gepa-dapo-grn 0.1.0

GEPA-aware DAPO reinforcement learning with optional Global Response Normalization

gepa-dapo-grn

gepa-dapo-grn is a standalone reinforcement learning engine for research workflows. It is GEPA-shaped but GEPA-agnostic, providing DAPO optimization, curriculum tracking, safety controls, and optional Global Response Normalization (GRN) without coupling to any external GEPA repositories.

What this library is

• Standalone RL engine with a stable public API under gepa_dapo_grn.*.
• GEPA-shaped but GEPA-agnostic: feedback is just structured reward/tag dictionaries.
• Supports DAPO + curriculum + safety + GRN with conservative defaults and GRN disabled unless explicitly enabled.

Install

pip install gepa-dapo-grn

Optional extras:

• gepa-dapo-grn[hf] adds HuggingFace integration helpers.
• gepa-dapo-grn[dev] installs test and formatting tools.

Minimal example (CPU-safe)

from gepa_dapo_grn import (
    DAPOTrainer,
    DAPOConfig,
    RewardMixerConfig,
    GEPAFeedback,
)

# minimal fake feedback example
fb = GEPAFeedback(
    rewards={"truth": 1.0, "harm": -0.5},
    tags={},
    meta={"task_id": "demo"},
    abstained=False,
)

Public API

Public API is defined by __init__.py exports. Anything not exported there is considered internal and may change without notice.

What this library intentionally does NOT do

• ship datasets
• provide prompt logic
• implement scoring logic
• encode opinions on ethics or safety policy

Versioning policy

This project follows semantic versioning:

• 0.x.y while interfaces are still evolving
• bump minor for interface changes
• bump patch for bugfixes only

See CHANGELOG.md for release notes.

PyPI publishing checklist

1. Install build and upload tooling:

   pip install build twine
   

2. Build artifacts:

   python -m build
   

3. Check distributions:

   twine check dist/*
   

4. Upload to TestPyPI first:

   twine upload --repository testpypi dist/*
   

5. Install from TestPyPI to validate:

   pip install -i https://test.pypi.org/simple/ gepa-dapo-grn
   

6. Upload to PyPI:

   twine upload dist/*
   

License

MIT (see LICENSE).