklanır ve yüksek zorlukta sorulara profesyonel yanıtlar sunar."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2, Qwen modelinin en yeni serisidir ve 128k bağlamı destekler. Mevcut en iyi açık kaynak modellerle karşılaştırıldığında, Qwen2-72B doğal dil anlama, bilgi, kod, matematik ve çok dilli yetenekler açısından mevcut lider modelleri önemli ölçüde aşmaktadır."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2, Qwen modelinin en yeni serisidir ve eşit ölçekli en iyi açık kaynak modelleri hatta daha büyük ölçekli modelleri aşabilmektedir. Qwen2 7B, birçok değerlendirmede belirgin bir avantaj elde etmiş, özellikle kod ve Çince anlama konusunda."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B, görüntü ve metin için çok modlu işleme desteği sunan güçlü bir görsel dil modelidir, görüntü içeriğini hassas bir şekilde tanıyabilir ve ilgili açıklamalar veya yanıtlar üretebilir."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct, 14 milyar parametreye sahip büyük bir dil modelidir. Performansı mükemmel olup, Çince ve çok dilli senaryoları optimize eder, akıllı soru-cevap, içerik üretimi gibi uygulamaları destekler."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct, 7 milyar parametreye sahip büyük bir dil modelidir. Fonksiyon çağrısı ile dış sistemlerle sorunsuz etkileşim destekler, esneklik ve ölçeklenebilirliği büyük ölçüde artırır. Çince ve çok dilli senaryoları optimize eder, akıllı soru-cevap, içerik üretimi gibi uygulamaları destekler."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct, büyük ölçekli önceden eğitilmiş bir programlama talimat modelidir, güçlü kod anlama ve üretme yeteneğine sahiptir, çeşitli programlama görevlerini verimli bir şekilde işleyebilir, özellikle akıllı kod yazma, otomatik betik oluşturma ve programlama sorunlarına yanıt verme için uygundur."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct, kod üretimi, kod anlama ve verimli geliştirme senaryoları için tasarlanmış büyük bir dil modelidir. Sektördeki en ileri 32B parametre ölçeğini kullanarak çeşitli programlama ihtiyaçlarını karşılayabilir."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "TeleMM çok modlu büyük model, Çin Telekom tarafından geliştirilen çok modlu anlama büyük modelidir. Metin, görüntü gibi çeşitli modlu girdileri işleyebilmekte ve görüntü anlama, grafik analizi gibi işlevleri desteklemektedir. Kullanıcılara çok modlu anlama hizmeti sunmakta ve kullanıcılarla çok modlu etkileşimde bulunarak, girdileri doğru bir şekilde anlamakta, soruları yanıtlamakta, yaratımda yardımcı olmakta ve çok modlu bilgi ve ilham desteği sunmaktadır. İnce ayrıntılı algılama, mantıksal akıl yürütme gibi çok modlu görevlerde mükemmel performans sergilemektedir."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large, sektördeki en büyük açık kaynaklı Transformer mimarisi MoE modelidir ve toplam 389 milyar parametre ile 52 milyar etkin parametreye sahiptir."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct, Qwen2 serisindeki talimat ince ayar büyük dil modelidir ve parametre ölçeği 72B'dir. Bu model, Transformer mimarisi temelinde, SwiGLU aktivasyon fonksiyonu, dikkat QKV önyargısı ve grup sorgu dikkati gibi teknikler kullanmaktadır. Büyük ölçekli girişleri işleyebilme yeteneğine sahiptir. Bu model, dil anlama, üretim, çok dilli yetenek, kodlama, matematik ve akıl yürütme gibi birçok standart testte mükemmel performans sergilemekte ve çoğu açık kaynak modelini geride bırakmakta, bazı görevlerde özel modellere karşı rekabet edebilir."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct, Alibaba Cloud tarafından yayınlanan en son büyük dil modeli serilerinden biridir. Bu 72B modeli, kodlama ve matematik gibi alanlarda önemli ölçüde geliştirilmiş yeteneklere sahiptir. Model ayrıca, Çince, İngilizce gibi 29'dan fazla dili kapsayan çok dilli destek sunmaktadır. Model, talimat takibi, yapılandırılmış verileri anlama ve yapılandırılmış çıktı (özellikle JSON) üretme konularında önemli iyileştirmeler göstermektedir."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Çin karakter diyalog senaryoları için optimize edilmiş, akıcı ve Çin ifade alışkanlıklarına uygun diyalog üretim yeteneği sunar."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Fireworks açık kaynak fonksiyon çağrı modeli, mükemmel talimat yürütme yetenekleri ve özelleştirilebilir özellikler sunar."
+ "abab7-chat-preview": {
+ "description": "abab6.5 serisi modellere kıyasla uzun metin, matematik, yazım gibi yeteneklerde büyük bir gelişim göstermektedir."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Fireworks şirketinin en son ürünü Firefunction-v2, Llama-3 tabanlı, fonksiyon çağrıları, diyalog ve talimat takibi gibi senaryolar için özel olarak optimize edilmiş yüksek performanslı bir modeldir."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1, güçlendirilmiş öğrenme ve soğuk başlangıç verileri ile optimize edilmiş, mükemmel akıl yürütme, matematik ve programlama performansına sahip en son teknoloji büyük bir dil modelidir."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b, hem görüntü hem de metin girdilerini alabilen, yüksek kaliteli verilerle eğitilmiş bir görsel dil modelidir ve çok modlu görevler için uygundur."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Deepseek tarafından sunulan güçlü Mixture-of-Experts (MoE) dil modeli, toplamda 671B parametreye sahiptir ve her bir etiket için 37B parametre etkinleştirilmektedir."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Llama 3 70B talimat modeli, çok dilli diyalog ve doğal dil anlama için optimize edilmiştir, çoğu rakip modelden daha iyi performans gösterir."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Llama 3 70B talimat modeli (HF versiyonu), resmi uygulama sonuçlarıyla uyumlu olup yüksek kaliteli talimat takibi görevleri için uygundur."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Llama 3 8B talimat modeli, diyalog ve çok dilli görevler için optimize edilmiştir, mükemmel ve etkili performans sunar."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Meta'nın 11B parametreli komut ayarlı görüntü akıl yürütme modelidir. Bu model, görsel tanıma, görüntü akıl yürütme, görüntü betimleme ve görüntü hakkında genel sorulara yanıt verme üzerine optimize edilmiştir. Bu model, grafikler ve resimler gibi görsel verileri anlayabilir ve görüntü detaylarını metin olarak betimleyerek görsel ile dil arasındaki boşluğu kapatır."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Llama 3.2 1B komut modeli, Meta tarafından sunulan hafif çok dilli bir modeldir. Bu model, verimliliği artırmak amacıyla daha büyük modellere göre gecikme ve maliyet açısından önemli iyileştirmeler sunar. Bu modelin örnek kullanım alanları arasında bilgi alma ve özetleme bulunmaktadır."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Llama 3.2 3B komut modeli, Meta tarafından sunulan hafif çok dilli bir modeldir. Bu model, verimliliği artırmak amacıyla daha büyük modellere göre gecikme ve maliyet açısından önemli iyileştirmeler sunar. Bu modelin örnek kullanım alanları arasında sorgulama, öneri yeniden yazma ve yazma desteği bulunmaktadır."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Meta'nın 90B parametreli komut ayarlı görüntü akıl yürütme modelidir. Bu model, görsel tanıma, görüntü akıl yürütme, görüntü betimleme ve görüntü hakkında genel sorulara yanıt verme üzerine optimize edilmiştir. Bu model, grafikler ve resimler gibi görsel verileri anlayabilir ve görüntü detaylarını metin olarak betimleyerek görsel ile dil arasındaki boşluğu kapatır."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct, Llama 3.1 70B'nin Aralık güncellemesi olan bir modeldir. Bu model, Llama 3.1 70B (2024 Temmuz'da piyasaya sürüldü) temel alınarak geliştirilmiş olup, araç çağrıları, çok dilli metin desteği, matematik ve programlama yeteneklerini artırmıştır. Model, akıl yürütme, matematik ve talimat takibi alanlarında sektördeki en yüksek standartlara ulaşmış olup, 3.1 405B ile benzer performans sunarken hız ve maliyet açısından önemli avantajlar sağlamaktadır."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "24B parametreli model, daha büyük modellerle karşılaştırılabilir en son teknoloji yeteneklerine sahiptir."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Mixtral MoE 8x22B talimat modeli, büyük ölçekli parametreler ve çok uzmanlı mimarisi ile karmaşık görevlerin etkili işlenmesini destekler."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Mixtral MoE 8x7B talimat modeli, çok uzmanlı mimarisi ile etkili talimat takibi ve yürütme sunar."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Mixtral MoE 8x7B talimat modeli (HF versiyonu), resmi uygulama ile uyumlu olup çeşitli yüksek verimli görev senaryoları için uygundur."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "MythoMax L2 13B modeli, yenilikçi birleşim teknolojileri ile hikaye anlatımı ve rol yapma konularında uzmandır."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "QwQ modeli, Qwen ekibi tarafından geliştirilen deneysel bir araştırma modelidir ve AI akıl yürütme yeteneklerini artırmaya odaklanmaktadır."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "Qwen-VL modelinin 72B versiyonu, Alibaba'nın en son iterasyonunun bir ürünüdür ve son bir yılın yeniliklerini temsil etmektedir."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5, Alibaba Cloud Qwen ekibi tarafından geliştirilen yalnızca kodlayıcı içeren bir dizi dil modelidir. Bu modeller, 0.5B, 1.5B, 3B, 7B, 14B, 32B ve 72B gibi farklı boyutları sunar ve temel (base) ve komut (instruct) versiyonlarına sahiptir."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct, Alibaba Cloud tarafından yayınlanan kod odaklı büyük dil modeli serisinin en son versiyonudur. Bu model, Qwen2.5 temelinde, 5.5 trilyon token ile eğitilerek kod üretimi, akıl yürütme ve düzeltme yeteneklerini önemli ölçüde artırmıştır. Hem kodlama yeteneklerini geliştirmiş hem de matematik ve genel yetenek avantajlarını korumuştur. Model, kod akıllı ajanları gibi pratik uygulamalar için daha kapsamlı bir temel sunmaktadır."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "StarCoder 15.5B modeli, ileri düzey programlama görevlerini destekler, çok dilli yetenekleri artırır ve karmaşık kod üretimi ve anlama için uygundur."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "StarCoder 7B modeli, 80'den fazla programlama dili için eğitilmiş olup, mükemmel programlama tamamlama yetenekleri ve bağlam anlama sunar."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Yi-Large modeli, mükemmel çok dilli işleme yetenekleri sunar ve her türlü dil üretimi ve anlama görevleri için uygundur."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus, Anthropic'in son derece karmaşık görevleri işlemek için en güçlü modelidir. Performans, zeka, akıcılık ve anlama açısından olağanüstü bir performans sergiler."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku, Anthropic'in en hızlı bir sonraki nesil modelidir. Claude 3 Haiku ile karşılaştırıldığında, Claude 3.5 Haiku, birçok beceride iyileşme göstermiştir ve birçok zeka kıyaslamasında bir önceki neslin en büyük modeli Claude 3 Opus'u geride bırakmıştır."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet, Opus'tan daha fazla yetenek ve Sonnet'ten daha hızlı bir hız sunar; aynı zamanda Sonnet ile aynı fiyatı korur. Sonnet, programlama, veri bilimi, görsel işleme ve ajan görevlerinde özellikle başarılıdır."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet, akıllı ve hızlı bir denge sunarak kurumsal iş yükleri için idealdir. Daha düşük bir fiyatla maksimum fayda sağlar, güvenilir ve büyük ölçekli dağıtım için uygundur."
},
- "code-raccoon-v1": {
- "description": "Kod Rakun, SenseTime büyük dil modeline dayanan bir yazılım akıllı geliştirme asistanıdır. Yazılım gereksinim analizi, mimari tasarım, kod yazımı, yazılım testi gibi aşamaları kapsar ve kullanıcıların kod yazma, programlama öğrenme gibi çeşitli ihtiyaçlarını karşılar. Kod Rakun, Python, Java, JavaScript, C++, Go, SQL gibi 90'dan fazla popüler programlama dilini ve VS Code, IntelliJ IDEA gibi popüler IDE'leri destekler. Gerçek uygulamalarda, Kod Rakun geliştiricilerin programlama verimliliğini %50'den fazla artırmasına yardımcı olabilir."
- },
"codegeex-4": {
"description": "CodeGeeX-4, çeşitli programlama dillerinde akıllı soru-cevap ve kod tamamlama desteği sunan güçlü bir AI programlama asistanıdır, geliştirme verimliliğini artırır."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+, gerçek işletme senaryoları ve karmaşık uygulamalar için tasarlanmış yüksek performanslı bir büyük dil modelidir."
},
+ "dall-e-2": {
+ "description": "İkinci nesil DALL·E modeli, daha gerçekçi ve doğru görüntü üretimi destekler, çözünürlüğü birinci neslin 4 katıdır."
+ },
+ "dall-e-3": {
+ "description": "En son DALL·E modeli, Kasım 2023'te piyasaya sürüldü. Daha gerçekçi ve doğru görüntü üretimi destekler, daha güçlü detay ifade yeteneğine sahiptir."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct, yüksek güvenilirlikte talimat işleme yetenekleri sunar ve çok çeşitli endüstri uygulamalarını destekler."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2, güçlü ve maliyet etkin bir karışık uzman (MoE) dil modelidir. 8.1 trilyon token yüksek kaliteli veri kümesi üzerinde önceden eğitilmiş ve denetimli ince ayar (SFT) ve pekiştirmeli öğrenme (RL) ile model yetenekleri daha da geliştirilmiştir. DeepSeek 67B ile karşılaştırıldığında, DeepSeek-V2 daha güçlü performans sunarken, eğitim maliyetlerini %42.5 oranında azaltmış, KV önbelleğini %93.3 oranında azaltmış ve maksimum üretim verimliliğini 5.76 kat artırmıştır. Bu model, 128k bağlam uzunluğunu desteklemekte ve standart testlerde ve açık üretim değerlendirmelerinde mükemmel performans sergilemektedir."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1, tekrarlayan öğrenme (RL) destekli bir çıkarım modelidir ve modeldeki tekrarlama ve okunabilirlik sorunlarını çözmektedir. RL'den önce, DeepSeek-R1 soğuk başlangıç verilerini tanıtarak çıkarım performansını daha da optimize etmiştir. Matematik, kod ve çıkarım görevlerinde OpenAI-o1 ile benzer bir performans sergilemekte ve özenle tasarlanmış eğitim yöntemleri ile genel etkisini artırmaktadır."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "DeepSeek-R1 damıtma modeli, pekiştirme öğrenimi ve soğuk başlatma verileri ile çıkarım performansını optimize eder, açık kaynak model çoklu görev standartlarını yeniler."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B, Llama-3.1-8B temel alınarak geliştirilmiş bir damıtma modelidir. Bu model, DeepSeek-R1 tarafından üretilen örneklerle ince ayar yapılmış, mükemmel çıkarım yeteneği sergilemektedir. Birçok referans testinde iyi performans göstermiş, MATH-500'de %89.1 doğruluk oranına, AIME 2024'te %50.4 geçiş oranına ulaşmış, CodeForces'ta 1205 puan alarak 8B ölçeğindeki model olarak güçlü matematik ve programlama yeteneğini göstermiştir."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "DeepSeek-R1 damıtma modeli, pekiştirme öğrenimi ve soğuk başlatma verileri ile çıkarım performansını optimize eder, açık kaynak model çoklu görev standartlarını yeniler."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "DeepSeek-R1 damıtma modeli, pekiştirme öğrenimi ve soğuk başlatma verileri ile çıkarım performansını optimize eder, açık kaynak model çoklu görev standartlarını yeniler."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B, Qwen2.5-32B temel alınarak bilgi damıtma ile elde edilen bir modeldir. Bu model, DeepSeek-R1 tarafından üretilen 800.000 seçkin örnek ile ince ayar yapılmış, matematik, programlama ve çıkarım gibi birçok alanda olağanüstü performans sergilemektedir. AIME 2024, MATH-500, GPQA Diamond gibi birçok referans testinde mükemmel sonuçlar elde etmiş, MATH-500'de %94.3 doğruluk oranına ulaşarak güçlü matematik çıkarım yeteneğini göstermiştir."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B, Qwen2.5-Math-7B temel alınarak bilgi damıtma ile elde edilen bir modeldir. Bu model, DeepSeek-R1 tarafından üretilen 800.000 seçkin örnek ile ince ayar yapılmış, mükemmel çıkarım yeteneği sergilemektedir. Birçok referans testinde öne çıkmış, MATH-500'de %92.8 doğruluk oranına, AIME 2024'te %55.5 geçiş oranına ulaşmış, CodeForces'ta 1189 puan alarak 7B ölçeğindeki model olarak güçlü matematik ve programlama yeteneğini göstermiştir."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5, önceki sürümlerin mükemmel özelliklerini bir araya getirir, genel ve kodlama yeteneklerini artırır."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3, 6710 milyar parametreye sahip bir karma uzman (MoE) dil modelidir. Çok başlı potansiyel dikkat (MLA) ve DeepSeekMoE mimarisini kullanarak, yardımcı kayıplar olmadan yük dengeleme stratejisi ile çıkarım ve eğitim verimliliğini optimize etmektedir. 14.8 trilyon yüksek kaliteli token üzerinde önceden eğitilmiş ve denetimli ince ayar ile tekrarlayan öğrenme gerçekleştirilmiştir; DeepSeek-V3, performans açısından diğer açık kaynaklı modelleri geride bırakmakta ve lider kapalı kaynaklı modellere yaklaşmaktadır."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B, yüksek karmaşıklıkta diyaloglar için eğitilmiş gelişmiş bir modeldir."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "En son teknolojiye sahip verimli LLM, akıl yürütme, matematik ve programlama konularında uzmandır."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2, DeepSeekMoE-27B tabanlı bir karma uzman (MoE) görsel dil modelidir. Seyrek etkinleştirilen MoE mimarisini kullanarak yalnızca 4.5B parametreyi etkinleştirerek olağanüstü performans sergilemektedir. Bu model, görsel soru yanıtlama, optik karakter tanıma, belge/tablolar/grafikler anlama ve görsel konumlandırma gibi birçok görevde mükemmel sonuçlar elde etmektedir."
+ },
"deepseek-chat": {
"description": "Genel ve kod yeteneklerini birleştiren yeni bir açık kaynak modeli, yalnızca mevcut Chat modelinin genel diyalog yeteneklerini ve Coder modelinin güçlü kod işleme yeteneklerini korumakla kalmaz, aynı zamanda insan tercihleri ile daha iyi hizalanmıştır. Ayrıca, DeepSeek-V2.5 yazım görevleri, talimat takibi gibi birçok alanda büyük iyileştirmeler sağlamıştır."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2, açık kaynaklı bir karışık uzman kod modelidir, kod görevlerinde mükemmel performans sergiler ve GPT4-Turbo ile karşılaştırılabilir."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1, tekrarlayan öğrenme (RL) destekli bir çıkarım modelidir ve modeldeki tekrarlama ve okunabilirlik sorunlarını çözmektedir. RL'den önce, DeepSeek-R1 soğuk başlangıç verilerini tanıtarak çıkarım performansını daha da optimize etmiştir. Matematik, kod ve çıkarım görevlerinde OpenAI-o1 ile benzer bir performans sergilemekte ve özenle tasarlanmış eğitim yöntemleri ile genel etkisini artırmaktadır."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 - DeepSeek paketindeki daha büyük ve daha akıllı model - Llama 70B mimarisine damıtılmıştır. Referans testleri ve insan değerlendirmelerine dayanarak, bu model orijinal Llama 70B'den daha akıllıdır, özellikle matematik ve gerçeklik doğruluğu gerektiren görevlerde mükemmel performans sergilemektedir."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "DeepSeek-R1-Distill serisi modeller, bilgi damıtma teknolojisi ile DeepSeek-R1 tarafından üretilen örneklerin Qwen, Llama gibi açık kaynak modeller üzerinde ince ayar yapılmasıyla elde edilmiştir."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "DeepSeek-R1-Distill serisi modeller, bilgi damıtma teknolojisi ile DeepSeek-R1 tarafından üretilen örneklerin Qwen, Llama gibi açık kaynak modeller üzerinde ince ayar yapılmasıyla elde edilmiştir."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "DeepSeek-R1-Distill serisi modeller, bilgi damıtma teknolojisi ile DeepSeek-R1 tarafından üretilen örneklerin Qwen, Llama gibi açık kaynak modeller üzerinde ince ayar yapılmasıyla elde edilmiştir."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "DeepSeek-R1-Distill serisi modeller, bilgi damıtma teknolojisi ile DeepSeek-R1 tarafından üretilen örneklerin Qwen, Llama gibi açık kaynak modeller üzerinde ince ayar yapılmasıyla elde edilmiştir."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "DeepSeek-R1-Distill serisi modeller, bilgi damıtma teknolojisi ile DeepSeek-R1 tarafından üretilen örneklerin Qwen, Llama gibi açık kaynak modeller üzerinde ince ayar yapılmasıyla elde edilmiştir."
+ },
+ "deepseek-reasoner": {
+ "description": "DeepSeek tarafından sunulan bir akıl yürütme modeli. Model, nihai yanıtı vermeden önce bir düşünce zinciri içeriği sunarak nihai cevabın doğruluğunu artırır."
+ },
"deepseek-v2": {
"description": "DeepSeek V2, ekonomik ve verimli işleme ihtiyaçları için uygun, etkili bir Mixture-of-Experts dil modelidir."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B, DeepSeek'in tasarım kodu modelidir, güçlü kod üretim yetenekleri sunar."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3, Hangzhou DeepSeek Yapay Zeka Temel Teknoloji Araştırma Şirketi tarafından geliştirilen MoE modelidir, birçok değerlendirme sonucunda öne çıkmakta ve ana akım listelerde açık kaynak modeller arasında birinci sırada yer almaktadır. V3, V2.5 modeline göre üretim hızında 3 kat artış sağlamış, kullanıcılara daha hızlı ve akıcı bir deneyim sunmuştur."
+ },
"deepseek/deepseek-chat": {
"description": "Genel ve kod yeteneklerini birleştiren yeni açık kaynak model, yalnızca mevcut Chat modelinin genel diyalog yeteneklerini ve Coder modelinin güçlü kod işleme yeteneklerini korumakla kalmaz, aynı zamanda insan tercihleriyle daha iyi hizalanmıştır. Ayrıca, DeepSeek-V2.5 yazma görevleri, talimat takibi gibi birçok alanda da büyük iyileştirmeler sağlamıştır."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1, yalnızca çok az etiketli veri ile modelin akıl yürütme yeteneğini büyük ölçüde artırır. Model, nihai yanıtı vermeden önce bir düşünce zinciri içeriği sunarak nihai yanıtın doğruluğunu artırır."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1, yalnızca çok az etiketli veri ile modelin akıl yürütme yeteneğini büyük ölçüde artırır. Model, nihai yanıtı vermeden önce bir düşünce zinciri içeriği sunarak nihai yanıtın doğruluğunu artırır."
+ },
"emohaa": {
"description": "Emohaa, duygusal sorunları anlamalarına yardımcı olmak için profesyonel danışmanlık yeteneklerine sahip bir psikolojik modeldir."
},
+ "ernie-3.5-128k": {
+ "description": "Baidu tarafından geliştirilen amiral gemisi büyük ölçekli dil modeli, geniş bir Çince ve İngilizce veri kümesini kapsar, güçlü genel yeteneklere sahiptir ve çoğu diyalog soru-cevap, yaratım, eklenti uygulama senaryolarını karşılayabilir; Baidu arama eklentisi ile otomatik entegrasyon desteği sunarak soru-cevap bilgilerini güncel tutar."
+ },
+ "ernie-3.5-8k": {
+ "description": "Baidu tarafından geliştirilen amiral gemisi büyük ölçekli dil modeli, geniş bir Çince ve İngilizce veri kümesini kapsar, güçlü genel yeteneklere sahiptir ve çoğu diyalog soru-cevap, yaratım, eklenti uygulama senaryolarını karşılayabilir; Baidu arama eklentisi ile otomatik entegrasyon desteği sunarak soru-cevap bilgilerini güncel tutar."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "Baidu tarafından geliştirilen amiral gemisi büyük ölçekli dil modeli, geniş bir Çince ve İngilizce veri kümesini kapsar, güçlü genel yeteneklere sahiptir ve çoğu diyalog soru-cevap, yaratım, eklenti uygulama senaryolarını karşılayabilir; Baidu arama eklentisi ile otomatik entegrasyon desteği sunarak soru-cevap bilgilerini güncel tutar."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "Baidu tarafından geliştirilen amiral gemisi ultra büyük ölçekli dil modeli, ERNIE 3.5'e göre model yeteneklerinde kapsamlı bir yükseltme gerçekleştirmiştir, çeşitli alanlardaki karmaşık görev senaryolarında geniş bir şekilde uygulanabilir; Baidu arama eklentisi ile otomatik entegrasyon desteği sunarak soru-cevap bilgilerini güncel tutar."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "Baidu tarafından geliştirilen amiral gemisi ultra büyük ölçekli dil modeli, ERNIE 3.5'e göre model yeteneklerinde kapsamlı bir yükseltme gerçekleştirmiştir, çeşitli alanlardaki karmaşık görev senaryolarında geniş bir şekilde uygulanabilir; Baidu arama eklentisi ile otomatik entegrasyon desteği sunarak soru-cevap bilgilerini güncel tutar."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "Baidu tarafından geliştirilen amiral gemisi ultra büyük ölçekli dil modeli, genel performansı mükemmel, çeşitli alanlardaki karmaşık görev senaryolarında geniş bir şekilde uygulanabilir; Baidu arama eklentisi ile otomatik entegrasyon desteği sunarak soru-cevap bilgilerini güncel tutar. ERNIE 4.0'a göre performans açısından daha üstündür."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "Baidu tarafından geliştirilen amiral gemisi ultra büyük ölçekli dil modeli, genel performansı mükemmel, çeşitli alanlardaki karmaşık görev senaryolarında geniş bir şekilde uygulanabilir; Baidu arama eklentisi ile otomatik entegrasyon desteği sunarak soru-cevap bilgilerini güncel tutar. ERNIE 4.0'a göre performans açısından daha üstündür."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "Baidu tarafından geliştirilen amiral gemisi ultra büyük ölçekli dil modeli, genel performansı mükemmel, çeşitli alanlardaki karmaşık görev senaryolarında geniş bir şekilde uygulanabilir; Baidu arama eklentisi ile otomatik entegrasyon desteği sunarak soru-cevap bilgilerini güncel tutar. ERNIE 4.0'a göre performans açısından daha üstündür."
+ },
+ "ernie-char-8k": {
+ "description": "Baidu tarafından geliştirilen dikey senaryo büyük dil modeli, oyun NPC'leri, müşteri hizmetleri diyalogları, diyalog karakter rolü gibi uygulama senaryolarına uygundur, karakter tarzı daha belirgin ve tutarlıdır, talimat takibi yeteneği daha güçlü, çıkarım performansı daha iyidir."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Baidu tarafından geliştirilen dikey senaryo büyük dil modeli, oyun NPC'leri, müşteri hizmetleri diyalogları, diyalog karakter rolü gibi uygulama senaryolarına uygundur, karakter tarzı daha belirgin ve tutarlıdır, talimat takibi yeteneği daha güçlü, çıkarım performansı daha iyidir."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite, Baidu tarafından geliştirilen hafif büyük dil modelidir, mükemmel model performansı ve çıkarım yeteneği ile düşük hesaplama gücüne sahip AI hızlandırıcı kartları için uygundur."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Baidu tarafından geliştirilen hafif büyük dil modeli, mükemmel model performansı ve çıkarım yeteneği ile ERNIE Lite'dan daha iyi sonuçlar verir, düşük hesaplama gücüne sahip AI hızlandırıcı kartları için uygundur."
+ },
+ "ernie-novel-8k": {
+ "description": "Baidu tarafından geliştirilen genel büyük dil modeli, roman devam ettirme yeteneğinde belirgin bir avantaja sahiptir, aynı zamanda kısa oyun, film gibi senaryolarda da kullanılabilir."
+ },
+ "ernie-speed-128k": {
+ "description": "Baidu'nun 2024 yılında yayımladığı en son yüksek performanslı büyük dil modeli, genel yetenekleri mükemmel, belirli senaryo sorunlarını daha iyi ele almak için temel model olarak ince ayar yapılabilir, aynı zamanda mükemmel çıkarım performansına sahiptir."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "Baidu'nun 2024 yılında yayımladığı en son yüksek performanslı büyük dil modeli, genel yetenekleri mükemmel, ERNIE Speed'den daha iyi sonuçlar verir, belirli senaryo sorunlarını daha iyi ele almak için temel model olarak ince ayar yapılabilir, aynı zamanda mükemmel çıkarım performansına sahiptir."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny, Baidu tarafından geliştirilen ultra yüksek performanslı büyük dil modelidir, dağıtım ve ince ayar maliyetleri Wenxin serisi modelleri arasında en düşüktür."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Tuning), kararlı ve ayarlanabilir bir performans sunar, karmaşık görev çözümleri için ideal bir seçimdir."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro, 2 milyon token'a kadar destekler, orta ölçekli çok modlu modeller için ideal bir seçimdir ve karmaşık görevler için çok yönlü destek sunar."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash, mükemmel hız, yerel araç kullanımı, çok modlu üretim ve 1M token bağlam penceresi dahil olmak üzere bir sonraki nesil özellikler ve iyileştirmeler sunar."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash, mükemmel hız, yerel araç kullanımı, çok modlu üretim ve 1M token bağlam penceresi dahil olmak üzere bir sonraki nesil özellikler ve iyileştirmeler sunar."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp, Google'ın en yeni deneysel çok modlu yapay zeka modelidir. Gelecek nesil özellikleri, olağanüstü hızı, yerel araç çağrısı ve çok modlu üretim ile donatılmıştır."
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "Maliyet etkinliği ve düşük gecikme gibi hedefler için optimize edilmiş bir Gemini 2.0 Flash modelidir."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Exp, Google'ın en son deneysel çok modlu AI modelidir, bir sonraki nesil özelliklere, olağanüstü hıza, yerel araç çağrısına ve çok modlu üretime sahiptir."
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Exp, Google'ın en son deneysel çok modlu AI modelidir, bir sonraki nesil özelliklere, üstün hıza, yerel araç çağrısına ve çok modlu üretime sahiptir."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114, Google'ın en son deneysel çok modlu AI modeli olup, hızlı işleme yeteneğine sahip, metin, görüntü ve video girişlerini desteklemekte ve çeşitli görevlerde verimli bir şekilde ölçeklenmektedir."
- },
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121, Google'un en yeni deneysel çok modlu AI modelidir. Hızlı işlem yeteneğine sahip olup, metin, görüntü ve video girişi destekler ve çeşitli görevler için verimli bir şekilde ölçeklenebilir."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Deneysel, Google'ın en son deneysel çok modlu AI modelidir ve önceki sürümlere göre belirli bir kalite artışı sağlamaktadır, özellikle dünya bilgisi, kod ve uzun bağlam için."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206, Google'ın en son deneysel çok modlu AI modelidir ve önceki sürümlere kıyasla belirli bir kalite artırımı sağlar."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus, video içeriği ve çoklu görüntüleri anlama yeteneğine sahiptir, çok modlu görevler için uygundur."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview, karmaşık akıl yürütme yeteneklerine sahip olup, mantıksal akıl yürütme, matematik, programlama gibi alanlarda mükemmel performans sergilemektedir."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash, mükemmel hız, yerel araç kullanımı, çok modlu üretim ve 1M token bağlam penceresi dahil olmak üzere bir sonraki nesil özellikler ve iyileştirmeler sunar."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Deneysel, Google'ın en son deneysel çok modlu AI modelidir ve önceki sürümlere göre belirli bir kalite artışı sağlamaktadır, özellikle dünya bilgisi, kod ve uzun bağlam için."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash, optimize edilmiş çok modlu işleme yetenekleri sunar ve çeşitli karmaşık görev senaryolarına uygundur."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro, en son optimize edilmiş teknolojileri birleştirerek daha verimli çok modlu veri işleme yetenekleri sunar."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2, Google tarafından sunulan verimli bir modeldir, küçük uygulamalardan karmaşık veri işleme senaryolarına kadar çeşitli uygulama alanlarını kapsar."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2, hafiflik ve verimlilik tasarım felsefesini sürdürmektedir."
},
"google/gemma-2-2b-it": {
"description": "Google'ın hafif talimat ayarlama modeli"
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2, Google tarafından sunulan verimli bir modeldir, küçük uygulamalardan karmaşık veri işleme senaryolarına kadar çeşitli uygulama alanlarını kapsar."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2, Google'ın hafif açık kaynak metin modeli serisidir."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo, çeşitli metin üretimi ve anlama görevleri için uygundur, şu anda gpt-3.5-turbo-0125'e işaret ediyor."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, yüksek kapasiteli metin üretim modeli, karmaşık görevler için uygundur."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo, çeşitli metin üretimi ve anlama görevleri için uygundur, şu anda gpt-3.5-turbo-0125'e işaret ediyor."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o, güncel en son sürümü korumak için gerçek zamanlı olarak güncellenen dinamik bir modeldir. Müşteri hizmetleri, eğitim ve teknik destek gibi büyük ölçekli uygulama senaryoları için güçlü dil anlama ve üretme yeteneklerini bir araya getirir."
},
+ "gpt-4o-audio-preview": {
+ "description": "GPT-4o Ses modeli, sesli giriş ve çıkış desteği sunar."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini, OpenAI'nin GPT-4 Omni'den sonra tanıttığı en yeni modeldir. Görsel ve metin girişi destekler ve metin çıktısı verir. En gelişmiş küçük model olarak, diğer son zamanlardaki öncü modellere göre çok daha ucuzdur ve GPT-3.5 Turbo'dan %60'tan fazla daha ucuzdur. En son teknolojiyi korurken, önemli bir maliyet etkinliği sunar. GPT-4o mini, MMLU testinde %82 puan almış olup, şu anda sohbet tercihleri açısından GPT-4'ün üzerinde yer almaktadır."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "GPT-4o-mini gerçek zamanlı versiyonu, ses ve metin için gerçek zamanlı giriş ve çıkış desteği sunar."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "GPT-4o gerçek zamanlı versiyonu, ses ve metin için gerçek zamanlı giriş ve çıkış desteği sunar."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "GPT-4o gerçek zamanlı versiyonu, ses ve metin için gerçek zamanlı giriş ve çıkış desteği sunar."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "GPT-4o gerçek zamanlı versiyonu, ses ve metin için gerçek zamanlı giriş ve çıkış desteği sunar."
+ },
"grok-2-1212": {
"description": "Bu model, doğruluk, talimat takibi ve çok dilli yetenekler açısından geliştirilmiştir."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "Hunyuan'ın en son MOE mimarisi FunctionCall modeli, yüksek kaliteli FunctionCall verisi ile eğitilmiş olup, bağlam penceresi 32K'ya ulaşmıştır ve birçok boyutta değerlendirme göstergelerinde lider konumdadır."
},
+ "hunyuan-large": {
+ "description": "Hunyuan-large modelinin toplam parametre sayısı yaklaşık 389B, etkin parametre sayısı yaklaşık 52B'dir; bu, mevcut endüstrideki en büyük parametre ölçeğine sahip ve en iyi performansı gösteren Transformer mimarisinin açık kaynaklı MoE modelidir."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Uzun metin görevlerini, örneğin belge özeti ve belge sorgulama gibi, işleme konusunda uzmandır; aynı zamanda genel metin oluşturma görevlerini de yerine getirme yeteneğine sahiptir. Uzun metinlerin analizi ve oluşturulmasında mükemmel bir performans sergiler, karmaşık ve ayrıntılı uzun metin içerik işleme ihtiyaçlarına etkili bir şekilde yanıt verebilir."
+ },
"hunyuan-lite": {
"description": "MOE yapısına yükseltilmiş, bağlam penceresi 256k, NLP, kod, matematik, endüstri gibi birçok değerlendirme setinde birçok açık kaynak modelden önde."
},
+ "hunyuan-lite-vision": {
+ "description": "Hunyuan'ın en son 7B çok modlu modeli, bağlam penceresi 32K, Çince ve İngilizce senaryolarında çok modlu diyalog, görüntü nesne tanıma, belge tablo anlama, çok modlu matematik vb. destekler; birçok boyutta değerlendirme kriterleri 7B rakip modellerden üstündür."
+ },
"hunyuan-pro": {
"description": "Trilyon seviyesinde parametre ölçeğine sahip MOE-32K uzun metin modeli. Çeşitli benchmarklarda kesin bir liderlik seviyesine ulaşarak, karmaşık talimatlar ve akıl yürütme yetenekleri ile karmaşık matematik yetenekleri sunar, functioncall desteği ile çok dilli çeviri, finans, hukuk ve sağlık gibi alanlarda önemli optimizasyonlar sağlar."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Daha iyi bir yönlendirme stratejisi kullanarak, yük dengeleme ve uzman yakınsaması sorunlarını hafifletir. Uzun metinlerde, iğne arama göstergesi %99.9'a ulaşmaktadır. MOE-256K, uzunluk ve etki açısından daha fazla bir sıçrama yaparak, girdi uzunluğunu büyük ölçüde genişletir."
},
+ "hunyuan-standard-vision": {
+ "description": "Hunyuan'ın en son çok modlu modeli, çok dilli yanıtları destekler, Çince ve İngilizce yetenekleri dengelidir."
+ },
"hunyuan-turbo": {
"description": "Hunyuan'ın yeni nesil büyük dil modelinin önizleme sürümü, tamamen yeni bir karma uzman modeli (MoE) yapısı kullanır ve hunyuan-pro'ya kıyasla daha hızlı çıkarım verimliliği ve daha güçlü performans sunar."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Hunyuan-turbo 2024 yılı 11 ay 20 günü sabit sürümü, hunyuan-turbo ve hunyuan-turbo-latest arasında bir versiyon."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "Bu sürümde yapılan optimizasyonlar: veri talimatı ölçeklendirme, modelin genel genelleme yeteneğini büyük ölçüde artırma; matematik, kodlama, mantıksal akıl yürütme yeteneklerini büyük ölçüde artırma; metin anlama ve kelime anlama ile ilgili yetenekleri optimize etme; metin oluşturma içerik üretim kalitesini optimize etme."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "Genel deneyim optimizasyonu, NLP anlama, metin oluşturma, sohbet, bilgi sorgulama, çeviri, alan vb. dahil; insan benzeri özellikleri artırma, modelin duygusal zekasını optimize etme; niyet belirsiz olduğunda modelin aktif olarak netleştirme yeteneğini artırma; kelime ve terim analizi ile ilgili sorunların işlenme yeteneğini artırma; yaratım kalitesini ve etkileşimliğini artırma; çoklu tur deneyimini geliştirme."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "Hunyuan'ın yeni nesil görsel dil amiral modeli, tamamen yeni bir karışık uzman modeli (MoE) yapısını benimser; metin ve görüntü anlama ile ilgili temel tanıma, içerik oluşturma, bilgi sorgulama, analiz ve akıl yürütme gibi yeteneklerde bir önceki nesil modele göre kapsamlı bir iyileştirme sağlar."
+ },
"hunyuan-vision": {
"description": "Hunyuan'ın en son çok modlu modeli, resim + metin girişi ile metin içeriği oluşturmayı destekler."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "En son model serimiz, olağanüstü çıkarım performansına sahiptir, 1M bağlam uzunluğunu destekler ve daha güçlü talimat takibi ve araç çağırma yetenekleri sunar."
},
+ "internlm3-latest": {
+ "description": "En son model serimiz, olağanüstü çıkarım performansına sahiptir ve aynı ölçekli açık kaynak modeller arasında liderdir. Varsayılan olarak en son yayımlanan InternLM3 serisi modellerine işaret eder."
+ },
+ "jina-deepsearch-v1": {
+ "description": "Derin arama, web araması, okuma ve akıl yürütmeyi birleştirerek kapsamlı bir araştırma yapar. Bunu, araştırma görevlerinizi kabul eden bir ajan olarak düşünebilirsiniz - geniş bir arama yapar ve birden fazla yineleme ile cevap verir. Bu süreç, sürekli araştırma, akıl yürütme ve sorunları çeşitli açılardan çözmeyi içerir. Bu, doğrudan önceden eğitilmiş verilerden cevaplar üreten standart büyük modellerle ve tek seferlik yüzey aramasına dayanan geleneksel RAG sistemleriyle temelde farklıdır."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM, öğrenme bilimleri ilkelerine uygun olarak eğitilmiş, görev odaklı deneysel bir dil modelidir. Eğitim ve öğrenim senaryolarında sistem talimatlarını takip edebilir ve uzman bir mentor olarak görev alabilir."
},
"lite": {
"description": "Spark Lite, son derece düşük gecikme süresi ve yüksek verimlilikle çalışan hafif bir büyük dil modelidir. Tamamen ücretsiz ve açık olup, gerçek zamanlı çevrimiçi arama işlevini desteklemektedir. Hızlı yanıt verme özelliği, düşük hesaplama gücüne sahip cihazlarda çıkarım uygulamaları ve model ince ayarlarında mükemmel performans sergileyerek, kullanıcılara maliyet etkinliği ve akıllı deneyim sunmakta, özellikle bilgi sorgulama, içerik oluşturma ve arama senaryolarında başarılı olmaktadır."
},
- "llama-3.1-70b-instruct": {
- "description": "Llama 3.1 70B Instruct modeli, 70B parametreye sahiptir ve büyük metin üretimi ve talimat görevlerinde mükemmel performans sunar."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B, daha güçlü AI akıl yürütme yeteneği sunar, karmaşık uygulamalar için uygundur ve yüksek verimlilik ve doğruluk sağlamak için çok sayıda hesaplama işlemini destekler."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B, hızlı metin üretim yeteneği sunan yüksek performanslı bir modeldir ve büyük ölçekli verimlilik ve maliyet etkinliği gerektiren uygulama senaryoları için son derece uygundur."
},
- "llama-3.1-8b-instruct": {
- "description": "Llama 3.1 8B Instruct modeli, 8B parametreye sahiptir ve görsel talimat görevlerinin etkili bir şekilde yürütülmesini sağlar, kaliteli metin üretim yetenekleri sunar."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Llama 3.1 Sonar Huge Online modeli, 405B parametreye sahiptir ve yaklaşık 127,000 belirteçlik bağlam uzunluğunu destekler, karmaşık çevrimiçi sohbet uygulamaları için tasarlanmıştır."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Llama 3.1 Sonar Large Chat modeli, 70B parametreye sahiptir ve yaklaşık 127,000 belirteçlik bağlam uzunluğunu destekler, karmaşık çevrimdışı sohbet görevleri için uygundur."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Llama 3.1 Sonar Large Online modeli, 70B parametreye sahiptir ve yaklaşık 127,000 belirteçlik bağlam uzunluğunu destekler, yüksek kapasiteli ve çeşitli sohbet görevleri için uygundur."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Llama 3.1 Sonar Small Chat modeli, 8B parametreye sahiptir ve çevrimdışı sohbet için tasarlanmıştır, yaklaşık 127,000 belirteçlik bağlam uzunluğunu destekler."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Llama 3.1 Sonar Small Online modeli, 8B parametreye sahiptir ve yaklaşık 127,000 belirteçlik bağlam uzunluğunu destekler, çevrimiçi sohbet için tasarlanmıştır ve çeşitli metin etkileşimlerini etkili bir şekilde işler."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2, görsel ve metin verilerini birleştiren görevleri işlemek için tasarlanmıştır. Görüntü tanımlama ve görsel soru-cevap gibi görevlerde mükemmel performans sergiler, dil üretimi ile görsel akıl yürütme arasındaki uçurumu aşar."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3, Llama serisinin en gelişmiş çok dilli açık kaynak büyük dil modelidir ve 405B modelinin performansını çok düşük maliyetle deneyimlemenizi sağlar. Transformer yapısına dayanmaktadır ve denetimli ince ayar (SFT) ve insan geri bildirimi ile güçlendirilmiş öğrenme (RLHF) ile faydalılığını ve güvenliğini artırmıştır. Talimat ayarlı versiyonu, çok dilli diyaloglar için optimize edilmiştir ve birçok endüstri kıyaslamasında birçok açık kaynak ve kapalı sohbet modelinden daha iyi performans göstermektedir. Bilgi kesim tarihi 2023 Aralık'tır."
+ },
"llama-3.3-70b-versatile": {
"description": "Meta Llama 3.3 çok dilli büyük dil modeli (LLM), 70B (metin girişi/metin çıkışı) içindeki önceden eğitilmiş ve talimat ayarlanmış bir üretim modelidir. Llama 3.3 talimat ayarlı saf metin modeli, çok dilli konuşma kullanım durumları için optimize edilmiştir ve yaygın endüstri kıyaslamalarında mevcut birçok açık kaynak ve kapalı sohbet modelinden daha üstündür."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2, görsel ve metin verilerini bir arada işleme amacıyla tasarlanmıştır. Görüntü betimleme ve görsel soru yanıtlama gibi görevlerde mükemmel performans sergiler, dil üretimi ve görsel akıl yürütme arasındaki boşluğu kapar."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3, Llama serisinin en gelişmiş çok dilli açık kaynak büyük dil modelidir, 405B modelinin performansını çok düşük maliyetle deneyimleme imkanı sunar. Transformer yapısına dayanır ve denetimli ince ayar (SFT) ve insan geri bildirimi ile güçlendirilmiş öğrenme (RLHF) ile kullanılabilirlik ve güvenliği artırılmıştır. Talimat ayarlı versiyonu çok dilli diyaloglar için optimize edilmiştir ve birçok endüstri standardında birçok açık kaynak ve kapalı sohbet modelinden daha iyi performans göstermektedir. Bilgi kesim tarihi 2023 Aralık'tır."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Meta Llama 3.3 çok dilli büyük dil modeli (LLM), 70B (metin girişi/metin çıkışı) içinde önceden eğitilmiş ve talimat ayarlı bir üretim modelidir. Llama 3.3 talimat ayarlı saf metin modeli, çok dilli diyalog kullanım durumları için optimize edilmiştir ve yaygın endüstri standartlarında birçok mevcut açık kaynak ve kapalı sohbet modelinden daha iyi performans göstermektedir."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2, görsel ve metin verilerini bir arada işleme amacıyla tasarlanmıştır. Görüntü betimleme ve görsel soru yanıtlama gibi görevlerde mükemmel performans sergiler, dil üretimi ve görsel akıl yürütme arasındaki boşluğu kapar."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "405B Llama 3.1 Turbo modeli, büyük veri işleme için devasa bağlam desteği sunar ve büyük ölçekli AI uygulamalarında öne çıkar."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1, Meta tarafından sunulan öncü bir modeldir, 405B parametreye kadar destekler ve karmaşık diyaloglar, çok dilli çeviri ve veri analizi alanlarında uygulanabilir."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B, çok dilli yüksek verimli diyalog desteği sunmaktadır."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2, görsel ve metin verilerini birleştiren görevleri işlemek için tasarlanmıştır. Görüntü tanımlama ve görsel soru yanıtlama gibi görevlerde mükemmel performans sergileyerek dil üretimi ve görsel akıl yürütme arasındaki boşluğu kapatmaktadır."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3, Llama serisinin en gelişmiş çok dilli açık kaynak büyük dil modelidir ve 405B modelinin performansını çok düşük maliyetle deneyimlemenizi sağlar. Transformer yapısına dayanmaktadır ve denetimli ince ayar (SFT) ve insan geri bildirimi ile güçlendirilmiş öğrenme (RLHF) ile faydalılığını ve güvenliğini artırmıştır. Talimat ayarlı versiyonu, çok dilli diyaloglar için optimize edilmiştir ve birçok endüstri kıyaslamasında birçok açık kaynak ve kapalı sohbet modelinden daha iyi performans göstermektedir. Bilgi kesim tarihi 2023 Aralık'tır."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3, Llama serisinin en gelişmiş çok dilli açık kaynak büyük dil modelidir ve 405B modelinin performansını çok düşük maliyetle deneyimlemenizi sağlar. Transformer yapısına dayanmaktadır ve denetimli ince ayar (SFT) ve insan geri bildirimi ile güçlendirilmiş öğrenme (RLHF) ile faydalılığını ve güvenliğini artırmıştır. Talimat ayarlı versiyonu, çok dilli diyaloglar için optimize edilmiştir ve birçok endüstri kıyaslamasında birçok açık kaynak ve kapalı sohbet modelinden daha iyi performans göstermektedir. Bilgi kesim tarihi 2023 Aralık'tır."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct, Llama 3.1 Instruct modelinin en büyük ve en güçlü versiyonudur. Bu, son derece gelişmiş bir diyalog akıl yürütme ve veri sentezleme modelidir ve belirli alanlarda uzmanlaşmış sürekli ön eğitim veya ince ayar için bir temel olarak da kullanılabilir. Llama 3.1, çok dilli büyük dil modelleri (LLM'ler) sunar ve 8B, 70B ve 405B boyutlarında önceden eğitilmiş, talimat ayarlı üretim modellerinden oluşur (metin girişi/çıkışı). Llama 3.1'in talimat ayarlı metin modelleri (8B, 70B, 405B), çok dilli diyalog kullanım durumları için optimize edilmiştir ve yaygın endüstri benchmark testlerinde birçok mevcut açık kaynaklı sohbet modelini geride bırakmıştır. Llama 3.1, çok dilli ticari ve araştırma amaçları için tasarlanmıştır. Talimat ayarlı metin modelleri, asistan benzeri sohbetler için uygundur, önceden eğitilmiş modeller ise çeşitli doğal dil üretim görevlerine uyum sağlayabilir. Llama 3.1 modeli, diğer modellerin çıktısını iyileştirmek için de kullanılabilir, bu da veri sentezleme ve rafine etme işlemlerini içerir. Llama 3.1, optimize edilmiş bir transformer mimarisi kullanarak oluşturulmuş bir otoregresif dil modelidir. Ayarlanmış versiyon, insan yardımseverliği ve güvenlik tercihleri ile uyumlu hale getirmek için denetimli ince ayar (SFT) ve insan geri bildirimi ile güçlendirilmiş öğrenme (RLHF) kullanır."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3, geliştiriciler, araştırmacılar ve işletmeler için açık bir büyük dil modelidir (LLM) ve onların üretken AI fikirlerini inşa etmelerine, denemelerine ve sorumlu bir şekilde genişletmelerine yardımcı olmak için tasarlanmıştır. Küresel topluluk yeniliğinin temel sistemlerinden biri olarak, sınırlı hesaplama gücü ve kaynaklara sahip, kenar cihazları ve daha hızlı eğitim süreleri için son derece uygundur."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "Gelişmiş LLM, sentetik veri üretimi, bilgi damıtma ve akıl yürütmeyi destekler, sohbet botları, programlama ve belirli alan görevleri için uygundur."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "Karmaşık diyalogları güçlendiren, mükemmel bağlam anlama, akıl yürütme yeteneği ve metin üretimi yeteneğine sahip."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "En son teknolojiye sahip model, dil anlama, mükemmel akıl yürütme yeteneği ve metin üretimi yeteneğine sahiptir."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "Gelişmiş görsel-dil modeli, görüntülerden yüksek kaliteli akıl yürütme yapma konusunda uzmandır."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "En son teknolojiye sahip küçük dil modeli, dil anlama, mükemmel akıl yürütme yeteneği ve metin üretimi yeteneğine sahiptir."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "En son teknolojiye sahip küçük dil modeli, dil anlama, mükemmel akıl yürütme yeteneği ve metin üretimi yeteneğine sahiptir."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "Gelişmiş görsel-dil modeli, görüntülerden yüksek kaliteli akıl yürütme yapma konusunda uzmandır."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "Akıllı LLM, akıl yürütme, matematik, genel bilgi ve fonksiyon çağrılarında uzmandır."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2, Microsoft AI tarafından sağlanan bir dil modelidir ve karmaşık diyaloglar, çok dilli destek, akıl yürütme ve akıllı asistan alanlarında özellikle başarılıdır."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K, ultra uzun bağlam işleme yeteneğine sahip bir modeldir, karmaşık üretim görevlerini karşılamak için ultra uzun metinler üretmekte kullanılabilir, 128,000 token'a kadar içeriği işleyebilir, araştırma, akademik ve büyük belgelerin üretilmesi gibi uygulama senaryoları için son derece uygundur."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Kimi görsel modeli (moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview gibi) resim içeriğini anlayabilir, resim metni, resim rengi ve nesne şekilleri gibi içerikleri kapsar."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K, orta uzunlukta bağlam işleme yeteneği sunar, 32,768 token'ı işleyebilir, çeşitli uzun belgeler ve karmaşık diyaloglar üretmek için özellikle uygundur, içerik oluşturma, rapor üretimi ve diyalog sistemleri gibi alanlarda kullanılabilir."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Kimi görsel modeli (moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview gibi) resim içeriğini anlayabilir, resim metni, resim rengi ve nesne şekilleri gibi içerikleri kapsar."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K, kısa metin görevleri için tasarlanmış, yüksek verimlilikte işleme performansı sunar, 8,192 token'ı işleyebilir, kısa diyaloglar, not alma ve hızlı içerik üretimi için son derece uygundur."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Kimi görsel modeli (moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview gibi) resim içeriğini anlayabilir, resim metni, resim rengi ve nesne şekilleri gibi içerikleri kapsar."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B, Nous Hermes 2'nin güncellenmiş versiyonudur ve en son iç geliştirme veri setlerini içermektedir."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B, NVIDIA tarafından özelleştirilmiş büyük bir dil modelidir, LLM tarafından üretilen yanıtların kullanıcı sorgularına daha iyi yardımcı olmasını sağlamak için tasarlanmıştır."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B, NVIDIA tarafından özelleştirilmiş büyük bir dil modelidir ve LLM tarafından üretilen yanıtların kullanıcı sorgularına yardımcı olma düzeyini artırmayı amaçlamaktadır. Bu model, Arena Hard, AlpacaEval 2 LC ve GPT-4-Turbo MT-Bench gibi standart testlerde mükemmel performans sergilemiştir ve 1 Ekim 2024 itibarıyla tüm üç otomatik hizalama testinde birinci sıradadır. Model, Llama-3.1-70B-Instruct modelinin temelinde RLHF (özellikle REINFORCE), Llama-3.1-Nemotron-70B-Reward ve HelpSteer2-Preference ipuçları kullanılarak eğitilmiştir."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "Eşsiz bir dil modeli, benzersiz doğruluk ve verimlilik sunar."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct, NVIDIA'nın özel olarak geliştirdiği büyük dil modelidir ve LLM tarafından üretilen yanıtların yardımcı olmasını artırmayı amaçlar."
+ },
"o1": {
"description": "Gelişmiş çıkarım ve karmaşık sorunları çözmeye odaklanır, matematik ve bilim görevlerini içerir. Derin bağlam anlayışı ve aracılık iş akışları gerektiren uygulamalar için son derece uygundur."
},
- "o1-2024-12-17": {
- "description": "o1, OpenAI'nin yeni çıkarım modelidir, metin ve görsel girişi destekler ve metin çıktısı verir, geniş genel bilgi gerektiren karmaşık görevler için uygundur. Bu model, 200K bağlam ve 2023 Ekim bilgi kesim tarihi ile donatılmıştır."
- },
"o1-mini": {
"description": "o1-mini, programlama, matematik ve bilim uygulama senaryoları için tasarlanmış hızlı ve ekonomik bir akıl yürütme modelidir. Bu model, 128K bağlam ve Ekim 2023 bilgi kesim tarihi ile donatılmıştır."
},
"o1-preview": {
"description": "o1, OpenAI'nin geniş genel bilgiye ihtiyaç duyan karmaşık görevler için uygun yeni bir akıl yürütme modelidir. Bu model, 128K bağlam ve Ekim 2023 bilgi kesim tarihi ile donatılmıştır."
