English | 简体中文
This is HepAI python library, the AI platform can accelerate scientific research in multidisciplinary scenarios, simplify model iteration and flow, and is a common infrastructure for the development of AI algorithms and applications.
The HepAI platform itself is a software system that carries AI algorithm models, provides AI computing power, connects data channels, and conducts AI training.
The HepAI framework integrates classic and state-of-the-art (SOTA) artificial intelligence algorithms in the field of high-energy physics. One can access related algorithm models, datasets, and computational resources through a unified interface, making the application of AI simple and efficient.
News
- [2024.05.16] v1.1.9 HepAI Client支持GPT-4o, 调用方法:
- [2024.03.26] v1.0.21 Make LLM request like OpenAI via HepAI object.
- [2023.10.24] v1.0.18 接入dalle文生图模型,调用方法教程见此处。
- [2023.04.21] v1.0.7通过hepai使用GPT-3.5,hepai_api.md.
- [2023.02.09] 基于ChatGPT的HaiChatGPT已上线,使用简单,无需梯子!详情查看:HaiChatGPT.
- [2023.01.16] 华为NPU服务器上架,如有算法国产化需求,请查阅NPU文档。
- [2022.10.20] HAI v1.0.6-Beta 第一个测试版本发布,4个算法和3个数据集
- [2022.08.23] HAI v1.0.0
Tutorials
Quick Start to Using HepAI on Computing Clusters
Reconstruction and identification of atmospheric neutrinos in JUNO experiments using PointNet
pip install hepai --upgrade
hai -V # 查看版本
-
命令行使用
hai train <model_name> # 训练模型, 例如: hai train particle_transformer hai eval <model_name>
-
python库使用
python库统一接口:
import hai model = hai.hub.load('<model_name>') # 加载模型 config = model.config # 获取模型配置 config.batch_size = 32 # 修改配置 model.trian() # 训练模型 model.eval() # 评估模型 model.infer('<data>') # 模型推理 hai.train('particle_transformer')
-
部署和远程调用
跨语言、跨平台的模型部署和远程调用
服务端:
hai start server # 启动服务
客户端
pip install hai-client
import hai_client hai = hai_client.HAI()
或其他支持gRPC的语言,详见deploy
note