Thank you for helping improve SpaceBio-Bench / GeneLab Benchmark. The most useful contributions are small, reproducible, and tied to the public release surface.
- Report a broken download path, stale link, or unclear dataset-card wording.
- Reproduce a public fold or release QA command and share the exact command and output.
- Improve documentation around task layout, submission format, or citation.
- Submit benchmark predictions using the public JSON format.
Prediction submissions should follow docs/submission_format.md. Include:
- task ID and fold IDs;
- model name and short model description;
- exact preprocessing or feature representation;
- the command used to validate or evaluate the submission.
Use a GitHub issue for discussion before opening a pull request with benchmark results or new public artifacts.
Keep pull requests focused. For documentation, metadata, or public-card changes, run:
make release-qa
make hpc-public-qaFor code changes that touch benchmark behavior, also run the relevant unit tests or scripts and describe what changed in the PR body.
Use the release labels in README.md and release/release_manifest.json. In short:
- v7.1 is the canonical historical result surface.
- v7.1.2 is a public-card and metadata patch over v7.1 results.
- v9 public bulk is a metadata catalog and baseline-summary surface.
Please avoid mixing future-lane claims into the v7.1 result surface.