MACE models

Effortlessly integrate pre-trained MACE models into your projects with the user-friendly mace-models package.

pip install mace-models

Loading models is simple with a few lines of code:

import mace_models

# Load a default MACE model
model = mace_models.load()
torch_model = model.get_model()
ase_calculator = model.get_calculator()
# get the model file 
model.get_file()

Customize your model selection by loading models from various repositories or specific revisions:

model = mace_models.load(
    "<model-name>",
    rev="<branch-or-sha>",
    remote="https://github.com/<user>/<repo>"
)

Models

The following models are available under the MIT license:

• ani500k_small Model: https://doi.org/10.1063/5.0155322 Training data: https://www.nature.com/articles/sdata2017193
• hydromace https://github.com/RokasEl/hydromace
• MACE-MP-0_large https://arxiv.org/abs/2401.00096
• MACE-MP-0 https://arxiv.org/abs/2401.00096
• MACE-MP-0_small https://arxiv.org/abs/2401.00096

Example usage with ASE

import mace_models
from ase.build import molecule

model = mace_models.load("MACE-MP-0_small")

water = molecule("H2O")
water.calc = model.get_calculator(dtype="float64")

print(water.get_potential_energy())
>>> -14.047933366728305

References

@article{kovacsEvaluationMACEForce2023,
  title = {Evaluation of the {{MACE}} Force Field Architecture: {{From}} Medicinal Chemistry to Materials Science},
  shorttitle = {Evaluation of the {{MACE}} Force Field Architecture},
  author = {Kov{\'a}cs, D{\'a}vid P{\'e}ter and Batatia, Ilyes and Arany, Eszter S{\'a}ra and Cs{\'a}nyi, G{\'a}bor},
  year = {2023},
  journal = {The Journal of Chemical Physics},
  volume = {159},
  number = {4},
  pages = {044118},
  issn = {0021-9606},
  doi = {10.1063/5.0155322},
  urldate = {2024-01-17}
}

@misc{batatiaFoundationModelAtomistic2023,
  title = {A Foundation Model for Atomistic Materials Chemistry},
  author = {Batatia, Ilyes and Benner, Philipp and Chiang, Yuan and Elena, Alin M. and Kov{\'a}cs, D{\'a}vid P. and Riebesell, Janosh and Advincula, Xavier R. and Asta, Mark and Baldwin, William J. and Bernstein, Noam and Bhowmik, Arghya and Blau, Samuel M. and C{\u a}rare, Vlad and Darby, James P. and De, Sandip and Della Pia, Flaviano and Deringer, Volker L. and Elijo{\v s}ius, Rokas and {El-Machachi}, Zakariya and Fako, Edvin and Ferrari, Andrea C. and {Genreith-Schriever}, Annalena and George, Janine and Goodall, Rhys E. A. and Grey, Clare P. and Han, Shuang and Handley, Will and Heenen, Hendrik H. and Hermansson, Kersti and Holm, Christian and Jaafar, Jad and Hofmann, Stephan and Jakob, Konstantin S. and Jung, Hyunwook and Kapil, Venkat and Kaplan, Aaron D. and Karimitari, Nima and Kroupa, Namu and Kullgren, Jolla and Kuner, Matthew C. and Kuryla, Domantas and Liepuoniute, Guoda and Margraf, Johannes T. and Magd{\u a}u, Ioan-Bogdan and Michaelides, Angelos and Moore, J. Harry and Naik, Aakash A. and Niblett, Samuel P. and Norwood, Sam Walton and O'Neill, Niamh and Ortner, Christoph and Persson, Kristin A. and Reuter, Karsten and Rosen, Andrew S. and Schaaf, Lars L. and Schran, Christoph and Sivonxay, Eric and Stenczel, Tam{\'a}s K. and Svahn, Viktor and Sutton, Christopher and {van der Oord}, Cas and {Varga-Umbrich}, Eszter and Vegge, Tejs and Vondr{\'a}k, Martin and Wang, Yangshuai and Witt, William C. and Zills, Fabian and Cs{\'a}nyi, G{\'a}bor},
  year = {2023},
  number = {arXiv:2401.00096},
  eprint = {2401.00096},
  primaryclass = {cond-mat, physics:physics},
  publisher = {{arXiv}},
  doi = {10.48550/arXiv.2401.00096},
  urldate = {2024-01-17},
  archiveprefix = {arxiv}
}

MACE is described in

@misc{batatiaMACEHigherOrder2022,
  title = {{{MACE}}: {{Higher Order Equivariant Message Passing Neural Networks}} for {{Fast}} and {{Accurate Force Fields}}},
  shorttitle = {{{MACE}}},
  author = {Batatia, Ilyes and Kov{\'a}cs, D{\'a}vid P{\'e}ter and Simm, Gregor N. C. and Ortner, Christoph and Cs{\'a}nyi, G{\'a}bor},
  year = {2022},
  number = {arXiv:2206.07697},
  eprint = {2206.07697},
  primaryclass = {cond-mat, physics:physics, stat},
  publisher = {{arXiv}},
  urldate = {2022-06-19},
  archiveprefix = {arxiv},
  langid = {english}
}