You need to agree to share your Hugging Face information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

These checkpoints are derivative works of Meta's UMA and AllScAIP models and are provided under the FAIR Chemistry License (see LICENSE), including the FAIR Chemistry Acceptable Use Policy. Publications that use them must acknowledge the use of the Meta materials.

Log in or Sign Up to review the conditions and access this model content.

UniHyena baseline checkpoints

Fine-tuned UMA and AllScAIP checkpoints used as baselines in the UniHyena project, which benchmarks a new machine-learning interatomic potential against them on OC20 S2EF, MatPES, MD22, OCELOT and QMOF.

Contents

One folder per run: <benchmark>/<benchmark>-<target>-<model>/

File What
inference_ckpt.pt final fine-tuned checkpoint (fairchem inference format, no optimizer state)
config/ the fairchem fine-tuning config the run used
results/ validation / test metrics, run summary, split metadata, training log

Load one with fairchem (commit 8dacc96ae56a7fa5aac158ba0559b039d1f8b727):

from fairchem.core import FAIRChemCalculator
from fairchem.core.units.mlip_unit import load_predict_unit

predictor = load_predict_unit("md22/md22-DHA-uma/inference_ckpt.pt")
calc = FAIRChemCalculator(predictor, task_name="omol")  # MD22 runs use the omol task

Base models

  • UMA-S 1.2.1 (uma-s-1p2p1, from facebook/UMA): Wood et al., UMA: A Family of Universal Models for Atoms, arXiv 2506.23971.
  • AllScAIP-md conserving (allscaip-md-conserving-all-omol, from facebook/OMol25): Qu et al., A recipe for scalable attention-based MLIPs, arXiv 2603.06567.

License

These checkpoints are derivative works of Meta's UMA and AllScAIP models and are distributed under the FAIR Chemistry License (included as LICENSE), subject to the FAIR Chemistry Acceptable Use Policy it contains. By using them you agree to those terms. Publications that use them must acknowledge the use of the Meta materials. The upstream models are not offered in comprehensively sanctioned jurisdictions, China, Russia or Belarus; the same restriction applies here.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support