Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
41 lines (32 loc) · 1.89 KB

CONTRIBUTING.md

File metadata and controls

41 lines (32 loc) · 1.89 KB
Raw

Contributing to FeatFuse

Thanks for helping build the open benchmark for feature integration in code language models. The fastest way to contribute is to add a plugin — a new engineered feature, fusion strategy, encoder, or dataset. Every plugin is a small class/function registered with a one-line decorator, so the benchmark picks it up automatically.

Quick start

git clone https://github.com/jorge-martinez-gil/graphcodebert-feature-integration
cd graphcodebert-feature-integration
pip install -e ".[dev]"
pytest -q                       # all tests should pass
featfuse run -c configs/smoke.yaml

Adding a plugin (the common case)

You want to add Subclass / register Guide
an engineered feature PairFeature → @FEATURES.register docs/adding_a_feature.md
a fusion strategy nn.Module → @FUSIONS.register docs/adding_a_fusion.md
a code encoder encoder class → @MODELS.register docs/adding_a_model.md
a dataset / task loader → @DATASETS.register docs/adding_a_model.md

Ground rules (scientific integrity)

  1. No fabricated results. Any number in code, docs, or a PR description must be reproducible from a committed config via featfuse run.
  2. Tests required. New plugins need at least one test; pytest -q must pass.
  3. Docs required. Update the relevant docs/ page and, if user-facing, the README.
  4. Keep the core light. The classical pipeline must keep running without torch/transformers (those live behind the neural extra).

Pull requests

  • Branch from main, keep PRs focused, and fill in the PR template checklist.
  • CI runs the test suite on Python 3.9–3.12 plus the CPU smoke benchmark.
  • By contributing you agree your work is released under the project's MIT License.