An ASE calculator interface for MACE models that can be used within Kinetica.jl's ASENEBCalculator
kinetic calculator. Implements a MACECalculator
builder, which can optionally be augmented with a TorchDFTD3Calculator
as in MACE's mace_mp
constructor.
This package comes with CondaPkg configuration for fetching the Python dependencies (mace-torch
and torch-dftd3
packages from PyPI). Assuming you already have Kinetica installed, if you only need to run MACE on a CPU then installation should be as simple as
pkg> add KineticaMACEBuilder
This will fetch the default version of PyTorch, which usually does not have CUDA capabilities. If MACE needs to run on a GPU then additional packages may need to be specified in your Julia project's CondaPkg.toml, depending on your hardware. See the PyTorch installation instructions for further information.
As with all ASE calculator builders in Kinetica, this can simply be instantiated and passed to the kinetic calculator:
using KineticaMACEBuilder
builder = MACEBuilder("/path/to/my.model", "cuda")
calc = ASENEBCalculator(builder, "./calc")