},
+ "o3-mini": {
+ "description": "o3-mini, aynı maliyet ve gecikme hedefleriyle yüksek zeka sunan en yeni küçük ölçekli çıkarım modelimizdir."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba, kod üretimine odaklanan Mamba 2 dil modelidir ve ileri düzey kod ve akıl yürütme görevlerine güçlü destek sunar."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini, OpenAI'nin GPT-4 Omni'den sonra sunduğu en son modeldir; görsel ve metin girişi destekler ve metin çıktısı verir. En gelişmiş küçük model olarak, diğer son zamanlardaki öncü modellere göre çok daha ucuzdur ve GPT-3.5 Turbo'dan %60'tan fazla daha ucuzdur. En son teknolojiyi korurken, önemli bir maliyet etkinliği sunar. GPT-4o mini, MMLU testinde %82 puan almış olup, şu anda sohbet tercihleri açısından GPT-4'ün üzerinde bir sıralamaya sahiptir."
},
- "openai/o1": {
- "description": "o1, OpenAI'nin yeni çıkarım modelidir, metin ve görsel girişi destekler ve metin çıktısı verir, geniş genel bilgi gerektiren karmaşık görevler için uygundur. Bu model, 200K bağlam ve 2023 Ekim bilgi kesim tarihi ile donatılmıştır."
- },
"openai/o1-mini": {
"description": "o1-mini, programlama, matematik ve bilim uygulama senaryoları için tasarlanmış hızlı ve ekonomik bir akıl yürütme modelidir. Bu model, 128K bağlam ve Ekim 2023 bilgi kesim tarihi ile donatılmıştır."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K, olağanüstü bağlam işleme yeteneği ile donatılmıştır ve 128K'ya kadar bağlam bilgilerini işleyebilir. Özellikle uzun metinlerin bütünsel analizi ve uzun vadeli mantıksal ilişkilerin işlenmesi gereken durumlar için uygundur ve karmaşık metin iletişiminde akıcı ve tutarlı bir mantık ile çeşitli alıntı desteği sunmaktadır."
},
+ "qvq-72b-preview": {
+ "description": "QVQ modeli, Qwen ekibi tarafından geliştirilen deneysel bir araştırma modelidir; görsel akıl yürütme yeteneğini artırmaya odaklanır, özellikle matematik akıl yürütme alanında."
+ },
"qwen-coder-plus-latest": {
"description": "Tongyi Qianwen kod modeli."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Tongyi Qianwen ultra büyük ölçekli görsel dil modeli. Geliştirilmiş versiyona kıyasla, görsel akıl yürütme yeteneğini ve talimatlara uyum yeteneğini bir kez daha artırır, daha yüksek görsel algı ve bilişsel seviyeler sunar."
},
+ "qwen-vl-ocr-latest": {
+ "description": "Tongyi Qianwen OCR, belge, tablo, sınav soruları, el yazısı gibi çeşitli görüntü türlerinden metin çıkarma yeteneğine odaklanan özel bir modeldir. Birçok yazıyı tanıyabilir, şu anda desteklenen diller: Çince, İngilizce, Fransızca, Japonca, Korece, Almanca, Rusça, İtalyanca, Vietnamca, Arapça."
+ },
"qwen-vl-plus-latest": {
"description": "Tongyi Qianwen büyük ölçekli görsel dil modelinin geliştirilmiş versiyonu. Detay tanıma ve metin tanıma yeteneklerini büyük ölçüde artırır, bir milyondan fazla piksel çözünürlüğü ve herhangi bir en-boy oranındaki görüntüleri destekler."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2, daha güçlü anlama ve üretme yeteneklerine sahip yeni bir büyük dil modeli serisidir."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "Çince ve İngilizce'ye yönelik LLM, dil, programlama, matematik, akıl yürütme gibi alanlara odaklanır."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "Gelişmiş LLM, kod üretimi, akıl yürütme ve düzeltme desteği sunar, ana akım programlama dillerini kapsar."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "Güçlü orta ölçekli kod modeli, 32K bağlam uzunluğunu destekler, çok dilli programlama konusunda uzmandır."
+ },
"qwen2": {
"description": "Qwen2, Alibaba'nın yeni nesil büyük ölçekli dil modelidir, mükemmel performans ile çeşitli uygulama ihtiyaçlarını destekler."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "Tongyi Qianwen 2.5, halka açık 14B ölçeğinde bir modeldir."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "Tongyi Qianwen 2.5, 72B ölçeğinde açık kaynak olarak sunulmuştur."
+ },
"qwen2.5-32b-instruct": {
"description": "Tongyi Qianwen 2.5, halka açık 32B ölçeğinde bir modeldir."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Qwen-Math modeli, güçlü matematik problem çözme yeteneklerine sahiptir."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "Talimat takibi, matematik, problem çözme, kodlama genelinde iyileştirme, her türlü nesneyi tanıma yeteneği artışı, çeşitli formatları doğrudan hassas bir şekilde görsel unsurları konumlandırma desteği, uzun video dosyalarını (en fazla 10 dakika) anlama ve saniye düzeyinde olay anlarını konumlandırma yeteneği, zaman sıralamasını ve hızını anlama, analiz ve konumlandırma yeteneğine dayanarak OS veya Mobil ajanları kontrol etme desteği, anahtar bilgileri çıkarma yeteneği ve Json formatında çıktı verme yeteneği güçlüdür, bu sürüm 72B versiyonudur, bu serinin en güçlü versiyonudur."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "Talimat takibi, matematik, problem çözme, kodlama genelinde iyileştirme, her türlü nesneyi tanıma yeteneği artışı, çeşitli formatları doğrudan hassas bir şekilde görsel unsurları konumlandırma desteği, uzun video dosyalarını (en fazla 10 dakika) anlama ve saniye düzeyinde olay anlarını konumlandırma yeteneği, zaman sıralamasını ve hızını anlama, analiz ve konumlandırma yeteneğine dayanarak OS veya Mobil ajanları kontrol etme desteği, anahtar bilgileri çıkarma yeteneği ve Json formatında çıktı verme yeteneği güçlüdür, bu sürüm 72B versiyonudur, bu serinin en güçlü versiyonudur."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5, Alibaba'nın yeni nesil büyük ölçekli dil modelidir ve mükemmel performansıyla çeşitli uygulama ihtiyaçlarını desteklemektedir."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro, Upstage tarafından sunulan yüksek akıllı LLM'dir, tek GPU talimat takibi yeteneğine odaklanır, IFEval puanı 80'in üzerindedir. Şu anda İngilizceyi desteklemekte olup, resmi versiyonu 2024 Kasım'da piyasaya sürülmesi planlanmaktadır ve dil desteği ile bağlam uzunluğunu genişletecektir."
},
+ "sonar": {
+ "description": "Arama bağlamına dayalı hafif bir arama ürünüdür, Sonar Pro'dan daha hızlı ve daha ucuzdur."
+ },
+ "sonar-pro": {
+ "description": "Gelişmiş sorgular ve takip desteği sunan, arama bağlamını destekleyen bir üst düzey arama ürünüdür."
+ },
+ "sonar-reasoning": {
+ "description": "DeepSeek akıl yürütme modeli tarafından desteklenen yeni API ürünü."
+ },
"step-1-128k": {
"description": "Performans ve maliyet arasında denge sağlar, genel senaryolar için uygundur."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "Bu model, güçlü bir video anlama yeteneğine sahiptir."
},
+ "step-1o-vision-32k": {
+ "description": "Bu model, güçlü bir görüntü anlama yeteneğine sahiptir. Step-1v serisi modellere kıyasla daha güçlü bir görsel performansa sahiptir."
+ },
"step-1v-32k": {
"description": "Görsel girdi desteği sunar, çok modlu etkileşim deneyimini artırır."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Büyük ölçekli bağlam etkileşimlerini destekler, karmaşık diyalog senaryoları için uygundur."
},
+ "step-2-mini": {
+ "description": "Yeni nesil kendi geliştirdiğimiz MFA Attention mimarisine dayanan hızlı büyük model, çok düşük maliyetle step1 ile benzer sonuçlar elde ederken, daha yüksek bir throughput ve daha hızlı yanıt süresi sağlıyor. Genel görevleri işleyebilme yeteneğine sahip olup, kodlama yeteneklerinde uzmanlık gösteriyor."
+ },
+ "taichu2_mm": {
+ "description": "Görüntü anlama, bilgi transferi, mantıksal atıf gibi yetenekleri birleştirerek, metin ve görüntü ile soru-cevap alanında öne çıkmaktadır."
+ },
"taichu_llm": {
"description": "Zidong Taichu dil büyük modeli, güçlü dil anlama yeteneği ile metin oluşturma, bilgi sorgulama, kod programlama, matematik hesaplama, mantıksal akıl yürütme, duygu analizi, metin özeti gibi yeteneklere sahiptir. Yenilikçi bir şekilde büyük veri ön eğitimi ile çok kaynaklı zengin bilgiyi birleştirir, algoritma teknolojisini sürekli olarak geliştirir ve büyük metin verilerinden kelime, yapı, dil bilgisi, anlam gibi yeni bilgileri sürekli olarak edinir, modelin performansını sürekli olarak evrimleştirir. Kullanıcılara daha kolay bilgi ve hizmetler sunar ve daha akıllı bir deneyim sağlar."
},
+ "text-embedding-3-large": {
+ "description": "En güçlü vektörleştirme modeli, İngilizce ve diğer dillerdeki görevler için uygundur."
+ },
+ "text-embedding-3-small": {
+ "description": "Verimli ve ekonomik yeni nesil Embedding modeli, bilgi arama, RAG uygulamaları gibi senaryolar için uygundur."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B), etkili stratejiler ve model mimarisi ile artırılmış hesaplama yetenekleri sunar."
},
+ "tts-1": {
+ "description": "En son metinden sese model, gerçek zamanlı senaryolar için hız optimizasyonu yapılmıştır."
+ },
+ "tts-1-hd": {
+ "description": "En son metinden sese model, kaliteyi optimize etmek için tasarlanmıştır."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B), ince ayar gerektiren talimat görevleri için uygundur ve mükemmel dil işleme yetenekleri sunar."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet, endüstri standartlarını yükselterek, rakip modelleri ve Claude 3 Opus'u aşan performans sergilemekte; geniş değerlendirmelerde mükemmel sonuçlar verirken, orta seviye modellerimizin hız ve maliyetine sahiptir."
},
+ "whisper-1": {
+ "description": "Genel ses tanıma modeli, çok dilli ses tanıma, ses çevirisi ve dil tanıma desteği sunar."
+ },
"wizardlm2": {
"description": "WizardLM 2, Microsoft AI tarafından sunulan bir dil modelidir, karmaşık diyaloglar, çok dilli, akıl yürütme ve akıllı asistan alanlarında özellikle başarılıdır."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Karmaşık görsel görevler için model, yüksek performanslı resim anlama ve analiz yetenekleri sunar."
+ },
+ "yi-vision-v2": {
+ "description": "Karmaşık görsel görevler için model, birden fazla resme dayalı yüksek performanslı anlama ve analiz yetenekleri sunar."
}
}
diff --git a/locales/tr-TR/providers.json b/locales/tr-TR/providers.json
index a1ecbc339b156..150309e5fc80c 100644
--- a/locales/tr-TR/providers.json
+++ b/locales/tr-TR/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure, GPT-3.5 ve en son GPT-4 serisi gibi çeşitli gelişmiş yapay zeka modelleri sunar. Farklı veri türlerini ve karmaşık görevleri destekleyerek güvenli, güvenilir ve sürdürülebilir yapay zeka çözümleri sağlamaya odaklanmaktadır."
},
+ "azureai": {
+ "description": "Azure, GPT-3.5 ve en son GPT-4 serisi dahil olmak üzere çeşitli gelişmiş AI modelleri sunar, çeşitli veri türlerini ve karmaşık görevleri destekler, güvenli, güvenilir ve sürdürülebilir AI çözümlerine odaklanır."
+ },
"baichuan": {
"description": "Baichuan Intelligent, yapay zeka büyük modellerinin geliştirilmesine odaklanan bir şirkettir. Modelleri, yerel bilgi ansiklopedisi, uzun metin işleme ve üretim gibi Çince görevlerde mükemmel performans sergilemekte ve uluslararası ana akım modelleri aşmaktadır. Baichuan Intelligent ayrıca sektördeki lider çok modlu yeteneklere sahiptir ve birçok otoriter değerlendirmede mükemmel sonuçlar elde etmiştir. Modelleri, Baichuan 4, Baichuan 3 Turbo ve Baichuan 3 Turbo 128k gibi farklı uygulama senaryolarına yönelik optimize edilmiş yüksek maliyet etkinliği çözümleri sunmaktadır."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek, yapay zeka teknolojisi araştırma ve uygulamalarına odaklanan bir şirkettir. En son modeli DeepSeek-V2.5, genel diyalog ve kod işleme yeteneklerini birleştirerek, insan tercihleriyle uyum, yazma görevleri ve talimat takibi gibi alanlarda önemli iyileştirmeler sağlamaktadır."
},
+ "doubao": {
+ "description": "ByteDance tarafından geliştirilen kendi büyük modeli. ByteDance içindeki 50'den fazla iş senaryosunda uygulama doğrulaması ile, günlük trilyon seviyesinde token kullanımı ile sürekli olarak geliştirilmekte, çeşitli modalite yetenekleri sunmakta ve kaliteli model performansı ile işletmelere zengin iş deneyimleri yaratmaktadır."
+ },
"fireworksai": {
"description": "Fireworks AI, işlev çağrısı ve çok modlu işleme üzerine odaklanan önde gelen bir gelişmiş dil modeli hizmet sağlayıcısıdır. En son modeli Firefunction V2, Llama-3 tabanlıdır ve işlev çağrısı, diyalog ve talimat takibi için optimize edilmiştir. Görsel dil modeli FireLLaVA-13B, görüntü ve metin karışık girişi desteklemektedir. Diğer dikkat çekici modeller arasında Llama serisi ve Mixtral serisi bulunmaktadır ve etkili çok dilli talimat takibi ve üretim desteği sunmaktadır."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "Büyük model araştırma ve geliştirme araç zincirine adanmış bir açık kaynak organizasyonu. Tüm AI geliştiricilerine verimli ve kullanımı kolay bir açık kaynak platformu sunarak en son büyük model ve algoritma teknolojilerine erişimi kolaylaştırır."
},
+ "jina": {
+ "description": "Jina AI, 2020 yılında kurulmuş, önde gelen bir arama AI şirketidir. Arama tabanlı platformumuz, işletmelerin güvenilir ve yüksek kaliteli üretken AI ve çok modlu arama uygulamaları geliştirmelerine yardımcı olan vektör modelleri, yeniden sıralayıcılar ve küçük dil modelleri içerir."
+ },
+ "lmstudio": {
+ "description": "LM Studio, bilgisayarınızda LLM'ler geliştirmek ve denemeler yapmak için bir masaüstü uygulamasıdır."
+ },
"minimax": {
"description": "MiniMax, 2021 yılında kurulan genel yapay zeka teknolojisi şirketidir ve kullanıcılarla birlikte akıllı çözümler yaratmayı hedeflemektedir. MiniMax, farklı modlarda genel büyük modeller geliştirmiştir. Bunlar arasında trilyon parametreli MoE metin büyük modeli, ses büyük modeli ve görüntü büyük modeli bulunmaktadır. Ayrıca, Conch AI gibi uygulamalar da sunmaktadır."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI, çeşitli büyük dil modelleri ve yapay zeka görüntü üretimi API hizmetleri sunan bir platformdur. Esnek, güvenilir ve maliyet etkin bir yapıya sahiptir. Llama3, Mistral gibi en son açık kaynak modelleri desteklemekte ve üretken yapay zeka uygulama geliştirme için kapsamlı, kullanıcı dostu ve otomatik ölçeklenebilir API çözümleri sunmaktadır. Bu, yapay zeka girişimlerinin hızlı gelişimi için uygundur."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™, bulut, veri merkezi, RTX™ AI kişisel bilgisayarlar ve iş istasyonlarında önceden eğitilmiş ve özelleştirilmiş AI modellerinin dağıtımını destekleyen, kendi kendine barındırılan GPU hızlandırmalı çıkarım mikro hizmetleri için konteynerler sunar."
+ },
"ollama": {
"description": "Ollama'nın sunduğu modeller, kod üretimi, matematiksel işlemler, çok dilli işleme ve diyalog etkileşimi gibi alanları kapsamaktadır. Kurumsal düzeyde ve yerelleştirilmiş dağıtım için çeşitli ihtiyaçları desteklemektedir."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "Çin Bilimler Akademisi Otomasyon Araştırma Enstitüsü ve Wuhan Yapay Zeka Araştırma Enstitüsü, çok modlu büyük modelin yeni neslini sunmaktadır. Çoklu soru-cevap, metin oluşturma, görüntü üretimi, 3D anlama, sinyal analizi gibi kapsamlı soru-cevap görevlerini desteklemekte ve daha güçlü bilişsel, anlama ve yaratma yetenekleri sunarak yeni bir etkileşim deneyimi sağlamaktadır."
},
+ "tencentcloud": {
+ "description": "Bilgi motoru atomik yetenekleri (LLM Knowledge Engine Atomic Power), bilgi motoru üzerine geliştirilmiş bilgi sorgulama tam zincir yetenekleri sunar. Bu yetenekler, işletmeler ve geliştiriciler için esnek model uygulamaları oluşturma ve geliştirme imkanı sağlar. Birden fazla atomik yeteneği kullanarak özel model hizmetlerinizi oluşturabilir, belge analizi, parçalama, embedding, çoklu yeniden yazım gibi hizmetleri bir araya getirerek işletmenize özel AI çözümleri tasarlayabilirsiniz."
+ },
"togetherai": {
"description": "Together AI, yenilikçi yapay zeka modelleri aracılığıyla lider performans elde etmeye odaklanmaktadır. Hızlı ölçeklenme desteği ve sezgisel dağıtım süreçleri dahil olmak üzere geniş özelleştirme yetenekleri sunarak işletmelerin çeşitli ihtiyaçlarını karşılamaktadır."
},
"upstage": {
"description": "Upstage, çeşitli ticari ihtiyaçlar için yapay zeka modelleri geliştirmeye odaklanmaktadır. Solar LLM ve belge AI gibi modeller, insan yapımı genel zeka (AGI) hedeflemektedir. Chat API aracılığıyla basit diyalog ajanları oluşturmakta ve işlev çağrısı, çeviri, gömme ve belirli alan uygulamalarını desteklemektedir."
},
+ "vllm": {
+ "description": "vLLM, LLM çıkarımı ve hizmetleri için hızlı ve kullanımı kolay bir kütüphanedir."
+ },
+ "volcengine": {
+ "description": "ByteDance tarafından sunulan büyük model hizmetleri geliştirme platformu, zengin özellikler, güvenlik ve rekabetçi fiyatlarla model çağırma hizmetleri sunar. Ayrıca model verileri, ince ayar, çıkarım, değerlendirme gibi uçtan uca işlevler sağlar ve AI uygulama geliştirme sürecinizi her yönüyle güvence altına alır."
+ },
"wenxin": {
"description": "Kurumsal düzeyde tek duraklı büyük model ve AI yerel uygulama geliştirme ve hizmet platformu, en kapsamlı ve kullanımı kolay üretken yapay zeka modeli geliştirme, uygulama geliştirme için tam süreç araç zinciri sunar."
},
diff --git a/locales/tr-TR/setting.json b/locales/tr-TR/setting.json
index 34fde2a7f1503..8b82dbb413fbe 100644
--- a/locales/tr-TR/setting.json
+++ b/locales/tr-TR/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Max Token Sınırlamasını Etkinleştir"
},
+ "enableReasoningEffort": {
+ "title": "Akıl yürütme yoğunluğunu ayarla"
+ },
"frequencyPenalty": {
- "desc": "Değer ne kadar yüksekse, tekrarlayan kelimeleri azaltma olasılığı o kadar yüksektir",
- "title": "Frequency Penalty"
+ "desc": "Değer ne kadar büyükse, kelime dağarcığı o kadar zengin ve çeşitli olur; değer ne kadar düşükse, kelimeler o kadar sade ve basit olur.",
+ "title": "Kelime Zenginliği"
},
"maxTokens": {
"desc": "Her etkileşim için kullanılan maksimum token sayısı",
@@ -212,19 +215,31 @@
"desc": "{{provider}} Model",
"title": "Model"
},
+ "params": {
+ "title": "Gelişmiş Parametreler"
+ },
"presencePenalty": {
- "desc": "Değer ne kadar yüksekse, yeni konulara genişleme olasılığı o kadar yüksektir",
- "title": "Presence Penalty"
+ "desc": "Değer ne kadar büyükse, farklı ifade biçimlerine yönelme eğilimi artar, kavram tekrarından kaçınılır; değer ne kadar küçükse, tekrar eden kavramlar veya anlatımlar kullanma eğilimi artar, ifade daha tutarlı olur.",
+ "title": "İfade Çeşitliliği"
+ },
+ "reasoningEffort": {
+ "desc": "Değer ne kadar yüksekse, akıl yürütme yeteneği o kadar güçlüdür, ancak yanıt süresi ve Token tüketimini artırabilir",
+ "options": {
+ "high": "Yüksek",
+ "low": "Düşük",
+ "medium": "Orta"
+ },
+ "title": "Akıl yürütme yoğunluğu"
},
"temperature": {
- "desc": "Değer ne kadar yüksekse, yanıt o kadar rastgele olur",
- "title": "Randomness",
- "titleWithValue": "temperature {{value}}"
+ "desc": "Değer ne kadar büyükse, cevap o kadar yaratıcı ve hayal gücü dolu olur; değer ne kadar küçükse, cevap o kadar titizdir.",
+ "title": "Yaratıcılık Aktifliği",
+ "warning": "Yaratıcılık aktifliği değeri çok büyükse, çıktı bozulabilir."
},
"title": "Model Ayarları",
"topP": {
- "desc": "temperature gibi, ancak temperature ile birlikte değişmez",
- "title": "Top P"
+ "desc": "Ne kadar olasılığı dikkate alır, değer ne kadar büyükse, daha fazla olası cevabı kabul eder; değer ne kadar küçükse, en olası cevabı seçme eğilimindedir. Yaratıcılık aktifliği ile birlikte değiştirilmesi önerilmez.",
+ "title": "Düşünce Açıklığı"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Genel Ayarlar",
"experiment": "Deney",
"llm": "Modeller",
+ "provider": "Yapay Zeka Hizmet Sağlayıcısı",
"sync": "Bulut Senkronizasyonu",
"system-agent": "Sistem Asistanı",
"tts": "Metin Seslendirme"
diff --git a/locales/vi-VN/changelog.json b/locales/vi-VN/changelog.json
index dea0fbeb1097b..4edf362e28aab 100644
--- a/locales/vi-VN/changelog.json
+++ b/locales/vi-VN/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "Xem tất cả nhật ký cập nhật",
"description": "Theo dõi các tính năng và cải tiến mới của {{appName}}",
"pagination": {
- "older": "Xem thay đổi lịch sử",
- "prev": "Trang trước"
+ "next": "Trang tiếp theo",
+ "older": "Xem thay đổi lịch sử"
},
"readDetails": "Đọc chi tiết",
"title": "Nhật ký cập nhật",
diff --git a/locales/vi-VN/common.json b/locales/vi-VN/common.json
index 42fab0ebb57c6..eab906f413dee 100644
--- a/locales/vi-VN/common.json
+++ b/locales/vi-VN/common.json
@@ -275,6 +275,7 @@
},
"temp": "Tạm thời",
"terms": "Điều khoản dịch vụ",
+ "update": "Cập nhật",
"updateAgent": "Cập nhật thông tin trợ lý",
"upgradeVersion": {
"action": "Nâng cấp",
@@ -285,6 +286,7 @@
"anonymousNickName": "Người dùng ẩn danh",
"billing": "Quản lý hóa đơn",
"cloud": "Trải nghiệm {{name}}",
+ "community": "Phiên bản cộng đồng",
"data": "Lưu trữ dữ liệu",
"defaultNickname": "Người dùng phiên bản cộng đồng",
"discord": "Hỗ trợ cộng đồng",
@@ -294,7 +296,6 @@
"help": "Trung tâm trợ giúp",
"moveGuide": "Đã di chuyển nút cài đặt đến đây",
"plans": "Kế hoạch đăng ký",
- "preview": "Phiên bản xem trước",
"profile": "Quản lý tài khoản",
"setting": "Cài đặt ứng dụng",
"usages": "Thống kê sử dụng"
diff --git a/locales/vi-VN/components.json b/locales/vi-VN/components.json
index 221493394473a..6dea1cd04cbf5 100644
--- a/locales/vi-VN/components.json
+++ b/locales/vi-VN/components.json
@@ -76,6 +76,7 @@
"custom": "Mô hình tùy chỉnh, mặc định hỗ trợ cả cuộc gọi hàm và nhận diện hình ảnh, vui lòng xác minh khả năng sử dụng của chúng theo tình hình cụ thể",
"file": "Mô hình này hỗ trợ tải lên và nhận diện tệp",
"functionCall": "Mô hình này hỗ trợ cuộc gọi hàm (Function Call)",
+ "reasoning": "Mô hình này hỗ trợ tư duy sâu sắc",
"tokens": "Mỗi phiên của mô hình này hỗ trợ tối đa {{tokens}} Tokens",
"vision": "Mô hình này hỗ trợ nhận diện hình ảnh"
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "Không có mô hình nào được kích hoạt, vui lòng điều chỉnh trong cài đặt",
"provider": "Nhà cung cấp"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "Do hạn chế bảo mật của trình duyệt, bạn cần cấu hình CORS cho Ollama để có thể sử dụng bình thường.",
+ "linux": {
+ "env": "Thêm `Environment` trong phần [Service], thêm biến môi trường OLLAMA_ORIGINS:",
+ "reboot": "Tải lại systemd và khởi động lại Ollama",
+ "systemd": "Gọi systemd để chỉnh sửa dịch vụ ollama:"
+ },
+ "macos": "Vui lòng mở ứng dụng «Terminal» và dán lệnh sau, sau đó nhấn Enter để chạy",
+ "reboot": "Vui lòng khởi động lại dịch vụ Ollama sau khi hoàn thành",
+ "title": "Cấu hình Ollama cho phép truy cập CORS",
+ "windows": "Trên Windows, nhấp vào «Control Panel», vào chỉnh sửa biến môi trường hệ thống. Tạo một biến môi trường mới có tên là «OLLAMA_ORIGINS» cho tài khoản người dùng của bạn, giá trị là *, nhấp «OK/Apply» để lưu"
+ },
+ "install": {
+ "description": "Vui lòng xác nhận rằng bạn đã mở Ollama, nếu chưa tải Ollama, hãy truy cập trang web chính thức <1>tải xuống",
+ "docker": "Nếu bạn thích sử dụng Docker, Ollama cũng cung cấp hình ảnh Docker chính thức, bạn có thể kéo xuống bằng lệnh sau:",
+ "linux": {
+ "command": "Cài đặt bằng lệnh sau:",
+ "manual": "Hoặc, bạn cũng có thể tham khảo <1>Hướng dẫn cài đặt thủ công trên Linux để tự cài đặt"
+ },
+ "title": "Cài đặt và khởi động ứng dụng Ollama trên máy tính",
+ "windowsTab": "Windows (phiên bản xem trước)"
+ }
+ },
+ "Thinking": {
+ "thinking": "Đang suy nghĩ sâu sắc...",
+ "thought": "Đã suy nghĩ sâu sắc (mất {{duration}} giây)",
+ "thoughtWithDuration": "Đã suy nghĩ sâu sắc"
}
}
diff --git a/locales/vi-VN/discover.json b/locales/vi-VN/discover.json
index 804488c247e77..726f4a2bfa475 100644
--- a/locales/vi-VN/discover.json
+++ b/locales/vi-VN/discover.json
@@ -126,6 +126,10 @@
"title": "Độ mới của chủ đề"
},
"range": "Phạm vi",
+ "reasoning_effort": {
+ "desc": "Cài đặt này được sử dụng để kiểm soát mức độ suy luận của mô hình trước khi tạo câu trả lời. Mức độ thấp ưu tiên tốc độ phản hồi và tiết kiệm Token, trong khi mức độ cao cung cấp suy luận đầy đủ hơn nhưng tiêu tốn nhiều Token hơn và làm giảm tốc độ phản hồi. Giá trị mặc định là trung bình, cân bằng giữa độ chính xác của suy luận và tốc độ phản hồi.",
+ "title": "Mức độ suy luận"
+ },
"temperature": {
"desc": "Cài đặt này ảnh hưởng đến sự đa dạng trong phản hồi của mô hình. Giá trị thấp hơn dẫn đến phản hồi dễ đoán và điển hình hơn, trong khi giá trị cao hơn khuyến khích phản hồi đa dạng và không thường gặp. Khi giá trị được đặt là 0, mô hình sẽ luôn đưa ra cùng một phản hồi cho đầu vào nhất định.",
"title": "Ngẫu nhiên"
diff --git a/locales/vi-VN/modelProvider.json b/locales/vi-VN/modelProvider.json
index e1df86d6f43db..27b0cc724fb65 100644
--- a/locales/vi-VN/modelProvider.json
+++ b/locales/vi-VN/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API Key"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Phiên bản API của Azure, theo định dạng YYYY-MM-DD, tham khảo [phiên bản mới nhất](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "Lấy danh sách",
+ "title": "Phiên bản API Azure"
+ },
+ "endpoint": {
+ "desc": "Tìm điểm kết thúc suy diễn mô hình Azure AI từ tổng quan dự án Azure AI",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Điểm kết thúc Azure AI"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "Tìm khóa API từ tổng quan dự án Azure AI",
+ "placeholder": "Khóa Azure",
+ "title": "Khóa"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "Nhập AWS Access Key Id",
@@ -63,6 +81,46 @@
"title": "ID tài khoản Cloudflare / địa chỉ API"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "Vui lòng nhập API Key của bạn",
+ "title": "API Key"
+ },
+ "basicTitle": "Thông tin cơ bản",
+ "configTitle": "Thông tin cấu hình",
+ "confirm": "Tạo mới",
+ "createSuccess": "Tạo mới thành công",
+ "description": {
+ "placeholder": "Giới thiệu về nhà cung cấp (tùy chọn)",
+ "title": "Giới thiệu về nhà cung cấp"
+ },
+ "id": {
+ "desc": "Là định danh duy nhất của nhà cung cấp dịch vụ, không thể sửa đổi sau khi tạo",
+ "format": "Chỉ có thể chứa số, chữ cái thường, dấu gạch ngang (-) và dấu gạch dưới (_) ",
+ "placeholder": "Nên viết toàn bộ bằng chữ thường, ví dụ openai, không thể sửa sau khi tạo",
+ "required": "Vui lòng nhập ID nhà cung cấp",
+ "title": "ID nhà cung cấp"
+ },
+ "logo": {
+ "required": "Vui lòng tải lên Logo nhà cung cấp hợp lệ",
+ "title": "Logo nhà cung cấp"
+ },
+ "name": {
+ "placeholder": "Vui lòng nhập tên hiển thị của nhà cung cấp",
+ "required": "Vui lòng nhập tên nhà cung cấp",
+ "title": "Tên nhà cung cấp"
+ },
+ "proxyUrl": {
+ "required": "Vui lòng nhập địa chỉ proxy",
+ "title": "Địa chỉ proxy"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "Vui lòng chọn loại SDK",
+ "title": "Định dạng yêu cầu"
+ },
+ "title": "Tạo nhà cung cấp AI tùy chỉnh"
+ },
"github": {
"personalAccessToken": {
"desc": "Nhập mã truy cập cá nhân Github của bạn, nhấp vào [đây](https://github.com/settings/tokens) để tạo",
@@ -77,6 +135,23 @@
"title": "Mã thông báo HuggingFace"
}
},
+ "list": {
+ "title": {
+ "disabled": "Nhà cung cấp chưa được kích hoạt",
+ "enabled": "Nhà cung cấp đã được kích hoạt"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "Thêm nhà cung cấp tùy chỉnh",
+ "all": "Tất cả",
+ "list": {
+ "disabled": "Chưa kích hoạt",
+ "enabled": "Đã kích hoạt"
+ },
+ "notFound": "Không tìm thấy kết quả tìm kiếm",
+ "searchProviders": "Tìm kiếm nhà cung cấp...",
+ "sort": "Sắp xếp tùy chỉnh"
+ },
"ollama": {
"checker": {
"desc": "Kiểm tra địa chỉ proxy có được nhập chính xác không",
@@ -94,33 +169,9 @@
"title": "Đang tải mô hình {{model}}"
},
"endpoint": {
- "desc": "Nhập địa chỉ proxy API của Ollama, có thể để trống nếu không chỉ định cụ thể",
+ "desc": "Phải bao gồm http(s)://, có thể để trống nếu không chỉ định thêm cho địa phương",
"title": "Địa chỉ proxy API"
},
- "setup": {
- "cors": {
- "description": "Do vấn đề về an ninh trình duyệt, bạn cần cấu hình CORS cho Ollama trước khi có thể sử dụng bình thường.",
- "linux": {
- "env": "Trong phần [Service], thêm `Environment`, thêm biến môi trường OLLAMA_ORIGINS:",
- "reboot": "Tải lại systemd và khởi động lại Ollama",
- "systemd": "Gọi systemd để chỉnh sửa dịch vụ ollama:"
- },
- "macos": "Vui lòng mở ứng dụng «Terminal», dán lệnh sau và nhấn Enter để chạy",
- "reboot": "Vui lòng khởi động lại dịch vụ Ollama sau khi hoàn thành",
- "title": "Cấu hình Ollama cho phép truy cập từ xa",
- "windows": "Trên Windows, nhấp vào «Control Panel», vào chỉnh sửa biến môi trường hệ thống. Tạo biến môi trường tên là «OLLAMA_ORIGINS» cho tài khoản người dùng của bạn, giá trị là * , nhấp vào «OK/Áp dụng» để lưu lại"
- },
- "install": {
- "description": "Vui lòng xác nhận rằng bạn đã bật Ollama. Nếu chưa tải Ollama, vui lòng truy cập trang web chính thức để <1>tải xuống",
- "docker": "Nếu bạn muốn sử dụng Docker, Ollama cũng cung cấp hình ảnh Docker chính thức, bạn có thể kéo theo lệnh sau:",
- "linux": {
- "command": "Cài đặt bằng lệnh sau:",
- "manual": "Hoặc bạn cũng có thể tham khảo <1>Hướng dẫn cài đặt thủ công trên Linux để tự cài đặt"
- },
- "title": "Cài đặt và mở Ollama ứng dụng trên máy cục bộ",
- "windowsTab": "Windows (Bản xem trước)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "Hủy tải xuống",
@@ -131,25 +182,145 @@
"title": "Tải xuống mô hình Ollama đã chỉ định"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "Nhập Access Key từ nền tảng Qianfan của Baidu",
- "placeholder": "Access Key Qianfan",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "Khóa của bạn và địa chỉ proxy sẽ được mã hóa bằng thuật toán <1>AES-GCM",
+ "apiKey": {
+ "desc": "Vui lòng nhập {{name}} API Key của bạn",
+ "placeholder": "{{name}} API Key",
+ "title": "API Key"
+ },
+ "baseURL": {
+ "desc": "Phải bao gồm http(s)://",
+ "invalid": "Vui lòng nhập một URL hợp lệ",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "Địa chỉ proxy API"
+ },
+ "checker": {
+ "button": "Kiểm tra",
+ "desc": "Kiểm tra xem API Key và địa chỉ proxy có được nhập đúng không",
+ "pass": "Kiểm tra thành công",
+ "title": "Kiểm tra kết nối"
+ },
+ "fetchOnClient": {
+ "desc": "Chế độ yêu cầu từ khách hàng sẽ phát động yêu cầu phiên trực tiếp từ trình duyệt, có thể cải thiện tốc độ phản hồi",
+ "title": "Sử dụng chế độ yêu cầu từ khách hàng"
+ },
+ "helpDoc": "Hướng dẫn cấu hình",
+ "waitingForMore": "Nhiều mô hình hơn đang <1>được lên kế hoạch, xin hãy chờ đợi"
},
- "checker": {
- "desc": "Kiểm tra xem AccessKey / SecretAccess có được nhập đúng không"
+ "createNew": {
+ "title": "Tạo mô hình AI tùy chỉnh"
},
- "secretKey": {
- "desc": "Nhập Secret Key từ nền tảng Qianfan của Baidu",
- "placeholder": "Secret Key Qianfan",
- "title": "Secret Key"
+ "item": {
+ "config": "Cấu hình mô hình",
+ "customModelCards": {
+ "addNew": "Tạo và thêm mô hình {{id}}",
+ "confirmDelete": "Sắp xóa mô hình tùy chỉnh này, sau khi xóa sẽ không thể khôi phục, xin hãy cẩn thận."
+ },
+ "delete": {
+ "confirm": "Xác nhận xóa mô hình {{displayName}}?",
+ "success": "Xóa thành công",
+ "title": "Xóa mô hình"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "Trường thực tế được yêu cầu trong Azure OpenAI",
+ "placeholder": "Vui lòng nhập tên triển khai mô hình trong Azure",
+ "title": "Tên triển khai mô hình"
+ },
+ "deployName": {
+ "extra": "Trường này sẽ được sử dụng làm ID mô hình khi gửi yêu cầu",
+ "placeholder": "Vui lòng nhập tên hoặc ID thực tế của mô hình đã triển khai",
+ "title": "Tên triển khai mô hình"
+ },
+ "displayName": {
+ "placeholder": "Vui lòng nhập tên hiển thị của mô hình, ví dụ ChatGPT, GPT-4, v.v.",
+ "title": "Tên hiển thị mô hình"
+ },
+ "files": {
+ "extra": "Hiện tại, việc tải lên tệp chỉ là một giải pháp Hack, chỉ dành cho thử nghiệm cá nhân. Vui lòng chờ đợi khả năng tải lên tệp hoàn chỉnh trong các bản cập nhật sau.",
+ "title": "Hỗ trợ tải lên tệp"
+ },
+ "functionCall": {
+ "extra": "Cấu hình này chỉ kích hoạt khả năng sử dụng công cụ của mô hình, từ đó có thể thêm các plugin loại công cụ cho mô hình. Tuy nhiên, việc hỗ trợ sử dụng công cụ thực sự hoàn toàn phụ thuộc vào chính mô hình, vui lòng tự kiểm tra tính khả dụng",
+ "title": "Hỗ trợ sử dụng công cụ"
+ },
+ "id": {
+ "extra": "Không thể sửa đổi sau khi tạo, sẽ được sử dụng làm id mô hình khi gọi AI",
+ "placeholder": "Vui lòng nhập id mô hình, ví dụ gpt-4o hoặc claude-3.5-sonnet",
+ "title": "ID mô hình"
+ },
+ "modalTitle": "Cấu hình mô hình tùy chỉnh",
+ "reasoning": {
+ "extra": "Cấu hình này sẽ chỉ kích hoạt khả năng suy nghĩ sâu của mô hình, hiệu quả cụ thể hoàn toàn phụ thuộc vào chính mô hình, vui lòng tự kiểm tra xem mô hình này có khả năng suy nghĩ sâu có thể sử dụng hay không",
+ "title": "Hỗ trợ suy nghĩ sâu"
+ },
+ "tokens": {
+ "extra": "Cài đặt số Token tối đa mà mô hình hỗ trợ",
+ "title": "Cửa sổ ngữ cảnh tối đa",
+ "unlimited": "Không giới hạn"
+ },
+ "vision": {
+ "extra": "Cấu hình này chỉ mở khả năng tải lên hình ảnh trong ứng dụng, việc hỗ trợ nhận diện hoàn toàn phụ thuộc vào mô hình, xin hãy tự kiểm tra khả năng nhận diện hình ảnh của mô hình này.",
+ "title": "Hỗ trợ nhận diện hình ảnh"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/Hình ảnh",
+ "inputCharts": "${{amount}}/Ký tự M",
+ "inputMinutes": "${{amount}}/Phút",
+ "inputTokens": "Nhập ${{amount}}/M",
+ "outputTokens": "Xuất ${{amount}}/M"
+ },
+ "releasedAt": "Phát hành vào {{releasedAt}}"
},
- "unlock": {
- "customRegion": "Khu vực dịch vụ tùy chỉnh",
- "description": "Nhập AccessKey / SecretKey của bạn để bắt đầu phiên trò chuyện. Ứng dụng sẽ không ghi lại cấu hình xác thực của bạn",
- "title": "Sử dụng thông tin xác thực của Wenxin Yiyan tùy chỉnh"
- }
+ "list": {
+ "addNew": "Thêm mô hình",
+ "disabled": "Chưa được kích hoạt",
+ "disabledActions": {
+ "showMore": "Hiển thị tất cả"
+ },
+ "empty": {
+ "desc": "Vui lòng tạo mô hình tùy chỉnh hoặc kéo mô hình để bắt đầu sử dụng",
+ "title": "Chưa có mô hình nào khả dụng"
+ },
+ "enabled": "Đã được kích hoạt",
+ "enabledActions": {
+ "disableAll": "Vô hiệu hóa tất cả",
+ "enableAll": "Kích hoạt tất cả",
+ "sort": "Sắp xếp mô hình tùy chỉnh"
+ },
+ "enabledEmpty": "Chưa có mô hình nào được kích hoạt, hãy kích hoạt mô hình bạn yêu thích từ danh sách bên dưới nhé~",
+ "fetcher": {
+ "clear": "Xóa mô hình đã lấy",
+ "fetch": "Lấy danh sách mô hình",
+ "fetching": "Đang lấy danh sách mô hình...",
+ "latestTime": "Thời gian cập nhật lần cuối: {{time}}",
+ "noLatestTime": "Chưa lấy danh sách"
+ },
+ "resetAll": {
+ "conform": "Xác nhận việc đặt lại tất cả các thay đổi của mô hình hiện tại? Sau khi đặt lại, danh sách mô hình hiện tại sẽ trở về trạng thái mặc định",
+ "success": "Đặt lại thành công",
+ "title": "Đặt lại tất cả các thay đổi"
+ },
+ "search": "Tìm kiếm mô hình...",
+ "searchResult": "Tìm thấy {{count}} mô hình",
+ "title": "Danh sách mô hình",
+ "total": "Có tổng cộng {{count}} mô hình khả dụng"
+ },
+ "searchNotFound": "Không tìm thấy kết quả tìm kiếm"
+ },
+ "sortModal": {
+ "success": "Cập nhật sắp xếp thành công",
+ "title": "Sắp xếp tùy chỉnh",
+ "update": "Cập nhật"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "Sắp xóa nhà cung cấp AI này, sau khi xóa sẽ không thể khôi phục, xác nhận có xóa không?",
+ "deleteSuccess": "Xóa thành công",
+ "tooltip": "Cập nhật cấu hình cơ bản của nhà cung cấp",
+ "updateSuccess": "Cập nhật thành công"
},
"zeroone": {
"title": "01.AI Zero One"
diff --git a/locales/vi-VN/models.json b/locales/vi-VN/models.json
index e3c136e3c9ca8..1985c78f77a96 100644
--- a/locales/vi-VN/models.json
+++ b/locales/vi-VN/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K nhấn mạnh an toàn ngữ nghĩa và định hướng trách nhiệm, được thiết kế đặc biệt cho các tình huống ứng dụng có yêu cầu cao về an toàn nội dung, đảm bảo độ chính xác và độ ổn định trong trải nghiệm người dùng."
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 sử dụng tìm kiếm cây để xây dựng chuỗi tư duy, và đưa vào cơ chế phản hồi, sử dụng học tăng cường để đào tạo, mô hình có khả năng tự phản hồi và sửa lỗi."
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro là mô hình xử lý ngôn ngữ tự nhiên cao cấp do công ty 360 phát hành, có khả năng tạo và hiểu văn bản xuất sắc, đặc biệt trong lĩnh vực tạo ra và sáng tạo, có thể xử lý các nhiệm vụ chuyển đổi ngôn ngữ phức tạp và diễn xuất vai trò."
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1 sử dụng tìm kiếm cây để xây dựng chuỗi tư duy, và giới thiệu cơ chế phản hồi, sử dụng học tăng cường để đào tạo, mô hình có khả năng tự phản hồi và sửa lỗi."
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra là phiên bản mạnh mẽ nhất trong dòng mô hình lớn Xinghuo, nâng cao khả năng hiểu và tóm tắt nội dung văn bản trong khi nâng cấp liên kết tìm kiếm trực tuyến. Đây là giải pháp toàn diện nhằm nâng cao năng suất văn phòng và đáp ứng chính xác nhu cầu, là sản phẩm thông minh dẫn đầu ngành."
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "Mô hình có khả năng hàng đầu trong nước, vượt trội hơn các mô hình chính thống nước ngoài trong các nhiệm vụ tiếng Trung như bách khoa toàn thư, văn bản dài và sáng tạo nội dung. Cũng có khả năng đa phương tiện hàng đầu trong ngành, thể hiện xuất sắc trong nhiều tiêu chuẩn đánh giá uy tín."
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Mô hình chưng cất DeepSeek-R1 dựa trên Qwen2.5-Math-1.5B, tối ưu hóa hiệu suất suy luận thông qua học tăng cường và dữ liệu khởi động lạnh, mô hình mã nguồn mở làm mới tiêu chuẩn đa nhiệm."
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Mô hình chưng cất DeepSeek-R1 dựa trên Qwen2.5-14B, tối ưu hóa hiệu suất suy luận thông qua học tăng cường và dữ liệu khởi động lạnh, mô hình mã nguồn mở làm mới tiêu chuẩn đa nhiệm."
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "Dòng DeepSeek-R1 tối ưu hóa hiệu suất suy luận thông qua học tăng cường và dữ liệu khởi động lạnh, mô hình mã nguồn mở làm mới tiêu chuẩn đa nhiệm, vượt qua mức OpenAI-o1-mini."
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "Mô hình chưng cất DeepSeek-R1 dựa trên Qwen2.5-Math-7B, tối ưu hóa hiệu suất suy luận thông qua học tăng cường và dữ liệu khởi động lạnh, mô hình mã nguồn mở làm mới tiêu chuẩn đa nhiệm."
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite có tốc độ phản hồi cực nhanh, giá trị tốt hơn, cung cấp sự lựa chọn linh hoạt cho khách hàng trong nhiều tình huống khác nhau. Hỗ trợ suy diễn và tinh chỉnh trong ngữ cảnh 128k."
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "Mô hình ngôn ngữ quy mô siêu lớn hàng đầu do Baidu tự phát triển, so với ERNIE 3.5 đã nâng cấp toàn diện khả năng của mô hình, phù hợp rộng rãi với các nhiệm vụ phức tạp trong nhiều lĩnh vực; hỗ trợ tự động kết nối với plugin tìm kiếm Baidu, đảm bảo thông tin hỏi đáp luôn cập nhật."
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "Mô hình ngôn ngữ lớn siêu cấp do Baidu tự phát triển, thể hiện hiệu suất xuất sắc, phù hợp rộng rãi với các tình huống nhiệm vụ phức tạp trong nhiều lĩnh vực; hỗ trợ tự động kết nối với plugin tìm kiếm Baidu, đảm bảo thông tin hỏi đáp kịp thời. So với ERNIE 4.0, hiệu suất tốt hơn."
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "Mô hình ngôn ngữ quy mô siêu lớn tự phát triển của Baidu, có hiệu suất tổng thể xuất sắc, phù hợp rộng rãi cho các tình huống tác vụ phức tạp trong nhiều lĩnh vực; hỗ trợ tự động kết nối với plugin tìm kiếm của Baidu, đảm bảo tính kịp thời của thông tin câu hỏi đáp. So với ERNIE 4.0, nó có hiệu suất tốt hơn."
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct là một trong những mô hình ngôn ngữ lớn mới nhất do Alibaba Cloud phát hành. Mô hình 7B này có khả năng cải thiện đáng kể trong các lĩnh vực mã hóa và toán học. Mô hình cũng cung cấp hỗ trợ đa ngôn ngữ, bao gồm hơn 29 ngôn ngữ, bao gồm tiếng Trung, tiếng Anh, v.v. Mô hình đã có sự cải thiện đáng kể trong việc tuân theo chỉ dẫn, hiểu dữ liệu có cấu trúc và tạo ra đầu ra có cấu trúc (đặc biệt là JSON)."
},
+ "MiniMax-Text-01": {
+ "description": "Trong dòng mô hình MiniMax-01, chúng tôi đã thực hiện những đổi mới táo bạo: lần đầu tiên hiện thực hóa quy mô lớn cơ chế chú ý tuyến tính, kiến trúc Transformer truyền thống không còn là lựa chọn duy nhất. Mô hình này có số lượng tham số lên tới 4560 tỷ, trong đó kích hoạt một lần là 45,9 tỷ. Hiệu suất tổng hợp của mô hình tương đương với các mô hình hàng đầu quốc tế, đồng thời có khả năng xử lý hiệu quả ngữ cảnh dài nhất toàn cầu lên tới 4 triệu token, gấp 32 lần GPT-4o và 20 lần Claude-3.5-Sonnet."
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO là một mô hình kết hợp đa dạng, nhằm cung cấp trải nghiệm sáng tạo xuất sắc."
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2 đã thể hiện hiệu suất xuất sắc trong nhiều tác vụ ngôn ngữ hình ảnh, bao gồm hiểu tài liệu và biểu đồ, hiểu văn bản trong cảnh, OCR, giải quyết vấn đề khoa học và toán học."
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2 đã thể hiện hiệu suất xuất sắc trong nhiều tác vụ ngôn ngữ hình ảnh, bao gồm hiểu tài liệu và biểu đồ, hiểu văn bản trong cảnh, OCR, giải quyết vấn đề khoa học và toán học."
- },
"Phi-3-medium-128k-instruct": {
"description": "Mô hình Phi-3-medium giống nhau, nhưng với kích thước ngữ cảnh lớn hơn cho RAG hoặc gợi ý ít."
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat là phiên bản mã nguồn mở trong loạt mô hình tiền huấn luyện GLM-4 do Zhizhu AI phát hành. Mô hình này thể hiện xuất sắc trong nhiều lĩnh vực như ngữ nghĩa, toán học, suy luận, mã và kiến thức. Ngoài việc hỗ trợ đối thoại nhiều vòng, GLM-4-9B-Chat còn có các tính năng nâng cao như duyệt web, thực thi mã, gọi công cụ tùy chỉnh (Function Call) và suy luận văn bản dài. Mô hình hỗ trợ 26 ngôn ngữ, bao gồm tiếng Trung, tiếng Anh, tiếng Nhật, tiếng Hàn và tiếng Đức. Trong nhiều bài kiểm tra chuẩn, GLM-4-9B-Chat đã thể hiện hiệu suất xuất sắc, như AlignBench-v2, MT-Bench, MMLU và C-Eval. Mô hình hỗ trợ độ dài ngữ cảnh tối đa 128K, phù hợp cho nghiên cứu học thuật và ứng dụng thương mại."
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 là một mô hình suy diễn được điều khiển bởi học tăng cường (RL), giải quyết các vấn đề về tính lặp lại và khả năng đọc trong mô hình. Trước khi áp dụng RL, DeepSeek-R1 đã giới thiệu dữ liệu khởi động lạnh, tối ưu hóa thêm hiệu suất suy diễn. Nó thể hiện hiệu suất tương đương với OpenAI-o1 trong các nhiệm vụ toán học, mã và suy diễn, và thông qua phương pháp đào tạo được thiết kế cẩn thận, nâng cao hiệu quả tổng thể."
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 là một mô hình ngôn ngữ hỗn hợp chuyên gia (MoE) với 6710 tỷ tham số, sử dụng chú ý tiềm ẩn đa đầu (MLA) và kiến trúc DeepSeekMoE, kết hợp chiến lược cân bằng tải không có tổn thất phụ trợ, tối ưu hóa hiệu suất suy diễn và đào tạo. Thông qua việc được tiền huấn luyện trên 14.8 triệu tỷ token chất lượng cao, và thực hiện tinh chỉnh giám sát và học tăng cường, DeepSeek-V3 vượt trội hơn các mô hình mã nguồn mở khác, gần với các mô hình đóng kín hàng đầu."
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma là một trong những loạt mô hình mở tiên tiến nhẹ của Google. Đây là một mô hình ngôn ngữ quy mô lớn chỉ có bộ giải mã, hỗ trợ tiếng Anh, cung cấp trọng số mở, biến thể tiền huấn luyện và biến thể tinh chỉnh theo chỉ dẫn. Mô hình Gemma phù hợp cho nhiều nhiệm vụ sinh văn bản, bao gồm hỏi đáp, tóm tắt và suy luận. Mô hình 9B này được đào tạo trên 8 triệu tỷ tokens. Quy mô tương đối nhỏ của nó cho phép triển khai trong các môi trường hạn chế tài nguyên, như máy tính xách tay, máy tính để bàn hoặc cơ sở hạ tầng đám mây của riêng bạn, giúp nhiều người hơn có thể tiếp cận các mô hình AI tiên tiến và thúc đẩy đổi mới."
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 là một phần của gia đình mô hình ngôn ngữ lớn đa ngôn ngữ do Meta phát triển, bao gồm các biến thể tiền huấn luyện và tinh chỉnh theo chỉ dẫn với quy mô tham số 8B, 70B và 405B. Mô hình 8B này được tối ưu hóa cho các tình huống đối thoại đa ngôn ngữ, thể hiện xuất sắc trong nhiều bài kiểm tra chuẩn ngành. Mô hình được đào tạo bằng hơn 15 triệu tỷ tokens từ dữ liệu công khai và sử dụng các kỹ thuật như tinh chỉnh giám sát và học tăng cường phản hồi của con người để nâng cao tính hữu ích và an toàn của mô hình. Llama 3.1 hỗ trợ sinh văn bản và sinh mã, với thời điểm cắt kiến thức là tháng 12 năm 2023."
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview là một mô hình xử lý ngôn ngữ tự nhiên độc đáo, có khả năng xử lý hiệu quả các nhiệm vụ tạo đối thoại phức tạp và hiểu ngữ cảnh."
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview là một mô hình nghiên cứu do đội ngũ Qwen phát triển, tập trung vào khả năng suy diễn hình ảnh, có lợi thế độc đáo trong việc hiểu các cảnh phức tạp và giải quyết các vấn đề toán học liên quan đến hình ảnh."
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview là mô hình nghiên cứu thử nghiệm mới nhất của Qwen, tập trung vào việc nâng cao khả năng suy luận của AI. Thông qua việc khám phá các cơ chế phức tạp như trộn ngôn ngữ và suy luận đệ quy, những lợi thế chính bao gồm khả năng phân tích suy luận mạnh mẽ, khả năng toán học và lập trình. Tuy nhiên, cũng có những vấn đề về chuyển đổi ngôn ngữ, vòng lặp suy luận, các vấn đề an toàn và sự khác biệt về các khả năng khác."
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct là phiên bản mới nhất trong loạt mô hình ngôn ngữ lớn chuyên biệt cho mã do Alibaba Cloud phát hành. Mô hình này được cải thiện đáng kể khả năng tạo mã, suy luận và sửa chữa thông qua việc đào tạo trên 5.5 triệu tỷ tokens, không chỉ nâng cao khả năng lập trình mà còn duy trì lợi thế về khả năng toán học và tổng quát. Mô hình cung cấp nền tảng toàn diện hơn cho các ứng dụng thực tế như tác nhân mã."
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math tập trung vào việc giải quyết các vấn đề trong lĩnh vực toán học, cung cấp giải pháp chuyên nghiệp cho các bài toán khó."
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 là dòng mô hình mới nhất của Qwen, hỗ trợ ngữ cảnh 128k, so với các mô hình mã nguồn mở tốt nhất hiện tại, Qwen2-72B vượt trội hơn hẳn trong nhiều khả năng như hiểu ngôn ngữ tự nhiên, kiến thức, mã, toán học và đa ngôn ngữ."
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 là dòng mô hình mới nhất của Qwen, có khả năng vượt qua các mô hình mã nguồn mở cùng quy mô hoặc thậm chí lớn hơn, Qwen2 7B đạt được lợi thế đáng kể trong nhiều bài kiểm tra, đặc biệt là trong việc hiểu mã và tiếng Trung."
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B là một mô hình ngôn ngữ hình ảnh mạnh mẽ, hỗ trợ xử lý đa phương thức giữa hình ảnh và văn bản, có khả năng nhận diện chính xác nội dung hình ảnh và sinh ra mô tả hoặc câu trả lời liên quan."
