Runs ML-driven chemistry predictions with UMA models via the FAIRChemCalculator for ASE workflows.
npx skills add https://github.com/jkitchin/skillz --skill fairchem
Please help me install this Agent Skill. Skill: fairchem Source: https://github.com/jkitchin/skillz/tree/main/skills/programming/fairchem Command: npx skills add https://github.com/jkitchin/skillz --skill fairchem
FAIRChem provides expert guidance for using Meta's FAIRChem library to run machine-learning-driven chemistry calculations across catalysis, materials, molecules, MOFs, and molecular crystals, delivering 100-1000× speedups over traditional DFT.
uma-s-1p1 (small) or uma-m-1p1 (medium) for universal predictions.Load UMA model (e.g., uma-m-1p1) and create FAIRChemCalculator for domain omat, then run an ASE LBFGS optimization on a bulk lattice.
High-intent search queries and answers about installing and using this skill.
FAIRChem delivers 100-1000× speedups using pretrained UMA models (uma-s-1p1 or uma-m-1p1) for energy and force predictions across catalysis, materials, molecules, MOFs, and crystals. Load a model, create a FAIRChemCalculator as an ASE-compatible calculator, and run your workflows—geometry optimization, MD, NEB, or high-throughput screening—without changing your ASE code.
Yes. UMA models are domain-agnostic universal models. Specify the domain task name (ocat, omat, omol, odac, omc) when initializing FAIRChemCalculator, and the same pretrained model adapts to your domain, eliminating the need for separate task-specific models.
FAIRChem supports multi-GPU inference for large-scale screening. Distribute batch inference across GPUs by leveraging ASE's parallelization or batching your structures, then pass them through FAIRChemCalculator to achieve linear scaling for high-throughput workflows.
Install fairchem-core, authenticate with Hugging Face (to access pretrained UMA models), have ASE installed for geometry optimization and MD, and optionally enable turbo mode for faster inference. No other dependencies are required.
Yes. FAIRChemCalculator integrates directly with ASE, so you can run LBFGS optimization, constrained geometry searches, NEB transitions, and MD simulations using standard ASE syntax without modifications.
Uma-s-1p1 is the small model for fast inference on resource-constrained systems; uma-m-1p1 is the medium model offering higher accuracy at moderate computational cost. Choose based on your accuracy-speed trade-off and available hardware.
Hongyu-yuApply FAIRChem UMA models with ASE integration for materials science simulations.
jkitchinBuild ASE atomistic structures and run simulations with calculators.
whichguyManage machine learning model and dataset discovery, download, and distribution via the Hugging Face Hub CLI.
pxlxingliangExtract DFT metrics from ABACUS, VASP, or Quantum ESPRESSO directories into JSON reports.