.pt or .model LAMMPS

[aminahmadish]

I have recently used the mace to run minimization with LAMMPS. I can see some newer updates introduced, but most of them just include the files with the ".model" extension, and not ".pt" extension. I found a few newer ones in this [https://huggingface.co/cyrusyc/mace-universal/tree/main/pretrained], but the files with ".pt" do not work!

[ilyes319]

.model files are the actual model and are the ones to use for lammps. .pt are training checkpoints.
all the mace foundation models for materials trained by our team is centralized here now.

[aminahmadish]

Thanks. I have trouble running LAMMPS with .model! Is there any sample config file or any documents to show how I can use it?
I could run it before with .pt files. Here is my config file:

dimension 3
boundary p p p
units real

atom_style charge
atom_modify map hash
read_data B80_t1buckyball_mace.data

pair_style mace no_domain_decomposition
pair_coeff 2023-12-03-mace-128-L1_epoch-199.model-lammps.pt B

neighbor 2 bin
neigh_modify every 10 delay 0 check no

thermo_style multi
thermo 10

timestep 0.25

restart 5000 output/min_B80_t1buckyball_mace.restart1 output/min_B80_t1buckyball_mace.restart2
dump 1 all dcd 10 output/min_B80_t1buckyball_mace.dcd

minimize 0.0 1.0e-8 1000 100000

[ilyes319]

You need to first compile the .model to use in lammps. Please see the doc: https://mace-docs.readthedocs.io/en/latest/guide/lammps.html.