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct là một mô hình ngôn ngữ lớn với 14 tỷ tham số, có hiệu suất xuất sắc, tối ưu cho các tình huống tiếng Trung và đa ngôn ngữ, hỗ trợ các ứng dụng như hỏi đáp thông minh, tạo nội dung."
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct là một mô hình ngôn ngữ lớn với 7 tỷ tham số, hỗ trợ gọi hàm và tương tác liền mạch với các hệ thống bên ngoài, nâng cao tính linh hoạt và khả năng mở rộng. Tối ưu cho các tình huống tiếng Trung và đa ngôn ngữ, hỗ trợ các ứng dụng như hỏi đáp thông minh, tạo nội dung."
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct là một mô hình hướng dẫn lập trình dựa trên đào tạo trước quy mô lớn, có khả năng hiểu và sinh mã mạnh mẽ, có thể xử lý hiệu quả các nhiệm vụ lập trình khác nhau, đặc biệt phù hợp cho việc viết mã thông minh, tạo kịch bản tự động và giải đáp các vấn đề lập trình."
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct là một mô hình ngôn ngữ lớn được thiết kế đặc biệt cho việc tạo mã, hiểu mã và các tình huống phát triển hiệu quả, với quy mô 32B tham số hàng đầu trong ngành, có thể đáp ứng nhu cầu lập trình đa dạng."
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "Mô hình đa phương tiện TeleMM là một mô hình hiểu đa phương tiện do China Telecom phát triển, có khả năng xử lý nhiều loại đầu vào như văn bản và hình ảnh, hỗ trợ các chức năng như hiểu hình ảnh, phân tích biểu đồ, cung cấp dịch vụ hiểu đa phương tiện cho người dùng. Mô hình có khả năng tương tác đa phương tiện với người dùng, hiểu chính xác nội dung đầu vào, trả lời câu hỏi, hỗ trợ sáng tạo và cung cấp thông tin và cảm hứng đa phương tiện một cách hiệu quả. Mô hình thể hiện xuất sắc trong các nhiệm vụ đa phương tiện như nhận thức chi tiết và suy luận logic."
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large là mô hình MoE kiến trúc Transformer mã nguồn mở lớn nhất trong ngành, với tổng số tham số là 3890 tỷ và 52 tỷ tham số kích hoạt."
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct là mô hình ngôn ngữ lớn được tinh chỉnh theo chỉ dẫn trong loạt Qwen2, với quy mô tham số là 72B. Mô hình này dựa trên kiến trúc Transformer, sử dụng hàm kích hoạt SwiGLU, độ lệch QKV trong chú ý và chú ý theo nhóm. Nó có khả năng xử lý đầu vào quy mô lớn. Mô hình thể hiện xuất sắc trong nhiều bài kiểm tra chuẩn về hiểu ngôn ngữ, sinh ngôn ngữ, khả năng đa ngôn ngữ, mã hóa, toán học và suy luận, vượt qua hầu hết các mô hình mã nguồn mở và thể hiện sức cạnh tranh tương đương với các mô hình độc quyền trong một số nhiệm vụ."
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct là một trong những mô hình ngôn ngữ lớn mới nhất do Alibaba Cloud phát hành. Mô hình 72B này có khả năng cải thiện đáng kể trong các lĩnh vực mã hóa và toán học. Mô hình cũng cung cấp hỗ trợ đa ngôn ngữ, bao gồm hơn 29 ngôn ngữ, bao gồm tiếng Trung, tiếng Anh, v.v. Mô hình đã có sự cải thiện đáng kể trong việc tuân theo chỉ dẫn, hiểu dữ liệu có cấu trúc và tạo ra đầu ra có cấu trúc (đặc biệt là JSON)."
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "Tối ưu hóa cho các tình huống đối thoại bằng tiếng Trung, cung cấp khả năng sinh đối thoại mượt mà và phù hợp với thói quen diễn đạt tiếng Trung."
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Mô hình gọi hàm mã nguồn mở của Fireworks, cung cấp khả năng thực hiện chỉ dẫn xuất sắc và tính năng tùy chỉnh mở."
+ "abab7-chat-preview": {
+ "description": "So với các mô hình trong series abab6.5, khả năng xử lý văn bản dài, toán học, viết lách, v.v. đã được cải thiện đáng kể."
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Firefunction-v2 mới nhất của công ty Fireworks là một mô hình gọi hàm hiệu suất cao, được phát triển dựa trên Llama-3 và được tối ưu hóa nhiều, đặc biệt phù hợp cho các tình huống gọi hàm, đối thoại và theo dõi chỉ dẫn."
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 là một mô hình ngôn ngữ lớn tiên tiến, được tối ưu hóa thông qua học tăng cường và dữ liệu khởi động lạnh, có hiệu suất suy luận, toán học và lập trình xuất sắc."
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b là một mô hình ngôn ngữ hình ảnh, có thể nhận cả hình ảnh và văn bản đầu vào, được huấn luyện bằng dữ liệu chất lượng cao, phù hợp cho các nhiệm vụ đa mô hình."
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Mô hình ngôn ngữ Mixture-of-Experts (MoE) mạnh mẽ do Deepseek cung cấp, với tổng số tham số là 671B, mỗi ký hiệu kích hoạt 37B tham số."
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Mô hình chỉ dẫn Llama 3 70B, được tối ưu hóa cho đối thoại đa ngôn ngữ và hiểu ngôn ngữ tự nhiên, hiệu suất vượt trội hơn nhiều mô hình cạnh tranh."
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Mô hình chỉ dẫn Llama 3 70B (phiên bản HF), giữ nguyên kết quả với thực hiện chính thức, phù hợp cho các nhiệm vụ theo dõi chỉ dẫn chất lượng cao."
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Mô hình chỉ dẫn Llama 3 8B, được tối ưu hóa cho đối thoại và các nhiệm vụ đa ngôn ngữ, thể hiện hiệu suất xuất sắc và hiệu quả."
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Mô hình suy luận hình ảnh chỉ dẫn với 11B tham số của Meta. Mô hình này được tối ưu hóa cho nhận diện hình ảnh, suy luận hình ảnh, mô tả hình ảnh và trả lời các câu hỏi chung liên quan đến hình ảnh. Mô hình có khả năng hiểu dữ liệu hình ảnh như biểu đồ và đồ thị, và thu hẹp khoảng cách giữa hình ảnh và ngôn ngữ thông qua việc tạo mô tả văn bản về chi tiết hình ảnh."
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Mô hình chỉ dẫn Llama 3.2 1B là một mô hình đa ngôn ngữ nhẹ mà Meta phát hành. Mô hình này được thiết kế để tăng cường hiệu quả, mang lại cải tiến đáng kể về độ trễ và chi phí so với các mô hình lớn hơn. Các trường hợp sử dụng ví dụ của mô hình này bao gồm truy xuất và tóm tắt."
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Mô hình chỉ dẫn Llama 3.2 3B là một mô hình đa ngôn ngữ nhẹ mà Meta phát hành. Mô hình này được thiết kế để tăng cường hiệu quả, mang lại cải tiến đáng kể về độ trễ và chi phí so với các mô hình lớn hơn. Các trường hợp sử dụng ví dụ của mô hình này bao gồm truy vấn, viết lại thông báo và hỗ trợ viết."
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Mô hình suy luận hình ảnh chỉ dẫn với 90B tham số của Meta. Mô hình này được tối ưu hóa cho nhận diện hình ảnh, suy luận hình ảnh, mô tả hình ảnh và trả lời các câu hỏi chung liên quan đến hình ảnh. Mô hình có khả năng hiểu dữ liệu hình ảnh như biểu đồ và đồ thị, và thu hẹp khoảng cách giữa hình ảnh và ngôn ngữ thông qua việc tạo mô tả văn bản về chi tiết hình ảnh."
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct là phiên bản cập nhật tháng 12 của Llama 3.1 70B. Mô hình này được cải tiến dựa trên Llama 3.1 70B (ra mắt vào tháng 7 năm 2024), nâng cao khả năng gọi công cụ, hỗ trợ văn bản đa ngôn ngữ, toán học và lập trình. Mô hình này đạt được trình độ hàng đầu trong ngành về suy luận, toán học và tuân thủ hướng dẫn, đồng thời có thể cung cấp hiệu suất tương tự như 3.1 405B, với lợi thế đáng kể về tốc độ và chi phí."
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "Mô hình 24B tham số, có khả năng tiên tiến tương đương với các mô hình lớn hơn."
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Mô hình chỉ dẫn Mixtral MoE 8x22B, với số lượng tham số lớn và kiến trúc nhiều chuyên gia, hỗ trợ toàn diện cho việc xử lý hiệu quả các nhiệm vụ phức tạp."
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Mô hình chỉ dẫn Mixtral MoE 8x7B, kiến trúc nhiều chuyên gia cung cấp khả năng theo dõi và thực hiện chỉ dẫn hiệu quả."
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Mô hình chỉ dẫn Mixtral MoE 8x7B (phiên bản HF), hiệu suất nhất quán với thực hiện chính thức, phù hợp cho nhiều tình huống nhiệm vụ hiệu quả."
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "Mô hình MythoMax L2 13B, kết hợp công nghệ hợp nhất mới, xuất sắc trong việc kể chuyện và đóng vai."
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "Mô hình QwQ là một mô hình nghiên cứu thử nghiệm được phát triển bởi đội ngũ Qwen, tập trung vào việc nâng cao khả năng suy luận của AI."
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "Phiên bản 72B của mô hình Qwen-VL là thành quả mới nhất của Alibaba, đại diện cho gần một năm đổi mới."
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 là một loạt mô hình ngôn ngữ chỉ chứa bộ giải mã do đội ngũ Qwen của Alibaba Cloud phát triển. Những mô hình này cung cấp các kích thước khác nhau, bao gồm 0.5B, 1.5B, 3B, 7B, 14B, 32B và 72B, và có hai biến thể: phiên bản cơ sở (base) và phiên bản chỉ dẫn (instruct)."
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct là phiên bản mới nhất trong loạt mô hình ngôn ngữ lớn chuyên biệt cho mã do Alibaba Cloud phát hành. Mô hình này được cải thiện đáng kể khả năng tạo mã, suy luận và sửa chữa thông qua việc đào tạo trên 5.5 triệu tỷ tokens, không chỉ nâng cao khả năng lập trình mà còn duy trì lợi thế về khả năng toán học và tổng quát. Mô hình cung cấp nền tảng toàn diện hơn cho các ứng dụng thực tế như tác nhân mã."
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "Mô hình StarCoder 15.5B, hỗ trợ các nhiệm vụ lập trình nâng cao, khả năng đa ngôn ngữ được cải thiện, phù hợp cho việc tạo và hiểu mã phức tạp."
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "Mô hình StarCoder 7B, được huấn luyện cho hơn 80 ngôn ngữ lập trình, có khả năng điền mã và hiểu ngữ cảnh xuất sắc."
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Mô hình Yi-Large, có khả năng xử lý đa ngôn ngữ xuất sắc, có thể được sử dụng cho nhiều nhiệm vụ sinh và hiểu ngôn ngữ."
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus là mô hình mạnh mẽ nhất của Anthropic, được sử dụng để xử lý các nhiệm vụ phức tạp cao. Nó thể hiện xuất sắc về hiệu suất, trí thông minh, sự trôi chảy và khả năng hiểu biết."
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku là mô hình thế hệ tiếp theo nhanh nhất của Anthropic. So với Claude 3 Haiku, Claude 3.5 Haiku có sự cải thiện trong nhiều kỹ năng và vượt qua mô hình lớn nhất thế hệ trước Claude 3 Opus trong nhiều bài kiểm tra trí tuệ."
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet cung cấp khả năng vượt trội hơn Opus và tốc độ nhanh hơn Sonnet, trong khi vẫn giữ giá tương tự. Sonnet đặc biệt xuất sắc trong lập trình, khoa học dữ liệu, xử lý hình ảnh và các nhiệm vụ đại lý."
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet cung cấp sự cân bằng lý tưởng giữa trí thông minh và tốc độ cho khối lượng công việc doanh nghiệp. Nó cung cấp hiệu suất tối đa với mức giá thấp hơn, đáng tin cậy và phù hợp cho triển khai quy mô lớn."
},
- "code-raccoon-v1": {
- "description": "Code Raccoon là một trợ lý phát triển phần mềm thông minh dựa trên mô hình ngôn ngữ lớn của SenseTime, bao phủ các giai đoạn phân tích yêu cầu phần mềm, thiết kế kiến trúc, viết mã, kiểm thử phần mềm, đáp ứng nhu cầu viết mã và học lập trình của người dùng. Code Raccoon hỗ trợ hơn 90 ngôn ngữ lập trình phổ biến như Python, Java, JavaScript, C++, Go, SQL và các IDE phổ biến như VS Code, IntelliJ IDEA. Trong thực tế, Code Raccoon có thể giúp các nhà phát triển nâng cao hiệu suất lập trình hơn 50%."
- },
"codegeex-4": {
"description": "CodeGeeX-4 là trợ lý lập trình AI mạnh mẽ, hỗ trợ nhiều ngôn ngữ lập trình với câu hỏi thông minh và hoàn thành mã, nâng cao hiệu suất phát triển."
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ là một mô hình ngôn ngữ lớn hiệu suất cao, được thiết kế cho các tình huống doanh nghiệp thực tế và ứng dụng phức tạp."
},
+ "dall-e-2": {
+ "description": "Mô hình DALL·E thế hệ thứ hai, hỗ trợ tạo hình ảnh chân thực và chính xác hơn, với độ phân giải gấp 4 lần thế hệ đầu tiên."
+ },
+ "dall-e-3": {
+ "description": "Mô hình DALL·E mới nhất, phát hành vào tháng 11 năm 2023. Hỗ trợ tạo hình ảnh chân thực và chính xác hơn, với khả năng thể hiện chi tiết mạnh mẽ hơn."
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct cung cấp khả năng xử lý chỉ dẫn đáng tin cậy, hỗ trợ nhiều ứng dụng trong ngành."
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 là một mô hình ngôn ngữ hỗn hợp chuyên gia (MoE) mạnh mẽ và tiết kiệm chi phí. Nó đã được tiền huấn luyện trên một tập dữ liệu chất lượng cao với 8.1 triệu tỷ tokens và được cải thiện thêm thông qua tinh chỉnh giám sát (SFT) và học tăng cường (RL). So với DeepSeek 67B, DeepSeek-V2 không chỉ mạnh mẽ hơn mà còn tiết kiệm 42.5% chi phí đào tạo, giảm 93.3% bộ nhớ KV và tăng gấp 5.76 lần thông lượng sinh tối đa. Mô hình hỗ trợ độ dài ngữ cảnh 128k và thể hiện xuất sắc trong các bài kiểm tra chuẩn và đánh giá sinh mở."
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 là một mô hình suy diễn được điều khiển bởi học tăng cường (RL), giải quyết các vấn đề về tính lặp lại và khả năng đọc hiểu trong mô hình. Trước khi áp dụng RL, DeepSeek-R1 đã giới thiệu dữ liệu khởi động lạnh, tối ưu hóa thêm hiệu suất suy diễn. Nó thể hiện hiệu suất tương đương với OpenAI-o1 trong các nhiệm vụ toán học, mã và suy diễn, và thông qua phương pháp đào tạo được thiết kế cẩn thận, nâng cao hiệu quả tổng thể."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "Mô hình chưng cất DeepSeek-R1, tối ưu hóa hiệu suất suy luận thông qua học tăng cường và dữ liệu khởi động lạnh, mô hình mã nguồn mở làm mới tiêu chuẩn đa nhiệm."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B là mô hình chưng cất phát triển từ Llama-3.1-8B. Mô hình này sử dụng các mẫu được tạo ra từ DeepSeek-R1 để tinh chỉnh, thể hiện khả năng suy luận xuất sắc. Trong nhiều bài kiểm tra chuẩn, nó đã thể hiện tốt, trong đó đạt 89.1% độ chính xác trên MATH-500, đạt 50.4% tỷ lệ vượt qua trên AIME 2024, và đạt điểm 1205 trên CodeForces, thể hiện khả năng toán học và lập trình mạnh mẽ cho mô hình quy mô 8B."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "Mô hình chưng cất DeepSeek-R1, tối ưu hóa hiệu suất suy luận thông qua học tăng cường và dữ liệu khởi động lạnh, mô hình mã nguồn mở làm mới tiêu chuẩn đa nhiệm."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "Mô hình chưng cất DeepSeek-R1, tối ưu hóa hiệu suất suy luận thông qua học tăng cường và dữ liệu khởi động lạnh, mô hình mã nguồn mở làm mới tiêu chuẩn đa nhiệm."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B là mô hình được tạo ra từ Qwen2.5-32B thông qua chưng cất kiến thức. Mô hình này sử dụng 800.000 mẫu được chọn lọc từ DeepSeek-R1 để tinh chỉnh, thể hiện hiệu suất xuất sắc trong nhiều lĩnh vực như toán học, lập trình và suy luận. Trong nhiều bài kiểm tra chuẩn như AIME 2024, MATH-500, GPQA Diamond, nó đã đạt được kết quả xuất sắc, trong đó đạt 94.3% độ chính xác trên MATH-500, thể hiện khả năng suy luận toán học mạnh mẽ."
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B là mô hình được tạo ra từ Qwen2.5-Math-7B thông qua chưng cất kiến thức. Mô hình này sử dụng 800.000 mẫu được chọn lọc từ DeepSeek-R1 để tinh chỉnh, thể hiện khả năng suy luận xuất sắc. Trong nhiều bài kiểm tra chuẩn, nó đã thể hiện xuất sắc, trong đó đạt 92.8% độ chính xác trên MATH-500, đạt 55.5% tỷ lệ vượt qua trên AIME 2024, và đạt điểm 1189 trên CodeForces, thể hiện khả năng toán học và lập trình mạnh mẽ cho mô hình quy mô 7B."
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 kết hợp các đặc điểm xuất sắc của các phiên bản trước, tăng cường khả năng tổng quát và mã hóa."
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 là một mô hình ngôn ngữ hỗn hợp chuyên gia (MoE) với 6710 tỷ tham số, sử dụng chú ý tiềm ẩn đa đầu (MLA) và kiến trúc DeepSeekMoE, kết hợp với chiến lược cân bằng tải không có tổn thất phụ trợ, tối ưu hóa hiệu suất suy diễn và đào tạo. Thông qua việc được tiền huấn luyện trên 14.8 triệu tỷ token chất lượng cao, và thực hiện tinh chỉnh giám sát và học tăng cường, DeepSeek-V3 vượt trội về hiệu suất so với các mô hình mã nguồn mở khác, gần gũi với các mô hình đóng nguồn hàng đầu."
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B là mô hình tiên tiến được huấn luyện cho các cuộc đối thoại phức tạp."
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "LLM hiệu quả tiên tiến, xuất sắc trong suy luận, toán học và lập trình."
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 là một mô hình ngôn ngữ hình ảnh hỗn hợp chuyên gia (MoE) được phát triển dựa trên DeepSeekMoE-27B, sử dụng kiến trúc MoE với kích hoạt thưa, đạt được hiệu suất xuất sắc chỉ với 4.5B tham số được kích hoạt. Mô hình này thể hiện xuất sắc trong nhiều nhiệm vụ như hỏi đáp hình ảnh, nhận diện ký tự quang học, hiểu tài liệu/bảng/biểu đồ và định vị hình ảnh."
+ },
"deepseek-chat": {
"description": "Mô hình mã nguồn mở mới kết hợp khả năng tổng quát và mã, không chỉ giữ lại khả năng đối thoại tổng quát của mô hình Chat ban đầu và khả năng xử lý mã mạnh mẽ của mô hình Coder, mà còn tốt hơn trong việc phù hợp với sở thích của con người. Hơn nữa, DeepSeek-V2.5 cũng đã đạt được sự cải thiện lớn trong nhiều khía cạnh như nhiệm vụ viết, theo dõi chỉ dẫn."
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 là mô hình mã nguồn mở hỗn hợp chuyên gia, thể hiện xuất sắc trong các nhiệm vụ mã, tương đương với GPT4-Turbo."
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 là một mô hình suy diễn được điều khiển bởi học tăng cường (RL), giải quyết các vấn đề về tính lặp lại và khả năng đọc hiểu trong mô hình. Trước khi áp dụng RL, DeepSeek-R1 đã giới thiệu dữ liệu khởi động lạnh, tối ưu hóa thêm hiệu suất suy diễn. Nó thể hiện hiệu suất tương đương với OpenAI-o1 trong các nhiệm vụ toán học, mã và suy diễn, và thông qua phương pháp đào tạo được thiết kế cẩn thận, nâng cao hiệu quả tổng thể."
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1 - mô hình lớn hơn và thông minh hơn trong bộ công cụ DeepSeek - đã được chưng cất vào kiến trúc Llama 70B. Dựa trên các bài kiểm tra chuẩn và đánh giá của con người, mô hình này thông minh hơn so với Llama 70B gốc, đặc biệt xuất sắc trong các nhiệm vụ yêu cầu độ chính xác về toán học và sự thật."
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "Mô hình DeepSeek-R1-Distill được tinh chỉnh từ các mẫu do DeepSeek-R1 tạo ra cho các mô hình mã nguồn mở như Qwen, Llama thông qua công nghệ chưng cất kiến thức."
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "Mô hình DeepSeek-R1-Distill được tinh chỉnh từ các mẫu do DeepSeek-R1 tạo ra cho các mô hình mã nguồn mở như Qwen, Llama thông qua công nghệ chưng cất kiến thức."
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "Mô hình DeepSeek-R1-Distill được tinh chỉnh từ các mẫu do DeepSeek-R1 tạo ra cho các mô hình mã nguồn mở như Qwen, Llama thông qua công nghệ chưng cất kiến thức."
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "Mô hình DeepSeek-R1-Distill được tinh chỉnh từ các mẫu do DeepSeek-R1 tạo ra cho các mô hình mã nguồn mở như Qwen, Llama thông qua công nghệ chưng cất kiến thức."
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "Mô hình DeepSeek-R1-Distill được tinh chỉnh từ các mẫu do DeepSeek-R1 tạo ra cho các mô hình mã nguồn mở như Qwen, Llama thông qua công nghệ chưng cất kiến thức."
+ },
+ "deepseek-reasoner": {
+ "description": "Mô hình suy diễn do DeepSeek phát triển. Trước khi đưa ra câu trả lời cuối cùng, mô hình sẽ xuất ra một đoạn nội dung chuỗi suy nghĩ để nâng cao độ chính xác của câu trả lời cuối."
+ },
"deepseek-v2": {
"description": "DeepSeek V2 là mô hình ngôn ngữ Mixture-of-Experts hiệu quả, phù hợp cho các nhu cầu xử lý tiết kiệm."
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B là mô hình mã thiết kế của DeepSeek, cung cấp khả năng sinh mã mạnh mẽ."
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 là mô hình MoE tự phát triển của Công ty Nghiên cứu Công nghệ AI Độ Sâu Hàng Châu, có nhiều thành tích xuất sắc trong các bài kiểm tra, đứng đầu bảng xếp hạng mô hình mã nguồn mở. V3 so với mô hình V2.5 đã cải thiện tốc độ tạo ra gấp 3 lần, mang đến trải nghiệm sử dụng nhanh chóng và mượt mà hơn cho người dùng."
+ },
"deepseek/deepseek-chat": {
"description": "Mô hình mã nguồn mở mới kết hợp khả năng tổng quát và mã, không chỉ giữ lại khả năng đối thoại tổng quát của mô hình Chat ban đầu và khả năng xử lý mã mạnh mẽ của mô hình Coder, mà còn tốt hơn trong việc phù hợp với sở thích của con người. Hơn nữa, DeepSeek-V2.5 cũng đã đạt được sự cải thiện lớn trong nhiều lĩnh vực như nhiệm vụ viết, theo dõi chỉ dẫn."
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 đã nâng cao khả năng suy luận của mô hình một cách đáng kể với rất ít dữ liệu được gán nhãn. Trước khi đưa ra câu trả lời cuối cùng, mô hình sẽ xuất ra một chuỗi suy nghĩ để nâng cao độ chính xác của câu trả lời cuối cùng."
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 đã nâng cao khả năng suy luận của mô hình một cách đáng kể với rất ít dữ liệu được gán nhãn. Trước khi đưa ra câu trả lời cuối cùng, mô hình sẽ xuất ra một chuỗi suy nghĩ để nâng cao độ chính xác của câu trả lời cuối cùng."
+ },
"emohaa": {
"description": "Emohaa là mô hình tâm lý, có khả năng tư vấn chuyên nghiệp, giúp người dùng hiểu các vấn đề cảm xúc."
},
+ "ernie-3.5-128k": {
+ "description": "Mô hình ngôn ngữ lớn quy mô lớn tự phát triển của Baidu, bao phủ một lượng lớn tài liệu tiếng Trung và tiếng Anh, có khả năng tổng quát mạnh mẽ, đáp ứng hầu hết các yêu cầu về đối thoại hỏi đáp, tạo nội dung, và ứng dụng plugin; hỗ trợ tự động kết nối với plugin tìm kiếm của Baidu, đảm bảo thông tin hỏi đáp kịp thời."
+ },
+ "ernie-3.5-8k": {
+ "description": "Mô hình ngôn ngữ lớn quy mô lớn tự phát triển của Baidu, bao phủ một lượng lớn tài liệu tiếng Trung và tiếng Anh, có khả năng tổng quát mạnh mẽ, đáp ứng hầu hết các yêu cầu về đối thoại hỏi đáp, tạo nội dung, và ứng dụng plugin; hỗ trợ tự động kết nối với plugin tìm kiếm của Baidu, đảm bảo thông tin hỏi đáp kịp thời."
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "Mô hình ngôn ngữ lớn quy mô lớn tự phát triển của Baidu, bao phủ một lượng lớn tài liệu tiếng Trung và tiếng Anh, có khả năng tổng quát mạnh mẽ, đáp ứng hầu hết các yêu cầu về đối thoại hỏi đáp, tạo nội dung, và ứng dụng plugin; hỗ trợ tự động kết nối với plugin tìm kiếm của Baidu, đảm bảo thông tin hỏi đáp kịp thời."
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "Mô hình ngôn ngữ lớn siêu quy mô tự phát triển của Baidu, so với ERNIE 3.5 đã thực hiện nâng cấp toàn diện về khả năng mô hình, phù hợp rộng rãi với các tình huống nhiệm vụ phức tạp trong nhiều lĩnh vực; hỗ trợ tự động kết nối với plugin tìm kiếm của Baidu, đảm bảo thông tin hỏi đáp kịp thời."
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "Mô hình ngôn ngữ lớn siêu quy mô tự phát triển của Baidu, so với ERNIE 3.5 đã thực hiện nâng cấp toàn diện về khả năng mô hình, phù hợp rộng rãi với các tình huống nhiệm vụ phức tạp trong nhiều lĩnh vực; hỗ trợ tự động kết nối với plugin tìm kiếm của Baidu, đảm bảo thông tin hỏi đáp kịp thời."
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "Mô hình ngôn ngữ lớn siêu quy mô tự phát triển của Baidu, có hiệu suất tổng thể xuất sắc, phù hợp rộng rãi với các tình huống nhiệm vụ phức tạp trong nhiều lĩnh vực; hỗ trợ tự động kết nối với plugin tìm kiếm của Baidu, đảm bảo thông tin hỏi đáp kịp thời. So với ERNIE 4.0, hiệu suất tốt hơn."
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "Mô hình ngôn ngữ lớn siêu quy mô tự phát triển của Baidu, có hiệu suất tổng thể xuất sắc, phù hợp rộng rãi với các tình huống nhiệm vụ phức tạp trong nhiều lĩnh vực; hỗ trợ tự động kết nối với plugin tìm kiếm của Baidu, đảm bảo thông tin hỏi đáp kịp thời. So với ERNIE 4.0, hiệu suất tốt hơn."
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "Mô hình ngôn ngữ lớn siêu quy mô tự phát triển của Baidu, có hiệu suất tổng thể xuất sắc, phù hợp rộng rãi với các tình huống nhiệm vụ phức tạp trong nhiều lĩnh vực; hỗ trợ tự động kết nối với plugin tìm kiếm của Baidu, đảm bảo thông tin hỏi đáp kịp thời. So với ERNIE 4.0, hiệu suất tốt hơn."
+ },
+ "ernie-char-8k": {
+ "description": "Mô hình ngôn ngữ lớn theo ngữ cảnh tự phát triển của Baidu, phù hợp cho các ứng dụng như NPC trong trò chơi, đối thoại dịch vụ khách hàng, và vai trò trong đối thoại, có phong cách nhân vật rõ ràng và nhất quán, khả năng tuân theo lệnh mạnh mẽ, hiệu suất suy luận tốt hơn."
+ },
+ "ernie-char-fiction-8k": {
+ "description": "Mô hình ngôn ngữ lớn theo ngữ cảnh tự phát triển của Baidu, phù hợp cho các ứng dụng như NPC trong trò chơi, đối thoại dịch vụ khách hàng, và vai trò trong đối thoại, có phong cách nhân vật rõ ràng và nhất quán, khả năng tuân theo lệnh mạnh mẽ, hiệu suất suy luận tốt hơn."
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite là mô hình ngôn ngữ lớn nhẹ tự phát triển của Baidu, kết hợp hiệu suất mô hình xuất sắc với hiệu suất suy luận, phù hợp cho việc sử dụng trên thẻ tăng tốc AI với công suất thấp."
+ },
+ "ernie-lite-pro-128k": {
+ "description": "Mô hình ngôn ngữ lớn nhẹ tự phát triển của Baidu, kết hợp hiệu suất mô hình xuất sắc với hiệu suất suy luận, hiệu suất tốt hơn ERNIE Lite, phù hợp cho việc sử dụng trên thẻ tăng tốc AI với công suất thấp."
+ },
+ "ernie-novel-8k": {
+ "description": "Mô hình ngôn ngữ lớn tổng quát tự phát triển của Baidu, có lợi thế rõ rệt trong khả năng viết tiếp tiểu thuyết, cũng có thể được sử dụng trong các tình huống như kịch ngắn, phim ảnh."
+ },
+ "ernie-speed-128k": {
+ "description": "Mô hình ngôn ngữ lớn hiệu suất cao tự phát triển của Baidu, được phát hành vào năm 2024, có khả năng tổng quát xuất sắc, phù hợp làm mô hình nền để tinh chỉnh, xử lý tốt hơn các vấn đề trong tình huống cụ thể, đồng thời có hiệu suất suy luận xuất sắc."
+ },
+ "ernie-speed-pro-128k": {
+ "description": "Mô hình ngôn ngữ lớn hiệu suất cao tự phát triển của Baidu, được phát hành vào năm 2024, có khả năng tổng quát xuất sắc, hiệu suất tốt hơn ERNIE Speed, phù hợp làm mô hình nền để tinh chỉnh, xử lý tốt hơn các vấn đề trong tình huống cụ thể, đồng thời có hiệu suất suy luận xuất sắc."
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny là mô hình ngôn ngữ lớn hiệu suất siêu cao tự phát triển của Baidu, có chi phí triển khai và tinh chỉnh thấp nhất trong dòng sản phẩm văn tâm."
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Tuning) cung cấp hiệu suất ổn định và có thể điều chỉnh, là lựa chọn lý tưởng cho các giải pháp nhiệm vụ phức tạp."
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro hỗ trợ lên đến 2 triệu tokens, là lựa chọn lý tưởng cho mô hình đa phương thức trung bình, phù hợp cho hỗ trợ đa diện cho các nhiệm vụ phức tạp."
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash cung cấp các tính năng và cải tiến thế hệ tiếp theo, bao gồm tốc độ vượt trội, sử dụng công cụ bản địa, tạo đa phương tiện và cửa sổ ngữ cảnh 1M token."
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash cung cấp các tính năng và cải tiến thế hệ tiếp theo, bao gồm tốc độ vượt trội, sử dụng công cụ bản địa, tạo đa phương tiện và cửa sổ ngữ cảnh 1M token."
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp là mô hình AI đa phương thức thử nghiệm mới nhất của Google, có các tính năng thế hệ tiếp theo, tốc độ vượt trội, khả năng gọi công cụ nguyên bản và tạo ra đa phương thức."
},
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "Một mô hình Gemini 2.0 Flash được tối ưu hóa cho hiệu quả chi phí và độ trễ thấp."
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Exp là mô hình AI đa phương thức thử nghiệm mới nhất của Google, sở hữu các tính năng thế hệ tiếp theo, tốc độ vượt trội, gọi công cụ bản địa và sinh ra đa phương thức."
+ },
"gemini-2.0-flash-thinking-exp-1219": {
"description": "Gemini 2.0 Flash Exp là mô hình AI đa phương thức thử nghiệm mới nhất của Google, với các tính năng thế hệ tiếp theo, tốc độ vượt trội, gọi công cụ bản địa và tạo ra đa phương thức."
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 là mô hình AI đa phương thức thử nghiệm mới nhất của Google, có khả năng xử lý nhanh, hỗ trợ đầu vào văn bản, hình ảnh và video, phù hợp với việc mở rộng hiệu quả cho nhiều nhiệm vụ."
- },
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 là mô hình AI đa phương thức thử nghiệm mới nhất của Google, có khả năng xử lý nhanh chóng, hỗ trợ đầu vào văn bản, hình ảnh và video, phù hợp cho việc mở rộng hiệu quả cho nhiều nhiệm vụ."
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental là mô hình AI đa phương tiện thử nghiệm mới nhất của Google, có sự cải thiện chất lượng nhất định so với các phiên bản trước, đặc biệt là về kiến thức thế giới, mã và ngữ cảnh dài."
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 là mô hình AI đa phương tiện thử nghiệm mới nhất của Google, có sự cải thiện về chất lượng so với các phiên bản trước."
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus có khả năng hiểu nội dung video và nhiều hình ảnh, phù hợp cho các nhiệm vụ đa phương tiện."
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview có khả năng suy luận phức tạp mạnh mẽ, thể hiện xuất sắc trong các lĩnh vực suy luận logic, toán học, lập trình."
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash cung cấp các tính năng và cải tiến thế hệ tiếp theo, bao gồm tốc độ vượt trội, sử dụng công cụ bản địa, tạo đa phương tiện và cửa sổ ngữ cảnh 1M token."
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental là mô hình AI đa phương tiện thử nghiệm mới nhất của Google, có sự cải thiện chất lượng nhất định so với các phiên bản trước, đặc biệt là về kiến thức thế giới, mã và ngữ cảnh dài."
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash cung cấp khả năng xử lý đa phương thức được tối ưu hóa, phù hợp cho nhiều tình huống nhiệm vụ phức tạp."
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro kết hợp công nghệ tối ưu hóa mới nhất, mang lại khả năng xử lý dữ liệu đa phương thức hiệu quả hơn."
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 là mô hình hiệu quả do Google phát hành, bao gồm nhiều ứng dụng từ ứng dụng nhỏ đến xử lý dữ liệu phức tạp."
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 tiếp tục triết lý thiết kế nhẹ và hiệu quả."
},
"google/gemma-2-2b-it": {
"description": "Mô hình tinh chỉnh hướng dẫn nhẹ của Google"
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 là mô hình hiệu quả do Google phát hành, bao gồm nhiều ứng dụng từ ứng dụng nhỏ đến xử lý dữ liệu phức tạp."
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 là một loạt mô hình văn bản mã nguồn mở nhẹ của Google."
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo, phù hợp cho nhiều nhiệm vụ sinh và hiểu văn bản, hiện tại trỏ đến gpt-3.5-turbo-0125."
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k, mô hình sinh văn bản dung lượng lớn, phù hợp cho các nhiệm vụ phức tạp."
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo, phù hợp cho nhiều nhiệm vụ sinh và hiểu văn bản, hiện tại trỏ đến gpt-3.5-turbo-0125."
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o là một mô hình động, được cập nhật liên tục để giữ phiên bản mới nhất. Nó kết hợp khả năng hiểu và tạo ngôn ngữ mạnh mẽ, phù hợp cho nhiều ứng dụng quy mô lớn, bao gồm dịch vụ khách hàng, giáo dục và hỗ trợ kỹ thuật."
},
+ "gpt-4o-audio-preview": {
+ "description": "Mô hình GPT-4o Audio, hỗ trợ đầu vào và đầu ra âm thanh."
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini là mô hình mới nhất do OpenAI phát hành sau GPT-4 Omni, hỗ trợ đầu vào hình ảnh và đầu ra văn bản. Là mô hình nhỏ gọn tiên tiến nhất của họ, nó rẻ hơn nhiều so với các mô hình tiên tiến gần đây khác và rẻ hơn hơn 60% so với GPT-3.5 Turbo. Nó giữ lại trí thông minh tiên tiến nhất trong khi có giá trị sử dụng đáng kể. GPT-4o mini đạt 82% điểm trong bài kiểm tra MMLU và hiện đứng cao hơn GPT-4 về sở thích trò chuyện."
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "Phiên bản thời gian thực của GPT-4o-mini, hỗ trợ đầu vào và đầu ra âm thanh và văn bản theo thời gian thực."
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "Phiên bản thời gian thực của GPT-4o, hỗ trợ đầu vào và đầu ra âm thanh và văn bản theo thời gian thực."
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "Phiên bản thời gian thực của GPT-4o, hỗ trợ đầu vào và đầu ra âm thanh và văn bản theo thời gian thực."
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "Phiên bản thời gian thực của GPT-4o, hỗ trợ đầu vào và đầu ra âm thanh và văn bản theo thời gian thực."
+ },
"grok-2-1212": {
"description": "Mô hình này đã được cải thiện về độ chính xác, khả năng tuân thủ hướng dẫn và khả năng đa ngôn ngữ."
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "Mô hình FunctionCall với cấu trúc MOE mới nhất của Hunyuan, được huấn luyện trên dữ liệu FunctionCall chất lượng cao, với cửa sổ ngữ cảnh đạt 32K, dẫn đầu trong nhiều chỉ số đánh giá."
},
+ "hunyuan-large": {
+ "description": "Mô hình Hunyuan-large có tổng số tham số khoảng 389B, số tham số kích hoạt khoảng 52B, là mô hình MoE mã nguồn mở có quy mô tham số lớn nhất và hiệu quả nhất trong ngành hiện nay."
+ },
+ "hunyuan-large-longcontext": {
+ "description": "Chuyên xử lý các nhiệm vụ văn bản dài như tóm tắt tài liệu và hỏi đáp tài liệu, đồng thời cũng có khả năng xử lý các nhiệm vụ tạo văn bản chung. Thể hiện xuất sắc trong phân tích và tạo nội dung văn bản dài, có thể đáp ứng hiệu quả các yêu cầu xử lý nội dung dài phức tạp và chi tiết."
+ },
"hunyuan-lite": {
"description": "Nâng cấp lên cấu trúc MOE, với cửa sổ ngữ cảnh 256k, dẫn đầu nhiều mô hình mã nguồn mở trong các bộ đánh giá NLP, mã, toán học, ngành nghề, v.v."
},
+ "hunyuan-lite-vision": {
+ "description": "Mô hình đa phương thức mới nhất 7B của Hunyuan, cửa sổ ngữ cảnh 32K, hỗ trợ đối thoại đa phương thức trong các tình huống tiếng Trung và tiếng Anh, nhận diện đối tượng hình ảnh, hiểu biết tài liệu và bảng biểu, toán học đa phương thức, v.v., với các chỉ số đánh giá vượt trội hơn các mô hình cạnh tranh 7B ở nhiều khía cạnh."
+ },
"hunyuan-pro": {
"description": "Mô hình văn bản dài MOE-32K với quy mô hàng triệu tham số. Đạt được mức độ dẫn đầu tuyệt đối trên nhiều benchmark, có khả năng xử lý các lệnh phức tạp và suy diễn, có khả năng toán học phức tạp, hỗ trợ functioncall, được tối ưu hóa cho các lĩnh vực dịch thuật đa ngôn ngữ, tài chính, pháp lý và y tế."
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "Sử dụng chiến lược định tuyến tốt hơn, đồng thời giảm thiểu vấn đề cân bằng tải và đồng nhất chuyên gia. Về mặt văn bản dài, chỉ số tìm kiếm đạt 99.9%. MOE-256K đã có bước đột phá về độ dài và hiệu quả, mở rộng đáng kể độ dài đầu vào có thể."
},
+ "hunyuan-standard-vision": {
+ "description": "Mô hình đa phương thức mới nhất của Hunyuan, hỗ trợ trả lời đa ngôn ngữ, khả năng tiếng Trung và tiếng Anh cân bằng."
+ },
"hunyuan-turbo": {
"description": "Phiên bản xem trước của thế hệ mới mô hình ngôn ngữ lớn Hunyuan, sử dụng cấu trúc mô hình chuyên gia hỗn hợp (MoE) hoàn toàn mới, so với hunyuan-pro, hiệu suất suy diễn nhanh hơn và hiệu quả mạnh mẽ hơn."
},
+ "hunyuan-turbo-20241120": {
+ "description": "Phiên bản cố định hunyuan-turbo ngày 20 tháng 11 năm 2024, là một phiên bản nằm giữa hunyuan-turbo và hunyuan-turbo-latest."
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "Phiên bản này tối ưu hóa: quy mô chỉ thị dữ liệu, nâng cao đáng kể khả năng tổng quát của mô hình; nâng cao đáng kể khả năng toán học, lập trình, và suy luận logic; tối ưu hóa khả năng hiểu biết văn bản và từ ngữ; tối ưu hóa chất lượng tạo nội dung văn bản."
+ },
+ "hunyuan-turbo-latest": {
+ "description": "Tối ưu hóa trải nghiệm chung, bao gồm hiểu biết NLP, sáng tạo văn bản, trò chuyện, hỏi đáp kiến thức, dịch thuật, và các lĩnh vực khác; nâng cao tính nhân văn, tối ưu hóa trí tuệ cảm xúc của mô hình; cải thiện khả năng làm rõ khi ý định không rõ ràng; nâng cao khả năng xử lý các vấn đề phân tích từ ngữ; nâng cao chất lượng và khả năng tương tác trong sáng tạo; cải thiện trải nghiệm đa vòng."
+ },
+ "hunyuan-turbo-vision": {
+ "description": "Mô hình ngôn ngữ hình ảnh thế hệ mới của Hunyuan, sử dụng cấu trúc mô hình chuyên gia hỗn hợp (MoE) hoàn toàn mới, nâng cao toàn diện khả năng nhận diện cơ bản, sáng tạo nội dung, hỏi đáp kiến thức, và phân tích suy luận so với mô hình thế hệ trước."
+ },
"hunyuan-vision": {
"description": "Mô hình đa phương thức mới nhất của Hunyuan, hỗ trợ đầu vào hình ảnh + văn bản để tạo ra nội dung văn bản."
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "Dòng mô hình mới nhất của chúng tôi, có hiệu suất suy luận xuất sắc, hỗ trợ độ dài ngữ cảnh 1M và khả năng theo dõi chỉ dẫn và gọi công cụ mạnh mẽ hơn."
},
+ "internlm3-latest": {
+ "description": "Dòng mô hình mới nhất của chúng tôi, có hiệu suất suy luận xuất sắc, dẫn đầu trong số các mô hình mã nguồn mở cùng cấp. Mặc định chỉ đến mô hình InternLM3 mới nhất mà chúng tôi đã phát hành."
+ },
+ "jina-deepsearch-v1": {
+ "description": "Tìm kiếm sâu kết hợp tìm kiếm trên mạng, đọc và suy luận, có thể thực hiện điều tra toàn diện. Bạn có thể coi nó như một đại lý, nhận nhiệm vụ nghiên cứu của bạn - nó sẽ thực hiện tìm kiếm rộng rãi và qua nhiều lần lặp lại trước khi đưa ra câu trả lời. Quá trình này liên quan đến nghiên cứu liên tục, suy luận và giải quyết vấn đề từ nhiều góc độ. Điều này khác biệt hoàn toàn với việc tạo ra câu trả lời trực tiếp từ dữ liệu đã được huấn luyện trước của các mô hình lớn tiêu chuẩn và các hệ thống RAG truyền thống dựa vào tìm kiếm bề mặt một lần."
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM là một mô hình ngôn ngữ thử nghiệm, chuyên biệt cho các nhiệm vụ, được đào tạo để tuân theo các nguyên tắc khoa học học tập, có thể tuân theo các chỉ dẫn hệ thống trong các tình huống giảng dạy và học tập, đóng vai trò như một người hướng dẫn chuyên gia."
},
"lite": {
"description": "Spark Lite là một mô hình ngôn ngữ lớn nhẹ, có độ trễ cực thấp và khả năng xử lý hiệu quả, hoàn toàn miễn phí và mở, hỗ trợ chức năng tìm kiếm trực tuyến theo thời gian thực. Đặc điểm phản hồi nhanh của nó giúp nó nổi bật trong các ứng dụng suy diễn trên thiết bị có công suất thấp và tinh chỉnh mô hình, mang lại hiệu quả chi phí và trải nghiệm thông minh xuất sắc cho người dùng, đặc biệt trong các tình huống hỏi đáp kiến thức, tạo nội dung và tìm kiếm."
},
- "llama-3.1-70b-instruct": {
- "description": "Mô hình Llama 3.1 70B Instruct, có 70B tham số, có thể cung cấp hiệu suất xuất sắc trong các nhiệm vụ sinh văn bản và chỉ dẫn lớn."
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B cung cấp khả năng suy luận AI mạnh mẽ hơn, phù hợp cho các ứng dụng phức tạp, hỗ trợ xử lý tính toán cực lớn và đảm bảo hiệu quả và độ chính xác cao."
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B là một mô hình hiệu suất cao, cung cấp khả năng sinh văn bản nhanh chóng, rất phù hợp cho các tình huống ứng dụng cần hiệu quả quy mô lớn và tiết kiệm chi phí."
},
- "llama-3.1-8b-instruct": {
- "description": "Mô hình Llama 3.1 8B Instruct, có 8B tham số, hỗ trợ thực hiện nhiệm vụ chỉ dẫn hình ảnh hiệu quả, cung cấp khả năng sinh văn bản chất lượng."
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Mô hình Llama 3.1 Sonar Huge Online, có 405B tham số, hỗ trợ độ dài ngữ cảnh khoảng 127,000 mã, được thiết kế cho các ứng dụng trò chuyện trực tuyến phức tạp."
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Mô hình Llama 3.1 Sonar Large Chat, có 70B tham số, hỗ trợ độ dài ngữ cảnh khoảng 127,000 mã, phù hợp cho các nhiệm vụ trò chuyện ngoại tuyến phức tạp."
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Mô hình Llama 3.1 Sonar Large Online, có 70B tham số, hỗ trợ độ dài ngữ cảnh khoảng 127,000 mã, phù hợp cho các nhiệm vụ trò chuyện có dung lượng lớn và đa dạng."
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Mô hình Llama 3.1 Sonar Small Chat, có 8B tham số, được thiết kế cho trò chuyện ngoại tuyến, hỗ trợ độ dài ngữ cảnh khoảng 127,000 mã."
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Mô hình Llama 3.1 Sonar Small Online, có 8B tham số, hỗ trợ độ dài ngữ cảnh khoảng 127,000 mã, được thiết kế cho trò chuyện trực tuyến, có khả năng xử lý hiệu quả các tương tác văn bản khác nhau."
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 được thiết kế để xử lý các nhiệm vụ kết hợp dữ liệu hình ảnh và văn bản. Nó thể hiện xuất sắc trong các nhiệm vụ mô tả hình ảnh và hỏi đáp hình ảnh, vượt qua rào cản giữa tạo ngôn ngữ và suy luận hình ảnh."
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 là mô hình ngôn ngữ lớn mã nguồn mở đa ngôn ngữ tiên tiến nhất trong dòng Llama, mang đến trải nghiệm hiệu suất tương đương với mô hình 405B với chi phí cực thấp. Dựa trên cấu trúc Transformer, và được cải thiện tính hữu ích và an toàn thông qua tinh chỉnh giám sát (SFT) và học tăng cường từ phản hồi của con người (RLHF). Phiên bản tinh chỉnh theo chỉ dẫn của nó được tối ưu hóa cho đối thoại đa ngôn ngữ, thể hiện tốt hơn nhiều mô hình trò chuyện mã nguồn mở và đóng kín trong nhiều tiêu chuẩn ngành. Ngày cắt đứt kiến thức là tháng 12 năm 2023."
+ },
"llama-3.3-70b-versatile": {
"description": "Mô hình ngôn ngữ lớn Meta Llama 3.3 (LLM) đa ngôn ngữ là mô hình tạo ra dựa trên 70B (đầu vào/đầu ra văn bản) đã được huấn luyện và điều chỉnh theo chỉ dẫn. Mô hình thuần văn bản Llama 3.3 được tối ưu hóa cho các trường hợp hội thoại đa ngôn ngữ và vượt trội hơn nhiều mô hình trò chuyện mã nguồn mở và đóng khác trên các tiêu chuẩn ngành thông thường."
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 được thiết kế để xử lý các tác vụ kết hợp dữ liệu hình ảnh và văn bản. Nó có khả năng xuất sắc trong các tác vụ mô tả hình ảnh và trả lời câu hỏi hình ảnh, vượt qua khoảng cách giữa tạo ngôn ngữ và suy luận hình ảnh."
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 là mô hình ngôn ngữ lớn mã nguồn mở đa ngôn ngữ tiên tiến nhất trong dòng Llama, mang đến trải nghiệm hiệu suất tương đương mô hình 405B với chi phí cực thấp. Dựa trên cấu trúc Transformer, và được cải thiện tính hữu ích và an toàn thông qua tinh chỉnh giám sát (SFT) và học tăng cường phản hồi từ con người (RLHF). Phiên bản tinh chỉnh theo chỉ dẫn được tối ưu hóa cho đối thoại đa ngôn ngữ, thể hiện tốt hơn nhiều mô hình trò chuyện mã nguồn mở và đóng trong nhiều tiêu chuẩn ngành. Ngày cắt kiến thức là tháng 12 năm 2023."
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Mô hình ngôn ngữ lớn đa ngôn ngữ Meta Llama 3.3 (LLM) là mô hình sinh ra từ 70B (đầu vào văn bản/đầu ra văn bản) với việc điều chỉnh trước và điều chỉnh theo lệnh. Mô hình điều chỉnh theo lệnh Llama 3.3 được tối ưu hóa cho các trường hợp sử dụng đối thoại đa ngôn ngữ và vượt trội hơn nhiều mô hình trò chuyện mã nguồn mở và đóng khác trên các bài kiểm tra chuẩn ngành phổ biến."
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 được thiết kế để xử lý các tác vụ kết hợp dữ liệu hình ảnh và văn bản. Nó có khả năng xuất sắc trong các tác vụ mô tả hình ảnh và trả lời câu hỏi hình ảnh, vượt qua khoảng cách giữa tạo ngôn ngữ và suy luận hình ảnh."
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "Mô hình Llama 3.1 Turbo 405B cung cấp hỗ trợ ngữ cảnh dung lượng lớn cho xử lý dữ liệu lớn, thể hiện xuất sắc trong các ứng dụng trí tuệ nhân tạo quy mô lớn."
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 là mô hình hàng đầu do Meta phát hành, hỗ trợ lên đến 405B tham số, có thể áp dụng cho cuộc đối thoại phức tạp, dịch đa ngôn ngữ và phân tích dữ liệu."
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B cung cấp hỗ trợ đối thoại hiệu quả đa ngôn ngữ."
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 được thiết kế để xử lý các nhiệm vụ kết hợp dữ liệu hình ảnh và văn bản. Nó thể hiện xuất sắc trong các nhiệm vụ mô tả hình ảnh và hỏi đáp hình ảnh, vượt qua ranh giới giữa sinh ngôn ngữ và suy diễn hình ảnh."
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 là mô hình ngôn ngữ lớn mã nguồn mở đa ngôn ngữ tiên tiến nhất trong dòng Llama, mang đến trải nghiệm hiệu suất tương đương với mô hình 405B với chi phí cực thấp. Dựa trên cấu trúc Transformer, và được cải thiện tính hữu ích và an toàn thông qua tinh chỉnh giám sát (SFT) và học tăng cường từ phản hồi của con người (RLHF). Phiên bản tinh chỉnh theo chỉ dẫn của nó được tối ưu hóa cho đối thoại đa ngôn ngữ, thể hiện tốt hơn nhiều mô hình trò chuyện mã nguồn mở và đóng kín trong nhiều tiêu chuẩn ngành. Ngày cắt đứt kiến thức là tháng 12 năm 2023."
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 là mô hình ngôn ngữ lớn mã nguồn mở đa ngôn ngữ tiên tiến nhất trong dòng Llama, mang đến trải nghiệm hiệu suất tương đương với mô hình 405B với chi phí cực thấp. Dựa trên cấu trúc Transformer, và được cải thiện tính hữu ích và an toàn thông qua tinh chỉnh giám sát (SFT) và học tăng cường từ phản hồi của con người (RLHF). Phiên bản tinh chỉnh theo chỉ dẫn của nó được tối ưu hóa cho đối thoại đa ngôn ngữ, thể hiện tốt hơn nhiều mô hình trò chuyện mã nguồn mở và đóng kín trong nhiều tiêu chuẩn ngành. Ngày cắt đứt kiến thức là tháng 12 năm 2023."
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct là mô hình lớn nhất và mạnh mẽ nhất trong mô hình Llama 3.1 Instruct, là một mô hình sinh dữ liệu và suy luận đối thoại tiên tiến, cũng có thể được sử dụng làm nền tảng cho việc tiền huấn luyện hoặc tinh chỉnh chuyên sâu trong các lĩnh vực cụ thể. Các mô hình ngôn ngữ lớn đa ngôn ngữ (LLMs) mà Llama 3.1 cung cấp là một tập hợp các mô hình sinh đã được tiền huấn luyện và điều chỉnh theo chỉ dẫn, bao gồm kích thước 8B, 70B và 405B (đầu vào/đầu ra văn bản). Các mô hình văn bản điều chỉnh theo chỉ dẫn của Llama 3.1 (8B, 70B, 405B) được tối ưu hóa cho các trường hợp đối thoại đa ngôn ngữ và đã vượt qua nhiều mô hình trò chuyện mã nguồn mở có sẵn trong các bài kiểm tra chuẩn ngành phổ biến. Llama 3.1 được thiết kế để sử dụng cho nhiều mục đích thương mại và nghiên cứu bằng nhiều ngôn ngữ. Các mô hình văn bản điều chỉnh theo chỉ dẫn phù hợp cho các cuộc trò chuyện giống như trợ lý, trong khi các mô hình đã được tiền huấn luyện có thể thích ứng với nhiều nhiệm vụ sinh ngôn ngữ tự nhiên khác nhau. Mô hình Llama 3.1 cũng hỗ trợ việc cải thiện các mô hình khác bằng cách sử dụng đầu ra của nó, bao gồm sinh dữ liệu tổng hợp và tinh chỉnh. Llama 3.1 là một mô hình ngôn ngữ tự hồi quy sử dụng kiến trúc biến áp tối ưu. Phiên bản điều chỉnh sử dụng tinh chỉnh có giám sát (SFT) và học tăng cường có phản hồi từ con người (RLHF) để phù hợp với sở thích của con người về tính hữu ích và an toàn."
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 là một mô hình ngôn ngữ lớn (LLM) mở dành cho các nhà phát triển, nhà nghiên cứu và doanh nghiệp, nhằm giúp họ xây dựng, thử nghiệm và mở rộng ý tưởng AI sinh một cách có trách nhiệm. Là một phần của hệ thống cơ sở hạ tầng đổi mới toàn cầu, nó rất phù hợp cho các thiết bị biên và thời gian huấn luyện nhanh hơn với khả năng tính toán và tài nguyên hạn chế."
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "LLM cao cấp, hỗ trợ tạo dữ liệu tổng hợp, chưng cất kiến thức và suy luận, phù hợp cho chatbot, lập trình và các nhiệm vụ chuyên biệt."
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "Tăng cường cuộc đối thoại phức tạp, có khả năng hiểu ngữ cảnh xuất sắc, suy luận và sinh văn bản."
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "Mô hình tiên tiến hàng đầu, có khả năng hiểu ngôn ngữ, suy luận xuất sắc và khả năng sinh văn bản."
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "Mô hình thị giác-ngôn ngữ tiên tiến, xuất sắc trong việc suy luận chất lượng cao từ hình ảnh."
