Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
50 lines (41 loc) · 1.67 KB

CONTRIBUTING.md

File metadata and controls

50 lines (41 loc) · 1.67 KB
Raw

Contributing

Thanks for your interest in improving this project!

Ways to contribute

  • Bug reports and reproduction cases
  • Docs improvements (README examples, platform notes)
  • Performance tweaks (better chunking, projector variants, caching strategies)
  • New retrieval/selection strategies (e.g., different policies, heuristics)
  • Tests (unit/smoke) and CI

Development setup

python -m venv .venv && source .venv/bin/activate
pip install --upgrade pip
# Choose your platform install (CUDA/ROCm/MPS/CPU) then:
pip install "transformers==4.43.3" accelerate sentencepiece sacrebleu numpy faiss-cpu

Style

  • Python ≥ 3.10, follow PEP 8.
  • Use type hints where practical.
  • Keep functions focused and documented with short docstrings.
  • Prefer explicitness over magic.

Commit messages

Use Conventional Commits where possible:

feat: add ROCm detection
fix: handle empty corpus gracefully
docs: expand README on FAISS choices
perf: cache chunk encodings across steps
refactor: split projector into MLP and linear
test: add smoke test for generate

Link issues with Fixes #123 when appropriate.

Pull requests

  • Create a topic branch; keep PRs focused and reasonably small.
  • Include a clear description and any platform-specific notes (CUDA/ROCm/MPS).
  • If your PR changes defaults or interfaces, update the README and examples.

Running quick smoke tests

  • Build a tiny index (data/wiki_lines.txt) and run generate end-to-end.
  • On CUDA boxes, ensure faiss-gpu is attempted; on ROCm/macOS/Windows, confirm faiss-cpu flows work.

Code of Conduct

By participating, you agree to follow our Code of Conduct (see CODE_OF_CONDUCT.md).

Happy hacking!