Convert-to-Onnx-Hf-Dir

Convert a Hugging Face model to ONNX & Upload Directly to Your Hf Model Repo

Onnx-to-Repo.mp4

Convert a Hugging Face model to ONNX format and directly upload the converted files to your Hugging Face model repository. This tool leverages Streamlit for a web-based UI and the Hugging Face Hub API for authentication and file uploads.

Overview

This application automates the following steps:

• Repository Setup: Downloads and extracts the Transformers.js repository from a specific version tag or branch.
• Model Conversion: Converts the specified Hugging Face model to ONNX format using a conversion script.
• Model Upload: Moves the generated ONNX files to an onnx/ subfolder and uploads them to your Hugging Face model repository.

Features

• Streamlit UI: Provides an easy-to-use web interface for inputting model details and tokens.
• Flexible Configuration: Uses environment variables and Streamlit secrets for configuration.
• Repository Management: Automatically handles repository download and extraction.
• Error Handling: Detailed logging and error messages for troubleshooting.

Requirements

• Python 3.7+
• Streamlit
• Hugging Face Hub
• Other dependencies listed in requirements.txt

Installation

1. Clone the repository:

   git clone https://github.com/PRITHIVSAKTHIUR/Convert-to-Onnx-Hf-Dir.git
   cd Convert-to-Onnx-Hf-Dir

2. Create a virtual environment and activate it:

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

3. Install dependencies:

   pip install -r requirements.txt

Configuration

The application requires a Hugging Face token for authentication. You can provide this token via environment variables or Streamlit secrets.

• Environment Variables:

  • HF_TOKEN: Your Hugging Face API token.
  • SPACE_AUTHOR_NAME (optional): Your Hugging Face username if not provided via token.

• Streamlit Secrets:

  • Create a file named .streamlit/secrets.toml with the following content:

    HF_TOKEN = "your_hf_api_token_here"

• User Token Input:

  • When running the application, you can also input your Hugging Face write token through the UI.

Usage

1. Run the Streamlit app:

   streamlit run app.py

2. Using the UI:

   • Enter the Hugging Face model ID you wish to convert (e.g., prithivMLmods/FastThink-0.5B-Tiny).
   • Optionally provide your Hugging Face write token if you wish to upload the model under your account.
   • Click on the "Proceed" button.
   • The application will:
     • Convert the model to ONNX format.
     • Upload the ONNX files to the onnx/ folder in your Hugging Face model repository.
   • A link will be provided to view your model on Hugging Face after a successful upload.

Code Structure

• app.py: Main application file that sets up the UI and orchestrates the conversion and upload process.
• Config class: Manages configuration and secrets.
• ModelConverter class: Contains methods for repository setup, model conversion, and upload.

Troubleshooting

• Conversion Errors: Check the error output provided by the app if conversion fails.
• Upload Errors: Ensure your Hugging Face token has write permissions for the target repository.
• Repository Setup: The app downloads and extracts the repository only if it does not already exist locally. If you encounter issues, try deleting the local repository folder (./transformers.js) and restarting the app.

Contributing

Contributions are welcome! Please follow these steps:

1. Fork the repository.
2. Create a feature branch: git checkout -b feature/YourFeature
3. Commit your changes and push: git push origin feature/YourFeature
4. Open a Pull Request with a clear description of your changes.