Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
121 lines (89 loc) · 3.56 KB

CONTRIBUTING.md

File metadata and controls

121 lines (89 loc) · 3.56 KB

Contributing to KernelAgent

Thanks for your interest in improving KernelAgent! We aim to make contributions straightforward, transparent, and welcoming.

Code of Conduct

By participating, you agree to abide by our Code of Conduct. See CODE_OF_CONDUCT.md.

Ways to Contribute

  • Report bugs and performance issues
  • Propose features and improvements
  • Improve documentation and examples
  • Add tests and fix bugs
  • Optimize kernels or refine prompts/templates

Development Setup

Supported Python: 3.8–3.12. A CUDA‑enabled GPU and an API key are only required for end‑to‑end runs; basic tests do not require them.

  1. Clone and install (choose one):
git clone https://github.com/pytorch-labs/KernelAgent.git
cd KernelAgent

# Using pip
python -m venv .venv && source .venv/bin/activate
pip install -e .
pip install -e ".[dev]"

# Or using uv (matches CI)
uv venv && source .venv/bin/activate
uv pip install -e ".[dev]"
  1. Optional: set environment for e2e runs (LLM + GPU):
cp .env.example .env   # then edit with your API key(s)

Running Tests

Unit tests:

uv run pytest tests/ -v

Coverage (optional):

uv run pytest tests/ -v --cov=triton_kernel_agent

End‑to‑end example (requires network + GPU):

uv run python e2e_test.py

Linting and Style

We follow PEP 8 with type hints and docstrings.

  • Indentation: 4 spaces
  • Line length: aim for ~100 chars where practical
  • Imports: standard library, third‑party, then local
  • Logging: prefer logging over print
  • Types: use annotations for public APIs

Ruff configuration is included in pyproject.toml. If you have ruff installed, run:

ruff check .

Pull Requests

We actively welcome your pull requests. Small, focused PRs are easier to review.

  1. Fork the repo and branch from main.
  2. Make your change with tests where appropriate.
  3. Update docs/README/examples if behavior or APIs change.
  4. Run tests locally and ensure CI passes.
  5. Optionally run ruff check . and address issues.
  6. Submit a clear PR description with context and rationale.

PR Checklist

  • Tests added or updated when behavior changes
  • Docs/examples updated when user‑facing behavior changes
  • No unrelated formatting or drive‑by refactors
  • CI green (see .github/workflows/ci.yml)

Issues

Use GitHub Issues for bugs and feature requests.

When reporting a bug, please include:

  • What you did (minimal reproducible example)
  • What you expected to happen
  • What actually happened (full error, logs, screenshots as relevant)
  • Environment details (OS, Python, GPU/CUDA, package versions)

Security

If you believe you’ve found a security vulnerability, do not open a public issue. Please follow the process described in Meta’s bug bounty program.

Contributor License Agreement (CLA)

Some contributions may require signing a CLA before they can be merged. If prompted by the bot during your PR, please complete the form here: https://code.facebook.com/cla

License

By contributing to KernelAgent, you agree that your contributions will be licensed under the LICENSE file in the root of this repository.

Notes for Kernel/Template Contributions

  • Changes to Jinja2 templates in triton_kernel_agent/templates/ should be validated with an end‑to‑end run (see e2e section) since they affect generation behavior.
  • For GPU kernels or performance‑sensitive code, include a short note about observed correctness and performance if available.