Installation

Model APIs

Install the package from PyPI:

pip install gpn

This base installation contains the supported model architectures, explicit AutoClass registration, NumPy, PyTorch, Transformers, and jaxtyping.

Inference and training

Install file-backed inference commands and their data dependencies with:

pip install "gpn[inference]"

For the maintained GPN and GPN-Star training paths:

pip install "gpn[train]"

PyTorch accelerator selection depends on your platform. For CUDA installations, follow the official PyTorch installer before or alongside GPN.

Development

GPN supports Python 3.13. Its Transformers version is pinned exactly in the package metadata, and the complete contributor environment is captured by the committed uv lockfile:

git clone https://github.com/songlab-cal/gpn.git
cd gpn
uv sync --extra train --group dev --group docs
uv run pytest

Normal tests are offline. Published-checkpoint validation is an explicit, networked audit and is never scheduled weekly.