Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
40 lines (28 loc) · 1.56 KB

CONTRIBUTING.md

File metadata and controls

40 lines (28 loc) · 1.56 KB
Raw

Contributing to MinerU-ROCm

Thanks for helping improve this adapter! A few ground rules.

Getting started

pip install -e ".[dev]"
pip install omnidocbench-rocm
make smoke-test        # runs pytest
make publish           # runs the conformance checker

The smoke backend (no GPU) should always pass — keep it working.

Wiring up a real model

  1. Implement _infer(img, platform, config) -> str in adapter/run_adapter.py (return markdown for one image).
  2. Set BACKEND in adapter/adapter_config.py to your backend name (or pass --backend <name>).
  3. Fill in adapter/setup/00-install-deps.{sh,ps1} so a fresh machine can provision weights + runtime.
  4. Run make eval-linux (or eval-windows), then make publish.

Before opening a PR

  • make smoke-test is green.
  • make publish reports CONFORMANT.
  • Model cards reflect reality (hardware, legacy badge, artifacts); result submissions must not assign central platform review or promote legacy badges.
  • Update docs/known-gaps.md — close resolved bullets, add new ones.

Reporting results

For a result produced in this repository, commit the engine-produced evidence bundle under results/omnidocbench/v16/<platform>/ and point the applicable model card at it. For an independent result, follow the short community benchmark checklist and provide an immutable evidence permalink. Raw predictions are useful but optional.

Code of conduct

Participation in this project is governed by the Code of Conduct. Be excellent to each other.