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "Mô hình ngôn ngữ nhỏ tiên tiến hàng đầu, có khả năng hiểu ngôn ngữ, suy luận xuất sắc và khả năng sinh văn bản."
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "Mô hình ngôn ngữ nhỏ tiên tiến hàng đầu, có khả năng hiểu ngôn ngữ, suy luận xuất sắc và khả năng sinh văn bản."
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "Mô hình thị giác-ngôn ngữ tiên tiến, xuất sắc trong việc suy luận chất lượng cao từ hình ảnh."
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "Mô hình LLM tiên tiến, xuất sắc trong suy luận, toán học, kiến thức chung và gọi hàm."
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 là mô hình ngôn ngữ do AI của Microsoft cung cấp, thể hiện xuất sắc trong các lĩnh vực đối thoại phức tạp, đa ngôn ngữ, suy luận và trợ lý thông minh."
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K là một mô hình có khả năng xử lý ngữ cảnh siêu dài, phù hợp cho việc sinh văn bản siêu dài, đáp ứng nhu cầu nhiệm vụ sinh phức tạp, có thể xử lý nội dung lên đến 128.000 tokens, rất phù hợp cho nghiên cứu, học thuật và sinh tài liệu lớn."
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Mô hình hình ảnh Kimi (bao gồm moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, v.v.) có khả năng hiểu nội dung hình ảnh, bao gồm văn bản hình ảnh, màu sắc hình ảnh và hình dạng vật thể."
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K cung cấp khả năng xử lý ngữ cảnh độ dài trung bình, có thể xử lý 32.768 tokens, đặc biệt phù hợp cho việc sinh các tài liệu dài và đối thoại phức tạp, ứng dụng trong sáng tạo nội dung, sinh báo cáo và hệ thống đối thoại."
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Mô hình hình ảnh Kimi (bao gồm moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, v.v.) có khả năng hiểu nội dung hình ảnh, bao gồm văn bản hình ảnh, màu sắc hình ảnh và hình dạng vật thể."
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K được thiết kế đặc biệt cho các nhiệm vụ sinh văn bản ngắn, có hiệu suất xử lý cao, có thể xử lý 8.192 tokens, rất phù hợp cho các cuộc đối thoại ngắn, ghi chú nhanh và sinh nội dung nhanh chóng."
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Mô hình hình ảnh Kimi (bao gồm moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview, v.v.) có khả năng hiểu nội dung hình ảnh, bao gồm văn bản hình ảnh, màu sắc hình ảnh và hình dạng vật thể."
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B là phiên bản nâng cấp của Nous Hermes 2, bao gồm bộ dữ liệu phát triển nội bộ mới nhất."
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B là mô hình ngôn ngữ lớn được tùy chỉnh bởi NVIDIA, được thiết kế để cải thiện mức độ hỗ trợ của phản hồi do LLM tạo ra đối với các truy vấn của người dùng."
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B là một mô hình ngôn ngữ quy mô lớn tùy chỉnh bởi NVIDIA, nhằm nâng cao mức độ hỗ trợ của phản hồi do LLM tạo ra đối với các truy vấn của người dùng. Mô hình này đã thể hiện xuất sắc trong các bài kiểm tra chuẩn như Arena Hard, AlpacaEval 2 LC và GPT-4-Turbo MT-Bench, đứng đầu trong cả ba bài kiểm tra tự động cho đến ngày 1 tháng 10 năm 2024. Mô hình sử dụng RLHF (đặc biệt là REINFORCE), Llama-3.1-Nemotron-70B-Reward và HelpSteer2-Preference để đào tạo trên cơ sở mô hình Llama-3.1-70B-Instruct."
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "Mô hình ngôn ngữ độc đáo, cung cấp độ chính xác và hiệu suất không thể sánh kịp."
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B là mô hình ngôn ngữ lớn tùy chỉnh của NVIDIA, nhằm nâng cao tính hữu ích của các phản hồi do LLM tạo ra."
+ },
"o1": {
"description": "Tập trung vào suy diễn nâng cao và giải quyết các vấn đề phức tạp, bao gồm các nhiệm vụ toán học và khoa học. Rất phù hợp cho các ứng dụng cần hiểu biết sâu sắc về ngữ cảnh và quy trình làm việc đại diện."
},
- "o1-2024-12-17": {
- "description": "o1 là mô hình suy diễn mới của OpenAI, hỗ trợ đầu vào hình ảnh và văn bản và xuất ra văn bản, phù hợp cho các nhiệm vụ phức tạp cần kiến thức chung rộng rãi. Mô hình này có ngữ cảnh 200K và thời điểm cắt đứt kiến thức vào tháng 10 năm 2023."
- },
"o1-mini": {
"description": "o1-mini là một mô hình suy diễn nhanh chóng và tiết kiệm chi phí, được thiết kế cho các ứng dụng lập trình, toán học và khoa học. Mô hình này có ngữ cảnh 128K và thời điểm cắt kiến thức vào tháng 10 năm 2023."
},
"o1-preview": {
"description": "o1 là mô hình suy diễn mới của OpenAI, phù hợp cho các nhiệm vụ phức tạp cần kiến thức tổng quát rộng rãi. Mô hình này có ngữ cảnh 128K và thời điểm cắt kiến thức vào tháng 10 năm 2023."
},
+ "o3-mini": {
+ "description": "o3-mini là mô hình suy diễn nhỏ gọn mới nhất của chúng tôi, cung cấp trí thông minh cao với chi phí và độ trễ tương tự như o1-mini."
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba là mô hình ngôn ngữ Mamba 2 tập trung vào sinh mã, cung cấp hỗ trợ mạnh mẽ cho các nhiệm vụ mã và suy luận tiên tiến."
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini là mô hình mới nhất của OpenAI, được phát hành sau GPT-4 Omni, hỗ trợ đầu vào hình ảnh và văn bản, và đầu ra văn bản. Là mô hình nhỏ tiên tiến nhất của họ, nó rẻ hơn nhiều so với các mô hình tiên tiến gần đây khác và rẻ hơn hơn 60% so với GPT-3.5 Turbo. Nó giữ lại trí thông minh tiên tiến nhất trong khi có giá trị sử dụng đáng kể. GPT-4o mini đạt 82% điểm trong bài kiểm tra MMLU và hiện đứng đầu về sở thích trò chuyện so với GPT-4."
},
- "openai/o1": {
- "description": "o1 là mô hình suy diễn mới của OpenAI, hỗ trợ đầu vào hình ảnh và văn bản và xuất ra văn bản, phù hợp cho các nhiệm vụ phức tạp cần kiến thức chung rộng rãi. Mô hình này có ngữ cảnh 200K và thời điểm cắt đứt kiến thức vào tháng 10 năm 2023."
- },
"openai/o1-mini": {
"description": "o1-mini là một mô hình suy diễn nhanh chóng và tiết kiệm chi phí, được thiết kế cho các ứng dụng lập trình, toán học và khoa học. Mô hình này có ngữ cảnh 128K và thời điểm cắt kiến thức vào tháng 10 năm 2023."
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K được cấu hình với khả năng xử lý ngữ cảnh cực lớn, có thể xử lý tới 128K thông tin ngữ cảnh, đặc biệt phù hợp cho việc phân tích toàn bộ và xử lý mối liên hệ logic lâu dài trong nội dung văn bản dài, có thể cung cấp logic mạch lạc và hỗ trợ trích dẫn đa dạng trong giao tiếp văn bản phức tạp."
},
+ "qvq-72b-preview": {
+ "description": "Mô hình QVQ là mô hình nghiên cứu thử nghiệm do đội ngũ Qwen phát triển, tập trung vào việc nâng cao khả năng suy luận hình ảnh, đặc biệt trong lĩnh vực suy luận toán học."
+ },
"qwen-coder-plus-latest": {
"description": "Mô hình mã Qwen."
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "Mô hình ngôn ngữ hình ảnh quy mô siêu lớn của Tongyi Qianwen. So với phiên bản nâng cao, nó lại nâng cao khả năng suy luận hình ảnh và khả năng tuân thủ chỉ dẫn, cung cấp mức độ nhận thức và cảm nhận hình ảnh cao hơn."
},
+ "qwen-vl-ocr-latest": {
+ "description": "Công Nghệ Qianwen OCR là mô hình chuyên dụng cho việc trích xuất văn bản, tập trung vào khả năng trích xuất văn bản từ các loại hình ảnh như tài liệu, bảng biểu, đề thi, chữ viết tay, v.v. Nó có thể nhận diện nhiều loại văn bản, hiện hỗ trợ các ngôn ngữ: tiếng Trung, tiếng Anh, tiếng Pháp, tiếng Nhật, tiếng Hàn, tiếng Đức, tiếng Nga, tiếng Ý, tiếng Việt, tiếng Ả Rập."
+ },
"qwen-vl-plus-latest": {
"description": "Mô hình ngôn ngữ hình ảnh quy mô lớn phiên bản nâng cao của Tongyi Qianwen. Nâng cao khả năng nhận diện chi tiết và nhận diện văn bản, hỗ trợ độ phân giải trên một triệu pixel và các tỷ lệ chiều dài và chiều rộng tùy ý."
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 là một loạt mô hình ngôn ngữ lớn hoàn toàn mới, có khả năng hiểu và sinh mạnh mẽ hơn."
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "LLM hướng đến tiếng Trung và tiếng Anh, tập trung vào ngôn ngữ, lập trình, toán học, suy luận và các lĩnh vực khác."
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "LLM cao cấp, hỗ trợ tạo mã, suy luận và sửa chữa, bao gồm các ngôn ngữ lập trình phổ biến."
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "Mô hình mã mạnh mẽ cỡ trung, hỗ trợ độ dài ngữ cảnh 32K, xuất sắc trong lập trình đa ngôn ngữ."
+ },
"qwen2": {
"description": "Qwen2 là mô hình ngôn ngữ quy mô lớn thế hệ mới của Alibaba, hỗ trợ các nhu cầu ứng dụng đa dạng với hiệu suất xuất sắc."
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "Mô hình 14B quy mô mở nguồn của Qwen 2.5."
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "Mô hình quy mô 72B được mở nguồn từ Qianwen 2.5."
+ },
"qwen2.5-32b-instruct": {
"description": "Mô hình 32B quy mô mở nguồn của Qwen 2.5."
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Mô hình Qwen-Math có khả năng giải quyết bài toán toán học mạnh mẽ."
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "Nâng cao khả năng theo dõi lệnh, toán học, giải quyết vấn đề, mã hóa, nâng cao khả năng nhận diện mọi thứ, hỗ trợ định vị chính xác các yếu tố thị giác từ nhiều định dạng khác nhau, hỗ trợ hiểu và định vị thời gian sự kiện trong các tệp video dài (tối đa 10 phút), có khả năng hiểu thứ tự thời gian và tốc độ, hỗ trợ điều khiển Agent trên OS hoặc Mobile dựa trên khả năng phân tích và định vị, khả năng trích xuất thông tin quan trọng và xuất định dạng Json mạnh mẽ, phiên bản này là phiên bản 72B, phiên bản mạnh nhất trong dòng sản phẩm này."
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "Nâng cao khả năng theo dõi lệnh, toán học, giải quyết vấn đề, mã hóa, nâng cao khả năng nhận diện mọi thứ, hỗ trợ định vị chính xác các yếu tố thị giác từ nhiều định dạng khác nhau, hỗ trợ hiểu và định vị thời gian sự kiện trong các tệp video dài (tối đa 10 phút), có khả năng hiểu thứ tự thời gian và tốc độ, hỗ trợ điều khiển Agent trên OS hoặc Mobile dựa trên khả năng phân tích và định vị, khả năng trích xuất thông tin quan trọng và xuất định dạng Json mạnh mẽ, phiên bản này là phiên bản 72B, phiên bản mạnh nhất trong dòng sản phẩm này."
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 là thế hệ mô hình ngôn ngữ quy mô lớn mới của Alibaba, hỗ trợ các nhu cầu ứng dụng đa dạng với hiệu suất xuất sắc."
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro là một LLM thông minh cao do Upstage phát hành, tập trung vào khả năng tuân theo hướng dẫn trên một GPU, đạt điểm IFEval trên 80. Hiện tại hỗ trợ tiếng Anh, phiên bản chính thức dự kiến ra mắt vào tháng 11 năm 2024, sẽ mở rộng hỗ trợ ngôn ngữ và độ dài ngữ cảnh."
},
+ "sonar": {
+ "description": "Sản phẩm tìm kiếm nhẹ dựa trên ngữ cảnh tìm kiếm, nhanh hơn và rẻ hơn so với Sonar Pro."
+ },
+ "sonar-pro": {
+ "description": "Sản phẩm tìm kiếm nâng cao hỗ trợ ngữ cảnh tìm kiếm, cho phép truy vấn và theo dõi nâng cao."
+ },
+ "sonar-reasoning": {
+ "description": "Sản phẩm API mới được hỗ trợ bởi mô hình suy luận của DeepSeek."
+ },
"step-1-128k": {
"description": "Cân bằng hiệu suất và chi phí, phù hợp cho các tình huống chung."
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "Mô hình này có khả năng hiểu video mạnh mẽ."
},
+ "step-1o-vision-32k": {
+ "description": "Mô hình này có khả năng hiểu hình ảnh mạnh mẽ. So với các mô hình trong series step-1v, nó có hiệu suất thị giác vượt trội hơn."
+ },
"step-1v-32k": {
"description": "Hỗ trợ đầu vào hình ảnh, tăng cường trải nghiệm tương tác đa mô hình."
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "Hỗ trợ tương tác ngữ cảnh quy mô lớn, phù hợp cho các tình huống đối thoại phức tạp."
},
+ "step-2-mini": {
+ "description": "Mô hình lớn siêu tốc dựa trên kiến trúc Attention tự nghiên cứu thế hệ mới MFA, đạt được hiệu quả tương tự như step1 với chi phí rất thấp, đồng thời duy trì thông lượng cao hơn và độ trễ phản hồi nhanh hơn. Có khả năng xử lý các nhiệm vụ chung, đặc biệt có năng lực trong lập trình."
+ },
+ "taichu2_mm": {
+ "description": "Kết hợp khả năng hiểu hình ảnh, chuyển giao kiến thức, suy luận logic, nổi bật trong lĩnh vực hỏi đáp hình ảnh và văn bản."
+ },
"taichu_llm": {
"description": "Mô hình ngôn ngữ lớn Taichu có khả năng hiểu ngôn ngữ mạnh mẽ và các khả năng như sáng tạo văn bản, trả lời câu hỏi kiến thức, lập trình mã, tính toán toán học, suy luận logic, phân tích cảm xúc, tóm tắt văn bản. Đổi mới kết hợp giữa đào tạo trước với dữ liệu phong phú từ nhiều nguồn, thông qua việc liên tục cải tiến công nghệ thuật toán và hấp thụ kiến thức mới từ dữ liệu văn bản khổng lồ, giúp mô hình ngày càng hoàn thiện. Cung cấp thông tin và dịch vụ tiện lợi hơn cho người dùng cùng trải nghiệm thông minh hơn."
},
+ "text-embedding-3-large": {
+ "description": "Mô hình vector hóa mạnh mẽ nhất, phù hợp cho các nhiệm vụ tiếng Anh và không phải tiếng Anh."
+ },
+ "text-embedding-3-small": {
+ "description": "Mô hình Embedding thế hệ mới hiệu quả và tiết kiệm, phù hợp cho tìm kiếm kiến thức, ứng dụng RAG và các tình huống khác."
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) cung cấp khả năng tính toán nâng cao thông qua chiến lược và kiến trúc mô hình hiệu quả."
},
+ "tts-1": {
+ "description": "Mô hình chuyển văn bản thành giọng nói mới nhất, tối ưu hóa tốc độ cho các tình huống thời gian thực."
+ },
+ "tts-1-hd": {
+ "description": "Mô hình chuyển văn bản thành giọng nói mới nhất, tối ưu hóa cho chất lượng."
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) phù hợp cho các nhiệm vụ chỉ dẫn tinh vi, cung cấp khả năng xử lý ngôn ngữ xuất sắc."
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet nâng cao tiêu chuẩn ngành, hiệu suất vượt trội so với các mô hình cạnh tranh và Claude 3 Opus, thể hiện xuất sắc trong nhiều đánh giá, đồng thời có tốc độ và chi phí tương đương với các mô hình tầm trung của chúng tôi."
},
+ "whisper-1": {
+ "description": "Mô hình nhận diện giọng nói đa năng, hỗ trợ nhận diện giọng nói đa ngôn ngữ, dịch giọng nói và nhận diện ngôn ngữ."
+ },
"wizardlm2": {
"description": "WizardLM 2 là mô hình ngôn ngữ do Microsoft AI cung cấp, đặc biệt xuất sắc trong các lĩnh vực đối thoại phức tạp, đa ngôn ngữ, suy luận và trợ lý thông minh."
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "Mô hình cho các nhiệm vụ hình ảnh phức tạp, cung cấp khả năng hiểu và phân tích hình ảnh hiệu suất cao."
+ },
+ "yi-vision-v2": {
+ "description": "Mô hình nhiệm vụ thị giác phức tạp, cung cấp khả năng hiểu và phân tích hiệu suất cao dựa trên nhiều hình ảnh."
}
}
diff --git a/locales/vi-VN/providers.json b/locales/vi-VN/providers.json
index ad77598444e8c..81b6bf2fa191e 100644
--- a/locales/vi-VN/providers.json
+++ b/locales/vi-VN/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure cung cấp nhiều mô hình AI tiên tiến, bao gồm GPT-3.5 và dòng GPT-4 mới nhất, hỗ trợ nhiều loại dữ liệu và nhiệm vụ phức tạp, cam kết cung cấp các giải pháp AI an toàn, đáng tin cậy và bền vững."
},
+ "azureai": {
+ "description": "Azure cung cấp nhiều mô hình AI tiên tiến, bao gồm GPT-3.5 và dòng GPT-4 mới nhất, hỗ trợ nhiều loại dữ liệu và nhiệm vụ phức tạp, cam kết cung cấp các giải pháp AI an toàn, đáng tin cậy và bền vững."
+ },
"baichuan": {
"description": "Baichuan Intelligent là công ty tập trung vào nghiên cứu phát triển mô hình ngôn ngữ lớn AI, mô hình của họ thể hiện xuất sắc trong các nhiệm vụ tiếng Trung như bách khoa toàn thư, xử lý văn bản dài và sáng tác, vượt trội hơn so với các mô hình chính thống quốc tế. Baichuan Intelligent còn có khả năng đa phương thức hàng đầu trong ngành, thể hiện xuất sắc trong nhiều bài kiểm tra uy tín. Các mô hình của họ bao gồm Baichuan 4, Baichuan 3 Turbo và Baichuan 3 Turbo 128k, được tối ưu hóa cho các tình huống ứng dụng khác nhau, cung cấp giải pháp hiệu quả về chi phí."
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek là một công ty tập trung vào nghiên cứu và ứng dụng công nghệ trí tuệ nhân tạo, mô hình mới nhất của họ, DeepSeek-V2.5, kết hợp khả năng đối thoại chung và xử lý mã, đồng thời đạt được sự cải thiện đáng kể trong việc căn chỉnh sở thích của con người, nhiệm vụ viết và tuân theo chỉ dẫn."
},
+ "doubao": {
+ "description": "Mô hình lớn tự phát triển do ByteDance phát triển. Được xác thực qua hơn 50 tình huống kinh doanh nội bộ của ByteDance, với việc sử dụng hàng nghìn tỷ token mỗi ngày để liên tục cải tiến, cung cấp nhiều khả năng đa phương thức, tạo ra trải nghiệm kinh doanh phong phú cho doanh nghiệp với hiệu quả mô hình chất lượng cao."
+ },
"fireworksai": {
"description": "Fireworks AI là nhà cung cấp dịch vụ mô hình ngôn ngữ cao cấp hàng đầu, tập trung vào gọi chức năng và xử lý đa phương thức. Mô hình mới nhất của họ, Firefunction V2, dựa trên Llama-3, được tối ưu hóa cho gọi chức năng, đối thoại và tuân theo chỉ dẫn. Mô hình ngôn ngữ hình ảnh FireLLaVA-13B hỗ trợ đầu vào hỗn hợp hình ảnh và văn bản. Các mô hình đáng chú ý khác bao gồm dòng Llama và dòng Mixtral, cung cấp hỗ trợ cho việc tuân theo và tạo ra chỉ dẫn đa ngôn ngữ hiệu quả."
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "Tổ chức mã nguồn mở chuyên nghiên cứu và phát triển công cụ cho mô hình lớn. Cung cấp nền tảng mã nguồn mở hiệu quả, dễ sử dụng cho tất cả các nhà phát triển AI, giúp tiếp cận công nghệ mô hình lớn và thuật toán tiên tiến nhất."
},
+ "jina": {
+ "description": "Jina AI được thành lập vào năm 2020, là một công ty hàng đầu trong lĩnh vực AI tìm kiếm. Nền tảng tìm kiếm của chúng tôi bao gồm các mô hình vector, bộ tái sắp xếp và các mô hình ngôn ngữ nhỏ, giúp các doanh nghiệp xây dựng các ứng dụng tìm kiếm sinh tạo và đa phương tiện đáng tin cậy và chất lượng cao."
+ },
+ "lmstudio": {
+ "description": "LM Studio là một ứng dụng máy tính để phát triển và thử nghiệm các LLM trên máy tính của bạn."
+ },
"minimax": {
"description": "MiniMax là công ty công nghệ trí tuệ nhân tạo tổng quát được thành lập vào năm 2021, cam kết cùng người dùng sáng tạo trí thông minh. MiniMax đã tự phát triển nhiều mô hình lớn đa phương thức, bao gồm mô hình văn bản MoE với một triệu tham số, mô hình giọng nói và mô hình hình ảnh. Họ cũng đã phát hành các ứng dụng như AI Hải Lý."
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI là một nền tảng cung cấp dịch vụ API cho nhiều mô hình ngôn ngữ lớn và tạo hình ảnh AI, linh hoạt, đáng tin cậy và hiệu quả về chi phí. Nó hỗ trợ các mô hình mã nguồn mở mới nhất như Llama3, Mistral, và cung cấp giải pháp API toàn diện, thân thiện với người dùng và tự động mở rộng cho phát triển ứng dụng AI, phù hợp cho sự phát triển nhanh chóng của các công ty khởi nghiệp AI."
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ cung cấp các container có thể được sử dụng để tự lưu trữ các dịch vụ vi mô suy diễn GPU tăng tốc, hỗ trợ triển khai các mô hình AI đã được huấn luyện trước và tùy chỉnh trên đám mây, trung tâm dữ liệu, máy tính cá nhân RTX™ AI và trạm làm việc."
+ },
"ollama": {
"description": "Mô hình do Ollama cung cấp bao quát rộng rãi các lĩnh vực như tạo mã, tính toán toán học, xử lý đa ngôn ngữ và tương tác đối thoại, hỗ trợ nhu cầu đa dạng cho triển khai doanh nghiệp và địa phương."
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "Viện Nghiên cứu Tự động hóa Trung Quốc và Viện Nghiên cứu Trí tuệ Nhân tạo Vũ Hán đã phát hành mô hình lớn đa phương thức thế hệ mới, hỗ trợ các nhiệm vụ hỏi đáp toàn diện như hỏi đáp nhiều vòng, sáng tác văn bản, tạo hình ảnh, hiểu 3D, phân tích tín hiệu, v.v., với khả năng nhận thức, hiểu biết và sáng tác mạnh mẽ hơn, mang đến trải nghiệm tương tác hoàn toàn mới."
},
+ "tencentcloud": {
+ "description": "Năng lực nguyên tử của động cơ tri thức (LLM Knowledge Engine Atomic Power) được phát triển dựa trên động cơ tri thức, cung cấp khả năng hỏi đáp toàn diện cho doanh nghiệp và nhà phát triển, cho phép xây dựng và phát triển ứng dụng mô hình một cách linh hoạt. Bạn có thể sử dụng nhiều năng lực nguyên tử để tạo ra dịch vụ mô hình riêng của mình, kết hợp các dịch vụ như phân tích tài liệu, tách rời, embedding, và viết lại nhiều vòng để tùy chỉnh các dịch vụ AI đặc thù cho doanh nghiệp."
+ },
"togetherai": {
"description": "Together AI cam kết đạt được hiệu suất hàng đầu thông qua các mô hình AI sáng tạo, cung cấp khả năng tùy chỉnh rộng rãi, bao gồm hỗ trợ mở rộng nhanh chóng và quy trình triển khai trực quan, đáp ứng nhiều nhu cầu của doanh nghiệp."
},
"upstage": {
"description": "Upstage tập trung vào việc phát triển các mô hình AI cho nhiều nhu cầu thương mại khác nhau, bao gồm Solar LLM và AI tài liệu, nhằm đạt được trí thông minh nhân tạo tổng quát (AGI) cho công việc. Tạo ra các đại lý đối thoại đơn giản thông qua Chat API, và hỗ trợ gọi chức năng, dịch thuật, nhúng và ứng dụng trong các lĩnh vực cụ thể."
},
+ "vllm": {
+ "description": "vLLM là một thư viện nhanh chóng và dễ sử dụng cho suy diễn và dịch vụ LLM."
+ },
+ "volcengine": {
+ "description": "Nền tảng phát triển dịch vụ mô hình lớn do ByteDance phát triển, cung cấp dịch vụ gọi mô hình phong phú, an toàn và có giá cả cạnh tranh, đồng thời cung cấp dữ liệu mô hình, tinh chỉnh, suy diễn, đánh giá và các chức năng đầu cuối khác, đảm bảo toàn diện cho việc phát triển ứng dụng AI của bạn."
+ },
"wenxin": {
"description": "Nền tảng phát triển và dịch vụ ứng dụng AI gốc với mô hình lớn một cửa dành cho doanh nghiệp, cung cấp chuỗi công cụ toàn diện và dễ sử dụng cho phát triển mô hình trí tuệ nhân tạo sinh sinh và phát triển ứng dụng."
},
diff --git a/locales/vi-VN/setting.json b/locales/vi-VN/setting.json
index 2d7cbbd24a246..bf985474faf10 100644
--- a/locales/vi-VN/setting.json
+++ b/locales/vi-VN/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "Bật giới hạn phản hồi một lần"
},
+ "enableReasoningEffort": {
+ "title": "Bật điều chỉnh cường độ suy luận"
+ },
"frequencyPenalty": {
- "desc": "Giá trị càng cao, càng có khả năng giảm sự lặp lại của từ/cụm từ",
- "title": "Hình phạt tần suất"
+ "desc": "Giá trị càng lớn, từ ngữ càng phong phú đa dạng; giá trị càng thấp, từ ngữ càng đơn giản mộc mạc",
+ "title": "Độ phong phú từ vựng"
},
"maxTokens": {
"desc": "Số lượng Token tối đa được sử dụng trong mỗi tương tác",
@@ -212,19 +215,31 @@
"desc": "Mô hình {{provider}}",
"title": "Mô hình"
},
+ "params": {
+ "title": "Tham số nâng cao"
+ },
"presencePenalty": {
- "desc": "Giá trị càng cao, càng có khả năng mở rộng đến chủ đề mới",
- "title": "Độ mới của chủ đề"
+ "desc": "Giá trị càng lớn, càng có xu hướng sử dụng các cách diễn đạt khác nhau, tránh lặp lại khái niệm; giá trị càng nhỏ, càng có xu hướng sử dụng các khái niệm hoặc mô tả lặp lại, thể hiện tính nhất quán cao hơn",
+ "title": "Độ phân tán trong diễn đạt"
+ },
+ "reasoningEffort": {
+ "desc": "Giá trị càng lớn, khả năng suy luận càng mạnh, nhưng có thể làm tăng thời gian phản hồi và tiêu tốn Token",
+ "options": {
+ "high": "Cao",
+ "low": "Thấp",
+ "medium": "Trung bình"
+ },
+ "title": "Cường độ suy luận"
},
"temperature": {
- "desc": "Giá trị càng cao, phản hồi càng ngẫu nhiên",
- "title": "Độ ngẫu nhiên",
- "titleWithValue": "Độ ngẫu nhiên {{value}}"
+ "desc": "Giá trị càng lớn, câu trả lời càng sáng tạo và giàu trí tưởng tượng; giá trị càng nhỏ, câu trả lời càng nghiêm ngặt",
+ "title": "Mức độ sáng tạo",
+ "warning": "Giá trị mức độ sáng tạo quá lớn có thể dẫn đến đầu ra bị lỗi"
},
"title": "Cài đặt mô hình",
"topP": {
- "desc": "Tương tự như độ ngẫu nhiên, nhưng không nên thay đổi cùng lúc với độ ngẫu nhiên",
- "title": "Lấy mẫu cốt lõi"
+ "desc": "Xem xét bao nhiêu khả năng, giá trị càng lớn, chấp nhận nhiều câu trả lời khả thi hơn; giá trị càng nhỏ, có xu hướng chọn câu trả lời khả thi nhất. Không khuyến nghị thay đổi cùng với mức độ sáng tạo",
+ "title": "Mức độ mở trong tư duy"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "Cài đặt chung",
"experiment": "Thử nghiệm",
"llm": "Mô hình ngôn ngữ",
+ "provider": "Nhà cung cấp AI",
"sync": "Đồng bộ trên đám mây",
"system-agent": "Trợ lý hệ thống",
"tts": "Dịch vụ giọng nói"
diff --git a/locales/zh-CN/auth.json b/locales/zh-CN/auth.json
index 422463902c045..3718a19d0a532 100644
--- a/locales/zh-CN/auth.json
+++ b/locales/zh-CN/auth.json
@@ -84,4 +84,4 @@
"security": "安全",
"stats": "数据统计"
}
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/changelog.json b/locales/zh-CN/changelog.json
index c139ed36954bf..59d8b53a262d7 100644
--- a/locales/zh-CN/changelog.json
+++ b/locales/zh-CN/changelog.json
@@ -8,11 +8,11 @@
"allChangelog": "查看所有更新日志",
"description": "持续追踪 {{appName}} 的新功能和改进",
"pagination": {
- "older": "查看历史变更",
- "prev": "上一页"
+ "next": "下一页",
+ "older": "查看历史变更"
},
"readDetails": "阅读详情",
"title": "更新日志",
"versionDetails": "版本详情",
"welcomeBack": "欢迎回来!"
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/chat.json b/locales/zh-CN/chat.json
index 9bd8da7119fee..e7bedd2625c0f 100644
--- a/locales/zh-CN/chat.json
+++ b/locales/zh-CN/chat.json
@@ -181,4 +181,4 @@
}
},
"zenMode": "专注模式"
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/clerk.json b/locales/zh-CN/clerk.json
index d2c2f4eafff0a..28669e61cfc00 100644
--- a/locales/zh-CN/clerk.json
+++ b/locales/zh-CN/clerk.json
@@ -766,4 +766,4 @@
"title": "添加 Web3 钱包"
}
}
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/common.json b/locales/zh-CN/common.json
index 67e1795a7caed..dec1331e893bb 100644
--- a/locales/zh-CN/common.json
+++ b/locales/zh-CN/common.json
@@ -275,6 +275,7 @@
},
"temp": "临时",
"terms": "服务条款",
+ "update": "更新",
"updateAgent": "更新助理信息",
"upgradeVersion": {
"action": "升级",
@@ -285,6 +286,7 @@
"anonymousNickName": "匿名用户",
"billing": "账单管理",
"cloud": "体验 {{name}}",
+ "community": "社区版",
"data": "数据存储",
"defaultNickname": "社区版用户",
"discord": "社区支持",
@@ -294,10 +296,9 @@
"help": "帮助中心",
"moveGuide": "设置按钮搬到这里啦",
"plans": "订阅方案",
- "preview": "预览版",
"profile": "账户管理",
"setting": "应用设置",
"usages": "用量统计"
},
"version": "版本"
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/components.json b/locales/zh-CN/components.json
index 2728ca27cd340..2a0a34d529056 100644
--- a/locales/zh-CN/components.json
+++ b/locales/zh-CN/components.json
@@ -76,6 +76,7 @@
"custom": "自定义模型，默认设定同时支持函数调用与视觉识别，请根据实际情况验证上述能力的可用性",
"file": "该模型支持上传文件读取与识别",
"functionCall": "该模型支持函数调用（Function Call）",
+ "reasoning": "该模型支持深度思考",
"tokens": "该模型单个会话最多支持 {{tokens}} Tokens",
"vision": "该模型支持视觉识别"
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "没有启用的模型，请前往设置开启",
"provider": "提供商"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "因浏览器安全限制，你需要为 Ollama 进行跨域配置后方可正常使用。",
+ "linux": {
+ "env": "在 [Service] 部分下添加 `Environment`，添加 OLLAMA_ORIGINS 环境变量：",
+ "reboot": "重载 systemd 并重启 Ollama",
+ "systemd": "调用 systemd 编辑 ollama 服务："
+ },
+ "macos": "请打开「终端」应用程序，并粘贴以下指令，并按回车运行",
+ "reboot": "请在执行完成后重启 Ollama 服务",
+ "title": "配置 Ollama 允许跨域访问",
+ "windows": "在 Windows 上，点击「控制面板」，进入编辑系统环境变量。为您的用户账户新建名为 「OLLAMA_ORIGINS」 的环境变量，值为 * ，点击 「OK/应用」 保存"
+ },
+ "install": {
+ "description": "请确认你已经开启 Ollama ，如果没有下载 Ollama ，请前往官网<1>下载",
+ "docker": "如果你更倾向于使用 Docker，Ollama 也提供了官方 Docker 镜像，你可以通过以下命令拉取：",
+ "linux": {
+ "command": "通过以下命令安装：",
+ "manual": "或者，你也可以参考 <1>Linux 手动安装指南 自行安装"
+ },
+ "title": "在本地安装并开启 Ollama 应用",
+ "windowsTab": "Windows (预览版)"
+ }
+ },
+ "Thinking": {
+ "thinking": "深度思考中...",
+ "thought": "已深度思考（用时 {{duration}} 秒）",
+ "thoughtWithDuration": "已深度思考"
}
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/discover.json b/locales/zh-CN/discover.json
index 10fd75bdc74c3..4eb7fa102f56e 100644
--- a/locales/zh-CN/discover.json
+++ b/locales/zh-CN/discover.json
@@ -126,6 +126,10 @@
"title": "话题新鲜度"
},
"range": "范围",
+ "reasoning_effort": {
+ "desc": "此设置用于控制模型在生成回答前的推理强度。低强度优先响应速度并节省 Token，高强度提供更完整的推理，但会消耗更多 Token 并降低响应速度。默认值为中，平衡推理准确性与响应速度。",
+ "title": "推理强度"
+ },
"temperature": {
"desc": "此设置影响模型回应的多样性。较低的值会导致更可预测和典型的回应，而较高的值则鼓励更多样化和不常见的回应。当值设为0时，模型对于给定的输入总是给出相同的回应。",
"title": "随机性"
@@ -201,4 +205,4 @@
"plugins": "插件",
"providers": "模型服务商"
}
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/error.json b/locales/zh-CN/error.json
index e35c9ca1896ff..899490642b64c 100644
--- a/locales/zh-CN/error.json
+++ b/locales/zh-CN/error.json
@@ -139,4 +139,4 @@
"unknownError": "错误原因: {{reason}}",
"uploadFailed": "文件上传失败"
}
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/file.json b/locales/zh-CN/file.json
index 236c7d7681b3e..134eed471f21e 100644
--- a/locales/zh-CN/file.json
+++ b/locales/zh-CN/file.json
@@ -91,4 +91,4 @@
"uploading": "正在上传"
}
}
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/knowledgeBase.json b/locales/zh-CN/knowledgeBase.json
index 63a5c45c244ae..2773110118b93 100644
--- a/locales/zh-CN/knowledgeBase.json
+++ b/locales/zh-CN/knowledgeBase.json
@@ -29,4 +29,4 @@
"testing": "召回测试"
},
"title": "知识库"
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/metadata.json b/locales/zh-CN/metadata.json
index c47cf8685fd5a..eaef354afb5a1 100644
--- a/locales/zh-CN/metadata.json
+++ b/locales/zh-CN/metadata.json
@@ -36,4 +36,4 @@
"description": "{{appName}} 带给你最好的 ChatGPT, Claude , Gemini, OLLaMA WebUI 使用体验",
"title": "欢迎使用 {{appName}}：个人 AI 效能工具，给自己一个更聪明的大脑"
}
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/migration.json b/locales/zh-CN/migration.json
index 3c92fe66e642f..1d88055c91d6c 100644
--- a/locales/zh-CN/migration.json
+++ b/locales/zh-CN/migration.json
@@ -42,4 +42,4 @@
"missVersion": "导入数据缺少版本号，请检查文件后重试",
"noMigration": "没有找到当前版本对应的迁移方案，请检查版本号后重试。如仍有问题请提交问题反馈"
}
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/modelProvider.json b/locales/zh-CN/modelProvider.json
index e40e224dc835e..03a5621b85777 100644
--- a/locales/zh-CN/modelProvider.json
+++ b/locales/zh-CN/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API Key"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Azure 的 API 版本，遵循 YYYY-MM-DD 格式，查阅[最新版本](https://learn.microsoft.com/zh-cn/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "获取列表",
+ "title": "Azure API Version"
+ },
+ "endpoint": {
+ "desc": "从 Azure AI 项目概述找到 Azure AI 模型推理终结点",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Azure AI 终结点"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "从 Azure AI 项目概述找到 API 密钥",
+ "placeholder": "Azure 密钥",
+ "title": "密钥"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "填入 AWS Access Key Id",
@@ -63,6 +81,46 @@
"title": "Cloudflare 账户 ID / API 地址"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "请填写你的 API Key",
+ "title": "API Key"
+ },
+ "basicTitle": "基本信息",
+ "configTitle": "配置信息",
+ "confirm": "新建",
+ "createSuccess": "新建成功",
+ "description": {
+ "placeholder": "服务商简介（选填）",
+ "title": "服务商简介"
+ },
+ "id": {
+ "desc": "作为服务商唯一标识，创建后将不可修改",
+ "format": "只能包含数字、小写字母、连字符（-）和下划线（_）",
+ "placeholder": "例如 openai、gemini 等",
+ "required": "请填写服务商 ID",
+ "title": "服务商 ID"
+ },
+ "logo": {
+ "required": "请上传正确的服务商 Logo",
+ "title": "服务商 Logo"
+ },
+ "name": {
+ "placeholder": "请输入服务商的展示名称",
+ "required": "请填写服务商名称",
+ "title": "服务商名称"
+ },
+ "proxyUrl": {
+ "required": "请填写代理地址",
+ "title": "代理地址"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "请选择 SDK 类型",
+ "title": "请求格式"
+ },
+ "title": "创建自定义 AI 服务商"
+ },
"github": {
"personalAccessToken": {
"desc": "填入你的 Github PAT，点击 [这里](https://github.com/settings/tokens) 创建",
@@ -77,6 +135,23 @@
"title": "HuggingFace Token"
}
},
+ "list": {
+ "title": {
+ "disabled": "未启用服务商",
+ "enabled": "已启用服务商"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "添加自定义服务商",
+ "all": "全部",
+ "list": {
+ "disabled": "未启用",
+ "enabled": "已启用"
+ },
+ "notFound": "未找到搜索结果",
+ "searchProviders": "搜索服务商...",
+ "sort": "自定义排序"
+ },
"ollama": {
"checker": {
"desc": "测试代理地址是否正确填写",
@@ -94,33 +169,9 @@
"title": "正在下载模型 {{model}} "
},
"endpoint": {
- "desc": "填入 Ollama 接口代理地址，本地未额外指定可留空",
+ "desc": "必须包含http(s)://，本地未额外指定可留空",
"title": "Ollama 服务地址"
},
- "setup": {
- "cors": {
- "description": "因浏览器安全限制，你需要为 Ollama 进行跨域配置后方可正常使用。",
- "linux": {
- "env": "在 [Service] 部分下添加 `Environment`，添加 OLLAMA_ORIGINS 环境变量：",
- "reboot": "重载 systemd 并重启 Ollama",
- "systemd": "调用 systemd 编辑 ollama 服务："
- },
- "macos": "请打开「终端」应用程序，并粘贴以下指令，并按回车运行",
- "reboot": "请在执行完成后重启 Ollama 服务",
- "title": "配置 Ollama 允许跨域访问",
- "windows": "在 Windows 上，点击「控制面板」，进入编辑系统环境变量。为您的用户账户新建名为 「OLLAMA_ORIGINS」 的环境变量，值为 * ，点击 「OK/应用」 保存"
- },
- "install": {
- "description": "请确认你已经开启 Ollama ，如果没有下载 Ollama ，请前往官网<1>下载",
- "docker": "如果你更倾向于使用 Docker，Ollama 也提供了官方 Docker 镜像，你可以通过以下命令拉取：",
- "linux": {
- "command": "通过以下命令安装：",
- "manual": "或者，你也可以参考 <1>Linux 手动安装指南 自行安装"
- },
- "title": "在本地安装并开启 Ollama 应用",
- "windowsTab": "Windows (预览版)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "取消下载",
@@ -131,25 +182,145 @@
"title": "下载指定的 Ollama 模型"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "填入百度千帆平台的 Access Key",
- "placeholder": "Qianfan Access Key",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "您的秘钥与代理地址等将使用 <1>AES-GCM 加密算法进行加密",
+ "apiKey": {
+ "desc": "请填写你的 {{name}} API Key",
+ "placeholder": "{{name}} API Key",
+ "title": "API Key"
+ },
+ "baseURL": {
+ "desc": "必须包含 http(s)://",
+ "invalid": "请输入合法的 URL",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "API 代理地址"
+ },
+ "checker": {
+ "button": "检查",
+ "desc": "测试 Api Key 与代理地址是否正确填写",
+ "pass": "检查通过",
+ "title": "连通性检查"
+ },
+ "fetchOnClient": {
+ "desc": "客户端请求模式将从浏览器直接发起会话请求，可提升响应速度",
+ "title": "使用客户端请求模式"
+ },
+ "helpDoc": "配置教程",
+ "waitingForMore": "更多模型正在 <1>计划接入 中，敬请期待"
},
- "checker": {
- "desc": "测试 AccessKey / SecretAccess 是否填写正确"
+ "createNew": {
+ "title": "创建自定义 AI 模型"
+ },
+ "item": {
+ "config": "配置模型",
+ "customModelCards": {
+ "addNew": "创建并添加 {{id}} 模型",
+ "confirmDelete": "即将删除该自定义模型，删除后将不可恢复，请谨慎操作。"
+ },
+ "delete": {
+ "confirm": "确认删除模型 {{displayName}}？",
+ "success": "删除成功",
+ "title": "删除模型"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "在 Azure OpenAI 中实际请求的字段",
+ "placeholder": "请输入 Azure 中的模型部署名称",
+ "title": "模型部署名称"
+ },
+ "deployName": {
+ "extra": "发送请求时会将该字段作为模型 ID",
+ "placeholder": "请输入模型实际部署的名称或 id",
+ "title": "模型部署名称"
+ },
+ "displayName": {
+ "placeholder": "请输入模型的展示名称，例如 ChatGPT、GPT-4 等",
+ "title": "模型展示名称"
+ },
+ "files": {
+ "extra": "当前文件上传实现仅为一种 Hack 方案，仅限自行尝试。完整文件上传能力请等待后续实现",
+ "title": "支持文件上传"
+ },
+ "functionCall": {
+ "extra": "此配置将仅开启模型使用工具的能力，进而可以为模型添加工具类的插件。但是否支持真正使用工具完全取决于模型本身，请自行测试的可用性",
+ "title": "支持工具使用"
+ },
+ "id": {
+ "extra": "创建后不可修改，调用 AI 时将作为模型 id 使用",
+ "placeholder": "请输入模型 id，例如 gpt-4o 或 claude-3.5-sonnet",
+ "title": "模型 ID"
+ },
+ "modalTitle": "自定义模型配置",
+ "reasoning": {
+ "extra": "此配置将仅开启模型深度思考的能力，具体效果完全取决于模型本身，请自行测试该模型是否具备可用的深度思考能力",
+ "title": "支持深度思考"
+ },
+ "tokens": {
+ "extra": "设置模型支持的最大 Token 数",
+ "title": "最大上下文窗口",
+ "unlimited": "无限制"
+ },
+ "vision": {
+ "extra": "此配置将仅开启应用中的图片上传配置，是否支持识别完全取决于模型本身，请自行测试该模型的视觉识别能力可用性",
+ "title": "支持视觉识别"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/图片",
+ "inputCharts": "${{amount}}/M 字符",
+ "inputMinutes": "${{amount}}/分钟",
+ "inputTokens": "输入 ${{amount}}/M",
+ "outputTokens": "输出 ${{amount}}/M"
+ },
+ "releasedAt": "发布于{{releasedAt}}"
},
- "secretKey": {
- "desc": "填入百度千帆平台 Secret Key",
- "placeholder": "Qianfan Secret Key",
- "title": "Secret Key"
+ "list": {
+ "addNew": "添加模型",
+ "disabled": "未启用",
+ "disabledActions": {
+ "showMore": "显示全部"
+ },
+ "empty": {
+ "desc": "请创建自定义模型或拉取模型后开始使用吧",
+ "title": "暂无可用模型"
+ },
+ "enabled": "已启用",
+ "enabledActions": {
+ "disableAll": "全部禁用",
+ "enableAll": "全部启用",
+ "sort": "自定义模型排序"
+ },
+ "enabledEmpty": "暂无启用模型，请从下方列表中启用心仪的模型吧~",
+ "fetcher": {
+ "clear": "清除获取的模型",
+ "fetch": "获取模型列表",
+ "fetching": "正在获取模型列表...",
+ "latestTime": "上次更新时间：{{time}}",
+ "noLatestTime": "暂未获取列表"
+ },
+ "resetAll": {
+ "conform": "确认重置当前模型的所有修改？重置后当前模型列表将会回到默认状态",
+ "success": "重置成功",
+ "title": "重置所有修改"
+ },
+ "search": "搜索模型...",
+ "searchResult": "搜索到 {{count}} 个模型",
+ "title": "模型列表",
+ "total": "共 {{count}} 个模型可用"
},
- "unlock": {
- "customRegion": "自定义服务区域",
- "description": "输入你的 AccessKey / SecretKey 即可开始会话。应用不会记录你的鉴权配置",
- "title": "使用自定义文心一言鉴权信息"
- }
+ "searchNotFound": "未找到搜索结果"
+ },
+ "sortModal": {
+ "success": "排序更新成功",
+ "title": "自定义排序",
+ "update": "更新"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "即将删除该 AI 服务商，删除后将无法找回，确认是否删除？",
+ "deleteSuccess": "删除成功",
+ "tooltip": "更新服务商基础配置",
+ "updateSuccess": "更新成功"
},
"zeroone": {
"title": "01.AI 零一万物"
diff --git a/locales/zh-CN/models.json b/locales/zh-CN/models.json
index af1c9eaf2a153..35f9cbd13bb61 100644
--- a/locales/zh-CN/models.json
+++ b/locales/zh-CN/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K 强调语义安全和责任导向，专为对内容安全有高度要求的应用场景设计，确保用户体验的准确性与稳健性。"
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 使用树搜索构建思维链，并引入了反思机制，使用强化学习训练，模型具备自我反思与纠错的能力。"
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro 是 360 公司推出的高级自然语言处理模型，具备卓越的文本生成和理解能力，尤其在生成与创作领域表现出色，能够处理复杂的语言转换和角色演绎任务。"
