This package provides a CLI for relaxing ASE Atoms objects using an ML force field model.

Clone with git then run:

pip install .

The package comes with a CLI:

ml_ff_tools [OPTIONS]

Arguments:

• model_path: Path to saved PyTorch model file. Should be a MACE model.

• data_path: Path to data file containing ASE Atoms objects in extended XYZ format.

Options:

• -s, --save-path: Path to save relaxed XYZ file. Default ./relaxed_atoms.xyz.

• --model-type: Only mace models currently supported.

• -b, --batch-size: Batch size for minibatch relaxation. Default 32.

• --device: Device to run model on. Either cpu or cuda. Default cpu.

• --max-num-iters: Max number of minimization iterations. Default 200.

• --max-step-size: Max step size for minimization. Default 1.0.

Example:

ml_ff_tools model.pth data.xyz -s relaxed.xyz --batch-size 64 --device cuda

This will relax the atoms in data.xyz using the model model.pth on a GPU, with a batch size of 64, saving the relaxed structures to relaxed.xyz.