See INSTALL.md for user vs contributor paths. Short form:
git clone https://github.com/bugkira/pararnn-torch
cd pararnn-torch
uv sync --group devOptional local hooks (same ruff gate as CI):
uvx pre-commit installuv run ruff check
uv run ruff format --check
uv run python -c "import pararnn; print(pararnn.__version__)"Full pytest (unit + numerics, including CUDA) runs in the maintainer
pararnn-lab tree against an editable checkout of this library.
Solver, Jacobian, and scan changes keep sequential-vs-parallel agreement tests green in lab. Document tolerances per dtype (float32 vs bfloat16) from unit roundoff.
User-visible changes go under [Unreleased] in CHANGELOG.md.
Use the PR template checklist.
Before a PyPI / tag cut: uv build and confirm the wheel contains only the
pararnn package (CI release job also greps for lab paths).
Keep examples/ as short onboarding scripts.