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1 使用树搜索构建思维链，并引入了反思机制，使用强化学习训练，模型具备自我反思与纠错的能力。"
+ },
"4.0Ultra": {
"description": "Spark Ultra 是星火大模型系列中最为强大的版本，在升级联网搜索链路同时，提升对文本内容的理解和总结能力。它是用于提升办公生产力和准确响应需求的全方位解决方案，是引领行业的智能产品。"
},
@@ -41,23 +47,35 @@
"Baichuan4-Turbo": {
"description": "模型能力国内第一，在知识百科、长文本、生成创作等中文任务上超越国外主流模型。还具备行业领先的多模态能力，多项权威评测基准表现优异。"
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "基于 Qwen2.5-Math-1.5B 的 DeepSeek-R1 蒸馏模型，通过强化学习与冷启动数据优化推理性能，开源模型刷新多任务标杆。"
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "基于 Qwen2.5-14B 的 DeepSeek-R1 蒸馏模型，通过强化学习与冷启动数据优化推理性能，开源模型刷新多任务标杆。"
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1 系列通过强化学习与冷启动数据优化推理性能，开源模型刷新多任务标杆，超越 OpenAI-o1-mini 水平。"
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "基于 Qwen2.5-Math-7B 的 DeepSeek-R1 蒸馏模型，通过强化学习与冷启动数据优化推理性能，开源模型刷新多任务标杆。"
+ },
"Doubao-lite-128k": {
- "description": "Doubao-lite 拥有极致的响应速度，更好的性价比，为客户不同场景提供更灵活的选择。支持128k上下文窗口的推理和精调。"
+ "description": "拥有极致的响应速度，更好的性价比，为客户不同场景提供更灵活的选择。支持 128k 上下文窗口的推理和精调。"
},
"Doubao-lite-32k": {
- "description": "Doubao-lite拥有极致的响应速度，更好的性价比，为客户不同场景提供更灵活的选择。支持32k上下文窗口的推理和精调。"
+ "description": "拥有极致的响应速度，更好的性价比，为客户不同场景提供更灵活的选择。支持 32k 上下文窗口的推理和精调。"
},
"Doubao-lite-4k": {
- "description": "Doubao-lite拥有极致的响应速度，更好的性价比，为客户不同场景提供更灵活的选择。支持4k上下文窗口的推理和精调。"
+ "description": "拥有极致的响应速度，更好的性价比，为客户不同场景提供更灵活的选择。支持 4k 上下文窗口的推理和精调。"
},
"Doubao-pro-128k": {
- "description": "效果最好的主力模型，适合处理复杂任务，在参考问答、总结摘要、创作、文本分类、角色扮演等场景都有很好的效果。支持128k上下文窗口的推理和精调。"
+ "description": "效果最好的主力模型，适合处理复杂任务，在参考问答、总结摘要、创作、文本分类、角色扮演等场景都有很好的效果。支持 128k 上下文窗口的推理和精调。"
},
"Doubao-pro-32k": {
- "description": "效果最好的主力模型，适合处理复杂任务，在参考问答、总结摘要、创作、文本分类、角色扮演等场景都有很好的效果。支持32k上下文窗口的推理和精调。"
+ "description": "效果最好的主力模型，适合处理复杂任务，在参考问答、总结摘要、创作、文本分类、角色扮演等场景都有很好的效果。支持 32k 上下文窗口的推理和精调。"
},
"Doubao-pro-4k": {
- "description": "效果最好的主力模型，适合处理复杂任务，在参考问答、总结摘要、创作、文本分类、角色扮演等场景都有很好的效果。支持4k上下文窗口的推理和精调。"
+ "description": "效果最好的主力模型，适合处理复杂任务，在参考问答、总结摘要、创作、文本分类、角色扮演等场景都有很好的效果。支持 4k 上下文窗口的推理和精调。"
},
"ERNIE-3.5-128K": {
"description": "百度自研的旗舰级大规模⼤语⾔模型，覆盖海量中英文语料，具有强大的通用能力，可满足绝大部分对话问答、创作生成、插件应用场景要求；支持自动对接百度搜索插件，保障问答信息时效。"
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "百度自研的旗舰级超大规模⼤语⾔模型，相较ERNIE 3.5实现了模型能力全面升级，广泛适用于各领域复杂任务场景；支持自动对接百度搜索插件，保障问答信息时效。"
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "百度自研的旗舰级超大规模⼤语⾔模型，综合效果表现出色，广泛适用于各领域复杂任务场景；支持自动对接百度搜索插件，保障问答信息时效。相较于ERNIE 4.0在性能表现上更优秀"
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "百度自研的旗舰级超大规模⼤语⾔模型，综合效果表现出色，广泛适用于各领域复杂任务场景；支持自动对接百度搜索插件，保障问答信息时效。相较于ERNIE 4.0在性能表现上更优秀"
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct 是阿里云发布的最新大语言模型系列之一。该 7B 模型在编码和数学等领域具有显著改进的能力。该模型还提供了多语言支持，覆盖超过 29 种语言，包括中文、英文等。模型在指令跟随、理解结构化数据以及生成结构化输出（尤其是 JSON）方面都有显著提升"
},
+ "MiniMax-Text-01": {
+ "description": "在 MiniMax-01系列模型中，我们做了大胆创新：首次大规模实现线性注意力机制，传统 Transformer架构不再是唯一的选择。这个模型的参数量高达4560亿，其中单次激活459亿。模型综合性能比肩海外顶尖模型，同时能够高效处理全球最长400万token的上下文，是GPT-4o的32倍，Claude-3.5-Sonnet的20倍。"
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO 是一款高度灵活的多模型合并，旨在提供卓越的创造性体验。"
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2-26B 是 InternVL 2.0 系列多模态大语言模型中的一员。该模型由 InternViT-6B-448px-V1-5 视觉模型、MLP 投影层和 internlm2-chat-20b 语言模型组成。它在各种视觉语言任务上展现出了卓越的性能，包括文档和图表理解、场景文本理解、OCR、科学和数学问题解决等。InternVL2-26B 使用 8K 上下文窗口训练，能够处理长文本、多图像和视频输入，显著提升了模型在这些任务上的处理能力"
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2-Llama3-76B 是 InternVL 2.0 系列中的大规模多模态模型。它由 InternViT-6B-448px-V1-5 视觉模型、MLP 投影层和 Hermes-2-Theta-Llama-3-70B 语言模型组成。该模型在各种视觉语言任务上表现出色，包括文档和图表理解、信息图表问答、场景文本理解和 OCR 任务等。InternVL2-Llama3-76B 使用 8K 上下文窗口训练，能够处理长文本、多图像和视频输入，显著提升了模型在这些任务上的处理能力，在多项基准测试中达到或接近最先进的商业模型水平"
- },
"Phi-3-medium-128k-instruct": {
"description": "相同的Phi-3-medium模型，但具有更大的上下文大小，适用于RAG或少量提示。"
},
@@ -167,14 +182,26 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat 是智谱 AI 推出的 GLM-4 系列预训练模型中的开源版本。该模型在语义、数学、推理、代码和知识等多个方面表现出色。除了支持多轮对话外，GLM-4-9B-Chat 还具备网页浏览、代码执行、自定义工具调用（Function Call）和长文本推理等高级功能。模型支持 26 种语言，包括中文、英文、日语、韩语和德语等。在多项基准测试中，GLM-4-9B-Chat 展现了优秀的性能，如 AlignBench-v2、MT-Bench、MMLU 和 C-Eval 等。该模型支持最大 128K 的上下文长度，适用于学术研究和商业应用"
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 是一款强化学习（RL）驱动的推理模型，解决了模型中的重复性和可读性问题。在 RL 之前，DeepSeek-R1 引入了冷启动数据，进一步优化了推理性能。它在数学、代码和推理任务中与 OpenAI-o1 表现相当，并且通过精心设计的训练方法，提升了整体效果。"
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 是一款拥有 6710 亿参数的混合专家（MoE）语言模型，采用多头潜在注意力（MLA）和 DeepSeekMoE 架构，结合无辅助损失的负载平衡策略，优化推理和训练效率。通过在 14.8 万亿高质量tokens上预训练，并进行监督微调和强化学习，DeepSeek-V3 在性能上超越其他开源模型，接近领先闭源模型。"
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma 是 Google 开发的轻量级、最先进的开放模型系列之一。它是一个仅解码器的大型语言模型，支持英语，提供开放权重、预训练变体和指令微调变体。Gemma 模型适用于各种文本生成任务，包括问答、摘要和推理。该 9B 模型是通过 8 万亿个 tokens 训练而成。其相对较小的规模使其可以在资源有限的环境中部署，如笔记本电脑、台式机或您自己的云基础设施，从而使更多人能够访问最先进的 AI 模型并促进创新"
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 是由 Meta 开发的多语言大型语言模型家族，包括 8B、70B 和 405B 三种参数规模的预训练和指令微调变体。该 8B 指令微调模型针对多语言对话场景进行了优化，在多项行业基准测试中表现优异。模型训练使用了超过 15 万亿个 tokens 的公开数据，并采用了监督微调和人类反馈强化学习等技术来提升模型的有用性和安全性。Llama 3.1 支持文本生成和代码生成，知识截止日期为 2023 年 12 月"
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview 是一款独具创新的自然语言处理模型，能够高效处理复杂的对话生成与上下文理解任务。"
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview 是由 Qwen 团队开发的专注于视觉推理能力的研究型模型，其在复杂场景理解和解决视觉相关的数学问题方面具有独特优势。"
+ },
"Qwen/QwQ-32B-Preview": {
- "description": "QwQ-32B-Preview是Qwen 最新的实验性研究模型，专注于提升AI推理能力。通过探索语言混合、递归推理等复杂机制，主要优势包括强大的推理分析能力、数学和编程能力。与此同时，也存在语言切换问题、推理循环、安全性考虑、其他能力方面的差异。"
+ "description": "Qwen QwQ 是由 Qwen 团队开发的实验研究模型，专注于提升AI推理能力。"
},
"Qwen/Qwen2-1.5B-Instruct": {
"description": "Qwen2-1.5B-Instruct 是 Qwen2 系列中的指令微调大语言模型，参数规模为 1.5B。该模型基于 Transformer 架构，采用了 SwiGLU 激活函数、注意力 QKV 偏置和组查询注意力等技术。它在语言理解、生成、多语言能力、编码、数学和推理等多个基准测试中表现出色，超越了大多数开源模型。与 Qwen1.5-1.8B-Chat 相比，Qwen2-1.5B-Instruct 在 MMLU、HumanEval、GSM8K、C-Eval 和 IFEval 等测试中均显示出显著的性能提升，尽管参数量略少"
@@ -183,7 +210,7 @@
"description": "Qwen 2 Instruct (72B) 为企业级应用提供精准的指令理解和响应。"
},
"Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct 是 Qwen2 系列中的指令微调大语言模型，参数规模为 72B。该模型基于 Transformer 架构，采用了 SwiGLU 激活函数、注意力 QKV 偏置和组查询注意力等技术。它能够处理大规模输入。该模型在语言理解、生成、多语言能力、编码、数学和推理等多个基准测试中表现出色，超越了大多数开源模型，并在某些任务上展现出与专有模型相当的竞争力"
+ "description": "Qwen2-7B-Instruct 是 Qwen2 系列中的指令微调大语言模型，参数规模为 7B。该模型基于 Transformer 架构，采用了 SwiGLU 激活函数、注意力 QKV 偏置和组查询注意力等技术。它能够处理大规模输入。该模型在语言理解、生成、多语言能力、编码、数学和推理等多个基准测试中表现出色，超越了大多数开源模型，并在某些任务上展现出与专有模型相当的竞争力。Qwen2-7B-Instruct 在多项评测中均优于 Qwen1.5-7B-Chat，显示出显著的性能提升"
},
"Qwen/Qwen2-VL-72B-Instruct": {
"description": "Qwen2-VL 是 Qwen-VL 模型的最新迭代版本，在视觉理解基准测试中达到了最先进的性能，包括 MathVista、DocVQA、RealWorldQA 和 MTVQA 等。Qwen2-VL 能够理解超过 20 分钟的视频，用于高质量的基于视频的问答、对话和内容创作。它还具备复杂推理和决策能力，可以与移动设备、机器人等集成，基于视觉环境和文本指令进行自动操作。除了英语和中文，Qwen2-VL 现在还支持理解图像中不同语言的文本，包括大多数欧洲语言、日语、韩语、阿拉伯语和越南语等"
@@ -210,20 +237,20 @@
"description": "Qwen2.5 是全新的大型语言模型系列，旨在优化指令式任务的处理。"
},
"Qwen/Qwen2.5-Coder-32B-Instruct": {
- "description": "Qwen2.5-Coder-32B-Instruct 是基于 Qwen2.5 开发的代码特定大语言模型。该模型通过 5.5 万亿 tokens 的训练，在代码生成、代码推理和代码修复方面都取得了显著提升。它是当前最先进的开源代码语言模型，编码能力可与 GPT-4 相媲美。模型不仅增强了编码能力，还保持了在数学和通用能力方面的优势，并支持长文本处理"
+ "description": "Qwen2.5 Coder 32B Instruct 是阿里云发布的代码特定大语言模型系列的最新版本。该模型在 Qwen2.5 的基础上，通过 5.5 万亿个 tokens 的训练，显著提升了代码生成、推理和修复能力。它不仅增强了编码能力，还保持了数学和通用能力的优势。模型为代码智能体等实际应用提供了更全面的基础"
},
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct 是阿里云发布的代码特定大语言模型系列的最新版本。该模型在 Qwen2.5 的基础上，通过 5.5 万亿个 tokens 的训练，显著提升了代码生成、推理和修复能力。它不仅增强了编码能力，还保持了数学和通用能力的优势。模型为代码智能体等实际应用提供了更全面的基础"
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math-72B 是阿里云发布的 Qwen2.5-Math 系列数学大语言模型之一。该模型支持使用思维链（CoT）和工具集成推理（TIR）方法解决中文和英文数学问题。相比前代 Qwen2-Math 系列，Qwen2.5-Math 系列在中英文数学基准测试中取得了显著的性能提升。该模型在处理精确计算、符号操作和算法操作方面表现出色，尤其适合解决复杂的数学和算法推理任务"
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 是 Qwen 模型的最新系列，支持 128k 上下文，对比当前最优的开源模型，Qwen2-72B 在自然语言理解、知识、代码、数学及多语言等多项能力上均显著超越当前领先的模型。"
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 是 Qwen 模型的最新系列，能够超越同等规模的最优开源模型甚至更大规模的模型，Qwen2 7B 在多个评测上取得显著的优势，尤其是代码及中文理解上。"
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B是一款强大的视觉语言模型，支持图像与文本的多模态处理，能够精确识别图像内容并生成相关描述或回答。"
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct 是一款 140 亿参数的大语言模型，性能表现优秀，优化中文和多语言场景，支持智能问答、内容生成等应用。"
},
@@ -231,11 +258,14 @@
"description": "Qwen2.5-32B-Instruct 是一款 320 亿参数的大语言模型，性能表现均衡，优化中文和多语言场景，支持智能问答、内容生成等应用。"
},
"Qwen2.5-72B-Instruct": {
- "description": "Qwen2.5-72B-Instruct 支持 16k 上下文, 生成长文本超过 8K 。支持 function call 与外部系统无缝交互，极大提升了灵活性和扩展性。模型知识明显增加，并且大大提高了编码和数学能力, 多语言支持超过 29 种"
+ "description": "Qwen2.5-72B-Instruct 支持 16k 上下文, 生成长文本超过 8K 。支持 function call 与外部系统无缝交互，极大提升了灵活性和扩展性。模型知识明显增加，并且大大提高了编码和数学能力, 多语言支持超过 29 种"
},
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct 是一款 70 亿参数的大语言模型，支持 function call 与外部系统无缝交互，极大提升了灵活性和扩展性。优化中文和多语言场景，支持智能问答、内容生成等应用。"
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct 是一款基于大规模预训练的编程指令模型，具备强大的代码理解和生成能力，能够高效地处理各种编程任务，特别适合智能代码编写、自动化脚本生成和编程问题解答。"
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct 是一款专为代码生成、代码理解和高效开发场景设计的大型语言模型，采用了业界领先的32B参数规模，能够满足多样化的编程需求。"
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "TeleMM多模态大模型是由中国电信自主研发的多模态理解大模型，能够处理文本、图像等多种模态输入，支持图像理解、图表分析等功能，为用户提供跨模态的理解服务。模型能够与用户进行多模态交互，准确理解输入内容，回答问题、协助创作，并高效提供多模态信息和灵感支持。在细粒度感知，逻辑推理等多模态任务上有出色表现"
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large 是业界最大的开源 Transformer 架构 MoE 模型，拥有 3890 亿总参数量和 520 亿激活参数量。"
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct 是 Qwen2 系列中的指令微调大语言模型，参数规模为 72B。该模型基于 Transformer 架构，采用了 SwiGLU 激活函数、注意力 QKV 偏置和组查询注意力等技术。它能够处理大规模输入。该模型在语言理解、生成、多语言能力、编码、数学和推理等多个基准测试中表现出色，超越了大多数开源模型，并在某些任务上展现出与专有模型相当的竞争力"
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct 是阿里云发布的最新大语言模型系列之一。该 72B 模型在编码和数学等领域具有显著改进的能力。该模型还提供了多语言支持，覆盖超过 29 种语言，包括中文、英文等。模型在指令跟随、理解结构化数据以及生成结构化输出（尤其是 JSON）方面都有显著提升"
},
@@ -317,80 +341,74 @@
"abab6.5t-chat": {
"description": "针对中文人设对话场景优化，提供流畅且符合中文表达习惯的对话生成能力。"
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Fireworks 开源函数调用模型，提供卓越的指令执行能力和开放可定制的特性。"
+ "abab7-chat-preview": {
+ "description": "相对于abab6.5系列模型在长文、数学、写作等能力有大幅度提升。"
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Fireworks 公司最新推出的 Firefunction-v2 是一款性能卓越的函数调用模型，基于 Llama-3 开发，并通过大量优化，特别适用于函数调用、对话及指令跟随等场景。"
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 是一款最先进的大型语言模型，经过强化学习和冷启动数据的优化，具有出色的推理、数学和编程性能。"
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b 是一款视觉语言模型，可以同时接收图像和文本输入，经过高质量数据训练，适合多模态任务。"
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Deepseek 提供的强大 Mixture-of-Experts (MoE) 语言模型，总参数量为 671B，每个标记激活 37B 参数。"
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
- "description": "Llama 3 70B 指令模型，专为多语言对话和自然语言理解优化，性能优于多数竞争模型。"
- },
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Llama 3 70B 指令模型（HF 版本），与官方实现结果保持一致，适合高质量的指令跟随任务。"
+ "description": "Meta 开发并发布了 Meta Llama 3 系列大语言模型（LLM），该系列包含 8B 和 70B 参数规模的预训练和指令微调生成文本模型。Llama 3 指令微调模型专为对话应用场景优化，并在常见的行业基准测试中优于许多现有的开源聊天模型。"
},
"accounts/fireworks/models/llama-v3-8b-instruct": {
- "description": "Llama 3 8B 指令模型，优化用于对话及多语言任务，表现卓越且高效。"
+ "description": "Meta 开发并发布了 Meta Llama 3 系列大语言模型（LLM），这是一个包含 8B 和 70B 参数规模的预训练和指令微调生成文本模型的集合。Llama 3 指令微调模型专为对话应用场景优化，并在常见的行业基准测试中优于许多现有的开源聊天模型。"
},
"accounts/fireworks/models/llama-v3-8b-instruct-hf": {
- "description": "Llama 3 8B 指令模型（HF 版本），与官方实现结果一致，具备高度一致性和跨平台兼容性。"
+ "description": "Meta Llama 3 指令微调模型专为对话应用场景优化，并在常见的行业基准测试中优于许多现有的开源聊天模型。Llama 3 8B Instruct（HF 版本）是 Llama 3 8B Instruct 的原始 FP16 版本，其结果应与官方 Hugging Face 实现一致。"
},
"accounts/fireworks/models/llama-v3p1-405b-instruct": {
- "description": "Llama 3.1 405B 指令模型，具备超大规模参数，适合复杂任务和高负载场景下的指令跟随。"
+ "description": "Meta Llama 3.1 系列是多语言大语言模型（LLM）集合，包含 8B、70B 和 405B 参数规模的预训练和指令微调生成模型。Llama 3.1 指令微调文本模型（8B、70B、405B）专为多语言对话场景优化，在常见的行业基准测试中优于许多现有的开源和闭源聊天模型。405B 是 Llama 3.1 家族中能力最强的模型。该模型采用 FP8 进行推理，与参考实现高度匹配。"
},
"accounts/fireworks/models/llama-v3p1-70b-instruct": {
- "description": "Llama 3.1 70B 指令模型，提供卓越的自然语言理解和生成能力，是对话及分析任务的理想选择。"
+ "description": "Meta Llama 3.1 系列是多语言大语言模型（LLM）集合，包含 8B、70B 和 405B 三种参数规模的预训练和指令微调生成模型。Llama 3.1 指令微调文本模型（8B、70B、405B）专为多语言对话应用优化，并在常见的行业基准测试中优于许多现有的开源和闭源聊天模型。"
},
"accounts/fireworks/models/llama-v3p1-8b-instruct": {
- "description": "Llama 3.1 8B 指令模型，专为多语言对话优化，能够在常见行业基准上超越多数开源及闭源模型。"
+ "description": "Meta Llama 3.1 系列是多语言大语言模型（LLM）集合，包含 8B、70B 和 405B 三种参数规模的预训练和指令微调生成模型。Llama 3.1 指令微调文本模型（8B、70B、405B）专为多语言对话应用优化，并在常见的行业基准测试中优于许多现有的开源和闭源聊天模型。"
},
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
- "description": "Meta的11B参数指令调整图像推理模型。该模型针对视觉识别、图像推理、图像描述和回答关于图像的一般性问题进行了优化。该模型能够理解视觉数据，如图表和图形，并通过生成文本描述图像细节来弥合视觉与语言之间的差距。"
- },
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Llama 3.2 1B 指令模型是Meta推出的一款轻量级多语言模型。该模型旨在提高效率，与更大型的模型相比，在延迟和成本方面提供了显著的改进。该模型的示例用例包括检索和摘要。"
+ "description": "Meta 推出的指令微调图像推理模型，拥有 110 亿参数。该模型针对视觉识别、图像推理、图片字幕生成以及图片相关的常规问答进行了优化。它能够理解视觉数据，如图表和图形，并通过生成文本描述图像细节，弥合视觉与语言之间的鸿沟。"
},
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
- "description": "Llama 3.2 3B 指令模型是Meta推出的一款轻量级多语言模型。该模型旨在提高效率，与更大型的模型相比，在延迟和成本方面提供了显著的改进。该模型的示例用例包括查询和提示重写以及写作辅助。"
+ "description": "Llama 3.2 3B Instruct 是 Meta 推出的轻量级多语言模型。该模型专为高效运行而设计，相较于更大型的模型，具有显著的延迟和成本优势。其典型应用场景包括查询和提示重写，以及写作辅助。"
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
- "description": "Meta的90B参数指令调整图像推理模型。该模型针对视觉识别、图像推理、图像描述和回答关于图像的一般性问题进行了优化。该模型能够理解视觉数据，如图表和图形，并通过生成文本描述图像细节来弥合视觉与语言之间的差距。"
+ "description": "Meta 推出的指令微调图像推理模型，拥有 900 亿参数。该模型针对视觉识别、图像推理、图片字幕生成以及图片相关的常规问答进行了优化。它能够理解视觉数据，如图表和图形，并通过生成文本描述图像细节，弥合视觉与语言之间的鸿沟。注意：该模型目前作为无服务器模型进行实验性提供。如果用于生产环境，请注意 Fireworks 可能会在短时间内取消部署该模型。"
+ },
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct 是 Llama 3.1 70B 的 12 月更新版本。该模型在 Llama 3.1 70B（于 2024 年 7 月发布）的基础上进行了改进，增强了工具调用、多语言文本支持、数学和编程能力。该模型在推理、数学和指令遵循方面达到了行业领先水平，并且能够提供与 3.1 405B 相似的性能，同时在速度和成本上具有显著优势。"
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "24B 参数模型，具备与更大型模型相当的最先进能力。"
},
"accounts/fireworks/models/mixtral-8x22b-instruct": {
- "description": "Mixtral MoE 8x22B 指令模型，大规模参数和多专家架构，全方位支持复杂任务的高效处理。"
+ "description": "Mixtral MoE 8x22B Instruct v0.1 是 Mixtral MoE 8x22B v0.1 的指令微调版本，已启用聊天完成功能 API。"
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
- "description": "Mixtral MoE 8x7B 指令模型，多专家架构提供高效的指令跟随及执行。"
- },
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Mixtral MoE 8x7B 指令模型（HF 版本），性能与官方实现一致，适合多种高效任务场景。"
+ "description": "Mixtral MoE 8x7B Instruct 是 Mixtral MoE 8x7B 的指令微调版本，已启用聊天完成功能 API。"
},
"accounts/fireworks/models/mythomax-l2-13b": {
- "description": "MythoMax L2 13B 模型，结合新颖的合并技术，擅长叙事和角色扮演。"
+ "description": "MythoMix 的改进版，可能是其更为完善的变体，是 MythoLogic-L2 和 Huginn 的合并，采用了高度实验性的张量类型合并技术。由于其独特的性质，该模型在讲故事和角色扮演方面表现出色。"
},
"accounts/fireworks/models/phi-3-vision-128k-instruct": {
- "description": "Phi-3-Vision-128K-Instruct 是一个轻量级的、最先进的开放多模态模型，它基于包括合成数据和经过筛选的公开网站在内的数据集构建，专注于非常高质量、推理密集型的数据，这些数据既包括文本也包括视觉。该模型属于 Phi-3 模型系列，其多模态版本支持 128K 的上下文长度（以标记为单位）。该模型经过严格的增强过程，结合了监督微调和直接偏好优化，以确保精确遵循指令和强大的安全措施。"
+ "description": "Phi-3-Vision-128K-Instruct 是一个轻量级的、最先进的开放多模态模型，基于包括合成数据和筛选后的公开网站数据集构建，重点关注文本和视觉方面的高质量、推理密集型数据。该模型属于 Phi-3 模型家族，其多模态版本支持 128K 上下文长度（以标记为单位）。该模型经过严格的增强过程，包括监督微调和直接偏好优化，以确保精确的指令遵循和强大的安全措施。"
},
"accounts/fireworks/models/qwen-qwq-32b-preview": {
- "description": "QwQ模型是由 Qwen 团队开发的实验性研究模型，专注于增强 AI 推理能力。"
+ "description": "Qwen QwQ 模型专注于推动 AI 推理，并展示了开放模型在推理能力上与闭源前沿模型匹敌的力量。QwQ-32B-Preview 是一个实验性发布版本，在 GPQA、AIME、MATH-500 和 LiveCodeBench 基准测试中，在分析和推理能力上可与 o1 相媲美，并超越 GPT-4o 和 Claude 3.5 Sonnet。注意：该模型目前作为无服务器模型进行实验性提供。如果用于生产环境，请注意 Fireworks 可能会在短时间内取消部署该模型。"
+ },
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "Qwen-VL 模型的 72B 版本是阿里巴巴最新迭代的成果，代表了近一年的创新。"
},
"accounts/fireworks/models/qwen2p5-72b-instruct": {
- "description": "Qwen2.5 是由阿里云 Qwen 团队开发的一系列仅包含解码器的语言模型。这些模型提供不同的大小，包括 0.5B、1.5B、3B、7B、14B、32B 和 72B，并且有基础版（base）和指令版（instruct）两种变体。"
+ "description": "Qwen2.5 是由 Qwen 团队和阿里云开发的一系列仅解码语言模型，提供 0.5B、1.5B、3B、7B、14B、32B 和 72B 不同参数规模，并包含基础版和指令微调版。"
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
- "description": "Qwen2.5 Coder 32B Instruct 是阿里云发布的代码特定大语言模型系列的最新版本。该模型在 Qwen2.5 的基础上，通过 5.5 万亿个 tokens 的训练，显著提升了代码生成、推理和修复能力。它不仅增强了编码能力，还保持了数学和通用能力的优势。模型为代码智能体等实际应用提供了更全面的基础"
- },
- "accounts/fireworks/models/starcoder-16b": {
- "description": "StarCoder 15.5B 模型，支持高级编程任务，多语言能力增强，适合复杂代码生成和理解。"
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "StarCoder 7B 模型，针对80多种编程语言训练，拥有出色的编程填充能力和语境理解。"
+ "description": "Qwen2.5-Coder 是最新一代专为代码设计的 Qwen 大型语言模型（前称为 CodeQwen）。注意：该模型目前作为无服务器模型进行实验性提供。如果用于生产环境，请注意 Fireworks 可能会在短时间内取消部署该模型。"
},
"accounts/yi-01-ai/models/yi-large": {
- "description": "Yi-Large 模型，具备卓越的多语言处理能力，可用于各类语言生成和理解任务。"
+ "description": "Yi-Large 是顶尖的大型语言模型之一，在 LMSYS 基准测试排行榜上，其表现仅次于 GPT-4、Gemini 1.5 Pro 和 Claude 3 Opus。它在多语言能力方面表现卓越，特别是在西班牙语、中文、日语、德语和法语方面。Yi-Large 还具有用户友好性，采用与 OpenAI 相同的 API 定义，便于集成。"
},
"ai21-jamba-1.5-large": {
"description": "一个398B参数（94B活跃）的多语言模型，提供256K长上下文窗口、函数调用、结构化输出和基于事实的生成。"
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus 是 Anthropic 用于处理高度复杂任务的最强大模型。它在性能、智能、流畅性和理解力方面表现卓越。"
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku 是 Anthropic 最快的下一代模型。与 Claude 3 Haiku 相比，Claude 3.5 Haiku 在各项技能上都有所提升，并在许多智力基准测试中超越了上一代最大的模型 Claude 3 Opus。"
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet 提供了超越 Opus 的能力和比 Sonnet 更快的速度，同时保持与 Sonnet 相同的价格。Sonnet 特别擅长编程、数据科学、视觉处理、代理任务。"
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet 在智能和速度方面为企业工作负载提供了理想的平衡。它以更低的价格提供最大效用，可靠且适合大规模部署。"
},
- "code-raccoon-v1": {
- "description": "代码小浣熊是基于商汤大语言模型的软件智能研发助手，覆盖软件需求分析、架构设计、代码编写、软件测试等环节，满足用户代码编写、编程学习等各类需求。代码小浣熊支持 Python、Java、JavaScript、C++、Go、SQL 等 90+主流编程语言和 VS Code、IntelliJ IDEA 等主流 IDE。在实际应用中，代码小浣熊可帮助开发者提升编程效率超 50%。"
- },
"codegeex-4": {
"description": "CodeGeeX-4 是强大的AI编程助手，支持多种编程语言的智能问答与代码补全，提升开发效率。"
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ 是一款高性能的大型语言模型，专为真实企业场景和复杂应用而设计。"
},
+ "dall-e-2": {
+ "description": "第二代 DALL·E 模型，支持更真实、准确的图像生成，分辨率是第一代的4倍"
+ },
+ "dall-e-3": {
+ "description": "最新的 DALL·E 模型，于2023年11月发布。支持更真实、准确的图像生成，具有更强的细节表现力"
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct 提供高可靠性的指令处理能力，支持多行业应用。"
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 是一个强大、经济高效的混合专家（MoE）语言模型。它在 8.1 万亿个 token 的高质量语料库上进行了预训练，并通过监督微调（SFT）和强化学习（RL）进一步提升了模型能力。与 DeepSeek 67B 相比， DeepSeek-V2 在性能更强的同时，节省了 42.5% 的训练成本，减少了 93.3% 的 KV 缓存，并将最大生成吞吐量提高到了 5.76 倍。该模型支持 128k 的上下文长度，在标准基准测试和开放式生成评估中都表现出色"
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 系列通过强化学习与冷启动数据优化推理性能，开源模型刷新多任务标杆，超越 OpenAI-o1-mini 水平。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "DeepSeek-R1 蒸馏模型，通过强化学习与冷启动数据优化推理性能，开源模型刷新多任务标杆。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B 是基于 Llama-3.1-8B 开发的蒸馏模型。该模型使用 DeepSeek-R1 生成的样本进行微调，展现出优秀的推理能力。在多个基准测试中表现不俗，其中在 MATH-500 上达到了 89.1% 的准确率，在 AIME 2024 上达到了 50.4% 的通过率，在 CodeForces 上获得了 1205 的评分，作为 8B 规模的模型展示了较强的数学和编程能力。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "DeepSeek-R1 蒸馏模型，通过强化学习与冷启动数据优化推理性能，开源模型刷新多任务标杆。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "DeepSeek-R1 蒸馏模型，通过强化学习与冷启动数据优化推理性能，开源模型刷新多任务标杆。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B 是基于 Qwen2.5-32B 通过知识蒸馏得到的模型。该模型使用 DeepSeek-R1 生成的 80 万个精选样本进行微调，在数学、编程和推理等多个领域展现出卓越的性能。在 AIME 2024、MATH-500、GPQA Diamond 等多个基准测试中都取得了优异成绩，其中在 MATH-500 上达到了 94.3% 的准确率，展现出强大的数学推理能力。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B 是基于 Qwen2.5-Math-7B 通过知识蒸馏得到的模型。该模型使用 DeepSeek-R1 生成的 80 万个精选样本进行微调，展现出优秀的推理能力。在多个基准测试中表现出色，其中在 MATH-500 上达到了 92.8% 的准确率，在 AIME 2024 上达到了 55.5% 的通过率，在 CodeForces 上获得了 1189 的评分，作为 7B 规模的模型展示了较强的数学和编程能力。"
},
"deepseek-ai/DeepSeek-V2.5": {
- "description": "DeepSeek-V2.5 是 DeepSeek-V2-Chat 和 DeepSeek-Coder-V2-Instruct 的升级版本，集成了两个先前版本的通用和编码能力。该模型在多个方面进行了优化，包括写作和指令跟随能力，更好地与人类偏好保持一致。DeepSeek-V2.5 在各种评估基准上都取得了显著的提升，如 AlpacaEval 2.0、ArenaHard、AlignBench 和 MT-Bench 等"
+ "description": "DeepSeek-V2.5 是 DeepSeek-V2-Chat 和 DeepSeek-Coder-V2-Instruct 的升级版本，集成了两个先前版本的通用和编码能力。该模型在多个方面进行了优化，包括写作和指令跟随能力，更好地与人类偏好保持一致。DeepSeek-V2.5 在各种评估基准上都取得了显著的提升，如 AlpacaEval 2.0、ArenaHard、AlignBench 和 MT-Bench 等。"
+ },
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 是一款拥有 6710 亿参数的混合专家（MoE）语言模型，采用多头潜在注意力（MLA）和 DeepSeekMoE 架构，结合无辅助损失的负载平衡策略，优化推理和训练效率。通过在 14.8 万亿高质量tokens上预训练，并进行监督微调和强化学习，DeepSeek-V3 在性能上超越其他开源模型，接近领先闭源模型。"
},
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek LLM Chat (67B) 是创新的 AI 模型 提供深度语言理解和互动能力。"
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "最先进的高效 LLM，擅长推理、数学和编程。"
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 是一个基于 DeepSeekMoE-27B 开发的混合专家（MoE）视觉语言模型，采用稀疏激活的 MoE 架构，在仅激活 4.5B 参数的情况下实现了卓越性能。该模型在视觉问答、光学字符识别、文档/表格/图表理解和视觉定位等多个任务中表现优异。"
+ },
"deepseek-chat": {
"description": "融合通用与代码能力的全新开源模型, 不仅保留了原有 Chat 模型的通用对话能力和 Coder 模型的强大代码处理能力，还更好地对齐了人类偏好。此外，DeepSeek-V2.5 在写作任务、指令跟随等多个方面也实现了大幅提升。"
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 是开源的混合专家代码模型，在代码任务方面表现优异，与 GPT4-Turbo 相媲美。"
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 在后训练阶段大规模使用了强化学习技术，在仅有极少标注数据的情况下，极大提升了模型推理能力。在数学、代码、自然语言推理等任务上，性能比肩 OpenAI o1 正式版。"
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek-R1-Distill-Llama-70B 是一个基于 Llama-3.3-70B-Instruct 的蒸馏大型语言模型，使用了 DeepSeek R1 的输出。"
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "DeepSeek-R1-Distill-Llama-8B 是一个基于 Llama-3.1-8B 的蒸馏大型语言模型，使用了 DeepSeek R1 的输出。"
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "DeepSeek-R1-Distill-Qwen-1.5B 是一个基于 Qwen2.5-Math-1.5B 的蒸馏大型语言模型，使用了 DeepSeek R1 的输出。"
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "DeepSeek-R1-Distill-Qwen-14B 是一个基于 Qwen2.5-14B 的蒸馏大型语言模型，使用了 DeepSeek R1 的输出。"
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B 是一个基于 Qwen2.5-32B 的蒸馏大型语言模型，使用了 DeepSeek R1 的输出。"
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B 是一个基于 Qwen2.5-Math-7B 的蒸馏大型语言模型，使用了 DeepSeek R1 的输出。"
+ },
+ "deepseek-reasoner": {
+ "description": "DeepSeek 推出的推理模型。在输出最终回答之前，模型会先输出一段思维链内容，以提升最终答案的准确性。"
+ },
"deepseek-v2": {
"description": "DeepSeek V2 是高效的 Mixture-of-Experts 语言模型，适用于经济高效的处理需求。"
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B 是 DeepSeek 的设计代码模型，提供强大的代码生成能力。"
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 为杭州深度求索人工智能基础技术研究有限公司自研的 MoE 模型，其多项评测成绩突出，在主流榜单中位列开源模型榜首。V3 相比 V2.5 模型生成速度实现 3 倍提升，为用户带来更加迅速流畅的使用体验。"
+ },
"deepseek/deepseek-chat": {
"description": "融合通用与代码能力的全新开源模型, 不仅保留了原有 Chat 模型的通用对话能力和 Coder 模型的强大代码处理能力，还更好地对齐了人类偏好。此外，DeepSeek-V2.5 在写作任务、指令跟随等多个方面也实现了大幅提升。"
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 在仅有极少标注数据的情况下，极大提升了模型推理能力。在输出最终回答之前，模型会先输出一段思维链内容，以提升最终答案的准确性。"
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 在仅有极少标注数据的情况下，极大提升了模型推理能力。在输出最终回答之前，模型会先输出一段思维链内容，以提升最终答案的准确性。"
+ },
"emohaa": {
"description": "Emohaa 是心理模型，具备专业咨询能力，帮助用户理解情感问题。"
},
+ "ernie-3.5-128k": {
+ "description": "百度自研的旗舰级大规模⼤语⾔模型，覆盖海量中英文语料，具有强大的通用能力，可满足绝大部分对话问答、创作生成、插件应用场景要求；支持自动对接百度搜索插件，保障问答信息时效。"
+ },
+ "ernie-3.5-8k": {
+ "description": "百度自研的旗舰级大规模⼤语⾔模型，覆盖海量中英文语料，具有强大的通用能力，可满足绝大部分对话问答、创作生成、插件应用场景要求；支持自动对接百度搜索插件，保障问答信息时效。"
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "百度自研的旗舰级大规模⼤语⾔模型，覆盖海量中英文语料，具有强大的通用能力，可满足绝大部分对话问答、创作生成、插件应用场景要求；支持自动对接百度搜索插件，保障问答信息时效。"
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "百度自研的旗舰级超大规模⼤语⾔模型，相较ERNIE 3.5实现了模型能力全面升级，广泛适用于各领域复杂任务场景；支持自动对接百度搜索插件，保障问答信息时效。"
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "百度自研的旗舰级超大规模⼤语⾔模型，相较ERNIE 3.5实现了模型能力全面升级，广泛适用于各领域复杂任务场景；支持自动对接百度搜索插件，保障问答信息时效。"
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "百度自研的旗舰级超大规模⼤语⾔模型，综合效果表现出色，广泛适用于各领域复杂任务场景；支持自动对接百度搜索插件，保障问答信息时效。相较于ERNIE 4.0在性能表现上更优秀"
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "百度自研的旗舰级超大规模⼤语⾔模型，综合效果表现出色，广泛适用于各领域复杂任务场景；支持自动对接百度搜索插件，保障问答信息时效。相较于ERNIE 4.0在性能表现上更优秀"
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "百度自研的旗舰级超大规模⼤语⾔模型，综合效果表现出色，广泛适用于各领域复杂任务场景；支持自动对接百度搜索插件，保障问答信息时效。相较于ERNIE 4.0在性能表现上更优秀"
+ },
+ "ernie-char-8k": {
+ "description": "百度自研的垂直场景大语言模型，适合游戏NPC、客服对话、对话角色扮演等应用场景，人设风格更为鲜明、一致，指令遵循能力更强，推理性能更优。"
+ },
+ "ernie-char-fiction-8k": {
+ "description": "百度自研的垂直场景大语言模型，适合游戏NPC、客服对话、对话角色扮演等应用场景，人设风格更为鲜明、一致，指令遵循能力更强，推理性能更优。"
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite是百度自研的轻量级大语言模型，兼顾优异的模型效果与推理性能，适合低算力AI加速卡推理使用。"
+ },
+ "ernie-lite-pro-128k": {
+ "description": "百度自研的轻量级大语言模型，兼顾优异的模型效果与推理性能，效果比ERNIE Lite更优，适合低算力AI加速卡推理使用。"
+ },
+ "ernie-novel-8k": {
+ "description": "百度自研通用大语言模型，在小说续写能力上有明显优势，也可用在短剧、电影等场景。"
+ },
+ "ernie-speed-128k": {
+ "description": "百度2024年最新发布的自研高性能大语言模型，通用能力优异，适合作为基座模型进行精调，更好地处理特定场景问题，同时具备极佳的推理性能。"
+ },
+ "ernie-speed-pro-128k": {
+ "description": "百度2024年最新发布的自研高性能大语言模型，通用能力优异，效果比ERNIE Speed更优，适合作为基座模型进行精调，更好地处理特定场景问题，同时具备极佳的推理性能。"
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny是百度自研的超高性能大语言模型，部署与精调成本在文心系列模型中最低。"
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Tuning) 提供稳定并可调优的性能，是复杂任务解决方案的理想选择。"
},
@@ -599,20 +731,29 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro 支持高达200万个tokens，是中型多模态模型的理想选择，适用于复杂任务的多方面支持。"
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash 提供下一代功能和改进，包括卓越的速度、原生工具使用、多模态生成和1M令牌上下文窗口。"
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash 提供下一代功能和改进，包括卓越的速度、原生工具使用、多模态生成和1M令牌上下文窗口。"
+ },
"gemini-2.0-flash-exp": {
- "description": "Gemini 2.0 Flash Exp 是 Google 最新的实验性多模态AI模型，拥有下一代特性，卓越的速度，原生工具调用以及多模态生成。"
+ "description": "Gemini 2.0 Flash Exp 是 Google 的实验性多模态AI模型，拥有下一代特性，卓越的速度，原生工具调用以及多模态生成。"
},
- "gemini-2.0-flash-thinking-exp-1219": {
- "description": "Gemini 2.0 Flash Exp 是 Google 最新的实验性多模态AI模型，拥有下一代特性，卓越的速度，原生工具调用以及多模态生成。"
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "一个 Gemini 2.0 Flash 模型，针对成本效益和低延迟等目标进行了优化。"
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
+ "description": "Gemini 2.0 Flash Thinking Exp 是 Google 的实验性多模态推理AI模型，能对复杂问题进行推理，拥有新的思维能力。"
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 是 Google 的实验性多模态AI模型，对输出质量有一定改进。"
+ "gemini-2.0-flash-thinking-exp-1219": {
+ "description": "Gemini 2.0 Flash Thinking Exp 是 Google 的实验性多模态推理AI模型，能对复杂问题进行推理，拥有新的思维能力。"
},
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 是 Google 的实验性多模态AI模型，拥有改进的编码、推理和视觉能力。"
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental 是 Google 最新的实验性多模态AI模型，与历史版本相比有一定的质量提升，特别是对于世界知识、代码和长上下文。"
},
"gemini-exp-1206": {
- "description": "Gemini Exp 1206 是 Google 最新的实验性多模态AI模型，与历史版本相比有一定的质量提升。"
+ "description": "Gemini Exp 1206 是 Google 的实验性多模态AI模型，与历史版本相比有一定的质量提升。"
},
"gemma-7b-it": {
"description": "Gemma 7B 适合中小规模任务处理，兼具成本效益。"
@@ -674,20 +815,35 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus 具备对视频内容及多图片的理解能力，适合多模态任务。"
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview具备强大的复杂推理能力，在逻辑推理、数学、编程等领域表现优异。"
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash 提供下一代功能和改进，包括卓越的速度、原生工具使用、多模态生成和1M令牌上下文窗口。"
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental 是 Google 最新的实验性多模态AI模型，与历史版本相比有一定的质量提升，特别是对于世界知识、代码和长上下文。"
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash 提供了优化后的多模态处理能力，适用多种复杂任务场景。"
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro 结合最新优化技术，带来更高效的多模态数据处理能力。"
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 是 Google 推出的高效模型，涵盖从小型应用到复杂数据处理的多种应用场景。"
+ },
"google/gemma-2-27b-it": {
- "description": "Gemma 是由 Google 开发的轻量级、最先进的开放模型系列，采用与 Gemini 模型相同的研究和技术构建。这些模型是仅解码器的大型语言模型，支持英语，提供预训练和指令微调两种变体的开放权重。Gemma 模型适用于各种文本生成任务，包括问答、摘要和推理。其相对较小的规模使其能够部署在资源有限的环境中，如笔记本电脑、台式机或个人云基础设施，从而让所有人都能获得最先进的 AI 模型，促进创新"
+ "description": "Gemma 2 27B 是一款通用大语言模型，具有优异的性能和广泛的应用场景。"
},
"google/gemma-2-2b-it": {
- "description": "Google的轻量级指令调优模型"
+ "description": "面向边缘应用的高级小型语言生成 AI 模型。"
+ },
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 是 Google 推出的高效模型，涵盖从小型应用到复杂数据处理的多种应用场景。"
},
"google/gemma-2-9b-it": {
- "description": "Gemma 是 Google 开发的轻量级、最先进的开放模型系列之一。它是一个仅解码器的大型语言模型，支持英语，提供开放权重、预训练变体和指令微调变体。Gemma 模型适用于各种文本生成任务，包括问答、摘要和推理。该 9B 模型是通过 8 万亿个 tokens 训练而成。其相对较小的规模使其可以在资源有限的环境中部署，如笔记本电脑、台式机或您自己的云基础设施，从而使更多人能够访问最先进的 AI 模型并促进创新"
+ "description": "Gemma 2 9B 由Google开发，提供高效的指令响应和综合能力。"
},
"google/gemma-2-9b-it:free": {
"description": "Gemma 2 是Google轻量化的开源文本模型系列。"
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo，适用于各种文本生成和理解任务，Currently points to gpt-3.5-turbo-0125"
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k，高容量文本生成模型，适合复杂任务。"
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo，适用于各种文本生成和理解任务，Currently points to gpt-3.5-turbo-0125"
},
@@ -714,7 +873,7 @@
"description": "GPT 3.5 Turbo 16k，高容量文本生成模型，适合复杂任务。"
},
"gpt-4": {
- "description": "GPT 4 Turbo，多模态模型，提供杰出的语言理解和生成能力，同时支持图像输入。"
+ "description": "GPT-4 提供了一个更大的上下文窗口，能够处理更长的文本输入，适用于需要广泛信息整合和数据分析的场景。"
},
"gpt-4-0125-preview": {
"description": "最新的 GPT-4 Turbo 模型具备视觉功能。现在，视觉请求可以使用 JSON 模式和函数调用。 GPT-4 Turbo 是一个增强版本，为多模态任务提供成本效益高的支持。它在准确性和效率之间找到平衡，适合需要进行实时交互的应用程序场景。"
@@ -744,7 +903,7 @@
"description": "GPT-4 视觉预览版，专为图像分析和处理任务设计。"
},
"gpt-4o": {
- "description": "OpenAI GPT-4系列中最先进的多模态模型，可以处理文本和图像输入。"
+ "description": "ChatGPT-4o 是一款动态模型，实时更新以保持当前最新版本。它结合了强大的语言理解与生成能力，适合于大规模应用场景，包括客户服务、教育和技术支持。"
},
"gpt-4o-2024-05-13": {
"description": "ChatGPT-4o 是一款动态模型，实时更新以保持当前最新版本。它结合了强大的语言理解与生成能力，适合于大规模应用场景，包括客户服务、教育和技术支持。"
@@ -755,8 +914,23 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o 是一款动态模型，实时更新以保持当前最新版本。它结合了强大的语言理解与生成能力，适合于大规模应用场景，包括客户服务、教育和技术支持。"
},
+ "gpt-4o-audio-preview": {
+ "description": "GPT-4o Audio 模型，支持音频输入输出"
+ },
"gpt-4o-mini": {
- "description": "一种经济高效的AI解决方案，适用于多种文本和图像任务。"
+ "description": "GPT-4o mini是OpenAI在GPT-4 Omni之后推出的最新模型，支持图文输入并输出文本。作为他们最先进的小型模型，它比其他近期的前沿模型便宜很多，并且比GPT-3.5 Turbo便宜超过60%。它保持了最先进的智能，同时具有显著的性价比。GPT-4o mini在MMLU测试中获得了 82% 的得分，目前在聊天偏好上排名高于 GPT-4。"
+ },
+ "gpt-4o-mini-realtime-preview": {
+ "description": "GPT-4o-mini 实时版本，支持音频和文本实时输入输出"
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "GPT-4o 实时版本，支持音频和文本实时输入输出"
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "GPT-4o 实时版本，支持音频和文本实时输入输出"
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "GPT-4o 实时版本，支持音频和文本实时输入输出"
},
"grok-2-1212": {
"description": "该模型在准确性、指令遵循和多语言能力方面有所改进。"
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "混元最新 MOE 架构 FunctionCall 模型，经过高质量的 FunctionCall 数据训练，上下文窗口达 32K，在多个维度的评测指标上处于领先。"
},
+ "hunyuan-large": {
+ "description": "Hunyuan-large 模型总参数量约 389B，激活参数量约 52B，是当前业界参数规模最大、效果最好的 Transformer 架构的开源 MoE 模型。"
+ },
+ "hunyuan-large-longcontext": {
+ "description": "擅长处理长文任务如文档摘要和文档问答等，同时也具备处理通用文本生成任务的能力。在长文本的分析和生成上表现优异，能有效应对复杂和详尽的长文内容处理需求。"
+ },
"hunyuan-lite": {
"description": "升级为 MOE 结构，上下文窗口为 256k ，在 NLP，代码，数学，行业等多项评测集上领先众多开源模型。"
},
+ "hunyuan-lite-vision": {
+ "description": "混元最新7B多模态模型，上下文窗口32K，支持中英文场景的多模态对话、图像物体识别、文档表格理解、多模态数学等，在多个维度上评测指标优于7B竞品模型。"
+ },
"hunyuan-pro": {
"description": "万亿级参数规模 MOE-32K 长文模型。在各种 benchmark 上达到绝对领先的水平，复杂指令和推理，具备复杂数学能力，支持 functioncall，在多语言翻译、金融法律医疗等领域应用重点优化。"
},
@@ -794,8 +977,23 @@
"hunyuan-standard-256K": {
"description": "采用更优的路由策略，同时缓解了负载均衡和专家趋同的问题。长文方面，大海捞针指标达到99.9%。MOE-256K 在长度和效果上进一步突破，极大的扩展了可输入长度。"
},
+ "hunyuan-standard-vision": {
+ "description": "混元最新多模态模型，支持多语种作答，中英文能力均衡。"
+ },
"hunyuan-turbo": {
- "description": "混元全新一代大语言模型的预览版，采用全新的混合专家模型（MoE）结构，相比hunyuan-pro推理效率更快，效果表现更强。"
+ "description": "本版本优化：数据指令scaling，大幅提升模型通用泛化能力；大幅提升数学、代码、逻辑推理能力；优化文本理解字词理解相关能力；优化文本创作内容生成质量"
+ },
+ "hunyuan-turbo-20241120": {
+ "description": "hunyuan-turbo 2024 年 11 月 20 日固定版本，介于 hunyuan-turbo 和 hunyuan-turbo-latest 之间的一个版本。"
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "本版本优化：数据指令scaling，大幅提升模型通用泛化能力；大幅提升数学、代码、逻辑推理能力；优化文本理解字词理解相关能力；优化文本创作内容生成质量"
+ },
+ "hunyuan-turbo-latest": {
+ "description": "通用体验优化，包括NLP理解、文本创作、闲聊、知识问答、翻译、领域等；提升拟人性，优化模型情商；提升意图模糊时模型主动澄清能力；提升字词解析类问题的处理能力；提升创作的质量和可互动性；提升多轮体验。"
+ },
+ "hunyuan-turbo-vision": {
+ "description": "混元新一代视觉语言旗舰大模型，采用全新的混合专家模型（MoE）结构，在图文理解相关的基础识别、内容创作、知识问答、分析推理等能力上相比前一代模型全面提升。"
},
"hunyuan-vision": {
"description": "混元最新多模态模型，支持图片+文本输入生成文本内容。"
@@ -807,10 +1005,16 @@
"description": "InternLM2.5-7B-Chat 是一个开源的对话模型，基于 InternLM2 架构开发。该 7B 参数规模的模型专注于对话生成任务，支持中英双语交互。模型采用了最新的训练技术，旨在提供流畅、智能的对话体验。InternLM2.5-7B-Chat 适用于各种对话应用场景，包括但不限于智能客服、个人助手等领域"
},
"internlm2-pro-chat": {
- "description": "我们仍在维护的老版本模型，有 7B、20B 多种模型参数量可选。"
+ "description": "InternLM2 版本最大的模型，专注于高度复杂的任务"
},
"internlm2.5-latest": {
- "description": "我们最新的模型系列，有着卓越的推理性能，支持 1M 的上下文长度以及更强的指令跟随和工具调用能力。"
+ "description": "我们仍在维护的老版本模型，经过多轮迭代有着极其优异且稳定的性能，包含 7B、20B 多种模型参数量可选，支持 1M 的上下文长度以及更强的指令跟随和工具调用能力。默认指向我们最新发布的 InternLM2.5 系列模型"
+ },
+ "internlm3-latest": {
+ "description": "我们最新的模型系列，有着卓越的推理性能，领跑同量级开源模型。默认指向我们最新发布的 InternLM3 系列模型"
+ },
+ "jina-deepsearch-v1": {
+ "description": "深度搜索结合了网络搜索、阅读和推理，可进行全面调查。您可以将其视为一个代理，接受您的研究任务 - 它会进行广泛搜索并经过多次迭代，然后才能给出答案。这个过程涉及持续的研究、推理和从各个角度解决问题。这与直接从预训练数据生成答案的标准大模型以及依赖一次性表面搜索的传统 RAG 系统有着根本的不同。"
},
"learnlm-1.5-pro-experimental": {
"description": "LearnLM 是一个实验性的、特定于任务的语言模型，经过训练以符合学习科学原则，可在教学和学习场景中遵循系统指令，充当专家导师等。"
@@ -818,30 +1022,18 @@
"lite": {
"description": "Spark Lite 是一款轻量级大语言模型，具备极低的延迟与高效的处理能力，完全免费开放，支持实时在线搜索功能。其快速响应的特性使其在低算力设备上的推理应用和模型微调中表现出色，为用户带来出色的成本效益和智能体验，尤其在知识问答、内容生成及搜索场景下表现不俗。"
},
- "llama-3.1-70b-instruct": {
- "description": "Llama 3.1 70B Instruct 模型，具备70B参数，能在大型文本生成和指示任务中提供卓越性能。"
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B 提供更强大的AI推理能力，适合复杂应用，支持超多的计算处理并保证高效和准确率。"
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B 是一款高效能模型，提供了快速的文本生成能力，非常适合需要大规模效率和成本效益的应用场景。"
},
- "llama-3.1-8b-instruct": {
- "description": "Llama 3.1 8B Instruct 模型，具备8B参数，支持画面指示任务的高效执行，提供优质的文本生成能力。"
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Llama 3.1 Sonar Huge Online 模型，具备405B参数，支持约127,000个标记的上下文长度，设计用于复杂的在线聊天应用。"
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Llama 3.1 Sonar Large Chat 模型，具备70B参数，支持约127,000个标记的上下文长度，适合于复杂的离线聊天任务。"
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Llama 3.1 Sonar Large Online 模型，具备70B参数，支持约127,000个标记的上下文长度，适用于高容量和多样化聊天任务。"
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Llama 3.1 Sonar Small Chat 模型，具备8B参数，专为离线聊天设计，支持约127,000个标记的上下文长度。"
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Llama 3.1 Sonar Small Online 模型，具备8B参数，支持约127,000个标记的上下文长度，专为在线聊天设计，能高效处理各种文本交互。"
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 旨在处理结合视觉和文本数据的任务。它在图像描述和视觉问答等任务中表现出色，跨越了语言生成和视觉推理之间的鸿沟。"
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 是 Llama 系列最先进的多语言开源大型语言模型，以极低成本体验媲美 405B 模型的性能。基于 Transformer 结构，并通过监督微调（SFT）和人类反馈强化学习（RLHF）提升有用性和安全性。其指令调优版本专为多语言对话优化，在多项行业基准上表现优于众多开源和封闭聊天模型。知识截止日期为 2023 年 12 月"
+ },
"llama-3.3-70b-versatile": {
"description": "Meta Llama 3.3 多语言大语言模型 ( LLM ) 是 70B（文本输入/文本输出）中的预训练和指令调整生成模型。 Llama 3.3 指令调整的纯文本模型针对多语言对话用例进行了优化，并且在常见行业基准上优于许多可用的开源和封闭式聊天模型。"
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 旨在处理结合视觉和文本数据的任务。它在图像描述和视觉问答等任务中表现出色，跨越了语言生成和视觉推理之间的鸿沟。"
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 是 Llama 系列最先进的多语言开源大型语言模型，以极低成本体验媲美 405B 模型的性能。基于 Transformer 结构，并通过监督微调（SFT）和人类反馈强化学习（RLHF）提升有用性和安全性。其指令调优版本专为多语言对话优化，在多项行业基准上表现优于众多开源和封闭聊天模型。知识截止日期为 2023 年 12 月"
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Meta Llama 3.3 多语言大语言模型 ( LLM ) 是 70B（文本输入/文本输出）中的预训练和指令调整生成模型。 Llama 3.3 指令调整的纯文本模型针对多语言对话用例进行了优化，并且在常见行业基准上优于许多可用的开源和封闭式聊天模型。"
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 旨在处理结合视觉和文本数据的任务。它在图像描述和视觉问答等任务中表现出色，跨越了语言生成和视觉推理之间的鸿沟。"
},
@@ -951,11 +1152,14 @@
"description": "Llama 3 8B Instruct Turbo 是一款高效能的大语言模型，支持广泛的应用场景。"
},
"meta-llama/Meta-Llama-3.1-405B-Instruct": {
- "description": "Meta Llama 3.1 是由 Meta 开发的多语言大型语言模型家族，包括 8B、70B 和 405B 三种参数规模的预训练和指令微调变体。该 405B 指令微调模型针对多语言对话场景进行了优化，在多项行业基准测试中表现优异。模型训练使用了超过 15 万亿个 tokens 的公开数据，并采用了监督微调和人类反馈强化学习等技术来提升模型的有用性和安全性。Llama 3.1 支持文本生成和代码生成，知识截止日期为 2023 年 12 月"
+ "description": "Llama 3.1 是 Meta 推出的领先模型，支持高达 405B 参数，可应用于复杂对话、多语言翻译和数据分析领域。"
},
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "405B 的 Llama 3.1 Turbo 模型，为大数据处理提供超大容量的上下文支持，在超大规模的人工智能应用中表现突出。"
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 是 Meta 推出的领先模型，支持高达 405B 参数，可应用于复杂对话、多语言翻译和数据分析领域。"
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "Meta Llama 3.1 是由 Meta 开发的多语言大型语言模型家族，包括 8B、70B 和 405B 三种参数规模的预训练和指令微调变体。该 70B 指令微调模型针对多语言对话场景进行了优化，在多项行业基准测试中表现优异。模型训练使用了超过 15 万亿个 tokens 的公开数据，并采用了监督微调和人类反馈强化学习等技术来提升模型的有用性和安全性。Llama 3.1 支持文本生成和代码生成，知识截止日期为 2023 年 12 月"
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 旨在处理结合视觉和文本数据的任务。它在图像描述和视觉问答等任务中表现出色，跨越了语言生成和视觉推理之间的鸿沟。"
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 是 Llama 系列最先进的多语言开源大型语言模型，以极低成本体验媲美 405B 模型的性能。基于 Transformer 结构，并通过监督微调（SFT）和人类反馈强化学习（RLHF）提升有用性和安全性。其指令调优版本专为多语言对话优化，在多项行业基准上表现优于众多开源和封闭聊天模型。知识截止日期为 2023 年 12 月"
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 是 Llama 系列最先进的多语言开源大型语言模型，以极低成本体验媲美 405B 模型的性能。基于 Transformer 结构，并通过监督微调（SFT）和人类反馈强化学习（RLHF）提升有用性和安全性。其指令调优版本专为多语言对话优化，在多项行业基准上表现优于众多开源和封闭聊天模型。知识截止日期为 2023 年 12 月"
