Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
51 lines (37 loc) · 1.84 KB

CONTRIBUTING.md

File metadata and controls

51 lines (37 loc) · 1.84 KB
Raw

Contributing

Development setup

uv sync --dev              # dev tools: ruff, pytest, pre-commit
uv run pre-commit install  # enable the git hooks (large-file guard)

Large files / data — keep them out of git

Raw data, model weights, and caches are gitignored (data/, *.h5, *.npy, *.pt, *.ckpt, scratch dirs) and must never be committed. The history was once bloated to ~67 GB by accidentally-committed multi-GB data files, and a 53 MB demo GIF had to be purged from history.

Two safeguards are in place:

  • A pre-commit hook (check-added-large-files) blocks staging any file over 10 MB locally — run uv run pre-commit install so it's active.
  • CI re-checks it on every PR to main (large-files job); the build fails if any tracked file exceeds 10 MB, so the guard holds even if someone skips the hook.

Large assets (demos, datasets, weights) belong in external hosting — Hugging Face, a GitHub Release, or Git LFS — not in the repo. Demo media should be a small compressed clip (e.g. VP9 .webm), not a multi-MB GIF.

If you use an editor "auto-checkpoint" / auto-commit extension, disable it or make sure it honors .gitignore — that was the original source of the bloat. (Note: those orphaned commits stay local; they are unreachable and never get pushed.)

If you cloned ablation-pipeline before June 2026

That branch's history was rewritten to remove the 53 MB GIF, so the commit IDs changed. If you have an older clone, reset — do not merge (a git pull/merge would drag the old file back and re-push it):

git fetch origin
git checkout ablation-pipeline
git reset --hard origin/ablation-pipeline   # stash/save any local work first

Or simply re-clone. main was not rewritten, so main-based work is unaffected.

Lint & test

uv run ruff check .
uv run pytest