Skip to content

gracemaker#

gracemaker is a tool for fitting interatomic potentials in a general non-linear Graph Atomic Cluster Expansion (GRACE) form.

Project GRACEmaker is a heavily modified and in large parts rewritten version of the PACEmaker software geared towards support for multi-component materials and graph architectures.

What's new#

07 July 2026: 0.6.0 Release#

  • Uncertainty estimates. Models can now report a per-atom uncertainty score (gamma) that flags when a prediction is an extrapolation — useful for spotting unreliable regions and for active learning. Comes with a new grace_uq command line tool.
  • LAMMPS Kokkos support. You can now export GRACE models for the fast Kokkos pair styles in LAMMPS. The uncertainty score can be carried along, so it's available during LAMMPS runs too.
  • Foundation models on HuggingFace. The released models now include uncertainty support out of the box,
  • GRACE-3L-OMAT/OAM. There are new, larger 3-layer models (GRACE-3L-OMAT-large and GRACE-3L-OMAT-large-ft-AM).
  • Faster and lighter by default. Foundation models now use fp32 precision by default, which roughly halves memory use and about doubles speed. (fp64 versions are still available under the -fp64 name.)
  • Better tooling. A new grace_dashboard shows training curves and metrics in the browser; the grace_models command was reworked to list, inspect, and download models (including Kokkos weights); and training can now save and resume in the middle of an epoch.

Series of ACE models#

Series of ACE models

Features#

  • Support for multi-component material systems with unlimited number of interacting elements.
  • Extension of the local ACE models to also include semi-local interactions (a.k.a. message passing).

  • Line of models targeting various performance/complexity regimes and hardware:

    • GRACE/FS: Fast, mildly non-linear local model with standalone C++ implementation enabling efficient multi-CPU parallelization using MPI in LAMMPS. Intended for large-scale simulations of systems containing up to several million atoms.

    • GRACE-1L: Local, non-linear model with improved accuracy intended to run on GPU. Utilizes TensorFlow library to run simulations in LAMMPS or in python and supports multi-GPU parallelization. Suitable for modeling both small systems and large-scale simulations of hundreds of thousands of atoms.

    • GRACE-2L: Semi-local, non-linear model offering state-of-the-art accuracy. Utilizes TensorFlow library to run simulations in LAMMPS or in python. Best applied for simulating molecular systems and materials with up to tens of thousand of atoms.

What's next ?#

Documentation#

Please use the navigation bar on the left to explore documentation

License#

This code and the foundation models are distributed under the Academic Software License.

Citation#

Please cite following papers if you use GRACEmkaer in your work:

BibTeX (click to expand) 
@article{PhysRevX.14.021036,
  title = {Graph Atomic Cluster Expansion for Semilocal Interactions beyond Equivariant Message Passing},
  author = {Bochkarev, Anton and Lysogorskiy, Yury and Drautz, Ralf},
  journal = {Phys. Rev. X},
  volume = {14},
  issue = {2},
  pages = {021036},
  numpages = {28},
  year = {2024},
  month = {Jun},
  publisher = {American Physical Society},
  doi = {10.1103/PhysRevX.14.021036},
  url = {https://link.aps.org/doi/10.1103/PhysRevX.14.021036}
}