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct 是 Llama 3.1 Instruct 模型中最大、最强大的模型，是一款高度先进的对话推理和合成数据生成模型，也可以用作在特定领域进行专业持续预训练或微调的基础。Llama 3.1 提供的多语言大型语言模型 (LLMs) 是一组预训练的、指令调整的生成模型，包括 8B、70B 和 405B 大小 (文本输入/输出)。Llama 3.1 指令调整的文本模型 (8B、70B、405B) 专为多语言对话用例进行了优化，并在常见的行业基准测试中超过了许多可用的开源聊天模型。Llama 3.1 旨在用于多种语言的商业和研究用途。指令调整的文本模型适用于类似助手的聊天，而预训练模型可以适应各种自然语言生成任务。Llama 3.1 模型还支持利用其模型的输出来改进其他模型，包括合成数据生成和精炼。Llama 3.1 是使用优化的变压器架构的自回归语言模型。调整版本使用监督微调 (SFT) 和带有人类反馈的强化学习 (RLHF) 来符合人类对帮助性和安全性的偏好。"
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 是一款面向开发者、研究人员和企业的开放大型语言模型 (LLM)，旨在帮助他们构建、实验并负责任地扩展他们的生成 AI 想法。作为全球社区创新的基础系统的一部分，它非常适合计算能力和资源有限、边缘设备和更快的训练时间。"
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "高级 LLM，支持合成数据生成、知识蒸馏和推理，适用于聊天机器人、编程和特定领域任务。"
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "赋能复杂对话，具备卓越的上下文理解、推理能力和文本生成能力。"
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "先进的最尖端模型，具备语言理解、卓越的推理能力和文本生成能力。"
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "尖端的视觉-语言模型，擅长从图像中进行高质量推理。"
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "先进的最尖端小型语言模型，具备语言理解、卓越的推理能力和文本生成能力。"
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "先进的最尖端小型语言模型，具备语言理解、卓越的推理能力和文本生成能力。"
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "尖端的视觉-语言模型，擅长从图像中进行高质量推理。"
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "先进的 LLM，擅长推理、数学、常识和函数调用。"
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 是微软AI提供的语言模型，在复杂对话、多语言、推理和智能助手领域表现尤为出色。"
},
@@ -1029,13 +1263,13 @@
"description": "Mistral 是 Mistral AI 发布的 7B 模型，适合多变的语言处理需求。"
},
"mistral-large": {
- "description": "Mistral的旗舰模型，适合需要大规模推理能力或高度专业化的复杂任务（合成文本生成、代码生成、RAG或代理）。"
+ "description": "Mixtral Large 是 Mistral 的旗舰模型，结合代码生成、数学和推理的能力，支持 128k 上下文窗口。"
},
"mistral-large-latest": {
"description": "Mistral Large是旗舰大模型，擅长多语言任务、复杂推理和代码生成，是高端应用的理想选择。"
},
"mistral-nemo": {
- "description": "Mistral Nemo是一种尖端的语言模型（LLM），在其尺寸类别中拥有最先进的推理、世界知识和编码能力。"
+ "description": "Mistral Nemo 由 Mistral AI 和 NVIDIA 合作推出，是高效性能的 12B 模型。"
},
"mistral-small": {
"description": "Mistral Small可用于任何需要高效率和低延迟的基于语言的任务。"
@@ -1050,7 +1284,7 @@
"description": "Mistral (7B) Instruct v0.2 提供改进的指令处理能力和更精确的结果。"
},
"mistralai/Mistral-7B-Instruct-v0.3": {
- "description": "Mistral AI的指令调优模型"
+ "description": "Mistral (7B) Instruct v0.3 提供高效的计算能力和自然语言理解，适合广泛的应用。"
},
"mistralai/Mistral-7B-v0.1": {
"description": "Mistral 7B是一款紧凑但高性能的模型，擅长批量处理和简单任务，如分类和文本生成，具有良好的推理能力。"
@@ -1082,32 +1316,44 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K 是一款拥有超长上下文处理能力的模型，适用于生成超长文本，满足复杂的生成任务需求，能够处理多达128,000个tokens的内容，非常适合科研、学术和大型文档生成等应用场景。"
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Kimi 视觉模型（包括 moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview 等）能够理解图片内容，包括图片文字、图片颜色和物体形状等内容。"
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K 提供中等长度的上下文处理能力，能够处理32,768个tokens，特别适合生成各种长文档和复杂对话，应用于内容创作、报告生成和对话系统等领域。"
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Kimi 视觉模型（包括 moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview 等）能够理解图片内容，包括图片文字、图片颜色和物体形状等内容。"
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K 专为生成短文本任务设计，具有高效的处理性能，能够处理8,192个tokens，非常适合简短对话、速记和快速内容生成。"
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Kimi 视觉模型（包括 moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview 等）能够理解图片内容，包括图片文字、图片颜色和物体形状等内容。"
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B 是 Nous Hermes 2的升级版本，包含最新的内部开发的数据集。"
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama-3.1-Nemotron-70B-Instruct 是由 NVIDIA 定制的大型语言模型，旨在提高 LLM 生成的响应对用户查询的帮助程度。该模型在 Arena Hard、AlpacaEval 2 LC 和 GPT-4-Turbo MT-Bench 等基准测试中表现出色，截至 2024 年 10 月 1 日，在所有三个自动对齐基准测试中排名第一。该模型使用 RLHF（特别是 REINFORCE）、Llama-3.1-Nemotron-70B-Reward 和 HelpSteer2-Preference 提示在 Llama-3.1-70B-Instruct 模型基础上进行训练"
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B 是由 NVIDIA 定制的大型语言模型，旨在提高 LLM 生成的响应对用户查询的帮助程度。该模型在 Arena Hard、AlpacaEval 2 LC 和 GPT-4-Turbo MT-Bench 等基准测试中表现出色，截至 2024 年 10 月 1 日，在所有三个自动对齐基准测试中排名第一。该模型使用 RLHF（特别是 REINFORCE）、Llama-3.1-Nemotron-70B-Reward 和 HelpSteer2-Preference 提示在 Llama-3.1-70B-Instruct 模型基础上进行训练"
},
- "o1": {
- "description": "专注于高级推理和解决复杂问题，包括数学和科学任务。非常适合需要深入上下文理解和代理工作流程的应用程序。"
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "独特的语言模型，提供无与伦比的准确性和效率表现。"
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct 是 NVIDIA 定制的大型语言模型，旨在提高 LLM 生成的响应的帮助性。"
},
- "o1-2024-12-17": {
+ "o1": {
"description": "o1是OpenAI新的推理模型，支持图文输入并输出文本，适用于需要广泛通用知识的复杂任务。该模型具有200K上下文和2023年10月的知识截止日期。"
},
"o1-mini": {
- "description": "比 o1-preview 更小、更快，成本低80%，在代码生成和小上下文操作方面表现良好。"
+ "description": "o1-mini是一款针对编程、数学和科学应用场景而设计的快速、经济高效的推理模型。该模型具有128K上下文和2023年10月的知识截止日期。"
},
"o1-preview": {
- "description": "专注于高级推理和解决复杂问题，包括数学和科学任务。非常适合需要深度上下文理解和自主工作流程的应用。"
+ "description": "o1是OpenAI新的推理模型，适用于需要广泛通用知识的复杂任务。该模型具有128K上下文和2023年10月的知识截止日期。"
+ },
+ "o3-mini": {
+ "description": "o3-mini 是我们最新的小型推理模型，在与 o1-mini 相同的成本和延迟目标下提供高智能。"
},
"open-codestral-mamba": {
"description": "Codestral Mamba是专注于代码生成的Mamba 2语言模型，为先进的代码和推理任务提供强力支持。"
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini是OpenAI在GPT-4 Omni之后推出的最新模型，支持图文输入并输出文本。作为他们最先进的小型模型，它比其他近期的前沿模型便宜很多，并且比GPT-3.5 Turbo便宜超过60%。它保持了最先进的智能，同时具有显著的性价比。GPT-4o mini在MMLU测试中获得了 82% 的得分，目前在聊天偏好上排名高于 GPT-4。"
},
- "openai/o1": {
- "description": "o1是OpenAI新的推理模型，支持图文输入并输出文本，适用于需要广泛通用知识的复杂任务。该模型具有200K上下文和2023年10月的知识截止日期。"
- },
"openai/o1-mini": {
"description": "o1-mini是一款针对编程、数学和科学应用场景而设计的快速、经济高效的推理模型。该模型具有128K上下文和2023年10月的知识截止日期。"
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K 配置了特大上下文处理能力，能够处理多达128K的上下文信息，特别适合需通篇分析和长期逻辑关联处理的长文内容，可在复杂文本沟通中提供流畅一致的逻辑与多样的引用支持。"
},
+ "qvq-72b-preview": {
+ "description": "QVQ模型是由 Qwen 团队开发的实验性研究模型，专注于提升视觉推理能力，尤其在数学推理领域。"
+ },
"qwen-coder-plus-latest": {
"description": "通义千问代码模型。"
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "通义千问超大规模视觉语言模型。相比增强版，再次提升视觉推理能力和指令遵循能力，提供更高的视觉感知和认知水平。"
},
+ "qwen-vl-ocr-latest": {
+ "description": "通义千问OCR是文字提取专有模型，专注于文档、表格、试题、手写体文字等类型图像的文字提取能力。它能够识别多种文字，目前支持的语言有：汉语、英语、法语、日语、韩语、德语、俄语、意大利语、越南语、阿拉伯语。"
+ },
"qwen-vl-plus-latest": {
"description": "通义千问大规模视觉语言模型增强版。大幅提升细节识别能力和文字识别能力，支持超百万像素分辨率和任意长宽比规格的图像。"
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 是全新的大型语言模型系列，具有更强的理解和生成能力。"
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "面向中文和英文的 LLM，针对语言、编程、数学、推理等领域。"
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "高级 LLM，支持代码生成、推理和修复，涵盖主流编程语言。"
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "强大的中型代码模型，支持 32K 上下文长度，擅长多语言编程。"
+ },
"qwen2": {
"description": "Qwen2 是阿里巴巴的新一代大规模语言模型，以优异的性能支持多元化的应用需求。"
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "通义千问2.5对外开源的14B规模的模型。"
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "通义千问2.5对外开源的72B规模的模型。"
+ },
"qwen2.5-32b-instruct": {
"description": "通义千问2.5对外开源的32B规模的模型。"
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Qwen-Math 模型具有强大的数学解题能力。"
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "指令跟随、数学、解题、代码整体提升，万物识别能力提升，支持多样格式直接精准定位视觉元素，支持对长视频文件（最长10分钟）进行理解和秒级别的事件时刻定位，能理解时间先后和快慢，基于解析和定位能力支持操控OS或Mobile的Agent，关键信息抽取能力和Json格式输出能力强，此版本为72B版本，本系列能力最强的版本。"
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "指令跟随、数学、解题、代码整体提升，万物识别能力提升，支持多样格式直接精准定位视觉元素，支持对长视频文件（最长10分钟）进行理解和秒级别的事件时刻定位，能理解时间先后和快慢，基于解析和定位能力支持操控OS或Mobile的Agent，关键信息抽取能力和Json格式输出能力强，此版本为72B版本，本系列能力最强的版本。"
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 是阿里巴巴的新一代大规模语言模型，以优异的性能支持多元化的应用需求。"
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro 是 Upstage 推出的一款高智能LLM，专注于单GPU的指令跟随能力，IFEval得分80以上。目前支持英语，正式版本计划于2024年11月推出，将扩展语言支持和上下文长度。"
},
+ "sonar": {
+ "description": "基于搜索上下文的轻量级搜索产品，比 Sonar Pro 更快、更便宜。"
+ },
+ "sonar-pro": {
+ "description": "支持搜索上下文的高级搜索产品，支持高级查询和跟进。"
+ },
+ "sonar-reasoning": {
+ "description": "由 DeepSeek 推理模型提供支持的新 API 产品。"
+ },
"step-1-128k": {
"description": "平衡性能与成本，适合一般场景。"
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "该模型拥有强大的视频理解能力。"
},
+ "step-1o-vision-32k": {
+ "description": "该模型拥有强大的图像理解能力。相比于 step-1v 系列模型，拥有更强的视觉性能。"
+ },
"step-1v-32k": {
"description": "支持视觉输入，增强多模态交互体验。"
},
@@ -1302,20 +1581,41 @@
"description": "小型视觉模型，适合基本的图文任务。"
},
"step-2-16k": {
- "description": "支持大规模上下文交互，适合复杂对话场景。"
+ "description": "step-2模型的实验版本，包含最新的特性，滚动更新中。不推荐在正式生产环境使用。"
+ },
+ "step-2-mini": {
+ "description": "基于新一代自研Attention架构MFA的极速大模型，用极低成本达到和step1类似的效果，同时保持了更高的吞吐和更快响应时延。能够处理通用任务，在代码能力上具备特长。"
+ },
+ "taichu2_mm": {
+ "description": "融合了图像理解、知识迁移、逻辑归因等能力，在图文问答领域表现突出"
},
"taichu_llm": {
- "description": "Taichu 2.0 基于海量高质数据训练，具有更强的文本理解、内容创作、对话问答等能力"
+ "description": "基于海量高质数据训练，具有更强的文本理解、内容创作、对话问答等能力"
+ },
+ "text-embedding-3-large": {
+ "description": "最强大的向量化模型，适用于英文和非英文任务"
+ },
+ "text-embedding-3-small": {
+ "description": "高效且经济的新一代 Embedding 模型，适用于知识检索、RAG 应用等场景"
},
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) 通过高效的策略和模型架构，提供增强的计算能力。"
},
+ "tts-1": {
+ "description": "最新的文本转语音模型，针对实时场景优化速度"
+ },
+ "tts-1-hd": {
+ "description": "最新的文本转语音模型，针对质量进行优化"
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) 适用于精细化指令任务，提供出色的语言处理能力。"
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet 提升了行业标准，性能超过竞争对手模型和 Claude 3 Opus，在广泛的评估中表现出色，同时具有我们中等层级模型的速度和成本。"
},
+ "whisper-1": {
+ "description": "通用语音识别模型，支持多语言语音识别、语音翻译和语言识别"
+ },
"wizardlm2": {
"description": "WizardLM 2 是微软AI提供的语言模型，在复杂对话、多语言、推理和智能助手领域表现尤为出色。"
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "复杂视觉任务模型，提供高性能图片理解、分析能力。"
+ },
+ "yi-vision-v2": {
+ "description": "复杂视觉任务模型，提供基于多张图片的高性能理解、分析能力。"
}
}
diff --git a/locales/zh-CN/plugin.json b/locales/zh-CN/plugin.json
index 2e34b6901c76b..3bc20a4aa92e7 100644
--- a/locales/zh-CN/plugin.json
+++ b/locales/zh-CN/plugin.json
@@ -163,4 +163,4 @@
"title": "插件商店"
},
"unknownPlugin": "未知插件"
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/portal.json b/locales/zh-CN/portal.json
index 666b4c3fc5403..c1146a0279aaa 100644
--- a/locales/zh-CN/portal.json
+++ b/locales/zh-CN/portal.json
@@ -32,4 +32,4 @@
"files": "文件",
"messageDetail": "消息详情",
"title": "工作区"
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/providers.json b/locales/zh-CN/providers.json
index dc0700acf5c81..49edc9b4f7473 100644
--- a/locales/zh-CN/providers.json
+++ b/locales/zh-CN/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure 提供多种先进的AI模型，包括GPT-3.5和最新的GPT-4系列，支持多种数据类型和复杂任务，致力于安全、可靠和可持续的AI解决方案。"
},
+ "azureai": {
+ "description": "Azure 提供多种先进的AI模型，包括GPT-3.5和最新的GPT-4系列，支持多种数据类型和复杂任务，致力于安全、可靠和可持续的AI解决方案。"
+ },
"baichuan": {
"description": "百川智能是一家专注于人工智能大模型研发的公司，其模型在国内知识百科、长文本处理和生成创作等中文任务上表现卓越，超越了国外主流模型。百川智能还具备行业领先的多模态能力，在多项权威评测中表现优异。其模型包括 Baichuan 4、Baichuan 3 Turbo 和 Baichuan 3 Turbo 128k 等，分别针对不同应用场景进行优化，提供高性价比的解决方案。"
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek 是一家专注于人工智能技术研究和应用的公司，其最新模型 DeepSeek-V3 多项评测成绩超越 Qwen2.5-72B 和 Llama-3.1-405B 等开源模型，性能对齐领军闭源模型 GPT-4o 与 Claude-3.5-Sonnet。"
},
+ "doubao": {
+ "description": "字节跳动推出的自研大模型。通过字节跳动内部50+业务场景实践验证，每日万亿级tokens大使用量持续打磨，提供多种模态能力，以优质模型效果为企业打造丰富的业务体验。"
+ },
"fireworksai": {
"description": "Fireworks AI 是一家领先的高级语言模型服务商，专注于功能调用和多模态处理。其最新模型 Firefunction V2 基于 Llama-3，优化用于函数调用、对话及指令跟随。视觉语言模型 FireLLaVA-13B 支持图像和文本混合输入。其他 notable 模型包括 Llama 系列和 Mixtral 系列，提供高效的多语言指令跟随与生成支持。"
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "致力于大模型研究与开发工具链的开源组织。为所有 AI 开发者提供高效、易用的开源平台，让最前沿的大模型与算法技术触手可及"
},
+ "jina": {
+ "description": "Jina AI 成立于 2020 年，是一家领先的搜索 AI 公司。我们的搜索底座平台包含了向量模型、重排器和小语言模型，可帮助企业构建可靠且高质量的生成式AI和多模态的搜索应用。"
+ },
+ "lmstudio": {
+ "description": "LM Studio 是一个用于在您的计算机上开发和实验 LLMs 的桌面应用程序。"
+ },
"minimax": {
"description": "MiniMax 是 2021 年成立的通用人工智能科技公司，致力于与用户共创智能。MiniMax 自主研发了不同模态的通用大模型，其中包括万亿参数的 MoE 文本大模型、语音大模型以及图像大模型。并推出了海螺 AI 等应用。"
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI 是一个提供多种大语言模型与 AI 图像生成的 API 服务的平台，灵活、可靠且具有成本效益。它支持 Llama3、Mistral 等最新的开源模型，并为生成式 AI 应用开发提供了全面、用户友好且自动扩展的 API 解决方案，适合 AI 初创公司的快速发展。"
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ 提供容器，可用于自托管 GPU 加速推理微服务，支持在云端、数据中心、RTX™ AI 个人电脑和工作站上部署预训练和自定义 AI 模型。"
+ },
"ollama": {
"description": "Ollama 提供的模型广泛涵盖代码生成、数学运算、多语种处理和对话互动等领域，支持企业级和本地化部署的多样化需求。"
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "中科院自动化研究所和武汉人工智能研究院推出新一代多模态大模型，支持多轮问答、文本创作、图像生成、3D理解、信号分析等全面问答任务，拥有更强的认知、理解、创作能力，带来全新互动体验。"
},
+ "tencentcloud": {
+ "description": "知识引擎原子能力（LLM Knowledge Engine Atomic Power）基于知识引擎研发的知识问答全链路能力，面向企业及开发者，提供灵活组建及开发模型应用的能力。您可通过多款原子能力组建您专属的模型服务，调用文档解析、拆分、embedding、多轮改写等服务进行组装，定制企业专属 AI 业务。"
+ },
"togetherai": {
"description": "Together AI 致力于通过创新的 AI 模型实现领先的性能，提供广泛的自定义能力，包括快速扩展支持和直观的部署流程，满足企业的各种需求。"
},
"upstage": {
"description": "Upstage 专注于为各种商业需求开发AI模型，包括 Solar LLM 和文档 AI，旨在实现工作的人造通用智能（AGI）。通过 Chat API 创建简单的对话代理，并支持功能调用、翻译、嵌入以及特定领域应用。"
},
+ "vllm": {
+ "description": "vLLM 是一个快速且易于使用的库，用于 LLM 推理和服务。"
+ },
+ "volcengine": {
+ "description": "字节跳动推出的大模型服务的开发平台，提供功能丰富、安全以及具备价格竞争力的模型调用服务，同时提供模型数据、精调、推理、评测等端到端功能，全方位保障您的 AI 应用开发落地。"
+ },
"wenxin": {
"description": "企业级一站式大模型与AI原生应用开发及服务平台，提供最全面易用的生成式人工智能模型开发、应用开发全流程工具链"
},
diff --git a/locales/zh-CN/ragEval.json b/locales/zh-CN/ragEval.json
index b6d8b1095890f..5ba5140fc23db 100644
--- a/locales/zh-CN/ragEval.json
+++ b/locales/zh-CN/ragEval.json
@@ -88,4 +88,4 @@
"title": "评测任务列表"
}
}
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/setting.json b/locales/zh-CN/setting.json
index ac69caac8a4c9..e7b19caa307a2 100644
--- a/locales/zh-CN/setting.json
+++ b/locales/zh-CN/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "开启单次回复限制"
},
+ "enableReasoningEffort": {
+ "title": "开启推理强度调整"
+ },
"frequencyPenalty": {
- "desc": "值越大，越有可能降低重复字词",
- "title": "频率惩罚度"
+ "desc": "值越大，用词越丰富多样；值越低，用词更朴实简单",
+ "title": "词汇丰富度"
},
"maxTokens": {
"desc": "单次交互所用的最大 Token 数",
@@ -212,19 +215,31 @@
"desc": "{{provider}} 模型",
"title": "模型"
},
+ "params": {
+ "title": "高级参数"
+ },
"presencePenalty": {
- "desc": "值越大，越有可能扩展到新话题",
- "title": "话题新鲜度"
+ "desc": "值越大，越倾向不同的表达方式，避免概念重复；值越小，越倾向使用重复的概念或叙述，表达更具一致性",
+ "title": "表述发散度"
+ },
+ "reasoningEffort": {
+ "desc": "值越大，推理能力越强，但可能会增加响应时间和 Token 消耗",
+ "options": {
+ "high": "高",
+ "low": "低",
+ "medium": "中"
+ },
+ "title": "推理强度"
},
"temperature": {
- "desc": "值越大，回复越随机",
- "title": "随机性",
- "titleWithValue": "随机性 {{value}}"
+ "desc": "数值越大，回答越有创意和想象力；数值越小，回答越严谨",
+ "title": "创意活跃度",
+ "warning": "创意活跃度数值过大，输出可能会产生乱码"
},
"title": "模型设置",
"topP": {
- "desc": "与随机性类似，但不要和随机性一起更改",
- "title": "核采样"
+ "desc": "考虑多少种可能性，值越大，接受更多可能的回答；值越小，倾向选择最可能的回答。不推荐和创意活跃度一起更改",
+ "title": "思维开放度"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "通用设置",
"experiment": "实验",
"llm": "语言模型",
+ "provider": "AI 服务商",
"sync": "云端同步",
"system-agent": "系统助手",
"tts": "语音服务"
diff --git a/locales/zh-CN/thread.json b/locales/zh-CN/thread.json
index 14e636782e2fa..06c7e69bccc50 100644
--- a/locales/zh-CN/thread.json
+++ b/locales/zh-CN/thread.json
@@ -7,4 +7,4 @@
"title": "开启新的子话题"
},
"notSupportMultiModals": "子话题暂不支持文件/图片上传，如有需求，欢迎留言：<1>💬 讨论区"
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/tool.json b/locales/zh-CN/tool.json
index 3f459aa66931b..eea196337d654 100644
--- a/locales/zh-CN/tool.json
+++ b/locales/zh-CN/tool.json
@@ -7,4 +7,4 @@
"images": "图片:",
"prompt": "提示词"
}
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/topic.json b/locales/zh-CN/topic.json
index cbc79ca1d194b..40e097cd182a3 100644
--- a/locales/zh-CN/topic.json
+++ b/locales/zh-CN/topic.json
@@ -34,4 +34,4 @@
"searchPlaceholder": "搜索话题...",
"temp": "临时",
"title": "话题"
-}
+}
\ No newline at end of file
diff --git a/locales/zh-CN/welcome.json b/locales/zh-CN/welcome.json
index 5a50378ffb809..54cf0a5c5e4e3 100644
--- a/locales/zh-CN/welcome.json
+++ b/locales/zh-CN/welcome.json
@@ -42,4 +42,4 @@
"desc2": "创建你的第一个助手，让我们开始吧~",
"title": "给自己一个更聪明的大脑"
}
-}
+}
\ No newline at end of file
diff --git a/locales/zh-TW/changelog.json b/locales/zh-TW/changelog.json
index e7d265ca8fe87..bdd1f70efc870 100644
--- a/locales/zh-TW/changelog.json
+++ b/locales/zh-TW/changelog.json
@@ -8,8 +8,8 @@
"allChangelog": "查看所有更新日誌",
"description": "持續追蹤 {{appName}} 的新功能和改進",
"pagination": {
- "older": "查看歷史變更",
- "prev": "上一頁"
+ "next": "下一頁",
+ "older": "查看歷史變更"
},
"readDetails": "閱讀詳情",
"title": "更新日誌",
diff --git a/locales/zh-TW/common.json b/locales/zh-TW/common.json
index 7ccfe99236317..5220c167fc155 100644
--- a/locales/zh-TW/common.json
+++ b/locales/zh-TW/common.json
@@ -275,6 +275,7 @@
},
"temp": "臨時",
"terms": "服務條款",
+ "update": "更新",
"updateAgent": "更新助理資訊",
"upgradeVersion": {
"action": "升級",
@@ -285,6 +286,7 @@
"anonymousNickName": "匿名使用者",
"billing": "帳單管理",
"cloud": "體驗 {{name}}",
+ "community": "社區版",
"data": "資料儲存",
"defaultNickname": "社群版使用者",
"discord": "社區支援",
@@ -294,7 +296,6 @@
"help": "幫助中心",
"moveGuide": "設置按鈕搬到這裡啦",
"plans": "訂閱方案",
- "preview": "預覽",
"profile": "帳戶管理",
"setting": "應用設定",
"usages": "用量統計"
diff --git a/locales/zh-TW/components.json b/locales/zh-TW/components.json
index ef7a0d63c1ba0..d126fee8c6481 100644
--- a/locales/zh-TW/components.json
+++ b/locales/zh-TW/components.json
@@ -76,6 +76,7 @@
"custom": "自訂模型，預設支援函式呼叫與視覺辨識，請根據實際情況驗證上述能力的可用性",
"file": "該模型支援上傳檔案讀取與辨識",
"functionCall": "該模型支援函式呼叫（Function Call）",
+ "reasoning": "該模型支持深度思考",
"tokens": "該模型單一會話最多支援 {{tokens}} Tokens",
"vision": "該模型支援視覺辨識"
},
@@ -84,5 +85,34 @@
"ModelSwitchPanel": {
"emptyModel": "沒有啟用的模型，請前往設定開啟",
"provider": "提供商"
+ },
+ "OllamaSetupGuide": {
+ "cors": {
+ "description": "因瀏覽器安全限制，你需要為 Ollama 進行跨域配置後方可正常使用。",
+ "linux": {
+ "env": "在 [Service] 部分下添加 `Environment`，添加 OLLAMA_ORIGINS 環境變數：",
+ "reboot": "重載 systemd 並重啟 Ollama",
+ "systemd": "調用 systemd 編輯 ollama 服務："
+ },
+ "macos": "請打開「終端」應用程式，並粘貼以下指令，然後按回車執行",
+ "reboot": "請在執行完成後重啟 Ollama 服務",
+ "title": "配置 Ollama 允許跨域訪問",
+ "windows": "在 Windows 上，點擊「控制面板」，進入編輯系統環境變數。為您的用戶帳戶新建名為「OLLAMA_ORIGINS」的環境變數，值為 *，點擊「確定/應用」保存"
+ },
+ "install": {
+ "description": "請確認你已經啟動 Ollama，如果沒有下載 Ollama，請前往官網<1>下載",
+ "docker": "如果你更傾向於使用 Docker，Ollama 也提供了官方 Docker 映像，你可以通過以下命令拉取：",
+ "linux": {
+ "command": "通過以下命令安裝：",
+ "manual": "或者，你也可以參考 <1>Linux 手動安裝指南 自行安裝"
+ },
+ "title": "在本地安裝並啟動 Ollama 應用",
+ "windowsTab": "Windows (預覽版)"
+ }
+ },
+ "Thinking": {
+ "thinking": "深度思考中...",
+ "thought": "已深度思考（用時 {{duration}} 秒）",
+ "thoughtWithDuration": "已深度思考"
}
}
diff --git a/locales/zh-TW/discover.json b/locales/zh-TW/discover.json
index 130f8b9c7a22c..cdb81fd51d114 100644
--- a/locales/zh-TW/discover.json
+++ b/locales/zh-TW/discover.json
@@ -126,6 +126,10 @@
"title": "話題新鮮度"
},
"range": "範圍",
+ "reasoning_effort": {
+ "desc": "此設定用於控制模型在生成回答前的推理強度。低強度優先響應速度並節省 Token，高強度提供更完整的推理，但會消耗更多 Token 並降低響應速度。預設值為中，平衡推理準確性與響應速度。",
+ "title": "推理強度"
+ },
"temperature": {
"desc": "此設置影響模型回應的多樣性。較低的值會導致更可預測和典型的回應，而較高的值則鼓勵更多樣化和不常見的回應。當值設為0時，模型對於給定的輸入總是給出相同的回應。",
"title": "隨機性"
diff --git a/locales/zh-TW/modelProvider.json b/locales/zh-TW/modelProvider.json
index f78a6cc5d16da..532772be2664d 100644
--- a/locales/zh-TW/modelProvider.json
+++ b/locales/zh-TW/modelProvider.json
@@ -19,6 +19,24 @@
"title": "API 金鑰"
}
},
+ "azureai": {
+ "azureApiVersion": {
+ "desc": "Azure 的 API 版本，遵循 YYYY-MM-DD 格式，查閱[最新版本](https://learn.microsoft.com/zh-tw/azure/ai-services/openai/reference#chat-completions)",
+ "fetch": "獲取列表",
+ "title": "Azure API 版本"
+ },
+ "endpoint": {
+ "desc": "從 Azure AI 專案概述找到 Azure AI 模型推理終結點",
+ "placeholder": "https://ai-userxxxxxxxxxx.services.ai.azure.com/models",
+ "title": "Azure AI 終結點"
+ },
+ "title": "Azure OpenAI",
+ "token": {
+ "desc": "從 Azure AI 專案概述找到 API 密鑰",
+ "placeholder": "Azure 密鑰",
+ "title": "密鑰"
+ }
+ },
"bedrock": {
"accessKeyId": {
"desc": "填入AWS Access Key Id",
@@ -63,6 +81,46 @@
"title": "Cloudflare 帳戶 ID / API 位址"
}
},
+ "createNewAiProvider": {
+ "apiKey": {
+ "placeholder": "請填寫你的 API Key",
+ "title": "API Key"
+ },
+ "basicTitle": "基本資訊",
+ "configTitle": "配置信息",
+ "confirm": "新建",
+ "createSuccess": "新建成功",
+ "description": {
+ "placeholder": "服務商簡介（選填）",
+ "title": "服務商簡介"
+ },
+ "id": {
+ "desc": "作為服務商唯一標識，創建後將不可修改",
+ "format": "只能包含數字、小寫字母、連字符（-）和底線（_）",
+ "placeholder": "建議全小寫，例如 openai，創建後將不可修改",
+ "required": "請填寫服務商 ID",
+ "title": "服務商 ID"
+ },
+ "logo": {
+ "required": "請上傳正確的服務商 Logo",
+ "title": "服務商 Logo"
+ },
+ "name": {
+ "placeholder": "請輸入服務商的展示名稱",
+ "required": "請填寫服務商名稱",
+ "title": "服務商名稱"
+ },
+ "proxyUrl": {
+ "required": "請填寫代理地址",
+ "title": "代理地址"
+ },
+ "sdkType": {
+ "placeholder": "openai/anthropic/azureai/ollama/...",
+ "required": "請選擇 SDK 類型",
+ "title": "請求格式"
+ },
+ "title": "創建自定義 AI 服務商"
+ },
"github": {
"personalAccessToken": {
"desc": "填入你的 Github 個人存取權杖，點擊[這裡](https://github.com/settings/tokens) 創建",
@@ -77,6 +135,23 @@
"title": "HuggingFace Token"
}
},
+ "list": {
+ "title": {
+ "disabled": "未啟用服務商",
+ "enabled": "已啟用服務商"
+ }
+ },
+ "menu": {
+ "addCustomProvider": "添加自定義服務商",
+ "all": "全部",
+ "list": {
+ "disabled": "未啟用",
+ "enabled": "已啟用"
+ },
+ "notFound": "未找到搜索結果",
+ "searchProviders": "搜索服務商...",
+ "sort": "自定義排序"
+ },
"ollama": {
"checker": {
"desc": "測試代理地址是否正確填寫",
@@ -94,33 +169,9 @@
"title": "正在下載模型 {{model}}"
},
"endpoint": {
- "desc": "填入 Ollama 接口代理地址，本地未額外指定可留空",
+ "desc": "必須包含http(s)://，本地未額外指定可留空",
"title": "接口代理地址"
},
- "setup": {
- "cors": {
- "description": "因瀏覽器安全限制，您需要為 Ollama 進行跨域配置後才能正常使用。",
- "linux": {
- "env": "在 [Service] 部分下添加 `Environment`，新增 OLLAMA_ORIGINS 環境變數：",
- "reboot": "重新載入 systemd 並重新啟動 Ollama",
- "systemd": "呼叫 systemd 編輯 ollama 服務："
- },
- "macos": "請開啟「終端」應用程式，貼上以下指令，然後按 Enter 執行",
- "reboot": "執行完成後請重新啟動 Ollama 服務",
- "title": "配置 Ollama 允許跨域訪問",
- "windows": "在 Windows 上，點擊「控制面板」，進入編輯系統環境變數。為您的使用者帳戶新增名為「OLLAMA_ORIGINS」的環境變數，值為 *，點擊「確定/應用」保存"
- },
- "install": {
- "description": "請確認您已經啟用 Ollama，如果尚未下載 Ollama，請前往官網<1>下載",
- "docker": "如果您更傾向於使用 Docker，Ollama 也提供了官方 Docker 映像，您可以透過以下命令拉取：",
- "linux": {
- "command": "透過以下命令安裝：",
- "manual": "或者，您也可以參考 <1>Linux 手動安裝指南 自行安裝"
- },
- "title": "在本地安裝並啟動 Ollama 應用",
- "windowsTab": "Windows (預覽版)"
- }
- },
"title": "Ollama",
"unlock": {
"cancel": "取消下載",
@@ -131,25 +182,145 @@
"title": "下載指定的 Ollama 模型"
}
},
- "wenxin": {
- "accessKey": {
- "desc": "填入百度千帆平台的 Access Key",
- "placeholder": "千帆 Access Key",
- "title": "Access Key"
+ "providerModels": {
+ "config": {
+ "aesGcm": "您的秘鑰與代理地址等將使用 <1>AES-GCM 加密算法進行加密",
+ "apiKey": {
+ "desc": "請填寫你的 {{name}} API Key",
+ "placeholder": "{{name}} API Key",
+ "title": "API Key"
+ },
+ "baseURL": {
+ "desc": "必須包含 http(s)://",
+ "invalid": "請輸入合法的 URL",
+ "placeholder": "https://your-proxy-url.com/v1",
+ "title": "API 代理地址"
+ },
+ "checker": {
+ "button": "檢查",
+ "desc": "測試 Api Key 與代理地址是否正確填寫",
+ "pass": "檢查通過",
+ "title": "連通性檢查"
+ },
+ "fetchOnClient": {
+ "desc": "客戶端請求模式將從瀏覽器直接發起會話請求，可提升響應速度",
+ "title": "使用客戶端請求模式"
+ },
+ "helpDoc": "配置教程",
+ "waitingForMore": "更多模型正在 <1>計劃接入 中，敬請期待"
},
- "checker": {
- "desc": "測試 AccessKey / SecretAccess 是否填寫正確"
+ "createNew": {
+ "title": "創建自定義 AI 模型"
},
- "secretKey": {
- "desc": "填入百度千帆平台 Secret Key",
- "placeholder": "千帆 Secret Key",
- "title": "Secret Key"
+ "item": {
+ "config": "配置模型",
+ "customModelCards": {
+ "addNew": "創建並添加 {{id}} 模型",
+ "confirmDelete": "即將刪除該自定義模型，刪除後將不可恢復，請謹慎操作。"
+ },
+ "delete": {
+ "confirm": "確認刪除模型 {{displayName}}？",
+ "success": "刪除成功",
+ "title": "刪除模型"
+ },
+ "modelConfig": {
+ "azureDeployName": {
+ "extra": "在 Azure OpenAI 中實際請求的字段",
+ "placeholder": "請輸入 Azure 中的模型部署名稱",
+ "title": "模型部署名稱"
+ },
+ "deployName": {
+ "extra": "發送請求時會將該字段作為模型 ID",
+ "placeholder": "請輸入模型實際部署的名稱或 id",
+ "title": "模型部署名稱"
+ },
+ "displayName": {
+ "placeholder": "請輸入模型的展示名稱，例如 ChatGPT、GPT-4 等",
+ "title": "模型展示名稱"
+ },
+ "files": {
+ "extra": "當前文件上傳實現僅為一種 Hack 方案，僅限自行嘗試。完整文件上傳能力請等待後續實現",
+ "title": "支持文件上傳"
+ },
+ "functionCall": {
+ "extra": "此配置將僅開啟模型使用工具的能力，進而可以為模型添加工具類的插件。但是否支持真正使用工具完全取決於模型本身，請自行測試其可用性",
+ "title": "支持工具使用"
+ },
+ "id": {
+ "extra": "創建後不可修改，調用 AI 時將作為模型 id 使用",
+ "placeholder": "請輸入模型 id，例如 gpt-4o 或 claude-3.5-sonnet",
+ "title": "模型 ID"
+ },
+ "modalTitle": "自定義模型配置",
+ "reasoning": {
+ "extra": "此配置將僅開啟模型深度思考的能力，具體效果完全取決於模型本身，請自行測試該模型是否具備可用的深度思考能力",
+ "title": "支持深度思考"
+ },
+ "tokens": {
+ "extra": "設定模型支持的最大 Token 數",
+ "title": "最大上下文窗口",
+ "unlimited": "無限制"
+ },
+ "vision": {
+ "extra": "此配置將僅開啟應用中的圖片上傳配置，是否支持識別完全取決於模型本身，請自行測試該模型的視覺識別能力可用性",
+ "title": "支持視覺識別"
+ }
+ },
+ "pricing": {
+ "image": "${{amount}}/圖片",
+ "inputCharts": "${{amount}}/M 字符",
+ "inputMinutes": "${{amount}}/分鐘",
+ "inputTokens": "輸入 ${{amount}}/M",
+ "outputTokens": "輸出 ${{amount}}/M"
+ },
+ "releasedAt": "發佈於{{releasedAt}}"
},
- "unlock": {
- "customRegion": "自訂服務區域",
- "description": "輸入你的 AccessKey / SecretKey 即可開始會話。應用不會記錄你的鑑權配置",
- "title": "使用自訂文心一言鑑權資訊"
- }
+ "list": {
+ "addNew": "新增模型",
+ "disabled": "未啟用",
+ "disabledActions": {
+ "showMore": "顯示全部"
+ },
+ "empty": {
+ "desc": "請創建自定義模型或拉取模型後開始使用吧",
+ "title": "暫無可用模型"
+ },
+ "enabled": "已啟用",
+ "enabledActions": {
+ "disableAll": "全部禁用",
+ "enableAll": "全部啟用",
+ "sort": "自訂模型排序"
+ },
+ "enabledEmpty": "暫無啟用模型，請從下方列表中啟用心儀的模型吧~",
+ "fetcher": {
+ "clear": "清除取得的模型",
+ "fetch": "取得模型列表",
+ "fetching": "正在取得模型列表...",
+ "latestTime": "上次更新時間：{{time}}",
+ "noLatestTime": "尚未取得列表"
+ },
+ "resetAll": {
+ "conform": "確認重置當前模型的所有修改？重置後當前模型列表將會回到預設狀態",
+ "success": "重置成功",
+ "title": "重置所有修改"
+ },
+ "search": "搜尋模型...",
+ "searchResult": "搜尋到 {{count}} 個模型",
+ "title": "模型列表",
+ "total": "共 {{count}} 個模型可用"
+ },
+ "searchNotFound": "未找到搜尋結果"
+ },
+ "sortModal": {
+ "success": "排序更新成功",
+ "title": "自定義排序",
+ "update": "更新"
+ },
+ "updateAiProvider": {
+ "confirmDelete": "即將刪除該 AI 服務商，刪除後將無法找回，確認是否刪除？",
+ "deleteSuccess": "刪除成功",
+ "tooltip": "更新服務商基礎配置",
+ "updateSuccess": "更新成功"
},
"zeroone": {
"title": "01.AI 零一萬物"
diff --git a/locales/zh-TW/models.json b/locales/zh-TW/models.json
index 8801bcea9a2eb..15f4cc1987dbb 100644
--- a/locales/zh-TW/models.json
+++ b/locales/zh-TW/models.json
@@ -17,9 +17,15 @@
"360gpt-turbo-responsibility-8k": {
"description": "360GPT Turbo Responsibility 8K 強調語義安全和責任導向，專為對內容安全有高度要求的應用場景設計，確保用戶體驗的準確性與穩健性。"
},
+ "360gpt2-o1": {
+ "description": "360gpt2-o1 使用樹搜索構建思維鏈，並引入了反思機制，使用強化學習訓練，模型具備自我反思與糾錯的能力。"
+ },
"360gpt2-pro": {
"description": "360GPT2 Pro 是 360 公司推出的高級自然語言處理模型，具備卓越的文本生成和理解能力，尤其在生成與創作領域表現出色，能夠處理複雜的語言轉換和角色演繹任務。"
},
+ "360zhinao2-o1": {
+ "description": "360zhinao2-o1 使用樹搜索構建思維鏈，並引入了反思機制，使用強化學習訓練，模型具備自我反思與糾錯的能力。"
+ },
"4.0Ultra": {
"description": "Spark4.0 Ultra 是星火大模型系列中最為強大的版本，在升級聯網搜索鏈路同時，提升對文本內容的理解和總結能力。它是用於提升辦公生產力和準確響應需求的全方位解決方案，是引領行業的智能產品。"
},
@@ -41,6 +47,18 @@
"Baichuan4-Turbo": {
"description": "模型能力國內第一，在知識百科、長文本、生成創作等中文任務上超越國外主流模型。還具備行業領先的多模態能力，多項權威評測基準表現優異。"
},
+ "DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "基於 Qwen2.5-Math-1.5B 的 DeepSeek-R1 蒸餾模型，通過強化學習與冷啟動數據優化推理性能，開源模型刷新多任務標杆。"
+ },
+ "DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "基於 Qwen2.5-14B 的 DeepSeek-R1 蒸餾模型，通過強化學習與冷啟動數據優化推理性能，開源模型刷新多任務標杆。"
+ },
+ "DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1 系列通過強化學習與冷啟動數據優化推理性能，開源模型刷新多任務標杆，超越 OpenAI-o1-mini 水平。"
+ },
+ "DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "基於 Qwen2.5-Math-7B 的 DeepSeek-R1 蒸餾模型，通過強化學習與冷啟動數據優化推理性能，開源模型刷新多任務標杆。"
+ },
"Doubao-lite-128k": {
"description": "Doubao-lite 擁有極致的回應速度，更好的性價比，為客戶不同場景提供更靈活的選擇。支持 128k 上下文窗口的推理和精調。"
},
@@ -74,9 +92,6 @@
"ERNIE-4.0-8K-Preview": {
"description": "百度自研的旗艦級超大規模語言模型，相較ERNIE 3.5實現了模型能力全面升級，廣泛適用於各領域複雜任務場景；支持自動對接百度搜索插件，保障問答信息時效。"
},
- "ERNIE-4.0-Turbo-128K": {
- "description": "百度自研的旗艦級超大規模大語言模型，綜合效果表現出色，廣泛適用於各領域複雜任務場景；支持自動對接百度搜索插件，保障問答信息時效。相較於ERNIE 4.0在性能表現上更優秀"
- },
"ERNIE-4.0-Turbo-8K-Latest": {
"description": "百度自研的旗艦級超大規模大語言模型，綜合效果表現優異，廣泛適用於各領域複雜任務場景；支持自動對接百度搜索插件，保障問答信息時效。相較於 ERNIE 4.0 在性能表現上更為優秀。"
},
@@ -110,6 +125,9 @@
"LoRA/Qwen/Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct 是阿里雲發布的最新大語言模型系列之一。該 7B 模型在編碼和數學等領域具有顯著改進的能力。該模型還提供了多語言支持，覆蓋超過 29 種語言，包括中文、英文等。模型在指令跟隨、理解結構化數據以及生成結構化輸出（尤其是 JSON）方面都有顯著提升"
},
+ "MiniMax-Text-01": {
+ "description": "在 MiniMax-01系列模型中，我們做了大膽創新：首次大規模實現線性注意力機制，傳統 Transformer架構不再是唯一的選擇。這個模型的參數量高達4560億，其中單次激活459億。模型綜合性能比肩海外頂尖模型，同時能夠高效處理全球最長400萬token的上下文，是GPT-4o的32倍，Claude-3.5-Sonnet的20倍。"
+ },
"Nous-Hermes-2-Mixtral-8x7B-DPO": {
"description": "Hermes 2 Mixtral 8x7B DPO 是一款高度靈活的多模型合併，旨在提供卓越的創造性體驗。"
},
@@ -119,9 +137,6 @@
"OpenGVLab/InternVL2-26B": {
"description": "InternVL2在各種視覺語言任務上展現出了卓越的性能，包括文檔和圖表理解、場景文本理解、OCR、科學和數學問題解決等。"
},
- "OpenGVLab/InternVL2-Llama3-76B": {
- "description": "InternVL2在各種視覺語言任務上展現出了卓越的性能，包括文檔和圖表理解、場景文本理解、OCR、科學和數學問題解決等。"
- },
"Phi-3-medium-128k-instruct": {
"description": "相同的Phi-3-medium模型，但具有更大的上下文大小，適用於RAG或少量提示。"
},
@@ -167,12 +182,24 @@
"Pro/THUDM/glm-4-9b-chat": {
"description": "GLM-4-9B-Chat 是智譜 AI 推出的 GLM-4 系列預訓練模型中的開源版本。該模型在語義、數學、推理、代碼和知識等多個方面表現出色。除了支持多輪對話外，GLM-4-9B-Chat 還具備網頁瀏覽、代碼執行、自定義工具調用（Function Call）和長文本推理等高級功能。模型支持 26 種語言，包括中文、英文、日文、韓文和德文等。在多項基準測試中，GLM-4-9B-Chat 展現了優秀的性能，如 AlignBench-v2、MT-Bench、MMLU 和 C-Eval 等。該模型支持最大 128K 的上下文長度，適用於學術研究和商業應用"
},
+ "Pro/deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 是一款強化學習（RL）驅動的推理模型，解決了模型中的重複性和可讀性問題。在 RL 之前，DeepSeek-R1 引入了冷啟動數據，進一步優化了推理性能。它在數學、代碼和推理任務中與 OpenAI-o1 表現相當，並且透過精心設計的訓練方法，提升了整體效果。"
+ },
+ "Pro/deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 是一款擁有 6710 億參數的混合專家（MoE）語言模型，採用多頭潛在注意力（MLA）和 DeepSeekMoE 架構，結合無輔助損失的負載平衡策略，優化推理和訓練效率。透過在 14.8 萬億高質量tokens上預訓練，並進行監督微調和強化學習，DeepSeek-V3 在性能上超越其他開源模型，接近領先閉源模型。"
+ },
"Pro/google/gemma-2-9b-it": {
"description": "Gemma 是 Google 開發的輕量級、最先進的開放模型系列之一。它是一個僅解碼器的大型語言模型，支持英語，提供開放權重、預訓練變體和指令微調變體。Gemma 模型適用於各種文本生成任務，包括問答、摘要和推理。該 9B 模型是通過 8 萬億個 tokens 訓練而成。其相對較小的規模使其可以在資源有限的環境中部署，如筆記本電腦、桌面電腦或您自己的雲基礎設施，從而使更多人能夠訪問最先進的 AI 模型並促進創新"
},
"Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": {
"description": "Meta Llama 3.1 是由 Meta 開發的多語言大型語言模型家族，包括 8B、70B 和 405B 三種參數規模的預訓練和指令微調變體。該 8B 指令微調模型針對多語言對話場景進行了優化，在多項行業基準測試中表現優異。模型訓練使用了超過 15 萬億個 tokens 的公開數據，並採用了監督微調和人類反饋強化學習等技術來提升模型的有用性和安全性。Llama 3.1 支持文本生成和代碼生成，知識截止日期為 2023 年 12 月"
},
+ "QwQ-32B-Preview": {
+ "description": "QwQ-32B-Preview 是一款獨具創新的自然語言處理模型，能夠高效處理複雜的對話生成與上下文理解任務。"
+ },
+ "Qwen/QVQ-72B-Preview": {
+ "description": "QVQ-72B-Preview 是由 Qwen 團隊開發的專注於視覺推理能力的研究型模型，其在複雜場景理解和解決視覺相關的數學問題方面具有獨特優勢。"
+ },
"Qwen/QwQ-32B-Preview": {
"description": "QwQ-32B-Preview是Qwen 最新的實驗性研究模型，專注於提升AI推理能力。通過探索語言混合、遞歸推理等複雜機制，主要優勢包括強大的推理分析能力、數學和編程能力。與此同時，也存在語言切換問題、推理循環、安全性考量、其他能力方面的差異。"
},
@@ -215,15 +242,15 @@
"Qwen/Qwen2.5-Coder-7B-Instruct": {
"description": "Qwen2.5-Coder-7B-Instruct 是阿里雲發布的代碼特定大語言模型系列的最新版本。該模型在 Qwen2.5 的基礎上，通過 5.5 萬億個 tokens 的訓練，顯著提升了代碼生成、推理和修復能力。它不僅增強了編碼能力，還保持了數學和通用能力的優勢。模型為代碼智能體等實際應用提供了更全面的基礎"
},
- "Qwen/Qwen2.5-Math-72B-Instruct": {
- "description": "Qwen2.5-Math專注於數學領域的問題求解，為高難度題提供專業解答。"
- },
"Qwen2-72B-Instruct": {
"description": "Qwen2 是 Qwen 模型的最新系列，支持 128k 上下文，對比當前最優的開源模型，Qwen2-72B 在自然語言理解、知識、代碼、數學及多語言等多項能力上均顯著超越當前領先的模型。"
},
"Qwen2-7B-Instruct": {
"description": "Qwen2 是 Qwen 模型的最新系列，能夠超越同等規模的最優開源模型甚至更大規模的模型，Qwen2 7B 在多個評測上取得顯著的優勢，尤其是在代碼及中文理解上。"
},
+ "Qwen2-VL-72B": {
+ "description": "Qwen2-VL-72B是一款強大的視覺語言模型，支持圖像與文本的多模態處理，能夠精確識別圖像內容並生成相關描述或回答。"
+ },
"Qwen2.5-14B-Instruct": {
"description": "Qwen2.5-14B-Instruct 是一款140億參數的大語言模型，性能表現優秀，優化中文和多語言場景，支持智能問答、內容生成等應用。"
},
@@ -236,6 +263,9 @@
"Qwen2.5-7B-Instruct": {
"description": "Qwen2.5-7B-Instruct 是一款70億參數的大語言模型，支持函數調用與外部系統無縫互動，極大提升了靈活性和擴展性。優化中文和多語言場景，支持智能問答、內容生成等應用。"
},
+ "Qwen2.5-Coder-14B-Instruct": {
+ "description": "Qwen2.5-Coder-14B-Instruct 是一款基於大規模預訓練的程式指令模型，具備強大的程式理解和生成能力，能夠高效地處理各種程式任務，特別適合智能程式編寫、自動化腳本生成和程式問題解答。"
+ },
"Qwen2.5-Coder-32B-Instruct": {
"description": "Qwen2.5-Coder-32B-Instruct 是一款專為代碼生成、代碼理解和高效開發場景設計的大型語言模型，採用了業界領先的32B參數規模，能夠滿足多樣化的程式需求。"
},
@@ -290,12 +320,6 @@
"TeleAI/TeleMM": {
"description": "TeleMM多模態大模型是由中國電信自主研發的多模態理解大模型，能夠處理文本、圖像等多種模態輸入，支持圖像理解、圖表分析等功能，為用戶提供跨模態的理解服務。模型能夠與用戶進行多模態互動，準確理解輸入內容，回答問題、協助創作，並高效提供多模態信息和靈感支持。在細粒度感知，邏輯推理等多模態任務上有出色表現"
},
- "Tencent/Hunyuan-A52B-Instruct": {
- "description": "Hunyuan-Large 是業界最大的開源 Transformer 架構 MoE 模型，擁有 3890 億總參數量和 520 億激活參數量。"
- },
- "Vendor-A/Qwen/Qwen2-7B-Instruct": {
- "description": "Qwen2-72B-Instruct 是 Qwen2 系列中的指令微調大語言模型，參數規模為 72B。該模型基於 Transformer 架構，採用了 SwiGLU 激活函數、注意力 QKV 偏置和組查詢注意力等技術。它能夠處理大規模輸入。該模型在語言理解、生成、多語言能力、編碼、數學和推理等多個基準測試中表現出色，超越了大多數開源模型，並在某些任務上展現出與專有模型相當的競爭力"
- },
"Vendor-A/Qwen/Qwen2.5-72B-Instruct": {
"description": "Qwen2.5-72B-Instruct 是阿里雲發布的最新大語言模型系列之一。該 72B 模型在編碼和數學等領域具有顯著改進的能力。該模型還提供了多語言支持，覆蓋超過 29 種語言，包括中文、英文等。模型在指令跟隨、理解結構化數據以及生成結構化輸出（尤其是 JSON）方面都有顯著提升"
},
@@ -317,21 +341,18 @@
"abab6.5t-chat": {
"description": "針對中文人設對話場景優化，提供流暢且符合中文表達習慣的對話生成能力。"
},
- "accounts/fireworks/models/firefunction-v1": {
- "description": "Fireworks 開源函數調用模型，提供卓越的指令執行能力和開放可定制的特性。"
+ "abab7-chat-preview": {
+ "description": "相對於abab6.5系列模型在長文、數學、寫作等能力有大幅度提升。"
},
- "accounts/fireworks/models/firefunction-v2": {
- "description": "Fireworks 公司最新推出的 Firefunction-v2 是一款性能卓越的函數調用模型，基於 Llama-3 開發，並通過大量優化，特別適用於函數調用、對話及指令跟隨等場景。"
+ "accounts/fireworks/models/deepseek-r1": {
+ "description": "DeepSeek-R1 是一款最先進的大型語言模型，經過強化學習和冷啟動數據的優化，具有出色的推理、數學和編程性能。"
},
- "accounts/fireworks/models/firellava-13b": {
- "description": "fireworks-ai/FireLLaVA-13b 是一款視覺語言模型，可以同時接收圖像和文本輸入，經過高質量數據訓練，適合多模態任務。"
+ "accounts/fireworks/models/deepseek-v3": {
+ "description": "Deepseek 提供的強大 Mixture-of-Experts (MoE) 語言模型，總參數量為 671B，每個標記激活 37B 參數。"
},
"accounts/fireworks/models/llama-v3-70b-instruct": {
"description": "Llama 3 70B 指令模型，專為多語言對話和自然語言理解優化，性能優於多數競爭模型。"
},
- "accounts/fireworks/models/llama-v3-70b-instruct-hf": {
- "description": "Llama 3 70B 指令模型（HF 版本），與官方實現結果保持一致，適合高質量的指令跟隨任務。"
- },
"accounts/fireworks/models/llama-v3-8b-instruct": {
"description": "Llama 3 8B 指令模型，優化用於對話及多語言任務，表現卓越且高效。"
},
@@ -350,24 +371,24 @@
"accounts/fireworks/models/llama-v3p2-11b-vision-instruct": {
"description": "Meta的11B參數指令調整圖像推理模型。該模型針對視覺識別、圖像推理、圖像描述和回答關於圖像的一般性問題進行了優化。該模型能夠理解視覺數據，如圖表和圖形，並通過生成文本描述圖像細節來弥合視覺與語言之間的差距。"
},
- "accounts/fireworks/models/llama-v3p2-1b-instruct": {
- "description": "Llama 3.2 1B 指令模型是Meta推出的一款輕量級多語言模型。該模型旨在提高效率，與更大型的模型相比，在延遲和成本方面提供了顯著的改進。該模型的示例用例包括檢索和摘要。"
- },
"accounts/fireworks/models/llama-v3p2-3b-instruct": {
"description": "Llama 3.2 3B 指令模型是Meta推出的一款輕量級多語言模型。該模型旨在提高效率，與更大型的模型相比，在延遲和成本方面提供了顯著的改進。該模型的示例用例包括查詢和提示重寫以及寫作輔助。"
},
"accounts/fireworks/models/llama-v3p2-90b-vision-instruct": {
"description": "Meta的90B參數指令調整圖像推理模型。該模型針對視覺識別、圖像推理、圖像描述和回答關於圖像的一般性問題進行了優化。該模型能夠理解視覺數據，如圖表和圖形，並通過生成文本描述圖像細節來弥合視覺與語言之間的差距。"
},
+ "accounts/fireworks/models/llama-v3p3-70b-instruct": {
+ "description": "Llama 3.3 70B Instruct 是 Llama 3.1 70B 的 12 月更新版本。該模型在 Llama 3.1 70B（於 2024 年 7 月發布）的基礎上進行了改進，增強了工具調用、多語言文本支持、數學和編程能力。該模型在推理、數學和指令遵循方面達到了行業領先水平，並且能夠提供與 3.1 405B 相似的性能，同時在速度和成本上具有顯著優勢。"
+ },
+ "accounts/fireworks/models/mistral-small-24b-instruct-2501": {
+ "description": "24B 參數模型，具備與更大型模型相當的最先進能力。"
+ },
"accounts/fireworks/models/mixtral-8x22b-instruct": {
"description": "Mixtral MoE 8x22B 指令模型，大規模參數和多專家架構，全方位支持複雜任務的高效處理。"
},
"accounts/fireworks/models/mixtral-8x7b-instruct": {
"description": "Mixtral MoE 8x7B 指令模型，多專家架構提供高效的指令跟隨及執行。"
},
- "accounts/fireworks/models/mixtral-8x7b-instruct-hf": {
- "description": "Mixtral MoE 8x7B 指令模型（HF 版本），性能與官方實現一致，適合多種高效任務場景。"
- },
"accounts/fireworks/models/mythomax-l2-13b": {
"description": "MythoMax L2 13B 模型，結合新穎的合併技術，擅長敘事和角色扮演。"
},
@@ -377,18 +398,15 @@
"accounts/fireworks/models/qwen-qwq-32b-preview": {
"description": "QwQ模型是由 Qwen 團隊開發的實驗性研究模型，專注於增強 AI 推理能力。"
},
+ "accounts/fireworks/models/qwen2-vl-72b-instruct": {
+ "description": "Qwen-VL 模型的 72B 版本是阿里巴巴最新迭代的成果，代表了近一年的創新。"
+ },
"accounts/fireworks/models/qwen2p5-72b-instruct": {
"description": "Qwen2.5 是由阿里雲 Qwen 團隊開發的一系列僅包含解碼器的語言模型。這些模型提供不同的大小，包括 0.5B、1.5B、3B、7B、14B、32B 和 72B，並且有基礎版（base）和指令版（instruct）兩種變體。"
},
"accounts/fireworks/models/qwen2p5-coder-32b-instruct": {
"description": "Qwen2.5 Coder 32B Instruct 是阿里雲發布的代碼特定大語言模型系列的最新版本。該模型在 Qwen2.5 的基礎上，通過 5.5 萬億個 tokens 的訓練，顯著提升了代碼生成、推理和修復能力。它不僅增強了編碼能力，還保持了數學和通用能力的優勢。模型為代碼智能體等實際應用提供了更全面的基礎"
},
- "accounts/fireworks/models/starcoder-16b": {
- "description": "StarCoder 15.5B 模型，支持高級編程任務，多語言能力增強，適合複雜代碼生成和理解。"
- },
- "accounts/fireworks/models/starcoder-7b": {
- "description": "StarCoder 7B 模型，針對 80 多種編程語言訓練，擁有出色的編程填充能力和語境理解。"
- },
"accounts/yi-01-ai/models/yi-large": {
"description": "Yi-Large 模型，具備卓越的多語言處理能力，可用於各類語言生成和理解任務。"
},
@@ -428,6 +446,9 @@
"anthropic/claude-3-opus": {
"description": "Claude 3 Opus 是 Anthropic 用於處理高度複雜任務的最強大模型。它在性能、智能、流暢性和理解力方面表現卓越。"
},
+ "anthropic/claude-3.5-haiku": {
+ "description": "Claude 3.5 Haiku 是 Anthropic 最快的下一代模型。與 Claude 3 Haiku 相比，Claude 3.5 Haiku 在各項技能上都有所提升，並在許多智力基準測試中超越了上一代最大的模型 Claude 3 Opus。"
+ },
"anthropic/claude-3.5-sonnet": {
"description": "Claude 3.5 Sonnet 提供了超越 Opus 的能力和比 Sonnet 更快的速度，同時保持與 Sonnet 相同的價格。Sonnet 特別擅長程式設計、數據科學、視覺處理、代理任務。"
},
@@ -467,9 +488,6 @@
"claude-3-sonnet-20240229": {
