unipkainfer 0.1.0

Standalone CPU and GPU inference runtime for Uni-pKa

unipkainfer

unipkainfer is a standalone, inference-only packaging of Uni-pKa, Uni-Mol, and the minimal pure-Python Uni-Core runtime they require. It supports CPU and single-GPU inference through PyTorch and does not build Uni-Core's optional fused CUDA extensions.

This distribution contains code derived from Uni-pKa (Apache-2.0), and code derived from Uni-Mol and Uni-Core (MIT). Copyright and license notices are preserved in the accompanying license and third-party notice files.

Installation

python -m pip install unipkainfer
unipka-download-model

By default, model checkpoints are downloaded to the user data directory. On Linux this is normally:

~/.local/share/unipkainfer/models

Choose another location when needed:

unipka-download-model --output-folder /path/to/models

Python API

from rdkit import Chem
from unipkainfer import predict_standard_free_energy

free_energy = predict_standard_free_energy(Chem.MolFromSmiles("CCO"))

Use UnipkaFreeEnergyConfig(model_dir=...) to override the default model location. A CUDA-enabled PyTorch installation and compatible NVIDIA driver are required for GPU inference.

See THIRD_PARTY_NOTICES.md, unipkainfer/unicoreinfer/UPSTREAM.md, and the included licenses for upstream provenance.