Resources

The repository includes scripts and small examples that complement the main package documentation.

Data Preparation

Dataset preparation helpers live under resources/data.

• resources/data/alexandria
• resources/data/alexandria+mptraj
• resources/data/mptraj
• resources/data/omat24

These scripts show how to download or convert source datasets and write Equitrain HDF5 files, including examples for Alexandria and MPTraj.

Training

Distributed MACE training examples live under resources/training.

• mace-alex-mptraj-multigpu.sh
• mace-alex-mptraj-multinode.sh
• mace-alex-mptraj-multinode-slurm.sh
• corresponding Accelerate YAML files

The scripts are intended as templates for single-node multi-GPU and multi-node jobs.

Model Resources

Initial model helpers and wrapper-specific examples live under resources/models.

• resources/models/ani: TorchANI export helper.
• resources/models/mace-jax: MACE foundation-to-JAX conversion helper.
• resources/models/m3gnet: M3GNet examples.
• resources/models/orb: ORB examples.
• resources/models/sevennet: SevenNet example configuration.

HDF5 Inspection

Use equitrain-hdf5-info to inspect Equitrain HDF5 files:

equitrain-hdf5-info data/train.h5

Use equitrain-hdf5-benchmark to measure sequential read throughput:

equitrain-hdf5-benchmark data/train.h5