Graph-pes

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.

Add support for ASE `.db` files for backing datasets.

Simplified the logic for distributed training set ups.

Added the `graph-pes-test` command.

Changed

Changed the implementation of `TensorNet` to exactly match that in `TorchMD`.

Fixed

Fixed test warnings.

Upgraded `locache` to `4.0.2` to fix some dataset caching issues.

Removed coloured logging for cleaner SLURM etc. output.

Added

Added the `graph-pes-resume` script.

Flushed logging for the `LoggedProgressBar` callback.

Changed

Use explicit `.cutoff`, `.batch` and `.ptr` properties on `AtomicGraph` objects

Aligned the `graph-pes` implementation of MACE with that of `ACEsuite/mace`.

Fixed

Fixed a bug whereby some parameters were being duplicated within a single parameter group.

Added

Added verbose logging to the early training callback.

Changed

[dev] moved ruff formatting to pre-commit hooks.

bumped `load-atoms` to 0.3.8.

Changed

Updated the minimum Python version to 3.9.

Un-pinned the `ase==3.22` dependency.

Migrate to using `data2objects>=0.1`