skills.rest
learningmatter-mit
learningmatter-mitOFFICIAL
Explore all skills by this author →
WebsiteGitHub

ml-mace-finetune

Fine-tune MACE interatomic potentials on labeled atomistic datasets.

144|21|Updated Jan 8, 2026
Search Tags:#fine-tuning#dataset-conversion#gpu-training#mace#interatomic-potentials#energy-forces-stress#materials-ai
One-click install
npx skills add https://github.com/learningmatter-mit/AtomisticSkills --skill ml-mace-finetune
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ml-mace-finetune
Source: https://github.com/learningmatter-mit/AtomisticSkills/tree/main/.agents/skills/ml-mace-finetune
Command: npx skills add https://github.com/learningmatter-mit/AtomisticSkills --skill ml-mace-finetune

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires ase, numpy, pyyaml, mace, and includes scripts (resource) components.

What problem does it solve?

MACE fine-tuning helps domain researchers adapt a foundation machine-learning interatomic potential to a specific chemical system or physical property when the out-of-the-box model is not accurate enough.

Core Features & Use Cases

  • Convert labeled JSON into MACE-ready training data: prepares .xyz files with energy, forces, and optional stress labels (including VASP kB → eV/ų stress conversion).
  • Generate a complete MACE finetune configuration: creates finetune_config.yaml compatible with mace_run_train, including options for freezing the backbone, reinitializing readout, and multi-head fine-tuning.
  • Benchmark and validate during the loop: supports running training, then extracting standardized training history metrics for comparison to foundation performance.

Quick Start

Use the ml-mace-finetune skill to fine-tune a foundation MACE model by asking it to run data preparation from your labeled JSON, generate finetune_config.yaml for your train/validation .xyz files, and execute mace_run_train on your chosen GPU device.

Frequently Asked Questions about ml-mace-finetune

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I fine-tune MACE interatomic potentials on a custom atomistic dataset?▼

To fine-tune MACE interatomic potentials, you need to convert labeled JSON data into MACE extxyz files with correct REF_* keys, generate a finetune_config.yaml aligned to mace_run_train arguments, and execute GPU-native training for energy, forces, and stress regression.

What is the process for converting labeled JSON structures into MACE training data?▼

Converting labeled JSON structures into MACE training data involves preparing .xyz files with energy, forces, and optional stress labels, including VASP kB to eV/ų stress conversion, to ensure compatibility with mace_run_train requirements.

Can I freeze the backbone or reinitialize the readout when fine-tuning MACE models?▼

Yes, you can freeze the backbone or reinitialize the readout during MACE fine-tuning by specifying these options in the generated finetune_config.yaml, allowing multi-head fine-tuning for target chemical systems and physical properties.

Do I need a GPU to run MACE fine-tuning for machine learning interatomic potentials?▼

Yes, MACE fine-tuning requires a GPU device to execute mace_run_train for machine learning interatomic potentials, ensuring efficient training and benchmarking for energy, forces, and stress regression scenarios.

How do I extract training metrics after running MACE fine-tuning?▼

After running MACE fine-tuning, you can extract standardized training history metrics to benchmark and validate the results, enabling direct comparison of energy, forces, and stress performance against the foundation model.

Related Skills

Explore More Education & Research Skills→
learningmatter-mitlearningmatter-mit
144

ml-matgl-finetune

Fine-tune MatGL interatomic potentials on labeled atomistic datasets with PyTorch Lightning.

Official
Advanced
learningmatter-mitlearningmatter-mit
144

ml-fairchem-finetune

Fine-tune Fairchem interatomic potentials on labeled structure datasets.

Official
Advanced
LiorZLiorZ
2

biotite

Read, manipulate, analyze, and fetch biomolecular structures and sequences in Python.

Community
Advanced
learningmatter-mitlearningmatter-mit
144

ml-mlip-benchmark

Benchmark machine learning interatomic potentials against labeled datasets using MAE/RMSE metrics.

Official
Intermediate