Tip

Now you can use a variety of popular models other than OpenAI's GPT2✨

Important

Main Repository: Zeta

---

Zeta-Tool - Easily create your own LLM

Fully Open-source LLM Tool

1. Select Pre-trained Model👐
2. Select Dataset🧠
3. Wait🕰️
4. Successfully Created Your Own LLM✨

How to use

Install

1. Install Git and Git-LFS
2. Clone This Repository (Example: git clone https://github.com/DiamondGotCat/Zeta-Tool.git)
3. Check and Install Python and PIP (Recommend: Miniconda)
4. Install Requiments using PIP (Example: pip install pandas transformers torch rich)

Training

1. Run training.py using Python
2. Answer Selection
3. Wait
4. Done

Test

1. Run execute.py using Python
2. If Loaded, then Enter Prompt
3. Get Answer
4. Enter /q to Exit Chat

Scripts

• training.py: Learning using Zeta-Formatted Dataset
• execute.py: Run Learned Model (You Need Move Model Folder to ./trained-model)

Built-in Dataset

• (Need Git-LFS to Clone This) OpenO1-SFT by OpenO1 Team I converted to Azuki-Formatted Dataset.
• Zeta-Classic 2n by DiamondGotCat

Selectable Base Models

OpenAI

• gpt2

  • Overview: Training from Scratch
  • Tokenizer: openai-community/gpt2

• gpt2-small

  • Tokenizer/Model: openai-community/gpt2

• gpt2-medium

  • Tokenizer/Model: openai-community/gpt2-medium

Google

NOTE: Model Access Permission Required

• gemma

  • Size 2b:
    • Tokenizer/Model: google/gemma-2b
  • Size 7b:
    • Tokenizer/Model: google/gemma-7b

• codegemma

  • Size 2b:
    • Tokenizer/Model: google/codegemma-2b
  • Size 7b:
    • Tokenizer/Model: google/codegemma-7b

• gemma1.1 (Inst)

  • Size 2b:
    • Tokenizer/Model: google/gemma-1.1-2b-it
  • Size 7b:
    • Tokenizer/Model: google/gemma-1.1-7b-it

• gemma2

  • Size 2b:
    • Tokenizer/Model: google/gemma-2-2b
  • Size 9b:
    • Tokenizer/Model: google/gemma-2-9b
  • Size 27b:
    • Tokenizer/Model: google/gemma-2-27b

Meta

NOTE: Model Access Permission Required

• llama2

  • Size 7b:
    • Tokenizer/Model: meta-llama/Llama-2-7b
  • Size 13b:
    • Tokenizer/Model: meta-llama/Llama-2-13b
  • Size 70b:
    • Tokenizer/Model: meta-llama/Llama-2-70b

• codellama

  • Size 7b:
    • Tokenizer/Model: meta-llama/CodeLlama-7b-hf
  • Size 13b:
    • Tokenizer/Model: meta-llama/CodeLlama-13b-hf
  • Size 34b:
    • Tokenizer/Model: meta-llama/CodeLlama-34b-hf
  • Size 70b:
    • Tokenizer/Model: meta-llama/CodeLlama-70b-hf

• llama3

  • Size 8b:
    • Tokenizer/Model: meta-llama/Meta-Llama-3-8B
  • Size 70b:
    • Tokenizer/Model: meta-llama/Meta-Llama-3-70B

• llama3.1

  • Size 8b:
    • Tokenizer/Model: meta-llama/Llama-3.1-8B
  • Size 70b:
    • Tokenizer/Model: meta-llama/Llama-3.1-70B
  • Size 405b:
    • Tokenizer/Model: meta-llama/Llama-3.1-405B

• llama3.2

  • Size 1b:
    • Tokenizer/Model: meta-llama/Llama-3.2-1B
  • Size 3b:
    • Tokenizer/Model: meta-llama/Llama-3.2-3B

• llama3.2-v (Vision)

  • Size 11b:
    • Tokenizer/Model: meta-llama/Llama-3.2-11B-Vision
  • Size 90b:
    • Tokenizer/Model: meta-llama/Llama-3.2-90B-Vision

• llama3.3

  • Size 70b: (Select confirm)
    • Tokenizer/Model: meta-llama/Llama-3.3-70B-Instruct

Alibaba

• qwen

  • Size 1.8b:
    • Tokenizer/Model: Qwen/Qwen-1_8B
  • Size 7b:
    • Tokenizer/Model: Qwen/Qwen-7B
  • Size 14b:
    • Tokenizer/Model: Qwen/Qwen-14B
  • Size 72b:
    • Tokenizer/Model: Qwen/Qwen-72B

• qwen1.5

  • Size 0.5b:
    • Tokenizer/Model: Qwen/Qwen1.5-0.5B
  • Size 1.8b:
    • Tokenizer/Model: Qwen/Qwen1.5-1.8B
  • Size 4b:
    • Tokenizer/Model: Qwen/Qwen1.5-4B
  • Size 14b:
    • Tokenizer/Model: Qwen/Qwen1.5-14B
  • Size 32b:
    • Tokenizer/Model: Qwen/Qwen1.5-32B
  • Size 72b:
    • Tokenizer/Model: Qwen/Qwen1.5-72B

Local Model

• trained_model
  • Overview: Your Pre-trained Model

---

Thank you for reading this.

Actually, Zeta-Tool is a personal project. There are still parts that haven't been developed yet.

If possible, please help in one of the following ways:

1. Simple: Please give it a star.
2. For programmers/engineers: Help with code fixes or testing. (See Ideas)
3. For those who can support Zeta-Tool's future: Publish the trained models on HuggingFace. However, please include information about the Zeta-Tool project. For more details, see Help with Trained Model.