Graph-pes

Latest version: v0.0.28

Safety actively analyzes 722491 Python packages for vulnerabilities to keep your Python projects secure.

Scan your dependencies

Page 1 of 4

0.0.26

Fixed a bug whereby gradients were not being propagated through the three-body angle and distance terms.

Improved handling of early stopping.

Fixed a bug where custom callbacks were causing graph-pes-train to fail.

0.0.24

Use [`vesin`](https://luthaf.fr/vesin/latest/index.html#) for accelerated neighbour list construction.

Add `ase_calculator` method to `GraphPESModel` for easy access to an ASE calculator wrapping the model.

Update the `mace` interfaces to use the default torch dtype if none is specified.

Add `ruff` check to CI.

0.0.22

Add support for the `MatterSim` potential.

**Breaking change**: removed the `WeightedLoss` class, and placed the weight directly on the `Loss` instance.

Added a `summary.yaml` file to the output of each training/testing run that stores results locally.

0.0.21

Fixed an inconsistency in the output shapes of `mace-torch` model predictions.

Added documentation for the `freeze` family of functions.

Fixed numerical instability in the `PaiNN` model.

0.0.19

Added

Added more fine-grained control over parameter freezing.

Fixed

Parameter counting bug

0.0.18

Added

Added support for using arbitrary ``mace-torch`` models within ``graph-pes``, including the ``MACE-MP`` and ``MACE-OFF`` foundation models.

Support for custom batchers for properties in the ``other`` field via the `register_custom_batcher` decorator.

Changed

Updated the documentation for the `graph-pes-train` command.

Page 1 of 4

© 2025 Safety CLI Cybersecurity Inc. All Rights Reserved.