Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
85 lines (57 loc) · 2.37 KB

CONTRIBUTING.md

File metadata and controls

85 lines (57 loc) · 2.37 KB
Raw

Contributing to tunelab

Quick links

Reporting bugs

Open an issue with:

  1. What you expected to happen
  2. What actually happened
  3. Steps to reproduce
  4. Your environment (OS, Python version, uv --version)

Suggesting features

Open an issue describing your use case and proposed solution. For larger changes, open a discussion first.

Development setup

Prerequisites: Python 3.10+, uv, Apple Silicon Mac (for training tests only — decide/data/eval tests run anywhere).

git clone https://github.com/rchaz/tunelab.git
cd tunelab

No pip install needed — every script uses uv with inline dependency declarations.

Running tests

bash tests/run_all.sh

First run downloads ~500MB of models into ~/.cache/huggingface and caches uv dependencies; later runs are fast.

Individual tests:

python3 tests/test_centroid_classify.py
python3 tests/test_split_data.py
# etc.

Project structure

skills/           # The five Claude Code skills (SKILL.md + scripts/)
  tune-decide/    # Front door — interviews, experiments, recommends
  tune-data/      # Dataset building, cleaning, splitting
  tune-train/     # Local LoRA/QLoRA/CPT training (MLX)
  tune-eval/      # Evaluation, cascade composition, LLM-as-judge
  tune-loop/      # Continuous improvement loop
concepts/         # Plain-English explainers for every idea tunelab uses
recipes/          # Worked end-to-end examples with real numbers
dogfood/          # Internal test runs and benchmark results
tests/            # Script-level tests (subprocess, real data)

Pull request guidelines

  1. Keep PRs focused — one change per PR
  2. bash tests/run_all.sh must pass
  3. Follow existing code patterns
  4. No new dependencies in scripts unless absolutely necessary (and declared inline via uv)
  5. If you add a new script, add a test for it

Commit messages

Use clear, descriptive commit messages in imperative style.

AI-assisted contributions

AI-assisted PRs are welcome. If you used AI tools, mention it in the PR description for transparency.

Security

If you discover a security vulnerability, please report it privately. See SECURITY.md.

License

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