"description": "Claude 3 Sonnet 在智能和速度方面為企業工作負載提供了理想的平衡。它以更低的價格提供最大效用，可靠且適合大規模部署。"
},
- "code-raccoon-v1": {
- "description": "代碼小浣熊是基於商湯大語言模型的軟體智能研發助手，覆蓋軟體需求分析、架構設計、代碼編寫、軟體測試等環節，滿足用戶代碼編寫、程式學習等各類需求。代碼小浣熊支持 Python、Java、JavaScript、C++、Go、SQL 等90+主流程式語言和 VS Code、IntelliJ IDEA 等主流 IDE。在實際應用中，代碼小浣熊可幫助開發者提升程式效率超過50%。"
- },
"codegeex-4": {
"description": "CodeGeeX-4是一個強大的AI編程助手，支持多種編程語言的智能問答與代碼補全，提升開發效率。"
},
@@ -521,18 +539,51 @@
"command-r-plus": {
"description": "Command R+ 是一款高性能的大型語言模型，專為真實企業場景和複雜應用而設計。"
},
+ "dall-e-2": {
+ "description": "第二代 DALL·E 模型，支持更真實、準確的圖像生成，解析度是第一代的4倍"
+ },
+ "dall-e-3": {
+ "description": "最新的 DALL·E 模型，於2023年11月發布。支持更真實、準確的圖像生成，具有更強的細節表現力"
+ },
"databricks/dbrx-instruct": {
"description": "DBRX Instruct 提供高可靠性的指令處理能力，支持多行業應用。"
},
- "deepseek-ai/DeepSeek-V2-Chat": {
- "description": "DeepSeek-V2 是一個強大、經濟高效的混合專家（MoE）語言模型。它在 8.1 萬億個 token 的高質量語料庫上進行了預訓練，並通過監督微調（SFT）和強化學習（RL）進一步提升了模型能力。與 DeepSeek 67B 相比， DeepSeek-V2 在性能更強的同時，節省了 42.5% 的訓練成本，減少了 93.3% 的 KV 緩存，並將最大生成吞吐量提高到了 5.76 倍。該模型支持 128k 的上下文長度，在標準基準測試和開放式生成評估中都表現出色"
+ "deepseek-ai/DeepSeek-R1": {
+ "description": "DeepSeek-R1 是一款強化學習（RL）驅動的推理模型，解決了模型中的重複性和可讀性問題。在 RL 之前，DeepSeek-R1 引入了冷啟動數據，進一步優化了推理性能。它在數學、程式碼和推理任務中與 OpenAI-o1 表現相當，並且通過精心設計的訓練方法，提升了整體效果。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": {
+ "description": "DeepSeek-R1 蒸餾模型，通過強化學習與冷啟動數據優化推理性能，開源模型刷新多任務標杆。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Llama-8B": {
+ "description": "DeepSeek-R1-Distill-Llama-8B 是基於 Llama-3.1-8B 開發的蒸餾模型。該模型使用 DeepSeek-R1 生成的樣本進行微調，展現出優秀的推理能力。在多個基準測試中表現不俗，其中在 MATH-500 上達到了 89.1% 的準確率，在 AIME 2024 上達到了 50.4% 的通過率，在 CodeForces 上獲得了 1205 的評分，作為 8B 規模的模型展示了較強的數學和編程能力。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": {
+ "description": "DeepSeek-R1 蒸餾模型，通過強化學習與冷啟動數據優化推理性能，開源模型刷新多任務標杆。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": {
+ "description": "DeepSeek-R1 蒸餾模型，通過強化學習與冷啟動數據優化推理性能，開源模型刷新多任務標杆。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": {
+ "description": "DeepSeek-R1-Distill-Qwen-32B 是基於 Qwen2.5-32B 通過知識蒸餾得到的模型。該模型使用 DeepSeek-R1 生成的 80 萬個精選樣本進行微調，在數學、編程和推理等多個領域展現出卓越的性能。在 AIME 2024、MATH-500、GPQA Diamond 等多個基準測試中都取得了優異成績，其中在 MATH-500 上達到了 94.3% 的準確率，展現出強大的數學推理能力。"
+ },
+ "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": {
+ "description": "DeepSeek-R1-Distill-Qwen-7B 是基於 Qwen2.5-Math-7B 通過知識蒸餾得到的模型。該模型使用 DeepSeek-R1 生成的 80 萬個精選樣本進行微調，展現出優秀的推理能力。在多個基準測試中表現出色，其中在 MATH-500 上達到了 92.8% 的準確率，在 AIME 2024 上達到了 55.5% 的通過率，在 CodeForces 上獲得了 1189 的評分，作為 7B 規模的模型展示了較強的數學和編程能力。"
},
"deepseek-ai/DeepSeek-V2.5": {
"description": "DeepSeek V2.5 集合了先前版本的優秀特徵，增強了通用和編碼能力。"
},
+ "deepseek-ai/DeepSeek-V3": {
+ "description": "DeepSeek-V3 是一款擁有 6710 億參數的混合專家（MoE）語言模型，採用多頭潛在注意力（MLA）和 DeepSeekMoE 架構，結合無輔助損失的負載平衡策略，優化推理和訓練效率。通過在 14.8 萬億高品質 tokens 上預訓練，並進行監督微調和強化學習，DeepSeek-V3 在性能上超越其他開源模型，接近領先閉源模型。"
+ },
"deepseek-ai/deepseek-llm-67b-chat": {
"description": "DeepSeek 67B 是為高複雜性對話訓練的先進模型。"
},
+ "deepseek-ai/deepseek-r1": {
+ "description": "最先進的高效 LLM，擅長推理、數學和編程。"
+ },
+ "deepseek-ai/deepseek-vl2": {
+ "description": "DeepSeek-VL2 是一個基於 DeepSeekMoE-27B 開發的混合專家（MoE）視覺語言模型，採用稀疏激活的 MoE 架構，在僅激活 4.5B 參數的情況下實現了卓越性能。該模型在視覺問答、光學字符識別、文檔/表格/圖表理解和視覺定位等多個任務中表現優異。"
+ },
"deepseek-chat": {
"description": "融合通用與代碼能力的全新開源模型，不僅保留了原有 Chat 模型的通用對話能力和 Coder 模型的強大代碼處理能力，還更好地對齊了人類偏好。此外，DeepSeek-V2.5 在寫作任務、指令跟隨等多個方面也實現了大幅提升。"
},
@@ -545,18 +596,99 @@
"deepseek-coder-v2:236b": {
"description": "DeepSeek Coder V2 是開源的混合專家代碼模型，在代碼任務方面表現優異，與 GPT4-Turbo 相媲美。"
},
+ "deepseek-r1": {
+ "description": "DeepSeek-R1 是一款強化學習（RL）驅動的推理模型，解決了模型中的重複性和可讀性問題。在 RL 之前，DeepSeek-R1 引入了冷啟動數據，進一步優化了推理性能。它在數學、程式碼和推理任務中與 OpenAI-o1 表現相當，並且通過精心設計的訓練方法，提升了整體效果。"
+ },
+ "deepseek-r1-distill-llama-70b": {
+ "description": "DeepSeek R1——DeepSeek 套件中更大更智能的模型——被蒸餾到 Llama 70B 架構中。基於基準測試和人工評估，該模型比原始 Llama 70B 更智能，尤其在需要數學和事實精確性的任務上表現出色。"
+ },
+ "deepseek-r1-distill-llama-8b": {
+ "description": "DeepSeek-R1-Distill 系列模型透過知識蒸餾技術，將 DeepSeek-R1 生成的樣本對 Qwen、Llama 等開源模型進行微調後得到。"
+ },
+ "deepseek-r1-distill-qwen-1.5b": {
+ "description": "DeepSeek-R1-Distill 系列模型透過知識蒸餾技術，將 DeepSeek-R1 生成的樣本對 Qwen、Llama 等開源模型進行微調後得到。"
+ },
+ "deepseek-r1-distill-qwen-14b": {
+ "description": "DeepSeek-R1-Distill 系列模型透過知識蒸餾技術，將 DeepSeek-R1 生成的樣本對 Qwen、Llama 等開源模型進行微調後得到。"
+ },
+ "deepseek-r1-distill-qwen-32b": {
+ "description": "DeepSeek-R1-Distill 系列模型透過知識蒸餾技術，將 DeepSeek-R1 生成的樣本對 Qwen、Llama 等開源模型進行微調後得到。"
+ },
+ "deepseek-r1-distill-qwen-7b": {
+ "description": "DeepSeek-R1-Distill 系列模型透過知識蒸餾技術，將 DeepSeek-R1 生成的樣本對 Qwen、Llama 等開源模型進行微調後得到。"
+ },
+ "deepseek-reasoner": {
+ "description": "DeepSeek 推出的推理模型。在輸出最終回答之前，模型會先輸出一段思維鏈內容，以提升最終答案的準確性。"
+ },
"deepseek-v2": {
"description": "DeepSeek V2 是高效的 Mixture-of-Experts 語言模型，適用於經濟高效的處理需求。"
},
"deepseek-v2:236b": {
"description": "DeepSeek V2 236B 是 DeepSeek 的設計代碼模型，提供強大的代碼生成能力。"
},
+ "deepseek-v3": {
+ "description": "DeepSeek-V3 為杭州深度求索人工智能基礎技術研究有限公司自研的 MoE 模型，其多項評測成績突出，在主流榜單中位列開源模型榜首。V3 相較 V2.5 模型生成速度實現 3 倍提升，為用戶帶來更加迅速流暢的使用體驗。"
+ },
"deepseek/deepseek-chat": {
"description": "融合通用與代碼能力的全新開源模型，不僅保留了原有 Chat 模型的通用對話能力和 Coder 模型的強大代碼處理能力，還更好地對齊了人類偏好。此外，DeepSeek-V2.5 在寫作任務、指令跟隨等多個方面也實現了大幅提升。"
},
+ "deepseek/deepseek-r1": {
+ "description": "DeepSeek-R1 在僅有極少標註數據的情況下，極大提升了模型推理能力。在輸出最終回答之前，模型會先輸出一段思維鏈內容，以提升最終答案的準確性。"
+ },
+ "deepseek/deepseek-r1:free": {
+ "description": "DeepSeek-R1 在僅有極少標註數據的情況下，極大提升了模型推理能力。在輸出最終回答之前，模型會先輸出一段思維鏈內容，以提升最終答案的準確性。"
+ },
"emohaa": {
"description": "Emohaa是一個心理模型，具備專業諮詢能力，幫助用戶理解情感問題。"
},
+ "ernie-3.5-128k": {
+ "description": "百度自研的旗艦級大規模大語言模型，覆蓋海量中英文語料，具有強大的通用能力，可滿足絕大部分對話問答、創作生成、插件應用場景要求；支持自動對接百度搜索插件，保障問答信息時效。"
+ },
+ "ernie-3.5-8k": {
+ "description": "百度自研的旗艦級大規模大語言模型，覆蓋海量中英文語料，具有強大的通用能力，可滿足絕大部分對話問答、創作生成、插件應用場景要求；支持自動對接百度搜索插件，保障問答信息時效。"
+ },
+ "ernie-3.5-8k-preview": {
+ "description": "百度自研的旗艦級大規模大語言模型，覆蓋海量中英文語料，具有強大的通用能力，可滿足絕大部分對話問答、創作生成、插件應用場景要求；支持自動對接百度搜索插件，保障問答信息時效。"
+ },
+ "ernie-4.0-8k-latest": {
+ "description": "百度自研的旗艦級超大規模大語言模型，相較ERNIE 3.5實現了模型能力全面升級，廣泛適用於各領域複雜任務場景；支持自動對接百度搜索插件，保障問答信息時效。"
+ },
+ "ernie-4.0-8k-preview": {
+ "description": "百度自研的旗艦級超大規模大語言模型，相較ERNIE 3.5實現了模型能力全面升級，廣泛適用於各領域複雜任務場景；支持自動對接百度搜索插件，保障問答信息時效。"
+ },
+ "ernie-4.0-turbo-128k": {
+ "description": "百度自研的旗艦級超大規模大語言模型，綜合效果表現出色，廣泛適用於各領域複雜任務場景；支持自動對接百度搜索插件，保障問答信息時效。相較於ERNIE 4.0在性能表現上更優秀"
+ },
+ "ernie-4.0-turbo-8k-latest": {
+ "description": "百度自研的旗艦級超大規模大語言模型，綜合效果表現出色，廣泛適用於各領域複雜任務場景；支持自動對接百度搜索插件，保障問答信息時效。相較於ERNIE 4.0在性能表現上更優秀"
+ },
+ "ernie-4.0-turbo-8k-preview": {
+ "description": "百度自研的旗艦級超大規模大語言模型，綜合效果表現出色，廣泛適用於各領域複雜任務場景；支持自動對接百度搜索插件，保障問答信息時效。相較於ERNIE 4.0在性能表現上更優秀"
+ },
+ "ernie-char-8k": {
+ "description": "百度自研的垂直場景大語言模型，適合遊戲NPC、客服對話、對話角色扮演等應用場景，人設風格更為鮮明、一致，指令遵循能力更強，推理性能更優。"
+ },
+ "ernie-char-fiction-8k": {
+ "description": "百度自研的垂直場景大語言模型，適合遊戲NPC、客服對話、對話角色扮演等應用場景，人設風格更為鮮明、一致，指令遵循能力更強，推理性能更優。"
+ },
+ "ernie-lite-8k": {
+ "description": "ERNIE Lite是百度自研的輕量級大語言模型，兼顧優異的模型效果與推理性能，適合低算力AI加速卡推理使用。"
+ },
+ "ernie-lite-pro-128k": {
+ "description": "百度自研的輕量級大語言模型，兼顧優異的模型效果與推理性能，效果比ERNIE Lite更優，適合低算力AI加速卡推理使用。"
+ },
+ "ernie-novel-8k": {
+ "description": "百度自研通用大語言模型，在小說續寫能力上有明顯優勢，也可用在短劇、電影等場景。"
+ },
+ "ernie-speed-128k": {
+ "description": "百度2024年最新發布的自研高性能大語言模型，通用能力優異，適合作為基座模型進行精調，更好地處理特定場景問題，同時具備極佳的推理性能。"
+ },
+ "ernie-speed-pro-128k": {
+ "description": "百度2024年最新發布的自研高性能大語言模型，通用能力優異，效果比ERNIE Speed更優，適合作為基座模型進行精調，更好地處理特定場景問題，同時具備極佳的推理性能。"
+ },
+ "ernie-tiny-8k": {
+ "description": "ERNIE Tiny是百度自研的超高性能大語言模型，部署與精調成本在文心系列模型中最低。"
+ },
"gemini-1.0-pro-001": {
"description": "Gemini 1.0 Pro 001 (Tuning) 提供穩定並可調優的性能，是複雜任務解決方案的理想選擇。"
},
@@ -599,17 +731,26 @@
"gemini-1.5-pro-latest": {
"description": "Gemini 1.5 Pro 支持高達 200 萬個 tokens，是中型多模態模型的理想選擇，適用於複雜任務的多方面支持。"
},
+ "gemini-2.0-flash": {
+ "description": "Gemini 2.0 Flash 提供下一代功能和改進，包括卓越的速度、原生工具使用、多模態生成和1M令牌上下文窗口。"
+ },
+ "gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash 提供下一代功能和改進，包括卓越的速度、原生工具使用、多模態生成和1M令牌上下文窗口。"
+ },
"gemini-2.0-flash-exp": {
"description": "Gemini 2.0 Flash Exp 是 Google 最新的實驗性多模態AI模型，擁有下一代特性，卓越的速度，原生工具調用以及多模態生成。"
},
- "gemini-2.0-flash-thinking-exp-1219": {
+ "gemini-2.0-flash-lite-preview-02-05": {
+ "description": "一個 Gemini 2.0 Flash 模型，針對成本效益和低延遲等目標進行了優化。"
+ },
+ "gemini-2.0-flash-thinking-exp-01-21": {
"description": "Gemini 2.0 Flash Exp 是 Google 最新的實驗性多模態AI模型，擁有下一代特性，卓越的速度，原生工具調用以及多模態生成。"
},
- "gemini-exp-1114": {
- "description": "Gemini Exp 1114 是 Google 最新的實驗性多模態 AI 模型，具備快速處理能力，支持文本、影像和影片輸入，適用於多種任務的高效擴展。"
+ "gemini-2.0-flash-thinking-exp-1219": {
+ "description": "Gemini 2.0 Flash Exp 是 Google 最新的實驗性多模態AI模型，擁有下一代特性，卓越的速度，原生工具調用以及多模態生成。"
},
- "gemini-exp-1121": {
- "description": "Gemini Exp 1121 是 Google 最新的實驗性多模態 AI 模型，具備快速處理能力，支持文本、圖像和視頻輸入，適用於多種任務的高效擴展。"
+ "gemini-2.0-pro-exp-02-05": {
+ "description": "Gemini 2.0 Pro Experimental 是 Google 最新的實驗性多模態AI模型，與歷史版本相比有一定的質量提升，特別是對於世界知識、代碼和長上下文。"
},
"gemini-exp-1206": {
"description": "Gemini Exp 1206 是 Google 最新的實驗性多模態 AI 模型，與歷史版本相比有一定的質量提升。"
@@ -674,18 +815,33 @@
"glm-4v-plus": {
"description": "GLM-4V-Plus具備對視頻內容及多圖片的理解能力，適合多模態任務。"
},
+ "glm-zero-preview": {
+ "description": "GLM-Zero-Preview具備強大的複雜推理能力，在邏輯推理、數學、程式設計等領域表現優異。"
+ },
+ "google/gemini-2.0-flash-001": {
+ "description": "Gemini 2.0 Flash 提供下一代功能和改進，包括卓越的速度、原生工具使用、多模態生成和1M令牌上下文窗口。"
+ },
+ "google/gemini-2.0-pro-exp-02-05:free": {
+ "description": "Gemini 2.0 Pro Experimental 是 Google 最新的實驗性多模態AI模型，與歷史版本相比有一定的質量提升，特別是對於世界知識、代碼和長上下文。"
+ },
"google/gemini-flash-1.5": {
"description": "Gemini 1.5 Flash 提供了優化後的多模態處理能力，適用於多種複雜任務場景。"
},
"google/gemini-pro-1.5": {
"description": "Gemini 1.5 Pro 結合最新的優化技術，帶來更高效的多模態數據處理能力。"
},
+ "google/gemma-2-27b": {
+ "description": "Gemma 2 是 Google 推出的高效模型，涵蓋從小型應用到複雜數據處理的多種應用場景。"
+ },
"google/gemma-2-27b-it": {
"description": "Gemma 2 延續了輕量化與高效的設計理念。"
},
"google/gemma-2-2b-it": {
"description": "Google的輕量級指令調優模型"
},
+ "google/gemma-2-9b": {
+ "description": "Gemma 2 是 Google 推出的高效模型，涵蓋從小型應用到複雜數據處理的多種應用場景。"
+ },
"google/gemma-2-9b-it": {
"description": "Gemma 2 是 Google 輕量化的開源文本模型系列。"
},
@@ -704,6 +860,9 @@
"gpt-3.5-turbo-1106": {
"description": "GPT 3.5 Turbo，適用於各種文本生成和理解任務，Currently points to gpt-3.5-turbo-0125"
},
+ "gpt-3.5-turbo-16k": {
+ "description": "GPT 3.5 Turbo 16k，高容量文本生成模型，適合複雜任務。"
+ },
"gpt-3.5-turbo-instruct": {
"description": "GPT 3.5 Turbo，適用於各種文本生成和理解任務，Currently points to gpt-3.5-turbo-0125"
},
@@ -755,9 +914,24 @@
"gpt-4o-2024-11-20": {
"description": "ChatGPT-4o 是一款動態模型，實時更新以保持當前最新版本。它結合了強大的語言理解與生成能力，適合於大規模應用場景，包括客戶服務、教育和技術支持。"
},
+ "gpt-4o-audio-preview": {
+ "description": "GPT-4o Audio 模型，支持音頻輸入輸出"
+ },
"gpt-4o-mini": {
"description": "GPT-4o mini是OpenAI在GPT-4 Omni之後推出的最新模型，支持圖文輸入並輸出文本。作為他們最先進的小型模型，它比其他近期的前沿模型便宜很多，並且比GPT-3.5 Turbo便宜超過60%。它保持了最先進的智能，同時具有顯著的性價比。GPT-4o mini在MMLU測試中獲得了82%的得分，目前在聊天偏好上排名高於GPT-4。"
},
+ "gpt-4o-mini-realtime-preview": {
+ "description": "GPT-4o-mini 實時版本，支持音頻和文本實時輸入輸出"
+ },
+ "gpt-4o-realtime-preview": {
+ "description": "GPT-4o 實時版本，支持音頻和文本實時輸入輸出"
+ },
+ "gpt-4o-realtime-preview-2024-10-01": {
+ "description": "GPT-4o 實時版本，支持音頻和文本實時輸入輸出"
+ },
+ "gpt-4o-realtime-preview-2024-12-17": {
+ "description": "GPT-4o 實時版本，支持音頻和文本實時輸入輸出"
+ },
"grok-2-1212": {
"description": "該模型在準確性、指令遵循和多語言能力方面有所改進。"
},
@@ -779,9 +953,18 @@
"hunyuan-functioncall": {
"description": "混元最新 MOE 架構 FunctionCall 模型，經過高質量的 FunctionCall 數據訓練，上下文窗口達 32K，在多個維度的評測指標上處於領先。"
},
+ "hunyuan-large": {
+ "description": "Hunyuan-large 模型總參數量約 389B，激活參數量約 52B，是當前業界參數規模最大、效果最好的 Transformer 架構的開源 MoE 模型。"
+ },
+ "hunyuan-large-longcontext": {
+ "description": "擅長處理長文任務如文檔摘要和文檔問答等，同時也具備處理通用文本生成任務的能力。在長文本的分析和生成上表現優異，能有效應對複雜和詳盡的長文內容處理需求。"
+ },
"hunyuan-lite": {
"description": "升級為 MOE 結構，上下文窗口為 256k，在 NLP、代碼、數學、行業等多項評測集上領先眾多開源模型。"
},
+ "hunyuan-lite-vision": {
+ "description": "混元最新7B多模態模型，上下文窗口32K，支持中英文場景的多模態對話、圖像物體識別、文檔表格理解、多模態數學等，在多個維度上評測指標優於7B競品模型。"
+ },
"hunyuan-pro": {
"description": "萬億級參數規模 MOE-32K 長文模型。在各種 benchmark 上達到絕對領先的水平，具備複雜指令和推理能力，支持 functioncall，在多語言翻譯、金融法律醫療等領域應用重點優化。"
},
@@ -794,9 +977,24 @@
"hunyuan-standard-256K": {
"description": "採用更優的路由策略，同時緩解了負載均衡和專家趨同的問題。長文方面，大海撈針指標達到 99.9%。MOE-256K 在長度和效果上進一步突破，極大地擴展了可輸入長度。"
},
+ "hunyuan-standard-vision": {
+ "description": "混元最新多模態模型，支持多語種作答，中英文能力均衡。"
+ },
"hunyuan-turbo": {
"description": "混元全新一代大語言模型的預覽版，採用全新的混合專家模型（MoE）結構，相較於 hunyuan-pro 推理效率更快，效果表現更強。"
},
+ "hunyuan-turbo-20241120": {
+ "description": "hunyuan-turbo 2024 年 11 月 20 日固定版本，介於 hunyuan-turbo 和 hunyuan-turbo-latest 之間的一個版本。"
+ },
+ "hunyuan-turbo-20241223": {
+ "description": "本版本優化：數據指令scaling，大幅提升模型通用泛化能力；大幅提升數學、程式碼、邏輯推理能力；優化文本理解字詞理解相關能力；優化文本創作內容生成質量"
+ },
+ "hunyuan-turbo-latest": {
+ "description": "通用體驗優化，包括NLP理解、文本創作、閒聊、知識問答、翻譯、領域等；提升擬人性，優化模型情商；提升意圖模糊時模型主動澄清能力；提升字詞解析類問題的處理能力；提升創作的質量和可互動性；提升多輪體驗。"
+ },
+ "hunyuan-turbo-vision": {
+ "description": "混元新一代視覺語言旗艦大模型，採用全新的混合專家模型（MoE）結構，在圖文理解相關的基礎識別、內容創作、知識問答、分析推理等能力上相比前一代模型全面提升。"
+ },
"hunyuan-vision": {
"description": "混元最新多模態模型，支持圖片 + 文本輸入生成文本內容。"
},
@@ -812,36 +1010,30 @@
"internlm2.5-latest": {
"description": "我們最新的模型系列，有著卓越的推理性能，支持 1M 的上下文長度以及更強的指令跟隨和工具調用能力。"
},
+ "internlm3-latest": {
+ "description": "我們最新的模型系列，有著卓越的推理性能，領跑同量級開源模型。默認指向我們最新發布的 InternLM3 系列模型"
+ },
+ "jina-deepsearch-v1": {
+ "description": "深度搜索結合了網路搜索、閱讀和推理，可進行全面調查。您可以將其視為一個代理，接受您的研究任務 - 它會進行廣泛搜索並經過多次迭代，然後才能給出答案。這個過程涉及持續的研究、推理和從各個角度解決問題。這與直接從預訓練數據生成答案的標準大模型以及依賴一次性表面搜索的傳統 RAG 系統有著根本的不同。"
+ },
"learnlm-1.5-pro-experimental": {
"description": "LearnLM 是一個實驗性的、特定於任務的語言模型，經過訓練以符合學習科學原則，可在教學和學習場景中遵循系統指令，充當專家導師等。"
},
"lite": {
"description": "Spark Lite 是一款輕量級大語言模型，具備極低的延遲與高效的處理能力，完全免費開放，支持即時在線搜索功能。其快速響應的特性使其在低算力設備上的推理應用和模型微調中表現出色，為用戶帶來出色的成本效益和智能體驗，尤其在知識問答、內容生成及搜索場景下表現不俗。"
},
- "llama-3.1-70b-instruct": {
- "description": "Llama 3.1 70B Instruct 模型，具備 70B 參數，能在大型文本生成和指示任務中提供卓越性能。"
- },
"llama-3.1-70b-versatile": {
"description": "Llama 3.1 70B 提供更強大的 AI 推理能力，適合複雜應用，支持超多的計算處理並保證高效和準確率。"
},
"llama-3.1-8b-instant": {
"description": "Llama 3.1 8B 是一款高效能模型，提供了快速的文本生成能力，非常適合需要大規模效率和成本效益的應用場景。"
},
- "llama-3.1-8b-instruct": {
- "description": "Llama 3.1 8B Instruct 模型，具備 8B 參數，支持畫面指示任務的高效執行，提供優質的文本生成能力。"
- },
"llama-3.1-sonar-huge-128k-online": {
"description": "Llama 3.1 Sonar Huge Online 模型，具備 405B 參數，支持約 127,000 個標記的上下文長度，設計用於複雜的在線聊天應用。"
},
- "llama-3.1-sonar-large-128k-chat": {
- "description": "Llama 3.1 Sonar Large Chat 模型，具備 70B 參數，支持約 127,000 個標記的上下文長度，適合於複雜的離線聊天任務。"
- },
"llama-3.1-sonar-large-128k-online": {
"description": "Llama 3.1 Sonar Large Online 模型，具備 70B 參數，支持約 127,000 個標記的上下文長度，適用於高容量和多樣化聊天任務。"
},
- "llama-3.1-sonar-small-128k-chat": {
- "description": "Llama 3.1 Sonar Small Chat 模型，具備 8B 參數，專為離線聊天設計，支持約 127,000 個標記的上下文長度。"
- },
"llama-3.1-sonar-small-128k-online": {
"description": "Llama 3.1 Sonar Small Online 模型，具備 8B 參數，支持約 127,000 個標記的上下文長度，專為在線聊天設計，能高效處理各種文本交互。"
},
@@ -857,6 +1049,9 @@
"llama-3.2-90b-vision-preview": {
"description": "Llama 3.2 旨在處理結合視覺和文本數據的任務。它在圖像描述和視覺問答等任務中表現出色，跨越了語言生成和視覺推理之間的鴻溝。"
},
+ "llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 是 Llama 系列最先進的多語言開源大型語言模型，以極低成本體驗媲美 405B 模型的性能。基於 Transformer 結構，並透過監督微調（SFT）和人類反饋強化學習（RLHF）提升有用性和安全性。其指令調優版本專為多語言對話優化，在多項行業基準上表現優於眾多開源和封閉聊天模型。知識截止日期為 2023 年 12 月"
+ },
"llama-3.3-70b-versatile": {
"description": "Meta Llama 3.3 多語言大語言模型 (LLM) 是 70B（文本輸入/文本輸出）中的預訓練和指令調整生成模型。Llama 3.3 指令調整的純文本模型針對多語言對話用例進行了優化，並且在常見行業基準上優於許多可用的開源和封閉式聊天模型。"
},
@@ -935,6 +1130,12 @@
"meta-llama/Llama-3.2-90B-Vision-Instruct-Turbo": {
"description": "LLaMA 3.2 旨在處理結合視覺和文本數據的任務。它在圖像描述和視覺問答等任務中表現出色，跨越了語言生成和視覺推理之間的鴻溝。"
},
+ "meta-llama/Llama-3.3-70B-Instruct": {
+ "description": "Llama 3.3 是 Llama 系列最先進的多語言開源大型語言模型，以極低成本體驗媲美 405B 模型的性能。基於 Transformer 結構，並通過監督微調（SFT）和人類反饋強化學習（RLHF）提升有用性和安全性。其指令調優版本專為多語言對話優化，在多項行業基準上表現優於眾多開源和封閉聊天模型。知識截止日期為 2023 年 12 月"
+ },
+ "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
+ "description": "Meta Llama 3.3 多語言大語言模型 ( LLM ) 是 70B（文本輸入/文本輸出）中的預訓練和指令調整生成模型。 Llama 3.3 指令調整的純文本模型針對多語言對話用例進行了優化，並且在常見行業基準上優於許多可用的開源和封閉式聊天模型。"
+ },
"meta-llama/Llama-Vision-Free": {
"description": "LLaMA 3.2 旨在處理結合視覺和文本數據的任務。它在圖像描述和視覺問答等任務中表現出色，跨越了語言生成和視覺推理之間的鴻溝。"
},
@@ -956,6 +1157,9 @@
"meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
"description": "405B 的 Llama 3.1 Turbo 模型，為大數據處理提供超大容量的上下文支持，在超大規模的人工智慧應用中表現突出。"
},
+ "meta-llama/Meta-Llama-3.1-70B": {
+ "description": "Llama 3.1 是 Meta 推出的領先模型，支持高達 405B 參數，可應用於複雜對話、多語言翻譯和數據分析領域。"
+ },
"meta-llama/Meta-Llama-3.1-70B-Instruct": {
"description": "LLaMA 3.1 70B 提供多語言的高效對話支持。"
},
@@ -992,6 +1196,12 @@
"meta-llama/llama-3.2-90b-vision-instruct": {
"description": "LLaMA 3.2 旨在處理結合視覺和文本數據的任務。它在圖像描述和視覺問答等任務中表現出色，跨越了語言生成和視覺推理之間的鴻溝。"
},
+ "meta-llama/llama-3.3-70b-instruct": {
+ "description": "Llama 3.3 是 Llama 系列最先進的多語言開源大型語言模型，以極低成本體驗媲美 405B 模型的性能。基於 Transformer 結構，並透過監督微調（SFT）和人類反饋強化學習（RLHF）提升有用性和安全性。其指令調優版本專為多語言對話優化，在多項行業基準上表現優於眾多開源和封閉聊天模型。知識截止日期為 2023 年 12 月"
+ },
+ "meta-llama/llama-3.3-70b-instruct:free": {
+ "description": "Llama 3.3 是 Llama 系列最先進的多語言開源大型語言模型，以極低成本體驗媲美 405B 模型的性能。基於 Transformer 結構，並透過監督微調（SFT）和人類反饋強化學習（RLHF）提升有用性和安全性。其指令調優版本專為多語言對話優化，在多項行業基準上表現優於眾多開源和封閉聊天模型。知識截止日期為 2023 年 12 月"
+ },
"meta.llama3-1-405b-instruct-v1:0": {
"description": "Meta Llama 3.1 405B Instruct 是 Llama 3.1 Instruct 模型中最大、最強大的模型，是一款高度先進的對話推理和合成數據生成模型，也可以用作在特定領域進行專業持續預訓練或微調的基礎。Llama 3.1 提供的多語言大型語言模型 (LLMs) 是一組預訓練的、指令調整的生成模型，包括 8B、70B 和 405B 大小 (文本輸入/輸出)。Llama 3.1 指令調整的文本模型 (8B、70B、405B) 專為多語言對話用例進行了優化，並在常見的行業基準測試中超過了許多可用的開源聊天模型。Llama 3.1 旨在用於多種語言的商業和研究用途。指令調整的文本模型適用於類似助手的聊天，而預訓練模型可以適應各種自然語言生成任務。Llama 3.1 模型還支持利用其模型的輸出來改進其他模型，包括合成數據生成和精煉。Llama 3.1 是使用優化的變壓器架構的自回歸語言模型。調整版本使用監督微調 (SFT) 和帶有人類反饋的強化學習 (RLHF) 來符合人類對幫助性和安全性的偏好。"
},
@@ -1007,6 +1217,30 @@
"meta.llama3-8b-instruct-v1:0": {
"description": "Meta Llama 3 是一款面向開發者、研究人員和企業的開放大型語言模型 (LLM)，旨在幫助他們構建、實驗並負責任地擴展他們的生成 AI 想法。作為全球社區創新的基礎系統的一部分，它非常適合計算能力和資源有限、邊緣設備和更快的訓練時間。"
},
+ "meta/llama-3.1-405b-instruct": {
+ "description": "高級 LLM，支持合成數據生成、知識蒸餾和推理，適用於聊天機器人、編程和特定領域任務。"
+ },
+ "meta/llama-3.1-70b-instruct": {
+ "description": "賦能複雜對話，具備卓越的上下文理解、推理能力和文本生成能力。"
+ },
+ "meta/llama-3.1-8b-instruct": {
+ "description": "先進的最尖端模型，具備語言理解、卓越的推理能力和文本生成能力。"
+ },
+ "meta/llama-3.2-11b-vision-instruct": {
+ "description": "尖端的視覺-語言模型，擅長從圖像中進行高品質推理。"
+ },
+ "meta/llama-3.2-1b-instruct": {
+ "description": "先進的最尖端小型語言模型，具備語言理解、卓越的推理能力和文本生成能力。"
+ },
+ "meta/llama-3.2-3b-instruct": {
+ "description": "先進的最尖端小型語言模型，具備語言理解、卓越的推理能力和文本生成能力。"
+ },
+ "meta/llama-3.2-90b-vision-instruct": {
+ "description": "尖端的視覺-語言模型，擅長從圖像中進行高品質推理。"
+ },
+ "meta/llama-3.3-70b-instruct": {
+ "description": "先進的 LLM，擅長推理、數學、常識和函數調用。"
+ },
"microsoft/WizardLM-2-8x22B": {
"description": "WizardLM 2 是微軟AI提供的語言模型，在複雜對話、多語言、推理和智能助手領域表現尤為出色。"
},
@@ -1082,33 +1316,45 @@
"moonshot-v1-128k": {
"description": "Moonshot V1 128K 是一款擁有超長上下文處理能力的模型，適用於生成超長文本，滿足複雜的生成任務需求，能夠處理多達 128,000 個 tokens 的內容，非常適合科研、學術和大型文檔生成等應用場景。"
},
+ "moonshot-v1-128k-vision-preview": {
+ "description": "Kimi 視覺模型（包括 moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview 等）能夠理解圖片內容，包括圖片文字、圖片顏色和物體形狀等內容。"
+ },
"moonshot-v1-32k": {
"description": "Moonshot V1 32K 提供中等長度的上下文處理能力，能夠處理 32,768 個 tokens，特別適合生成各種長文檔和複雜對話，應用於內容創作、報告生成和對話系統等領域。"
},
+ "moonshot-v1-32k-vision-preview": {
+ "description": "Kimi 視覺模型（包括 moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview 等）能夠理解圖片內容，包括圖片文字、圖片顏色和物體形狀等內容。"
+ },
"moonshot-v1-8k": {
"description": "Moonshot V1 8K 專為生成短文本任務設計，具有高效的處理性能，能夠處理 8,192 個 tokens，非常適合簡短對話、速記和快速內容生成。"
},
+ "moonshot-v1-8k-vision-preview": {
+ "description": "Kimi 視覺模型（包括 moonshot-v1-8k-vision-preview/moonshot-v1-32k-vision-preview/moonshot-v1-128k-vision-preview 等）能夠理解圖片內容，包括圖片文字、圖片顏色和物體形狀等內容。"
+ },
"nousresearch/hermes-2-pro-llama-3-8b": {
"description": "Hermes 2 Pro Llama 3 8B 是 Nous Hermes 2 的升級版本，包含最新的內部開發的數據集。"
},
- "nvidia/Llama-3.1-Nemotron-70B-Instruct": {
- "description": "Llama 3.1 Nemotron 70B 是由 NVIDIA 定製的大型語言模型，旨在提升 LLM 生成的回應對用戶查詢的幫助程度。"
- },
"nvidia/Llama-3.1-Nemotron-70B-Instruct-HF": {
"description": "Llama 3.1 Nemotron 70B 是由 NVIDIA 定制的大型語言模型，旨在提高 LLM 生成的回應對用戶查詢的幫助程度。該模型在 Arena Hard、AlpacaEval 2 LC 和 GPT-4-Turbo MT-Bench 等基準測試中表現出色，截至 2024 年 10 月 1 日，在所有三個自動對齊基準測試中排名第一。該模型使用 RLHF（特別是 REINFORCE）、Llama-3.1-Nemotron-70B-Reward 和 HelpSteer2-Preference 提示在 Llama-3.1-70B-Instruct 模型基礎上進行訓練"
},
+ "nvidia/llama-3.1-nemotron-51b-instruct": {
+ "description": "獨特的語言模型，提供無與倫比的準確性和效率表現。"
+ },
+ "nvidia/llama-3.1-nemotron-70b-instruct": {
+ "description": "Llama-3.1-Nemotron-70B-Instruct 是 NVIDIA 定制的大型語言模型，旨在提高 LLM 生成的響應的幫助性。"
+ },
"o1": {
"description": "專注於高級推理和解決複雜問題，包括數學和科學任務。非常適合需要深入上下文理解和代理工作流程的應用程序。"
},
- "o1-2024-12-17": {
- "description": "o1是OpenAI新的推理模型，支持圖文輸入並輸出文本，適用於需要廣泛通用知識的複雜任務。該模型具有200K上下文和2023年10月的知識截止日期。"
- },
"o1-mini": {
"description": "o1-mini是一款針對程式設計、數學和科學應用場景而設計的快速、經濟高效的推理模型。該模型具有128K上下文和2023年10月的知識截止日期。"
},
"o1-preview": {
"description": "o1是OpenAI新的推理模型，適用於需要廣泛通用知識的複雜任務。該模型具有128K上下文和2023年10月的知識截止日期。"
},
+ "o3-mini": {
+ "description": "o3-mini 是我們最新的小型推理模型，在與 o1-mini 相同的成本和延遲目標下提供高智能。"
+ },
"open-codestral-mamba": {
"description": "Codestral Mamba 是專注於代碼生成的 Mamba 2 語言模型，為先進的代碼和推理任務提供強力支持。"
},
@@ -1130,9 +1376,6 @@
"openai/gpt-4o-mini": {
"description": "GPT-4o mini是OpenAI在GPT-4 Omni之後推出的最新模型，支持圖文輸入並輸出文本。作為他們最先進的小型模型，它比其他近期的前沿模型便宜很多，並且比GPT-3.5 Turbo便宜超過60%。它保持了最先進的智能，同時具有顯著的性價比。GPT-4o mini在MMLU測試中獲得了82%的得分，目前在聊天偏好上排名高於GPT-4。"
},
- "openai/o1": {
- "description": "o1是OpenAI新的推理模型，支持圖文輸入並輸出文本，適用於需要廣泛通用知識的複雜任務。該模型具有200K上下文和2023年10月的知識截止日期。"
- },
"openai/o1-mini": {
"description": "o1-mini是一款針對程式設計、數學和科學應用場景而設計的快速、經濟高效的推理模型。該模型具有128K上下文和2023年10月的知識截止日期。"
},
@@ -1160,6 +1403,9 @@
"pro-128k": {
"description": "Spark Pro 128K 配置了特大上下文處理能力，能夠處理多達128K的上下文信息，特別適合需通篇分析和長期邏輯關聯處理的長文內容，可在複雜文本溝通中提供流暢一致的邏輯與多樣的引用支持。"
},
+ "qvq-72b-preview": {
+ "description": "QVQ模型是由 Qwen 團隊開發的實驗性研究模型，專注於提升視覺推理能力，尤其在數學推理領域。"
+ },
"qwen-coder-plus-latest": {
"description": "通義千問代碼模型。"
},
@@ -1199,6 +1445,9 @@
"qwen-vl-max-latest": {
"description": "通義千問超大規模視覺語言模型。相比增強版，再次提升視覺推理能力和指令遵循能力，提供更高的視覺感知和認知水平。"
},
+ "qwen-vl-ocr-latest": {
+ "description": "通義千問OCR是文字提取專有模型，專注於文檔、表格、試題、手寫體文字等類型圖像的文字提取能力。它能夠識別多種文字，目前支持的語言有：漢語、英語、法語、日語、韓語、德語、俄語、意大利語、越南語、阿拉伯語。"
+ },
"qwen-vl-plus-latest": {
"description": "通義千問大規模視覺語言模型增強版。大幅提升細節識別能力和文字識別能力，支持超百萬像素解析度和任意長寬比規格的圖像。"
},
@@ -1208,6 +1457,15 @@
"qwen/qwen-2-7b-instruct:free": {
"description": "Qwen2 是全新的大型語言模型系列，具有更強的理解和生成能力。"
},
+ "qwen/qwen2.5-7b-instruct": {
+ "description": "面向中文和英文的 LLM，針對語言、編程、數學、推理等領域。"
+ },
+ "qwen/qwen2.5-coder-32b-instruct": {
+ "description": "高級 LLM，支持代碼生成、推理和修復，涵蓋主流編程語言。"
+ },
+ "qwen/qwen2.5-coder-7b-instruct": {
+ "description": "強大的中型代碼模型，支持 32K 上下文長度，擅長多語言編程。"
+ },
"qwen2": {
"description": "Qwen2 是阿里巴巴的新一代大規模語言模型，以優異的性能支持多元化的應用需求。"
},
@@ -1217,6 +1475,9 @@
"qwen2.5-14b-instruct": {
"description": "通義千問2.5對外開源的14B規模的模型。"
},
+ "qwen2.5-14b-instruct-1m": {
+ "description": "通義千問2.5對外開源的72B規模的模型。"
+ },
"qwen2.5-32b-instruct": {
"description": "通義千問2.5對外開源的32B規模的模型。"
},
@@ -1244,6 +1505,12 @@
"qwen2.5-math-7b-instruct": {
"description": "Qwen-Math模型具有強大的數學解題能力。"
},
+ "qwen2.5-vl-72b-instruct": {
+ "description": "指令跟隨、數學、解題、代碼整體提升，萬物識別能力提升，支持多樣格式直接精準定位視覺元素，支持對長視頻文件（最長10分鐘）進行理解和秒級別的事件時刻定位，能理解時間先後和快慢，基於解析和定位能力支持操控OS或Mobile的Agent，關鍵信息抽取能力和Json格式輸出能力強，此版本為72B版本，本系列能力最強的版本。"
+ },
+ "qwen2.5-vl-7b-instruct": {
+ "description": "指令跟隨、數學、解題、代碼整體提升，萬物識別能力提升，支持多樣格式直接精準定位視覺元素，支持對長視頻文件（最長10分鐘）進行理解和秒級別的事件時刻定位，能理解時間先後和快慢，基於解析和定位能力支持操控OS或Mobile的Agent，關鍵信息抽取能力和Json格式輸出能力強，此版本為72B版本，本系列能力最強的版本。"
+ },
"qwen2.5:0.5b": {
"description": "Qwen2.5 是阿里巴巴的新一代大規模語言模型，以優異的性能支持多元化的應用需求。"
},
@@ -1277,6 +1544,15 @@
"solar-pro": {
"description": "Solar Pro 是 Upstage 推出的一款高智能LLM，專注於單GPU的指令跟隨能力，IFEval得分80以上。目前支持英語，正式版本計劃於2024年11月推出，將擴展語言支持和上下文長度。"
},
+ "sonar": {
+ "description": "基於搜索上下文的輕量級搜索產品，比 Sonar Pro 更快、更便宜。"
+ },
+ "sonar-pro": {
+ "description": "支持搜索上下文的高級搜索產品，支持高級查詢和跟進。"
+ },
+ "sonar-reasoning": {
+ "description": "由 DeepSeek 推理模型提供支持的新 API 產品。"
+ },
"step-1-128k": {
"description": "平衡性能與成本，適合一般場景。"
},
@@ -1295,6 +1571,9 @@
"step-1.5v-mini": {
"description": "該模型擁有強大的視頻理解能力。"
},
+ "step-1o-vision-32k": {
+ "description": "該模型擁有強大的圖像理解能力。相比於 step-1v 系列模型，擁有更強的視覺性能。"
+ },
"step-1v-32k": {
"description": "支持視覺輸入，增強多模態交互體驗。"
},
@@ -1304,18 +1583,39 @@
"step-2-16k": {
"description": "支持大規模上下文交互，適合複雜對話場景。"
},
+ "step-2-mini": {
+ "description": "基於新一代自研Attention架構MFA的極速大模型，用極低成本達到和step1類似的效果，同時保持了更高的吞吐和更快響應時延。能夠處理通用任務，在程式碼能力上具備特長。"
+ },
+ "taichu2_mm": {
+ "description": "融合了圖像理解、知識遷移、邏輯歸因等能力，在圖文問答領域表現突出"
+ },
"taichu_llm": {
"description": "紫東太初語言大模型具備超強語言理解能力以及文本創作、知識問答、代碼編程、數學計算、邏輯推理、情感分析、文本摘要等能力。創新性地將大數據預訓練與多源豐富知識相結合，通過持續打磨算法技術，並不斷吸收海量文本數據中詞彙、結構、語法、語義等方面的新知識，實現模型效果不斷進化。為用戶提供更加便捷的信息和服務以及更為智能化的體驗。"
},
+ "text-embedding-3-large": {
+ "description": "最強大的向量化模型，適用於英文和非英文任務"
+ },
+ "text-embedding-3-small": {
+ "description": "高效且經濟的新一代 Embedding 模型，適用於知識檢索、RAG 應用等場景"
+ },
"togethercomputer/StripedHyena-Nous-7B": {
"description": "StripedHyena Nous (7B) 通過高效的策略和模型架構，提供增強的計算能力。"
},
+ "tts-1": {
+ "description": "最新的文本轉語音模型，針對即時場景優化速度"
+ },
+ "tts-1-hd": {
+ "description": "最新的文本轉語音模型，針對品質進行優化"
+ },
"upstage/SOLAR-10.7B-Instruct-v1.0": {
"description": "Upstage SOLAR Instruct v1 (11B) 適用於精細化指令任務，提供出色的語言處理能力。"
},
"us.anthropic.claude-3-5-sonnet-20241022-v2:0": {
"description": "Claude 3.5 Sonnet 提升了行業標準，性能超越競爭對手模型和 Claude 3 Opus，在廣泛的評估中表現出色，同時具備我們中等層級模型的速度和成本。"
},
+ "whisper-1": {
+ "description": "通用語音識別模型，支持多語言語音識別、語音翻譯和語言識別"
+ },
"wizardlm2": {
"description": "WizardLM 2 是微軟 AI 提供的語言模型，在複雜對話、多語言、推理和智能助手領域表現尤為出色。"
},
@@ -1354,5 +1654,8 @@
},
"yi-vision": {
"description": "複雜視覺任務模型，提供高性能圖片理解、分析能力。"
+ },
+ "yi-vision-v2": {
+ "description": "複雜視覺任務模型，提供基於多張圖片的高性能理解、分析能力。"
}
}
diff --git a/locales/zh-TW/providers.json b/locales/zh-TW/providers.json
index 7803cda4fa736..26a753b452220 100644
--- a/locales/zh-TW/providers.json
+++ b/locales/zh-TW/providers.json
@@ -11,6 +11,9 @@
"azure": {
"description": "Azure 提供多種先進的 AI 模型，包括 GPT-3.5 和最新的 GPT-4 系列，支持多種數據類型和複雜任務，致力於安全、可靠和可持續的 AI 解決方案。"
},
+ "azureai": {
+ "description": "Azure 提供多種先進的 AI 模型，包括 GPT-3.5 和最新的 GPT-4 系列，支持多種數據類型和複雜任務，致力於安全、可靠和可持續的 AI 解決方案。"
+ },
"baichuan": {
"description": "百川智能是一家專注於人工智慧大模型研發的公司，其模型在國內知識百科、長文本處理和生成創作等中文任務上表現卓越，超越了國外主流模型。百川智能還具備行業領先的多模態能力，在多項權威評測中表現優異。其模型包括 Baichuan 4、Baichuan 3 Turbo 和 Baichuan 3 Turbo 128k 等，分別針對不同應用場景進行優化，提供高性價比的解決方案。"
},
@@ -23,6 +26,9 @@
"deepseek": {
"description": "DeepSeek 是一家專注於人工智慧技術研究和應用的公司，其最新模型 DeepSeek-V2.5 融合了通用對話和代碼處理能力，並在人類偏好對齊、寫作任務和指令跟隨等方面實現了顯著提升。"
},
+ "doubao": {
+ "description": "字節跳動推出的自研大模型。透過字節跳動內部50+業務場景實踐驗證，每日萬億級tokens大使用量持續打磨，提供多種模態能力，以優質模型效果為企業打造豐富的業務體驗。"
+ },
"fireworksai": {
"description": "Fireworks AI 是一家領先的高級語言模型服務商，專注於功能調用和多模態處理。其最新模型 Firefunction V2 基於 Llama-3，優化用於函數調用、對話及指令跟隨。視覺語言模型 FireLLaVA-13B 支持圖像和文本混合輸入。其他 notable 模型包括 Llama 系列和 Mixtral 系列，提供高效的多語言指令跟隨與生成支持。"
},
@@ -50,6 +56,12 @@
"internlm": {
"description": "致力於大模型研究與開發工具鏈的開源組織。為所有 AI 開發者提供高效、易用的開源平台，讓最前沿的大模型與算法技術觸手可及"
},
+ "jina": {
+ "description": "Jina AI 成立於 2020 年，是一家領先的搜索 AI 公司。我們的搜索底座平台包含了向量模型、重排器和小語言模型，可幫助企業構建可靠且高品質的生成式 AI 和多模態的搜索應用。"
+ },
+ "lmstudio": {
+ "description": "LM Studio 是一個用於在您的電腦上開發和實驗 LLMs 的桌面應用程式。"
+ },
"minimax": {
"description": "MiniMax 是 2021 年成立的通用人工智慧科技公司，致力於與用戶共創智能。MiniMax 自主研發了不同模態的通用大模型，其中包括萬億參數的 MoE 文本大模型、語音大模型以及圖像大模型。並推出了海螺 AI 等應用。"
},
@@ -62,6 +74,9 @@
"novita": {
"description": "Novita AI 是一個提供多種大語言模型與 AI 圖像生成的 API 服務的平台，靈活、可靠且具有成本效益。它支持 Llama3、Mistral 等最新的開源模型，並為生成式 AI 應用開發提供了全面、用戶友好且自動擴展的 API 解決方案，適合 AI 初創公司的快速發展。"
},
+ "nvidia": {
+ "description": "NVIDIA NIM™ 提供容器，可用於自托管 GPU 加速推理微服務，支持在雲端、數據中心、RTX™ AI 個人電腦和工作站上部署預訓練和自定義 AI 模型。"
+ },
"ollama": {
"description": "Ollama 提供的模型廣泛涵蓋代碼生成、數學運算、多語種處理和對話互動等領域，支持企業級和本地化部署的多樣化需求。"
},
@@ -92,12 +107,21 @@
"taichu": {
"description": "中科院自動化研究所和武漢人工智慧研究院推出新一代多模態大模型，支持多輪問答、文本創作、圖像生成、3D理解、信號分析等全面問答任務，擁有更強的認知、理解、創作能力，帶來全新互動體驗。"
},
+ "tencentcloud": {
+ "description": "知識引擎原子能力（LLM Knowledge Engine Atomic Power）基於知識引擎研發的知識問答全鏈路能力，面向企業及開發者，提供靈活組建及開發模型應用的能力。您可透過多款原子能力組建您專屬的模型服務，調用文檔解析、拆分、embedding、多輪改寫等服務進行組裝，定制企業專屬 AI 業務。"
+ },
"togetherai": {
"description": "Together AI 致力於透過創新的 AI 模型實現領先的性能，提供廣泛的自定義能力，包括快速擴展支持和直觀的部署流程，滿足企業的各種需求。"
},
"upstage": {
"description": "Upstage 專注於為各種商業需求開發 AI 模型，包括 Solar LLM 和文檔 AI，旨在實現工作的人工通用智能（AGI）。通過 Chat API 創建簡單的對話代理，並支持功能調用、翻譯、嵌入以及特定領域應用。"
},
+ "vllm": {
+ "description": "vLLM 是一個快速且易於使用的庫，用於 LLM 推理和服務。"
+ },
+ "volcengine": {
+ "description": "字節跳動推出的大模型服務的開發平台，提供功能豐富、安全以及具備價格競爭力的模型調用服務，同時提供模型數據、精調、推理、評測等端到端功能，全方位保障您的 AI 應用開發落地。"
+ },
"wenxin": {
"description": "企業級一站式大模型與AI原生應用開發及服務平台，提供最全面易用的生成式人工智慧模型開發、應用開發全流程工具鏈"
},
diff --git a/locales/zh-TW/setting.json b/locales/zh-TW/setting.json
index 21e35c85a6f5d..0271d271bd87f 100644
--- a/locales/zh-TW/setting.json
+++ b/locales/zh-TW/setting.json
@@ -200,9 +200,12 @@
"enableMaxTokens": {
"title": "啟用單次回覆限制"
},
+ "enableReasoningEffort": {
+ "title": "開啟推理強度調整"
+ },
"frequencyPenalty": {
- "desc": "數值越大，越有可能降低重複字詞",
- "title": "頻率懲罰度"
+ "desc": "值越大，用詞越豐富多樣；值越低，用詞更樸實簡單",
+ "title": "詞彙豐富度"
},
"maxTokens": {
"desc": "單次互動所使用的最大 Token 數",
@@ -212,19 +215,31 @@
"desc": "{{provider}} 模型",
"title": "模型"
},
+ "params": {
+ "title": "高級參數"
+ },
"presencePenalty": {
- "desc": "數值越大，越有可能擴展到新話題",
- "title": "話題新鮮度"
+ "desc": "值越大，越傾向不同的表達方式，避免概念重複；值越小，越傾向使用重複的概念或敘述，表達更具一致性",
+ "title": "表述發散度"
+ },
+ "reasoningEffort": {
+ "desc": "值越大，推理能力越強，但可能會增加回應時間和 Token 消耗",
+ "options": {
+ "high": "高",
+ "low": "低",
+ "medium": "中"
+ },
+ "title": "推理強度"
},
"temperature": {
- "desc": "數值越大，回覆越隨機",
- "title": "隨機性",
- "titleWithValue": "隨機性 {{value}}"
+ "desc": "數值越大，回答越有創意和想像力；數值越小，回答越嚴謹",
+ "title": "創意活躍度",
+ "warning": "創意活躍度數值過大，輸出可能會產生亂碼"
},
"title": "模型設定",
"topP": {
- "desc": "與隨機性類似，但不要和隨機性一起更改",
- "title": "核採樣"
+ "desc": "考慮多少種可能性，值越大，接受更多可能的回答；值越小，傾向選擇最可能的回答。不推薦和創意活躍度一起更改",
+ "title": "思維開放度"
}
},
"settingPlugin": {
@@ -411,6 +426,7 @@
"common": "通用設置",
"experiment": "實驗",
"llm": "語言模型",
+ "provider": "AI 服務商",
"sync": "雲端同步",
"system-agent": "系統助手",
"tts": "語音服務"
diff --git a/netlify.toml b/netlify.toml
index 6090fb20c00d2..0546e7e2c2ee5 100644
--- a/netlify.toml
+++ b/netlify.toml
@@ -1,9 +1,9 @@
[build]
-command = "pnpm run build"
+command = "rm -rf .next node_modules/.cache && pnpm run build"
publish = ".next"
[build.environment]
-NODE_OPTIONS = "--max_old_space_size=4096"
+NODE_OPTIONS = "--max-old-space-size=4096"
[template.environment]
OPENAI_API_KEY = "set your OpenAI API Key"
diff --git a/next.config.ts b/next.config.ts
index 35095f78efe47..cdbfcc93e7364 100644
--- a/next.config.ts
+++ b/next.config.ts
@@ -10,7 +10,6 @@ const enableReactScan = !!process.env.REACT_SCAN_MONITOR_API_KEY;
const isUsePglite = process.env.NEXT_PUBLIC_CLIENT_DB === 'pglite';
// if you need to proxy the api endpoint to remote server
-const API_PROXY_ENDPOINT = process.env.API_PROXY_ENDPOINT || '';
const basePath = process.env.NEXT_PUBLIC_BASE_PATH;
@@ -27,8 +26,8 @@ const nextConfig: NextConfig = {
'gpt-tokenizer',
],
webVitalsAttribution: ['CLS', 'LCP'],
+ webpackMemoryOptimizations: true,
},
-
async headers() {
return [
{
@@ -105,6 +104,12 @@ const nextConfig: NextConfig = {
},
];
},
+ logging: {
+ fetches: {
+ fullUrl: true,
+ hmrRefreshes: true,
+ },
+ },
output: buildWithDocker ? 'standalone' : undefined,
reactStrictMode: true,
redirects: async () => [
@@ -165,14 +170,22 @@ const nextConfig: NextConfig = {
permanent: true,
source: '/welcome',
},
+ {
+ destination: '/settings/provider/volcengine',
+ permanent: true,
+ source: '/settings/provider/doubao',
+ },
+ // we need back /repos url in the further
+ {
+ destination: '/files',
+ permanent: false,
+ source: '/repos',
+ },
],
- rewrites: async () => [
- // due to google api not work correct in some countries
- // we need a proxy to bypass the restriction
- { destination: `${API_PROXY_ENDPOINT}/api/chat/google`, source: '/api/chat/google' },
- ],
+ // when external packages in dev mode with turbopack, this config will lead to bundle error
+ serverExternalPackages: isProd ? ['@electric-sql/pglite'] : undefined,
- serverExternalPackages: ['@electric-sql/pglite'],
+ transpilePackages: ['pdfjs-dist', 'mermaid'],
webpack(config) {
config.experiments = {
diff --git a/package.json b/package.json
index 8c66144dd079f..02746c373af05 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
{
"name": "@lobehub/chat",
- "version": "1.43.3",
+ "version": "1.59.0",
"description": "Lobe Chat - an open-source, high-performance chatbot framework that supports speech synthesis, multimodal, and extensible Function Call plugin system. Supports one-click free deployment of your private ChatGPT/LLM web application.",
"keywords": [
"framework",
@@ -27,11 +27,11 @@
"sideEffects": false,
"scripts": {
"build": "next build",
+ "build:analyze": "ANALYZE=true next build",
+ "build:docker": "DOCKER=true next build && npm run build-sitemap",
"postbuild": "npm run build-sitemap && npm run build-migrate-db",
"build-migrate-db": "bun run db:migrate",
"build-sitemap": "tsx ./scripts/buildSitemapIndex/index.ts",
- "build:analyze": "ANALYZE=true next build",
- "build:docker": "DOCKER=true next build && npm run build-sitemap",
"db:generate": "drizzle-kit generate && npm run db:generate-client",
"db:generate-client": "tsx ./scripts/migrateClientDB/compile-migrations.ts",
"db:migrate": "MIGRATION_DB=1 tsx ./scripts/migrateServerDB/index.ts",
@@ -39,14 +39,14 @@
"db:push-test": "NODE_ENV=test drizzle-kit push",
"db:studio": "drizzle-kit studio",
"db:z-pull": "drizzle-kit introspect",
- "dev": "next dev --turbo -p 3010",
- "docs:i18n": "lobe-i18n md && npm run lint:mdx",
+ "dev": "next dev --turbopack -p 3010",
+ "docs:i18n": "lobe-i18n md && npm run lint:md && npm run lint:mdx",
"docs:seo": "lobe-seo && npm run lint:mdx",
"i18n": "npm run workflow:i18n && lobe-i18n",
"lint": "npm run lint:ts && npm run lint:style && npm run type-check && npm run lint:circular",
"lint:circular": "dpdm src/**/*.ts --no-warning --no-tree --exit-code circular:1 --no-progress -T true --skip-dynamic-imports circular",
- "lint:md": "remark . --quiet --frail --output",
- "lint:mdx": "npm run workflow:mdx-with-lint && prettier -c --write \"{src,docs}/**/*.mdx\" && npm run workflow:mdx-with-lint",
+ "lint:md": "remark . --silent --output",
+ "lint:mdx": "npm run workflow:mdx && remark \"docs/**/*.mdx\" -r ./.remarkrc.mdx.js --silent --output && eslint \"docs/**/*.mdx\" --quiet --fix",
"lint:style": "stylelint \"{src,tests}/**/*.{js,jsx,ts,tsx}\" --fix",
"lint:ts": "eslint \"{src,tests}/**/*.{js,jsx,ts,tsx}\" --fix",
"prepare": "husky",
@@ -59,11 +59,11 @@
"start": "next start -p 3210",
"stylelint": "stylelint \"src/**/*.{js,jsx,ts,tsx}\" --fix",
"test": "npm run test-app && npm run test-server",
+ "test:update": "vitest -u",
"test-app": "vitest run --config vitest.config.ts",
"test-app:coverage": "vitest run --config vitest.config.ts --coverage",
"test-server": "vitest run --config vitest.server.config.ts",
"test-server:coverage": "vitest run --config vitest.server.config.ts --coverage",
- "test:update": "vitest -u",
"type-check": "tsc --noEmit",
"webhook:ngrok": "ngrok http http://localhost:3011",
"workflow:cdn": "tsx ./scripts/cdnWorkflow/index.ts",
@@ -72,16 +72,15 @@
"workflow:docs": "tsx ./scripts/docsWorkflow/index.ts",
"workflow:i18n": "tsx ./scripts/i18nWorkflow/index.ts",
"workflow:mdx": "tsx ./scripts/mdxWorkflow/index.ts",
- "workflow:mdx-with-lint": "tsx ./scripts/mdxWorkflow/index.ts && eslint \"docs/**/*.mdx\" --quiet --fix",
"workflow:readme": "tsx ./scripts/readmeWorkflow/index.ts"
},
"lint-staged": {
"*.md": [
- "remark --quiet --output --",
+ "remark --silent --output --",
"prettier --write --no-error-on-unmatched-pattern"
],
"*.mdx": [
- "npm run workflow:mdx",
+ "remark -r ./.remarkrc.mdx.js --silent --output --",
"eslint --quiet --fix"
],
"*.json": [
@@ -103,224 +102,232 @@
]
},
"dependencies": {
- "@ant-design/icons": "^5.5.1",
- "@ant-design/pro-components": "^2.7.18",
- "@anthropic-ai/sdk": "^0.33.0",
- "@auth/core": "^0.37.0",
- "@aws-sdk/client-bedrock-runtime": "^3.675.0",
- "@aws-sdk/client-s3": "^3.675.0",
- "@aws-sdk/s3-request-presigner": "^3.675.0",
- "@azure/core-rest-pipeline": "1.16.0",
- "@azure/openai": "1.0.0-beta.12",
- "@baiducloud/qianfan": "^0.1.9",
- "@cfworker/json-schema": "^4.0.0",
- "@clerk/localizations": "^3.3.0",
- "@clerk/nextjs": "^6.9.6",
- "@clerk/themes": "^2.1.37",
- "@codesandbox/sandpack-react": "^2.19.9",
+ "@ant-design/icons": "^5.5.2",
+ "@ant-design/pro-components": "^2.8.3",
