Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
51 lines (36 loc) · 1.67 KB

CONTRIBUTING.md

File metadata and controls

51 lines (36 loc) · 1.67 KB
Raw

Contributing

Development setup

Use Python 3.11 for local tests:

python -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt ".[test]"
python -m pytest

Tests must use generated fixtures. Do not commit robot recordings, datasets, checkpoints, or other user data.

Pull requests

  • Keep dataset conversion and training changes independently testable.
  • Add a regression test for bug fixes.
  • Preserve the /data/input, /data/output, checkpoint, and manifest contracts unless the change is explicitly documented as breaking.
  • Run all tests and at least one Docker build before requesting review.

Dependencies and CUDA images

Runtime dependencies are pinned in requirements.txt. Update them deliberately and run pip check in every CUDA image.

The supported build matrix is defined in .github/workflows/docker.yml. A CUDA variant must use an official, digest-pinned PyTorch runtime image and pass the same test and import checks as the other variants. cuda remains an alias for cuda121.

Updating ACT

ACT is pinned by ACT_GIT_REF in Dockerfile. To update it:

  1. Review upstream changes between the old and new commits.
  2. Rebase src/act_overrides/detr/models/detr_vae.py.
  3. Update the commit in Dockerfile and THIRD_PARTY_NOTICES.md.
  4. Build all CUDA variants and run a GPU smoke training.

Do not point published images at a mutable upstream branch.

Releases

Pushes to main publish the floating CUDA tags. A Git tag matching vMAJOR.MINOR.PATCH publishes versioned tags such as MAJOR.MINOR.PATCH-cuda121. Docker Hub credentials are provided through the DOCKERHUB_USERNAME and DOCKERHUB_TOKEN repository secrets.