Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
85 lines (65 loc) · 3.24 KB

CONTRIBUTING.md

File metadata and controls

85 lines (65 loc) · 3.24 KB
Raw

Contributing to SLMRouter

Thank you for helping. Bug reports, questions, documentation fixes and code are all welcome.

Reporting a bug or asking for a feature

Open an issue using one of the templates. For a bug, include the SLMRouter version, Python version, runtime (Ollama or llama.cpp, with its version), the model, and the smallest code that reproduces the problem. If it is a routing problem, run the CLI with --verbose and paste the printed prompt and raw model output.

Security problems must not be reported in public issues. See SECURITY.md.

Development setup

SLMRouter has no runtime dependencies and no development dependencies beyond Python 3.10 or newer.

git clone https://github.com/DYNAMOatGHUB/SLM-Router.git
cd SLM-Router
python -m venv .venv && source .venv/bin/activate
pip install -e .

Run the same checks CI runs:

python -m unittest discover -s tests -t .   # all tests
python -m eval.run_eval --mode sim          # simulated evaluation, exits 1 on failure
python -m bench.bench_parser                # parser latency benchmark

No model is needed for any of these. To try a live model, install Ollama, run ollama pull llama3:8b, then python -m slmrouter --backend ollama "What is 2+3?" --verbose.

Rules every change must follow

These are design guarantees, and CI enforces most of them. A pull request that breaks one will not be merged.

  • No runtime dependencies. Standard library only (tests/test_no_deps.py).
  • Model text never becomes code. No eval, exec, compile, __import__, ast.literal_eval, pickle or importlib in slmrouter/ (tests/test_static_gate.py).
  • Bounded model calls. A query uses at most 1 + max_repairs model calls.
  • Fail closed. Ambiguous model output is rejected with an error code, never guessed.
  • Tools are never retried after they raise or time out.
  • The public signatures and the contracts in slmrouter/types.py and slmrouter/errors.py are listed in Section 8 of ARCHITECTURE.md. Open an issue to discuss a change to them before writing code.
  • Changes to the prompt template need a live evaluation run showing no regression (python -m eval.run_eval --mode live ...).

Pull requests

  1. Fork the repository and create a branch from main.
  2. Keep the change focused on one thing, and match the style of the surrounding code.
  3. Add or update tests. Bug fixes should include a test that fails without the fix.
  4. Update README.md if behaviour visible to users changes, and add a line under "Unreleased" in CHANGELOG.md.
  5. Make sure all three checks above pass, then open the pull request and fill in the template.

Releasing (maintainers)

  1. Move the "Unreleased" entries in CHANGELOG.md under the new version.
  2. Set the same version in pyproject.toml and slmrouter/__init__.py.
  3. Merge to main, then tag and push: git tag v0.2.0 && git push origin v0.2.0.

The Release workflow checks that the tag matches the version, runs the tests, publishes to PyPI and creates a GitHub release.

License

By contributing, you agree that your contributions are licensed under the Apache License 2.0, as described in Section 5 of that license.