data/md_simulation/models/_uma.py

import torch
from fairchem.core import pretrained_mlip, FAIRChemCalculator
from fairchem.core.datasets import data_list_collater
from types import MethodType
from ase.stress import full_3x3_to_voigt_6_stress, voigt_6_to_full_3x3_stress
import os

# set hf token here
os.environ["HF_TOKEN"] = ""

def load_pretrained_uma(model_name="uma-s-1p1", device="cpu", task_name="omat"):
    predictor = pretrained_mlip.get_predict_unit(model_name, device=device)#, cache_dir="/userhome/home/aymaheshwari/hf-cache")
    calc = FAIRChemCalculator(predictor, task_name=task_name)

    # Use predictor as the model-like object
    uma_model = predictor

    def forward(self, atoms):
        # Convert atoms to data object using the calculator's a2g converter
        data_object = calc.a2g(atoms)
        batch = data_list_collater([data_object], otf_graph=True)
        
        # Get predictions directly from the model
        output = uma_model.predict(batch)
        
        results = {
            "energy": output["energy"],
            "forces": output["forces"],
            "stress": torch.tensor(full_3x3_to_voigt_6_stress(output["stress"].reshape(3,3).detach().cpu().numpy())),  # as voigt notation),,
        }
        return results

    uma_model.forward = MethodType(forward, uma_model)

    return uma_model