Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
76 lines (54 loc) · 1.87 KB

CONTRIBUTING.md

File metadata and controls

76 lines (54 loc) · 1.87 KB

Contributing

Thanks for contributing.

This project is a research-style codebase, so reproducibility and clear experiment hygiene matter more than volume of code changes.

Local Setup

make setup
make check
make test

If you work on GPU paths, install GPU extras:

make setup-gpu

Development Workflow

  1. Create a branch from main.
  2. Keep changes scoped to one concern (feature, fix, docs, or results).
  3. Run checks before opening a PR.
  4. Include rationale in commit messages and PR description.

Required Checks Before PR

make check
make test

For benchmark-related changes, also run at least one smoke benchmark:

make bench-toy OUT=/tmp/bench_toy.jsonl
.venv/bin/python scripts/validate_results_jsonl.py --path /tmp/bench_toy.jsonl --strict

Benchmark and Artifact Policy

  • datasets/ is for local data and run outputs (gitignored by default).
  • reports/ contains tracked summary artifacts intended for sharing.
  • logs/ is local runtime noise and is ignored.
  • Prefer new output filenames for long runs to avoid mixing old and new records.

AutoJudge Notes

  • Keep paper-aligned C-grid: 1e-7,1e-6,1e-5,1e-4,1e-3,1e-2,1e-1,1e0.
  • Do not introduce 1e1 or 1e2 into preset defaults.
  • If behavior changes, update tests in tests/test_autojudge.py.

Code Style

  • Follow existing style in touched files.
  • Keep interfaces explicit; avoid hidden behavior changes.
  • Add comments only for non-obvious logic.

Pull Request Checklist

  • Scope is clear and focused
  • make check passes
  • make test passes
  • Relevant docs/configs updated
  • Added or updated tests for logic changes
  • Benchmark outputs validated when applicable

Reporting Bugs

Please use the bug report template and include:

  • exact command used,
  • dataset/model preset,
  • stack trace or log snippet,
  • expected vs actual behavior.