Custom LLM Training and Fine-tuning System

This project provides a system for training, fine-tuning, and using custom language models. It supports both pre-training from scratch and fine-tuning of existing models.

Features

• Pre-train models from scratch
• Fine-tune existing pre-trained models
• Use pre-trained or fine-tuned models for inference
• Interactive text generation
• Support for custom datasets

Dataset Formats

Fine-tuning Dataset Format

The fine-tuning dataset should be in JSON format with the following structure:

{
    "instruction": "The task or instruction to be performed",
    "input": "The input text or prompt",
    "output": "The expected output or response"
}

Example:

{
    "instruction": "Evaluate the following phrase by transforming it into the spelling given.",
    "input": "freind --> friend",
    "output": "The spelling of the given phrase \"freind\" is incorrect, the correct spelling is \"friend\"."
}

Pre-training Dataset Format

For pre-training, you can use:

• Plain text files (.txt)
• Text data from URLs (e.g., Gutenberg books)

Usage

1. Run the main script:

python main.py

2. Choose from the following options:

   • Use pretrained model
   • Use existing fine-tuned model
   • Train a custom model from scratch
   • Fine-tune an existing pretrained model

3. Follow the prompts to:

   • Enter model names
   • Provide dataset paths/URLs
   • Set training parameters (batch size, epochs)
   • Interact with the model

Model Files

The system uses the following file naming conventions:

• Pre-trained models: {model_name}_pretrained_model.pth
• Fine-tuned models: instruct_{model_name}.pth

Requirements

• Python 3.x
• PyTorch
• tiktoken
• CUDA (optional, for GPU acceleration)

Notes

• Make sure your dataset follows the specified format
• Model files should be in the current working directory
• For fine-tuning, ensure you have a pre-trained model available
• The system automatically handles device selection (CPU/GPU)

Prerequisites

• Python 3.8 or higher
• CUDA-capable GPU (recommended for faster training)
• pip (Python package installer)

Installation

1. Clone the repository:

git clone https://github.com/RohitPawar001/Build_your_custom_LLM.git
cd custom_llm

2. Create and activate a virtual environment (recommended):

# Windows
python -m venv AI_env
.\AI_env\Scripts\activate

# Linux/Mac
python -m venv AI_env
source AI_env/bin/activate

3. Install the required dependencies:

pip install -r requirements.txt

Project Structure

.
├── requirements.txt
├── .gitignore
├── instruct_{model_name}.pth  # Fine-tuned model weights
└── {model_name}_pretrained_model.pth   # Pretrained model weights

Dependencies

The project requires the following main packages:

• PyTorch (>=2.0.0)
• NumPy (>=1.21.0)
• Pandas (>=1.3.0)
• scikit-learn (>=1.0.0)
• Matplotlib (>=3.4.0)
• tqdm (>=4.62.0)
• datasets (>=2.12.0)
• accelerate (>=0.20.0)

Contributing

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.