What's new

0.6.2 (October 2026)

• Uncertainty-driven exploration (HAL). grace_uq hal runs molecular dynamics biased towards high uncertainty and collects the structures where the extrapolation grade γ is high, as new training data. Step by step: the UQ workflows tutorial.

0.6.1 (3 September 2026)

• Foundation models re-exported. The distributed models were rebuilt to fix a NaN in virial_sigma (the uncertainty contribution to the virial). Cached models now record where they were downloaded from, and the new grace_models update re-downloads only the copies that are out of date.
• Tunable uncertainty threshold. grace_uq build --threshold-percentile Q puts γ = 1 at the Q-th percentile of the training distribution instead of the default robust outlier fence, for training sets where many atoms read γ > 1. grace_uq info reports which calibration was used.
• Bulk prediction. A new predict_structures helper evaluates a whole list of structures with a single calculator instance; grace_predict and grace_uq predict now both use it.

0.6.0 (7 July 2026)

• Uncertainty estimates. Models can report a per-atom uncertainty score γ that flags extrapolation, for spotting unreliable predictions and for active learning. New command-line tool: grace_uq.
• LAMMPS Kokkos support. GRACE models can be exported for the fast Kokkos pair styles in LAMMPS, with the uncertainty score γ.
• Foundation models on HuggingFace. The released models 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 use fp32 by default: about half the memory and twice the speed. The fp64 versions have the suffix -fp64.
• Better tooling. grace_dashboard shows training curves and metrics in the browser; grace_models lists, inspects and downloads models (including Kokkos weights); a fit can save and resume in the middle of an epoch.

Important Note

If a model was fitted with gracemaker version < 0.5.1, it will not be compatible with newer versions due to a format change.
You can convert it to the new format using the following command:

grace_utils -p seed/1/model.yaml -c seed/1/checkpoints/checkpoint.best_test_loss.index update_model

This will generate new files with the "-converted" suffix, which you can replace the old files (model.yaml and checkpoints) with.

GRACE - GRaph Atomic Cluster Expansion

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.

Documentation

Please see documentation for installation instructions and examples.

Tutorial

• Fitting tutorials and their materials: here.
• Uncertainty and active learning (CALM, HAL, selection for DFT): UQ workflows.

Also in a video format

Support

Also, you may join ACE support Zulip channel for additional resources: https://acesupport.zulipchat.com/join/xtwxu2grjbtg64m3vnhypi6p/

Reference

Please see

• Y.Lysogorskiy, A. Bochkarev, R.Drautz, Graph atomic cluster expansion for foundational machine learning interatomic potentials. npj Computational Materials 12(1), 114 (2026)

@article{lysogorskiy2026graph,
  title={Graph atomic cluster expansion for foundational machine learning interatomic potentials},
  author={Lysogorskiy, Yury and Bochkarev, Anton and Drautz, Ralf},
  journal={npj Computational Materials},
  volume={12},
  number={1},
  pages={114},
  year={2026},
  doi={10.1038/s41524-026-01979-1},
  url={https://www.nature.com/articles/s41524-026-01979-1}
}

• For the CALM uncertainty indication (extrapolation grade gamma): Y. Lysogorskiy, A. Bochkarev, R. Drautz, A latent-space extrapolation grade built into graph atomic cluster expansion foundation potentials. arXiv:2609.40060 (2026)

@article{lysogorskiy2026calm,
  title={A latent-space extrapolation grade built into graph atomic cluster expansion foundation potentials},
  author={Lysogorskiy, Yury and Bochkarev, Anton and Drautz, Ralf},
  journal={arXiv preprint arXiv:2609.40060},
  year={2026},
  url={https://arxiv.org/abs/2609.40060}
}

• Anton Bochkarev, Yury Lysogorskiy, and Ralf Drautz Graph Atomic Cluster Expansion for Semilocal Interactions beyond Equivariant Message Passing. Phys. Rev. X 14, 021036 (2024)

@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}
}