CONTRIBUTING.md

Contributing to TurboQuant

Thanks for your interest in contributing! TurboQuant is the first open-source implementation of the TurboQuant paper (ICLR 2026).

Getting Started

git clone https://github.com/OnlyTerp/turboquant.git
cd turboquant
pip install -e ".[dev]"
pytest src/test_turboquant.py -v

What We Need Help With

• Benchmarking on more models — We've validated on Mistral-7B and Nemotron-Nano-4B. More models = better.
• Triton kernel correctness — kernels.py is experimental and uses Rademacher S matrices (see IMPLEMENTATION_NOTES.md). Needs validation.
• vLLM integration — The plugin scaffold in vllm_plugin/ needs real-world testing.
• Performance optimization — The pure PyTorch path is correct but slow. GPU acceleration welcome.
• More bit-width configurations — The paper shows results at 2.5-bit and 3.5-bit. We support both but need more testing.

Code Style

• Python 3.10+
• Type hints on public API functions
• Docstrings on all public classes and functions
• pytest for tests — run pytest src/test_turboquant.py -v before submitting

Pull Request Process

1. Fork the repo and create a feature branch
2. Add tests for new functionality
3. Ensure all tests pass: pytest src/test_turboquant.py -v
4. Update documentation if needed
5. Open a PR with a clear description of what changed and why

Reporting Issues

Please include:

• Python and PyTorch versions
• GPU model (if applicable)
• Minimal reproduction code
• Full error traceback

License

By contributing, you agree that your contributions will be licensed under the MIT License.