LLM Finetuner

This project provides a user-friendly interface for fine-tuning Large Language Models (LLMs) using the Unsloth library. It includes features for dataset preparation, synthetic dataset creation, model training, testing, and GGUF conversion.

Features

• Load and fine-tune various pre-trained models
• Prepare existing datasets or create synthetic datasets
• Fine-tune models with customizable hyperparameters
• Test fine-tuned models
• Convert models to GGUF format for deployment

Prerequisites

• Python 3.8 or higher
• CUDA-capable GPU (for efficient training)

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/llm-finetuner.git
   cd llm-finetuner
   

2. Create a virtual environment (optional but recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`
   

3. Install the required packages:

   pip install -r requirements.txt
   

Usage

1. Run the application:

   python main.py
   

2. Open the provided URL in your web browser to access the Gradio interface.

3. Follow these steps in the interface: a. Settings: Enter your Hugging Face token and select a model. b. Dataset: Prepare an existing dataset or create a synthetic one. c. Training: Set hyperparameters and start the fine-tuning process. d. Test: Test your fine-tuned model with custom inputs. e. GGUF Conversion: Convert your model to GGUF format if needed.

Notes

• Ensure you have the necessary API keys for OpenAI or Anthropic if you plan to use them for synthetic dataset creation.
• If using Ollama for local generation, make sure it's installed and running on your machine.
• Fine-tuning can be computationally intensive. Ensure you have adequate GPU resources available.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License.

Comprehensive Python Setup Guide

This guide will walk you through setting up Python, creating a virtual environment, and running your LLM Finetuner project on a new system.

1. Install Python

Windows:

1. Go to https://www.python.org/downloads/windows/
2. Download the latest Python 3.x installer (64-bit version recommended)
3. Run the installer
4. Check "Add Python to PATH" during installation
5. Click "Install Now"

macOS:

1. Install Homebrew if you haven't already:

   /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
   

2. Install Python using Homebrew:

   brew install python
   

Linux (Ubuntu/Debian):

1. Update package list:

   sudo apt update
   

2. Install Python:

   sudo apt install python3 python3-pip python3-venv
   

2. Verify Python Installation

Open a terminal (Command Prompt on Windows) and run:

python --version

You should see the Python version number. If not, try python3 --version.

3. Install Git

Windows:

1. Go to https://git-scm.com/download/win
2. Download and run the installer
3. Use the default settings during installation

macOS:

If you installed Homebrew earlier:

brew install git

Linux (Ubuntu/Debian):

sudo apt install git

4. Clone the Repository

1. Open a terminal
2. Navigate to where you want to store the project
3. Clone the repository:

   git clone https://github.com/yourusername/llm-finetuner.git
   cd llm-finetuner
   

5. Create and Activate a Virtual Environment

Windows:

python -m venv venv
venv\Scripts\activate

macOS and Linux:

python3 -m venv venv
source venv/bin/activate

Your prompt should change to indicate that the virtual environment is active.

6. Install Required Packages

With the virtual environment activated:

pip install -r requirements.txt

This may take a while as it installs all necessary dependencies.

7. Set Up CUDA (for GPU support)

If you have an NVIDIA GPU and want to use it for training:

1. Go to https://developer.nvidia.com/cuda-downloads
2. Download and install the CUDA Toolkit appropriate for your system
3. Install the cuDNN library:
   • Go to https://developer.nvidia.com/cudnn
   • Download cuDNN (you may need to create an NVIDIA account)
   • Follow the installation instructions for your system

8. Run the Application

With the virtual environment still activated:

python main.py

This will start the Gradio interface. Open the provided URL in your web browser.

9. Using the LLM Finetuner

1. In the "Settings" tab:

   • Enter your Hugging Face token
   • Select a model

2. In the "Dataset" tab:

   • Prepare an existing dataset or create a synthetic one

3. In the "Training" tab:

   • Set hyperparameters and start training

4. In the "Test" tab:

   • Test your fine-tuned model

5. In the "GGUF Conversion" tab:

   • Convert your model to GGUF format if needed

Troubleshooting

• If python doesn't work, try python3
• Ensure your GPU drivers are up to date for CUDA support
• If you encounter "command not found" errors, ensure the relevant programs are in your system's PATH

Closing Notes

• Always activate the virtual environment before running the project
• To deactivate the virtual environment, simply type deactivate in the terminal
• Keep your Python packages updated with pip install --upgrade -r requirements.txt

Remember to keep your API keys and tokens secure. Happy fine-tuning!

Alternative, installation

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121 pip install triton pip install unsloth gradio transformers datasets tqdm