+ "@anthropic-ai/sdk": "^0.36.3",
+ "@auth/core": "^0.37.4",
+ "@aws-sdk/client-bedrock-runtime": "^3.723.0",
+ "@aws-sdk/client-s3": "^3.723.0",
+ "@aws-sdk/s3-request-presigner": "^3.723.0",
+ "@azure-rest/ai-inference": "1.0.0-beta.5",
+ "@azure/core-auth": "^1.9.0",
+ "@cfworker/json-schema": "^4.1.0",
+ "@clerk/localizations": "^3.9.6",
+ "@clerk/nextjs": "^6.10.6",
+ "@clerk/themes": "^2.2.4",
+ "@codesandbox/sandpack-react": "^2.19.10",
"@cyntler/react-doc-viewer": "^1.17.0",
"@electric-sql/pglite": "0.2.13",
"@google/generative-ai": "^0.21.0",
"@huggingface/inference": "^2.8.1",
"@icons-pack/react-simple-icons": "9.6.0",
- "@khmyznikov/pwa-install": "^0.3.9",
- "@langchain/community": "^0.3.0",
- "@lobehub/charts": "^1.9.12",
+ "@khmyznikov/pwa-install": "0.3.9",
+ "@langchain/community": "^0.3.22",
+ "@lobehub/charts": "^1.12.0",
"@lobehub/chat-plugin-sdk": "^1.32.4",
"@lobehub/chat-plugins-gateway": "^1.9.0",
- "@lobehub/icons": "^1.56.0",
- "@lobehub/tts": "^1.25.1",
- "@lobehub/ui": "^1.156.3",
- "@neondatabase/serverless": "^0.10.1",
- "@next/third-parties": "^15.0.0",
+ "@lobehub/icons": "^1.73.1",
+ "@lobehub/tts": "^1.28.0",
+ "@lobehub/ui": "^1.164.15",
+ "@neondatabase/serverless": "^0.10.4",
+ "@next/third-parties": "^15.1.4",
"@react-spring/web": "^9.7.5",
- "@sentry/nextjs": "^7.119.2",
- "@serwist/next": "^9.0.9",
- "@t3-oss/env-nextjs": "^0.11.1",
- "@tanstack/react-query": "^5.59.15",
+ "@sentry/nextjs": "^7.120.2",
+ "@serwist/next": "^9.0.11",
+ "@t3-oss/env-nextjs": "^0.12.0",
+ "@tanstack/react-query": "^5.62.16",
"@trpc/client": "next",
"@trpc/next": "next",
"@trpc/react-query": "next",
"@trpc/server": "next",
- "@vercel/analytics": "^1.3.1",
- "@vercel/edge-config": "^1.3.0",
- "@vercel/speed-insights": "^1.0.12",
- "ahooks": "^3.8.1",
- "ai": "^3.4.16",
- "antd": "^5.22.6",
+ "@vercel/analytics": "^1.4.1",
+ "@vercel/edge-config": "^1.4.0",
+ "@vercel/functions": "^2",
+ "@vercel/speed-insights": "^1.1.0",
+ "ahooks": "^3.8.4",
+ "ai": "^3.4.33",
+ "antd": "^5.23.0",
"antd-style": "^3.7.1",
"brotli-wasm": "^3.0.1",
"chroma-js": "^3.1.2",
+ "countries-and-timezones": "^3.7.2",
"dayjs": "^1.11.13",
- "debug": "^4.3.7",
+ "debug": "^4.4.0",
"dexie": "^3.2.7",
- "diff": "^5.2.0",
- "drizzle-orm": "^0.38.0",
+ "diff": "^7.0.0",
+ "drizzle-orm": "^0.39.0",
"drizzle-zod": "^0.5.1",
"fast-deep-equal": "^3.1.3",
"file-type": "^19.6.0",
- "framer-motion": "^11.11.9",
- "gpt-tokenizer": "^2.5.0",
- "i18next": "^23.16.1",
- "i18next-browser-languagedetector": "^7.2.1",
+ "framer-motion": "^11.16.0",
+ "gpt-tokenizer": "^2.8.1",
+ "i18next": "^24.2.1",
+ "i18next-browser-languagedetector": "^8.0.2",
"i18next-resources-to-backend": "^1.2.1",
"idb-keyval": "^6.2.1",
"immer": "^10.1.1",
- "jose": "^5.9.4",
+ "jose": "^5.9.6",
"js-sha256": "^0.11.0",
"jsonl-parse-stringify": "^1.0.3",
- "langchain": "^0.3.2",
+ "langchain": "^0.3.10",
"langfuse": "3.29.1",
"langfuse-core": "3.29.1",
"lodash-es": "^4.17.21",
- "lucide-react": "latest",
- "mammoth": "^1.8.0",
- "modern-screenshot": "^4.4.39",
- "nanoid": "^5.0.7",
- "next": "^15.1.3",
+ "lucide-react": "^0.475.0",
+ "mammoth": "^1.9.0",
+ "mdast-util-to-markdown": "^2.1.2",
+ "modern-screenshot": "^4.5.5",
+ "nanoid": "^5.0.9",
+ "next": "^15.1.4",
"next-auth": "beta",
- "next-mdx-remote": "^4.4.1",
+ "next-mdx-remote": "^5.0.0",
"nextjs-toploader": "^3.7.15",
"numeral": "^2.0.6",
"nuqs": "^1.20.0",
- "officeparser": "^4.2.0",
- "ollama": "^0.5.10",
- "openai": "^4.68.1",
+ "officeparser": "^5.1.1",
+ "ollama": "^0.5.11",
+ "openai": "^4.77.3",
"openapi-fetch": "^0.9.8",
"partial-json": "^0.1.7",
"pdf-parse": "^1.1.1",
"pdfjs-dist": "4.8.69",
- "pg": "^8.13.0",
- "pino": "^9.5.0",
+ "pg": "^8.13.1",
+ "pino": "^9.6.0",
"plaiceholder": "^3.0.0",
"polished": "^4.3.1",
- "posthog-js": "^1.174.2",
- "pwa-install-handler": "^2.6.1",
+ "posthog-js": "^1.205.0",
+ "pwa-install-handler": "^2.6.2",
"query-string": "^9.1.1",
"random-words": "^2.0.1",
"react": "^19.0.0",
- "react-confetti": "^6.1.0",
+ "react-confetti": "^6.2.2",
"react-dom": "^19.0.0",
"react-fast-marquee": "^1.6.5",
- "react-hotkeys-hook": "^4.5.1",
- "react-i18next": "14.0.2",
- "react-layout-kit": "^1.9.0",
+ "react-hotkeys-hook": "^4.6.1",
+ "react-i18next": "^15.4.0",
+ "react-layout-kit": "^1.9.1",
"react-lazy-load": "^4.0.1",
"react-pdf": "^9.2.1",
- "react-scan": "^0.0.51",
- "react-virtuoso": "^4.12.0",
+ "react-rnd": "^10.4.14",
+ "react-scan": "^0.1.0",
+ "react-virtuoso": "^4.12.3",
"react-wrap-balancer": "^1.1.1",
- "remark": "^14.0.3",
- "remark-gfm": "^3.0.1",
- "remark-html": "^15.0.2",
+ "remark": "^15.0.1",
+ "remark-gfm": "^4.0.0",
+ "remark-html": "^16.0.1",
"request-filtering-agent": "^2.0.1",
- "resolve-accept-language": "^3.1.8",
+ "resolve-accept-language": "^3.1.10",
"rtl-detect": "^1.1.2",
"semver": "^7.6.3",
"sharp": "^0.33.5",
- "shiki": "^1.22.0",
- "stripe": "^15.12.0",
- "superjson": "^2.2.1",
- "svix": "^1.37.0",
- "swr": "^2.2.5",
+ "shiki": "^1.26.1",
+ "stripe": "^16.12.0",
+ "superjson": "^2.2.2",
+ "svix": "^1.45.1",
+ "swr": "^2.3.0",
"systemjs": "^6.15.1",
"ts-md5": "^1.3.1",
- "ua-parser-js": "^1.0.39",
- "unstructured-client": "^0.18.1",
+ "ua-parser-js": "^1.0.40",
+ "unstructured-client": "^0.19.0",
"url-join": "^5.0.0",
"use-merge-value": "^1.2.0",
"utility-types": "^3.11.0",
- "uuid": "^11.0.0",
+ "uuid": "^11.0.4",
"ws": "^8.18.0",
"y-protocols": "^1.0.6",
"y-webrtc": "^10.3.0",
- "yaml": "^2.6.0",
- "yjs": "^13.6.20",
- "zod": "^3.23.8",
- "zustand": "5.0.1",
- "zustand-utils": "^1.3.2"
+ "yaml": "^2.7.0",
+ "yjs": "^13.6.21",
+ "zod": "^3.24.1",
+ "zustand": "^5.0.3",
+ "zustand-utils": "^2"
},
"devDependencies": {
- "@commitlint/cli": "^19.5.0",
+ "@commitlint/cli": "^19.6.1",
"@edge-runtime/vm": "^5.0.0",
- "@huggingface/tasks": "^0.12.22",
+ "@huggingface/tasks": "^0.15.0",
"@lobehub/i18n-cli": "^1.20.3",
- "@lobehub/lint": "^1.24.4",
- "@lobehub/seo-cli": "^1.4.2",
- "@next/bundle-analyzer": "^15.0.0",
- "@next/eslint-plugin-next": "^15.0.0",
+ "@lobehub/lint": "^1.25.5",
+ "@lobehub/seo-cli": "^1.4.3",
+ "@next/bundle-analyzer": "^15.1.4",
+ "@next/eslint-plugin-next": "^15.1.4",
"@peculiar/webcrypto": "^1.5.0",
"@semantic-release/exec": "^6.0.3",
- "@testing-library/jest-dom": "^6.6.2",
- "@testing-library/react": "^16.0.1",
- "@types/chroma-js": "^2.4.4",
+ "@testing-library/jest-dom": "^6.6.3",
+ "@testing-library/react": "^16.1.0",
+ "@testing-library/user-event": "^14.6.1",
+ "@types/chroma-js": "^3.1.0",
"@types/crypto-js": "^4.2.2",
"@types/debug": "^4.1.12",
- "@types/diff": "^6.0.0",
+ "@types/diff": "^7.0.0",
"@types/fs-extra": "^11.0.4",
"@types/ip": "^1.1.3",
"@types/json-schema": "^7.0.15",
- "@types/lodash": "^4.17.12",
+ "@types/lodash": "^4.17.14",
"@types/lodash-es": "^4.17.12",
- "@types/node": "^20.16.13",
+ "@types/node": "^22.10.5",
"@types/numeral": "^2.0.5",
"@types/pg": "^8.11.10",
- "@types/react": "18.3.13",
- "@types/react-dom": "^19.0.0",
+ "@types/react": "^19.0.3",
+ "@types/react-dom": "^19.0.2",
"@types/rtl-detect": "^1.0.3",
"@types/semver": "^7.5.8",
"@types/systemjs": "^6.15.1",
"@types/ua-parser-js": "^0.7.39",
"@types/unist": "^3.0.3",
"@types/uuid": "^10.0.0",
- "@types/ws": "^8.5.12",
+ "@types/ws": "^8.5.13",
"@vitest/coverage-v8": "~1.2.2",
"ajv-keywords": "^5.1.0",
- "commitlint": "^19.5.0",
- "consola": "^3.2.3",
+ "commitlint": "^19.6.1",
+ "consola": "^3.3.3",
"crypto-js": "^4.2.0",
- "dotenv": "^16.4.5",
- "dpdm-fast": "^1.0.4",
- "drizzle-kit": "^0.30.0",
+ "dotenv": "^16.4.7",
+ "dpdm-fast": "^1.0.7",
+ "drizzle-kit": "^0.30.1",
"eslint": "^8.57.1",
- "eslint-plugin-mdx": "^2.3.4",
- "eslint-plugin-unused-imports": "4.0.1",
+ "eslint-plugin-mdx": "^3.1.5",
"fake-indexeddb": "^6.0.0",
"fs-extra": "^11.2.0",
"glob": "^11.0.0",
"gray-matter": "^4.0.3",
- "happy-dom": "^15.7.4",
- "husky": "^9.1.6",
+ "happy-dom": "^17.0.0",
+ "husky": "^9.1.7",
"just-diff": "^6.0.2",
- "lint-staged": "^15.2.10",
+ "lint-staged": "^15.3.0",
"lodash": "^4.17.21",
- "markdown-table": "^3.0.3",
+ "markdown-table": "^3.0.4",
"markdown-to-txt": "^2.0.1",
- "mime": "^4.0.4",
+ "mime": "^4.0.6",
"node-fetch": "^3.3.2",
"node-gyp": "^11.0.0",
- "openapi-typescript": "^6.7.6",
- "p-map": "^7.0.2",
- "prettier": "^3.3.3",
- "remark-cli": "^11.0.0",
- "remark-parse": "^10.0.2",
+ "openapi-typescript": "^7.5.2",
+ "p-map": "^7.0.3",
+ "prettier": "^3.4.2",
+ "remark-cli": "^12.0.1",
+ "remark-frontmatter": "^5.0.0",
+ "remark-mdx": "^3.1.0",
+ "remark-parse": "^11.0.0",
"semantic-release": "^21.1.2",
- "serwist": "^9.0.9",
+ "serwist": "^9.0.11",
"stylelint": "^15.11.0",
- "tsx": "^4.19.1",
- "typescript": "^5.6.3",
+ "tsx": "^4.19.2",
+ "typescript": "^5.7.2",
"unified": "^11.0.5",
"unist-util-visit": "^5.0.0",
- "vite": "^5.4.9",
+ "vite": "^5.4.11",
"vitest": "~1.2.2",
"vitest-canvas-mock": "^0.3.3"
},
- "packageManager": "pnpm@9.15.2",
+ "packageManager": "pnpm@9.15.5",
"publishConfig": {
"access": "public",
"registry": "https://registry.npmjs.org"
},
"pnpm": {
+ "overrides": {
+ "mdast-util-gfm-autolink-literal": "2.0.0"
+ },
"packageExtensions": {
"@inkjs/ui": {
"dependencies": {
@@ -328,5 +335,8 @@
}
}
}
+ },
+ "overrides": {
+ "mdast-util-gfm-autolink-literal": "2.0.0"
}
}
diff --git a/scripts/mdxWorkflow/index.ts b/scripts/mdxWorkflow/index.ts
index a3e40ce6d6c77..36b7970e78e15 100644
--- a/scripts/mdxWorkflow/index.ts
+++ b/scripts/mdxWorkflow/index.ts
@@ -37,6 +37,13 @@ const run = () => {
.replaceAll(` \\*\\*`, ' **')
.replaceAll(/\n{2,}/g, '\n\n');
+ if (!data?.title) {
+ const regex = /^#\s(.+)/;
+ const match = regex.exec(formatedContent.trim());
+ const title = match ? match[1] : '';
+ data.title = title;
+ }
+
writeFileSync(post, matter.stringify(formatedContent, data));
} catch (error) {
consola.error(post);
diff --git a/scripts/readmeWorkflow/syncAgentIndex.ts b/scripts/readmeWorkflow/syncAgentIndex.ts
index a46422adea8f2..3656aeb4b5fa4 100644
--- a/scripts/readmeWorkflow/syncAgentIndex.ts
+++ b/scripts/readmeWorkflow/syncAgentIndex.ts
@@ -51,5 +51,4 @@ const runAgentTable = async (lang?: string) => {
export default async () => {
await runAgentTable();
await runAgentTable('zh-CN');
- await runAgentTable('ja-JP');
};
diff --git a/scripts/readmeWorkflow/syncPluginIndex.ts b/scripts/readmeWorkflow/syncPluginIndex.ts
index 53a35ce250673..0d8997d8d5013 100644
--- a/scripts/readmeWorkflow/syncPluginIndex.ts
+++ b/scripts/readmeWorkflow/syncPluginIndex.ts
@@ -51,5 +51,4 @@ const runPluginTable = async (lang?: string) => {
export default async () => {
await runPluginTable();
await runPluginTable('zh-CN');
- await runPluginTable('ja-JP');
};
diff --git a/scripts/readmeWorkflow/syncProviderIndex.ts b/scripts/readmeWorkflow/syncProviderIndex.ts
index c9f789d404aa5..b038dc3a8ff11 100644
--- a/scripts/readmeWorkflow/syncProviderIndex.ts
+++ b/scripts/readmeWorkflow/syncProviderIndex.ts
@@ -47,5 +47,4 @@ const runProviderTable = async (lang?: string) => {
export default async () => {
await runProviderTable();
await runProviderTable('zh-CN');
- await runProviderTable('ja-JP');
};
diff --git a/scripts/serverLauncher/startServer.js b/scripts/serverLauncher/startServer.js
index 3d7ddda2dfbdc..a7ec5241723b2 100644
--- a/scripts/serverLauncher/startServer.js
+++ b/scripts/serverLauncher/startServer.js
@@ -1,6 +1,6 @@
-const dns = require('dns').promises;
-const fs = require('fs').promises;
-const { spawn } = require('child_process');
+const dns = require('node:dns').promises;
+const fs = require('node:fs').promises;
+const { spawn } = require('node:child_process');
// Set file paths
const DB_MIGRATION_SCRIPT_PATH = '/app/docker.cjs';
@@ -9,23 +9,28 @@ const PROXYCHAINS_CONF_PATH = '/etc/proxychains4.conf';
// Function to check if a string is a valid IP address
const isValidIP = (ip, version = 4) => {
- const ipv4Regex = /^(25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)(\.(25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)){3}$/;
- const ipv6Regex = /^(([0-9a-f]{1,4}:){7,7}[0-9a-f]{1,4}|([0-9a-f]{1,4}:){1,7}:|([0-9a-f]{1,4}:){1,6}:[0-9a-f]{1,4}|([0-9a-f]{1,4}:){1,5}(:[0-9a-f]{1,4}){1,2}|([0-9a-f]{1,4}:){1,4}(:[0-9a-f]{1,4}){1,3}|([0-9a-f]{1,4}:){1,3}(:[0-9a-f]{1,4}){1,4}|([0-9a-f]{1,4}:){1,2}(:[0-9a-f]{1,4}){1,5}|[0-9a-f]{1,4}:((:[0-9a-f]{1,4}){1,6})|:((:[0-9a-f]{1,4}){1,7}|:)|fe80:(:[0-9a-f]{0,4}){0,4}%[0-9a-z]{1,}|::(ffff(:0{1,4}){0,1}:){0,1}((25[0-5]|(2[0-4]|1{0,1}[0-9]){0,1}[0-9])\.){3,3}(25[0-5]|(2[0-4]|1{0,1}[0-9]){0,1}[0-9])|([0-9a-f]{1,4}:){1,4}:((25[0-5]|(2[0-4]|1{0,1}[0-9]){0,1}[0-9])\.){3,3}(25[0-5]|(2[0-4]|1{0,1}[0-9]){0,1}[0-9]))$/;
+ const ipv4Regex =
+ /^(25[0-5]|2[0-4]\d|[01]?\d{1,2})(\.(25[0-5]|2[0-4]\d|[01]?\d{1,2})){3}$/;
+ const ipv6Regex =
+ /^(([\da-f]{1,4}:){7}[\da-f]{1,4}|([\da-f]{1,4}:){1,7}:|([\da-f]{1,4}:){1,6}:[\da-f]{1,4}|([\da-f]{1,4}:){1,5}(:[\da-f]{1,4}){1,2}|([\da-f]{1,4}:){1,4}(:[\da-f]{1,4}){1,3}|([\da-f]{1,4}:){1,3}(:[\da-f]{1,4}){1,4}|([\da-f]{1,4}:){1,2}(:[\da-f]{1,4}){1,5}|[\da-f]{1,4}:((:[\da-f]{1,4}){1,6})|:((:[\da-f]{1,4}){1,7}|:)|fe80:(:[\da-f]{0,4}){0,4}%[\da-z]+|::(ffff(:0{1,4}){0,1}:){0,1}((25[0-5]|(2[0-4]|1{0,1}\d){0,1}\d)\.){3}(25[0-5]|(2[0-4]|1{0,1}\d){0,1}\d)|([\da-f]{1,4}:){1,4}:((25[0-5]|(2[0-4]|1{0,1}\d){0,1}\d)\.){3}(25[0-5]|(2[0-4]|1{0,1}\d){0,1}\d))$/;
switch (version) {
- case 4:
+ case 4: {
return ipv4Regex.test(ip);
- case 6:
+ }
+ case 6: {
return ipv6Regex.test(ip);
- default:
+ }
+ default: {
return ipv4Regex.test(ip) || ipv6Regex.test(ip);
+ }
}
};
// Function to parse protocol, host and port from a URL
const parseUrl = (url) => {
- const { protocol, hostname: host, port } = new URL(url);
- return { protocol: protocol.replace(':', ''), host, port: port || 443 };
+ const { protocol, hostname: host, port, username: user, password: pass } = new URL(url);
+ return { host, pass, port: port || 443, protocol: protocol.replace(':', ''), user };
};
// Function to resolve host IP via DNS
@@ -34,7 +39,9 @@ const resolveHostIP = async (host, version = 4) => {
const { address } = await dns.lookup(host, { family: version });
if (!isValidIP(address, version)) {
- console.error(`❌ DNS Error: Invalid resolved IP: ${address}. IP address must be IPv${version}.`);
+ console.error(
+ `❌ DNS Error: Invalid resolved IP: ${address}. IP address must be IPv${version}.`,
+ );
process.exit(1);
}
@@ -48,16 +55,20 @@ const resolveHostIP = async (host, version = 4) => {
// Function to generate proxychains configuration
const runProxyChainsConfGenerator = async (url) => {
- const { protocol, host, port } = parseUrl(url);
+ const { protocol, host, port, user, pass } = parseUrl(url);
if (!['http', 'socks4', 'socks5'].includes(protocol)) {
- console.error(`❌ ProxyChains: Invalid protocol (${protocol}). Protocol must be 'http', 'socks4' and 'socks5'.`);
+ console.error(
+ `❌ ProxyChains: Invalid protocol (${protocol}). Protocol must be 'http', 'socks4' and 'socks5'.`,
+ );
process.exit(1);
}
const validPort = parseInt(port, 10);
- if (isNaN(validPort) || validPort <= 0 || validPort > 65535) {
- console.error(`❌ ProxyChains: Invalid port (${port}). Port must be a number between 1 and 65535.`);
+ if (isNaN(validPort) || validPort <= 0 || validPort > 65_535) {
+ console.error(
+ `❌ ProxyChains: Invalid port (${port}). Port must be a number between 1 and 65535.`,
+ );
process.exit(1);
}
@@ -65,6 +76,9 @@ const runProxyChainsConfGenerator = async (url) => {
const configContent = `
localnet 127.0.0.0/255.0.0.0
+localnet 10.0.0.0/255.0.0.0
+localnet 172.16.0.0/255.240.0.0
+localnet 192.168.0.0/255.255.0.0
localnet ::1/128
proxy_dns
remote_dns_subnet 224
@@ -72,20 +86,24 @@ strict_chain
tcp_connect_time_out 8000
tcp_read_time_out 15000
[ProxyList]
-${protocol} ${ip} ${port}
+${protocol} ${ip} ${port} ${user} ${pass}
`.trim();
await fs.writeFile(PROXYCHAINS_CONF_PATH, configContent);
- console.log(`✅ ProxyChains: All outgoing traffic routed via ${protocol}://${ip}:${port}.`);
+ console.log(`✅ ProxyChains: All outgoing traffic routed via ${url}.`);
console.log('-------------------------------------');
};
// Function to execute a script with child process spawn
const runScript = (scriptPath, useProxy = false) => {
- const command = useProxy ? ['/bin/proxychains', '-q', '/bin/node', scriptPath] : ['/bin/node', scriptPath];
+ const command = useProxy
+ ? ['/bin/proxychains', '-q', '/bin/node', scriptPath]
+ : ['/bin/node', scriptPath];
return new Promise((resolve, reject) => {
const process = spawn(command.shift(), command, { stdio: 'inherit' });
- process.on('close', (code) => (code === 0 ? resolve() : reject(new Error(`🔴 Process exited with code ${code}`))));
+ process.on('close', (code) =>
+ code === 0 ? resolve() : reject(new Error(`🔴 Process exited with code ${code}`)),
+ );
});
};
@@ -112,7 +130,9 @@ const runServer = async () => {
await runScript(DB_MIGRATION_SCRIPT_PATH);
} catch (err) {
if (err.code === 'ENOENT') {
- console.log(`⚠️ DB Migration: Not found ${DB_MIGRATION_SCRIPT_PATH}. Skipping DB migration. Ensure to migrate database manually.`);
+ console.log(
+ `⚠️ DB Migration: Not found ${DB_MIGRATION_SCRIPT_PATH}. Skipping DB migration. Ensure to migrate database manually.`,
+ );
console.log('-------------------------------------');
} else {
console.error('❌ Error during DB migration:');
diff --git a/scripts/vercelIgnoredBuildStep.js b/scripts/vercelIgnoredBuildStep.js
new file mode 100644
index 0000000000000..35f50038c37ec
--- /dev/null
+++ b/scripts/vercelIgnoredBuildStep.js
@@ -0,0 +1,41 @@
+const { execSync } = require('node:child_process');
+
+// 获取当前分支名
+const branchName = process.env.VERCEL_GIT_COMMIT_REF || '';
+
+function shouldProceedBuild() {
+ // 如果是 lighthouse 分支或以 testgru 开头的分支，取消构建
+ if (branchName === 'lighthouse' || branchName.startsWith('gru/')) {
+ return false;
+ }
+
+ try {
+ // 检查文件变更，排除特定文件和目录
+ const diffCommand =
+ 'git diff HEAD^ HEAD --quiet -- \
+ ":!./*.md" \
+ ":!./Dockerfile" \
+ ":!./.github" \
+ ":!./.husky" \
+ ":!./scripts"';
+
+ execSync(diffCommand);
+
+ return false;
+ } catch {
+ return true;
+ }
+}
+
+const shouldBuild = shouldProceedBuild();
+
+console.log('shouldBuild:', shouldBuild);
+if (shouldBuild) {
+ console.log('✅ - Build can proceed');
+ // eslint-disable-next-line unicorn/no-process-exit
+ process.exit(1);
+} else {
+ console.log('🛑 - Build cancelled');
+ // eslint-disable-next-line unicorn/no-process-exit
+ process.exit(0);
+}
diff --git a/src/app/(backend)/api/auth/error/page.tsx b/src/app/(backend)/api/auth/error/page.tsx
deleted file mode 100644
index f2173c573d1f7..0000000000000
--- a/src/app/(backend)/api/auth/error/page.tsx
+++ /dev/null
@@ -1,5 +0,0 @@
-'use client';
-
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('./AuthErrorPage'));
diff --git a/src/app/(backend)/webapi/chat/minimax/route.test.ts b/src/app/(backend)/webapi/chat/minimax/route.test.ts
deleted file mode 100644
index 1ae4933ba2a4f..0000000000000
--- a/src/app/(backend)/webapi/chat/minimax/route.test.ts
+++ /dev/null
@@ -1,26 +0,0 @@
-// @vitest-environment edge-runtime
-import { describe, expect, it, vi } from 'vitest';
-
-import { POST as UniverseRoute } from '../[provider]/route';
-import { POST, runtime } from './route';
-
-// 模拟 '../[provider]/route'
-vi.mock('../[provider]/route', () => ({
- POST: vi.fn().mockResolvedValue('mocked response'),
-}));
-
-describe('Configuration tests', () => {
- it('should have runtime set to "edge"', () => {
- expect(runtime).toBe('nodejs');
- });
-});
-
-describe('Minimax POST function tests', () => {
- it('should call UniverseRoute with correct parameters', async () => {
- const mockRequest = new Request('https://example.com', { method: 'POST' });
- await POST(mockRequest);
- expect(UniverseRoute).toHaveBeenCalledWith(mockRequest, {
- params: Promise.resolve({ provider: 'minimax' }),
- });
- });
-});
diff --git a/src/app/(backend)/webapi/chat/minimax/route.ts b/src/app/(backend)/webapi/chat/minimax/route.ts
deleted file mode 100644
index d3eafa591d9ee..0000000000000
--- a/src/app/(backend)/webapi/chat/minimax/route.ts
+++ /dev/null
@@ -1,6 +0,0 @@
-import { POST as UniverseRoute } from '../[provider]/route';
-
-export const runtime = 'nodejs';
-
-export const POST = async (req: Request) =>
- UniverseRoute(req, { params: Promise.resolve({ provider: 'minimax' }) });
diff --git a/src/app/(backend)/webapi/chat/models/[provider]/route.ts b/src/app/(backend)/webapi/chat/models/[provider]/route.ts
index 1f6c4e5b24987..b99a05f5efa0f 100644
--- a/src/app/(backend)/webapi/chat/models/[provider]/route.ts
+++ b/src/app/(backend)/webapi/chat/models/[provider]/route.ts
@@ -9,7 +9,7 @@ import { createErrorResponse } from '@/utils/errorResponse';
export const runtime = 'edge';
const noNeedAPIKey = (provider: string) =>
- [ModelProvider.OpenRouter, ModelProvider.TogetherAI].includes(provider as any);
+ [ModelProvider.OpenRouter].includes(provider as any);
export const GET = checkAuth(async (req, { params, jwtPayload }) => {
const { provider } = await params;
diff --git a/src/app/(backend)/webapi/chat/wenxin/route.test.ts b/src/app/(backend)/webapi/chat/wenxin/route.test.ts
deleted file mode 100644
index cb25976b70fd4..0000000000000
--- a/src/app/(backend)/webapi/chat/wenxin/route.test.ts
+++ /dev/null
@@ -1,27 +0,0 @@
-// @vitest-environment edge-runtime
-import { describe, expect, it, vi } from 'vitest';
-
-import { POST as UniverseRoute } from '../[provider]/route';
-import { POST, runtime } from './route';
-
-// 模拟 '../[provider]/route'
-vi.mock('../[provider]/route', () => ({
- POST: vi.fn().mockResolvedValue('mocked response'),
-}));
-
-describe('Configuration tests', () => {
- it('should have runtime set to "edge"', () => {
- expect(runtime).toBe('nodejs');
- });
-});
-
-describe('Wenxin POST function tests', () => {
- it('should call UniverseRoute with correct parameters', async () => {
- const mockRequest = new Request('https://example.com', { method: 'POST' });
- await POST(mockRequest);
- expect(UniverseRoute).toHaveBeenCalledWith(mockRequest, {
- createRuntime: expect.anything(),
- params: Promise.resolve({ provider: 'wenxin' }),
- });
- });
-});
diff --git a/src/app/(backend)/webapi/chat/wenxin/route.ts b/src/app/(backend)/webapi/chat/wenxin/route.ts
deleted file mode 100644
index b048f62e5ec4b..0000000000000
--- a/src/app/(backend)/webapi/chat/wenxin/route.ts
+++ /dev/null
@@ -1,30 +0,0 @@
-import { getLLMConfig } from '@/config/llm';
-import { AgentRuntime, ModelProvider } from '@/libs/agent-runtime';
-import LobeWenxinAI from '@/libs/agent-runtime/wenxin';
-
-import { POST as UniverseRoute } from '../[provider]/route';
-
-export const runtime = 'nodejs';
-
-export const maxDuration = 30;
-
-export const POST = async (req: Request) =>
- UniverseRoute(req, {
- createRuntime: (payload) => {
- const { WENXIN_ACCESS_KEY, WENXIN_SECRET_KEY } = getLLMConfig();
- let accessKey: string | undefined = WENXIN_ACCESS_KEY;
- let secretKey: string | undefined = WENXIN_SECRET_KEY;
-
- // if the payload has the api key, use user
- if (payload.apiKey) {
- accessKey = payload?.wenxinAccessKey;
- secretKey = payload?.wenxinSecretKey;
- }
-
- const params = { accessKey, secretKey };
- const instance = new LobeWenxinAI(params);
-
- return new AgentRuntime(instance);
- },
- params: Promise.resolve({ provider: ModelProvider.Wenxin }),
- });
diff --git a/src/app/(backend)/webapi/revalidate/route.ts b/src/app/(backend)/webapi/revalidate/route.ts
new file mode 100644
index 0000000000000..3f7bf3e4f7123
--- /dev/null
+++ b/src/app/(backend)/webapi/revalidate/route.ts
@@ -0,0 +1,24 @@
+import { revalidateTag } from 'next/cache';
+import { NextRequest, NextResponse } from 'next/server';
+
+export const GET = async (request: NextRequest) => {
+ if (!process.env.REVALIDATE_SECRET) {
+ return NextResponse.json('REVALIDATE_SECRET is not set', { status: 501 });
+ }
+
+ const authToken = request.headers.get('Authorization');
+
+ if (!authToken || authToken !== process.env.REVALIDATE_SECRET) {
+ return NextResponse.json('Unauthorized', { status: 401 });
+ }
+
+ const tag = request.nextUrl.searchParams.get('tag');
+
+ if (!tag) {
+ return NextResponse.json('tag query parameter is required', { status: 400 });
+ }
+
+ revalidateTag(tag);
+
+ return Response.json({ now: Date.now(), revalidated: true });
+};
diff --git a/src/app/(main)/(mobile)/me/profile/page.tsx b/src/app/(main)/(mobile)/me/profile/page.tsx
deleted file mode 100644
index 32f1f57704761..0000000000000
--- a/src/app/(main)/(mobile)/me/profile/page.tsx
+++ /dev/null
@@ -1,28 +0,0 @@
-import { redirect } from 'next/navigation';
-
-import { metadataModule } from '@/server/metadata';
-import { translation } from '@/server/translation';
-import { isMobileDevice } from '@/utils/server/responsive';
-
-import Category from './features/Category';
-
-export const generateMetadata = async () => {
- const { t } = await translation('auth');
- return metadataModule.generate({
- description: t('header.desc'),
- title: t('header.title'),
- url: '/me/profile',
- });
-};
-
-const Page = async () => {
- const mobile = await isMobileDevice();
-
- if (!mobile) return redirect('/profile');
-
- return ;
-};
-
-Page.displayName = 'MeProfile';
-
-export default Page;
diff --git a/src/app/(main)/(mobile)/me/settings/page.tsx b/src/app/(main)/(mobile)/me/settings/page.tsx
deleted file mode 100644
index 794ac05835299..0000000000000
--- a/src/app/(main)/(mobile)/me/settings/page.tsx
+++ /dev/null
@@ -1,28 +0,0 @@
-import { redirect } from 'next/navigation';
-
-import { metadataModule } from '@/server/metadata';
-import { translation } from '@/server/translation';
-import { isMobileDevice } from '@/utils/server/responsive';
-
-import Category from './features/Category';
-
-export const generateMetadata = async () => {
- const { t } = await translation('setting');
- return metadataModule.generate({
- description: t('header.desc'),
- title: t('header.title'),
- url: '/me/settings',
- });
-};
-
-const Page = async () => {
- const mobile = await isMobileDevice();
-
- if (!mobile) return redirect('/settings/common');
-
- return ;
-};
-
-Page.displayName = 'MeSettings';
-
-export default Page;
diff --git a/src/app/(main)/@nav/default.tsx b/src/app/(main)/@nav/default.tsx
deleted file mode 100644
index 452ebffe79d6a..0000000000000
--- a/src/app/(main)/@nav/default.tsx
+++ /dev/null
@@ -1,10 +0,0 @@
-import ServerLayout from '@/components/server/ServerLayout';
-
-import Desktop from './_layout/Desktop';
-import Mobile from './_layout/Mobile';
-
-const Nav = ServerLayout({ Desktop, Mobile });
-
-Nav.displayName = 'Nav';
-
-export default Nav;
diff --git a/src/app/(main)/_layout/type.ts b/src/app/(main)/_layout/type.ts
deleted file mode 100644
index 49de1b261e147..0000000000000
--- a/src/app/(main)/_layout/type.ts
+++ /dev/null
@@ -1,6 +0,0 @@
-import { ReactNode } from 'react';
-
-export interface LayoutProps {
- children: ReactNode;
- nav: ReactNode;
-}
diff --git a/src/app/(main)/changelog/error.tsx b/src/app/(main)/changelog/error.tsx
deleted file mode 100644
index 071491038c704..0000000000000
--- a/src/app/(main)/changelog/error.tsx
+++ /dev/null
@@ -1,5 +0,0 @@
-'use client';
-
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/Error'));
diff --git a/src/app/(main)/changelog/not-found.tsx b/src/app/(main)/changelog/not-found.tsx
deleted file mode 100644
index 02503bc7fa461..0000000000000
--- a/src/app/(main)/changelog/not-found.tsx
+++ /dev/null
@@ -1,3 +0,0 @@
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/404'));
diff --git a/src/app/(main)/chat/(workspace)/@portal/error.tsx b/src/app/(main)/chat/(workspace)/@portal/error.tsx
deleted file mode 100644
index 071491038c704..0000000000000
--- a/src/app/(main)/chat/(workspace)/@portal/error.tsx
+++ /dev/null
@@ -1,5 +0,0 @@
-'use client';
-
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/Error'));
diff --git a/src/app/(main)/chat/error.tsx b/src/app/(main)/chat/error.tsx
deleted file mode 100644
index 071491038c704..0000000000000
--- a/src/app/(main)/chat/error.tsx
+++ /dev/null
@@ -1,5 +0,0 @@
-'use client';
-
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/Error'));
diff --git a/src/app/(main)/chat/not-found.tsx b/src/app/(main)/chat/not-found.tsx
deleted file mode 100644
index 02503bc7fa461..0000000000000
--- a/src/app/(main)/chat/not-found.tsx
+++ /dev/null
@@ -1,3 +0,0 @@
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/404'));
diff --git a/src/app/(main)/chat/settings/error.tsx b/src/app/(main)/chat/settings/error.tsx
deleted file mode 100644
index 071491038c704..0000000000000
--- a/src/app/(main)/chat/settings/error.tsx
+++ /dev/null
@@ -1,5 +0,0 @@
-'use client';
-
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/Error'));
diff --git a/src/app/(main)/chat/settings/features/EditPage.tsx b/src/app/(main)/chat/settings/features/EditPage.tsx
deleted file mode 100644
index c844f28c11a0e..0000000000000
--- a/src/app/(main)/chat/settings/features/EditPage.tsx
+++ /dev/null
@@ -1,45 +0,0 @@
-'use client';
-
-import isEqual from 'fast-deep-equal';
-import { memo } from 'react';
-import { useTranslation } from 'react-i18next';
-
-import PageTitle from '@/components/PageTitle';
-import { AgentSettings } from '@/features/AgentSetting';
-import { useAgentStore } from '@/store/agent';
-import { agentSelectors } from '@/store/agent/selectors';
-import { useSessionStore } from '@/store/session';
-import { sessionMetaSelectors } from '@/store/session/selectors';
-
-const EditPage = memo(() => {
- const { t } = useTranslation('setting');
- const id = useSessionStore((s) => s.activeId);
- const config = useAgentStore(agentSelectors.currentAgentConfig, isEqual);
- const meta = useSessionStore(sessionMetaSelectors.currentAgentMeta, isEqual);
- const [useFetchAgentConfig, updateAgentConfig] = useAgentStore((s) => [
- s.useFetchAgentConfig,
- s.updateAgentConfig,
- ]);
-
- const [updateAgentMeta, title] = useSessionStore((s) => [
- s.updateSessionMeta,
- sessionMetaSelectors.currentAgentTitle(s),
- ]);
-
- useFetchAgentConfig(id);
-
- return (
- <>
-
-
-
- );
-});
-
-export default EditPage;
diff --git a/src/app/(main)/chat/settings/modal/page.tsx b/src/app/(main)/chat/settings/modal/page.tsx
deleted file mode 100644
index 8863499c1fab7..0000000000000
--- a/src/app/(main)/chat/settings/modal/page.tsx
+++ /dev/null
@@ -1,23 +0,0 @@
-'use client';
-
-import { useLayoutEffect } from 'react';
-
-import { useQueryRoute } from '@/hooks/useQueryRoute';
-
-/**
- * @description: Chat Settings Modal (intercepting routes fallback when hard refresh)
- * @example: /chat/settings/modal?tab=prompt => /chat/settings
- * @refs: https://github.com/lobehub/lobe-chat/discussions/2295#discussioncomment-9290942
- */
-
-const ChatSettingsModalFallback = () => {
- const router = useQueryRoute();
-
- useLayoutEffect(() => {
- router.replace('/chat/settings', { query: { tab: '' } });
- }, []);
-
- return null;
-};
-
-export default ChatSettingsModalFallback;
diff --git a/src/app/(main)/chat/settings/not-found.tsx b/src/app/(main)/chat/settings/not-found.tsx
deleted file mode 100644
index 02503bc7fa461..0000000000000
--- a/src/app/(main)/chat/settings/not-found.tsx
+++ /dev/null
@@ -1,3 +0,0 @@
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/404'));
diff --git a/src/app/(main)/chat/settings/page.tsx b/src/app/(main)/chat/settings/page.tsx
deleted file mode 100644
index 63b66be34dc97..0000000000000
--- a/src/app/(main)/chat/settings/page.tsx
+++ /dev/null
@@ -1,7 +0,0 @@
-import EditPage from './features/EditPage';
-
-const Page = () => {
- return ;
-};
-
-export default Page;
diff --git a/src/app/(main)/discover/(detail)/error.tsx b/src/app/(main)/discover/(detail)/error.tsx
deleted file mode 100644
index 071491038c704..0000000000000
--- a/src/app/(main)/discover/(detail)/error.tsx
+++ /dev/null
@@ -1,5 +0,0 @@
-'use client';
-
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/Error'));
diff --git a/src/app/(main)/discover/(detail)/not-found.tsx b/src/app/(main)/discover/(detail)/not-found.tsx
deleted file mode 100644
index 02503bc7fa461..0000000000000
--- a/src/app/(main)/discover/(detail)/not-found.tsx
+++ /dev/null
@@ -1,3 +0,0 @@
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/404'));
diff --git a/src/app/(main)/discover/(list)/error.tsx b/src/app/(main)/discover/(list)/error.tsx
deleted file mode 100644
index 071491038c704..0000000000000
--- a/src/app/(main)/discover/(list)/error.tsx
+++ /dev/null
@@ -1,5 +0,0 @@
-'use client';
-
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/Error'));
diff --git a/src/app/(main)/discover/(list)/not-found.tsx b/src/app/(main)/discover/(list)/not-found.tsx
deleted file mode 100644
index 02503bc7fa461..0000000000000
--- a/src/app/(main)/discover/(list)/not-found.tsx
+++ /dev/null
@@ -1,3 +0,0 @@
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/404'));
diff --git a/src/app/(main)/discover/error.tsx b/src/app/(main)/discover/error.tsx
deleted file mode 100644
index 071491038c704..0000000000000
--- a/src/app/(main)/discover/error.tsx
+++ /dev/null
@@ -1,5 +0,0 @@
-'use client';
-
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/Error'));
diff --git a/src/app/(main)/discover/not-found.tsx b/src/app/(main)/discover/not-found.tsx
deleted file mode 100644
index 02503bc7fa461..0000000000000
--- a/src/app/(main)/discover/not-found.tsx
+++ /dev/null
@@ -1,3 +0,0 @@
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/404'));
diff --git a/src/app/(main)/error.tsx b/src/app/(main)/error.tsx
deleted file mode 100644
index 071491038c704..0000000000000
--- a/src/app/(main)/error.tsx
+++ /dev/null
@@ -1,5 +0,0 @@
-'use client';
-
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/Error'));
diff --git a/src/app/(main)/files/(content)/@menu/features/KnowledgeBase/EmptyStatus.tsx b/src/app/(main)/files/(content)/@menu/features/KnowledgeBase/EmptyStatus.tsx
deleted file mode 100644
index f06be327f7db1..0000000000000
--- a/src/app/(main)/files/(content)/@menu/features/KnowledgeBase/EmptyStatus.tsx
+++ /dev/null
@@ -1,53 +0,0 @@
-import { createStyles } from 'antd-style';
-import React from 'react';
-import { Trans } from 'react-i18next';
-import { Flexbox } from 'react-layout-kit';
-
-const useStyles = createStyles(({ css, token }) => ({
- container: css`
- font-size: 12px;
- color: ${token.colorTextTertiary};
- `,
-
- paragraph: css`
- justify-content: center;
- width: 100%;
-
- kbd {
- margin-inline: 2px;
- padding-inline: 6px;
- background: ${token.colorFillTertiary};
- border-radius: 4px;
- }
- `,
-}));
-const EmptyStatus = () => {
- const { styles } = useStyles();
- return (
-
-
-
-
- 点击 + 开始创建知识库
-
-
-
- );
-};
-
-export default EmptyStatus;
diff --git a/src/app/(main)/not-found.tsx b/src/app/(main)/not-found.tsx
deleted file mode 100644
index 02503bc7fa461..0000000000000
--- a/src/app/(main)/not-found.tsx
+++ /dev/null
@@ -1,3 +0,0 @@
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/404'));
diff --git a/src/app/(main)/profile/error.tsx b/src/app/(main)/profile/error.tsx
deleted file mode 100644
index 071491038c704..0000000000000
--- a/src/app/(main)/profile/error.tsx
+++ /dev/null
@@ -1,5 +0,0 @@
-'use client';
-
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/Error'));
diff --git a/src/app/(main)/profile/not-found.tsx b/src/app/(main)/profile/not-found.tsx
deleted file mode 100644
index 02503bc7fa461..0000000000000
--- a/src/app/(main)/profile/not-found.tsx
+++ /dev/null
@@ -1,3 +0,0 @@
-import dynamic from 'next/dynamic';
-
-export default dynamic(() => import('@/components/404'));
diff --git a/src/app/(main)/profile/security/page.tsx b/src/app/(main)/profile/security/page.tsx
deleted file mode 100644
index 95a233c60cfe0..0000000000000
--- a/src/app/(main)/profile/security/page.tsx
+++ /dev/null
@@ -1,34 +0,0 @@
-import { Skeleton } from 'antd';
-import dynamic from 'next/dynamic';
-import { notFound } from 'next/navigation';
-
-import { enableClerk } from '@/const/auth';
-import { metadataModule } from '@/server/metadata';
-import { translation } from '@/server/translation';
-import { isMobileDevice } from '@/utils/server/responsive';
-
-const ClerkProfile = dynamic(() => import('../features/ClerkProfile'), {
- loading: () => (
-
- 
- 
- 
- 
- 
| **AiHubMix** | 使用 OpenAI 企业接口，全站模型价格为官方 **86 折**（含 GPT-4 、Cluade 3.5 等） | `https://aihubmix.com/v1` | [获取](https://lobe.li/CnsM6fH) |