Gemma models Fine-tuning UI

A user-friendly web interface for fine-tuning Google's Gemma language models, allowing users with limited technical expertise to customize models for specific use cases.

Features

• Dataset Management: Upload, preprocess, and validate your training data
• Model Configuration: Configure Gemma models with optimized hyperparameters
• Training Visualization: Monitor training progress with real-time metrics
• Model Evaluation: Evaluate model performance with comprehensive metrics
• Export Options: Export fine-tuned models in various formats

Getting Started

Prerequisites

• Python 3.8+
• CUDA-compatible GPU (recommended for training)
• Google API credentials (for Gemma model access)

Installation

1. Clone this repository:

   git clone https://github.com/yourusername/gemma-finetuning-ui.git
   cd gemma-finetuning-ui
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Set up your Google API credentials:

   export GOOGLE_API_KEY=your_api_key_here
   

Running the Application

Start the Gradio web interface:

python app.py

The application will be available at http://localhost:7860

Usage Guide

1. Dataset Upload

• Upload your dataset in CSV, JSONL, or text format
• Configure preprocessing options
• Validate and preview your processed data

2. Model Configuration

• Select Gemma model variant (2B or 7B)
• Configure training hyperparameters
• Choose fine-tuning approach (full fine-tuning or parameter-efficient methods)

3. Training

• Start and monitor training progress
• View real-time metrics and visualizations
• Save checkpoints during training

4. Evaluation

• Evaluate model performance on test data
• View comprehensive metrics
• Test model with custom inputs

5. Export

• Export your fine-tuned model in various formats
• Configure optimization options
• Deploy to your preferred platform

Project Structure

gemma-finetuning-ui/
├── README.md                   # Project documentation
├── requirements.txt            # Dependencies
├── app.py                      # Main Gradio application entry point
├── frontend/                   # Frontend components
│   ├── components.py           # Reusable UI components
│   ├── pages.py                # Page layouts
│   └── utils.py                # Frontend utilities
├── backend/
│   ├── data_processing/        # Data handling modules
│   │   ├── __init__.py
│   │   ├── preprocessor.py     # Data preprocessing
│   │   ├── validation.py       # Input validation
│   │   └── augmentation.py     # Data augmentation (basic)
│   ├── model/                  # Model-related code
│   │   ├── __init__.py
│   │   ├── loader.py           # Model loading utilities
│   │   ├── trainer.py          # Training functionality
│   │   └── evaluation.py       # Model evaluation
│   └── utils/                  # Backend utilities
│       ├── __init__.py
│       ├── config.py           # Configuration handling
│       └── visualization.py    # Training metrics visualization
├── examples/                   # Example datasets
│   └── healthcare_sample.jsonl # Sample healthcare dataset
└── notebooks/                  # Development notebooks
    └── prototype.ipynb         # Prototype implementation

Contributing

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

Acknowledgements

• Google for the Gemma language models
• Hugging Face for transformer libraries
• Gradio for the web interface framework