FeNNol Pre-trained Models Collection

This repository contains a collection of pre-trained models compatible with the FeNNol library.

Models

• FeNNix-Bio1 (small, medium)
• ANI1x, ANI1ccx, ANI2x
• MACE-OFF23 (small, medium, large)
• MACE-MP (small, medium, large)

Installation

In order to get the model files (files with .fnx extensions), you can either clone this repository or download the desired model directly from the github interface.

Warning: Models larger than 100MB are stored using Git LFS. If you clone the repository and want to use these models, make sure to have Git LFS installed on your system.

Usage

To use the models, simply download the desired model and load it using the FeNNol library. For example, to load the MACE-OFF23 small model in an ASE calculator, use the following code:

from fennol.ase import FENNIXCalculator
from ase import build

atoms = build.molecule('H2O')
calc = FENNIXCalculator(model='MACE-OFF23/mace_off_small.fnx')
atoms.calc = calc
print("energy =",atoms.get_potential_energy())

Citation

If you use any of the models provided in this repository, please cite the corresponding papers. Please also cite this paper if you use the FeNNol library.

T. Plé, O. Adjoua, L. Lagardère and J-P. Piquemal. FeNNol: an Efficient and Flexible Library for Building Force-field-enhanced Neural Network Potentials. arXiv preprint arXiv:2405.01491 (2024)

@article{ple2024fennol,
  title={FeNNol: an Efficient and Flexible Library for Building Force-field-enhanced Neural Network Potentials},
  author={Pl{\'e}, Thomas and Adjoua, Olivier and Lagard{\`e}re, Louis and Piquemal, Jean-Philip},
  journal={arXiv preprint arXiv:2405.01491},
  year={2024}
}

Licence

Each model family is distributed under its own licence. Please refer to the corresponding model folder for more information.