Thanks for contributing to PyTorch Training Inspector.
- Clone the repository.
- Create a virtual environment.
- Install dependencies:
pip install -r requirements.txt
pip install -e .[dev]Run these checks before opening a pull request:
python -m pytest -v
python -m black --check .
python -m isort --check-only .
python -m mypy inspector- Keep PRs focused and well-scoped.
- Add or update tests for behavioral changes.
- Update documentation when APIs or behavior change.
- Include a clear summary of motivation, approach, and validation.
Use imperative, concise messages, for example:
Add OOM detector lookahead calibrationFix gradient hook cleanup in edge caseDocument dashboard export workflow
Open a GitHub issue with:
- Exact reproduction steps
- Expected vs observed behavior
- Environment details (
python,torch, OS, GPU) - Minimal reproducible snippet if possible
Do not open public issues for sensitive vulnerabilities. See SECURITY.md.
All contributors are expected to follow CODE_OF_CONDUCT.md.