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

ml-mlip-automl

Automate MLIP hyperparameter tuning via LLM-driven iterative search over learning rate, freezing, and loss weighting.

144|21|Updated Jan 8, 2026
Search Tags:#hyperparameter tuning#llm optimization#automl#fairchem#mace#mlip#matgl
One-click install
npx skills add https://github.com/learningmatter-mit/AtomisticSkills --skill ml-mlip-automl
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ml-mlip-automl
Source: https://github.com/learningmatter-mit/AtomisticSkills/tree/main/.agents/skills/ml-mlip-automl
Command: npx skills add https://github.com/learningmatter-mit/AtomisticSkills --skill ml-mlip-automl

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It helps you avoid slow, expensive trial-and-error when training machine learning interatomic potentials by automatically searching for effective hyperparameters.

Core Features & Use Cases

  • LLM-driven iterative hyperparameter search: Uses validation MAE trends to decide the next parameter set to try instead of exhaustive grid search.
  • Cross-framework fine-tuning orchestration: Coordinates foundation fine-tuning runs for MACE, MatGL, and FairChem to evaluate configurations consistently.
  • Practical early stopping search strategy: Runs short trials (about ~10 epochs on small datasets) and stops after performance plateaus or repeated degradation.

Quick Start

Ask your agent to run a ~10-trial LLM-guided hyperparameter search for your MLIP dataset, read each run’s training_history.json to choose the next configuration, and then retrain the best setup for many epochs with early stopping.

Frequently Asked Questions about ml-mlip-automl

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

FAQPage Schema
How do I automate hyperparameter tuning for machine learning interatomic potentials?▼

Automate MLIP hyperparameter tuning by orchestrating an LLM-driven iterative search over learning rate, freezing strategy, and loss weighting for MACE, MatGL, and FairChem models.

How does LLM optimization work for MLIP fine-tuning?▼

LLM optimization for MLIP fine-tuning uses validation MAE trends from training_history.json to intelligently select the next hyperparameter configuration instead of relying on exhaustive grid search.

What is the best way to run a hyperparameter search for MACE or MatGL models?▼

Run a short trial of about 10 epochs on small datasets, parse the validation loss signals to choose the next configuration, and retrain the best setup for many epochs with early stopping.

Does this AutoML approach support foundation model fine-tuning for FairChem datasets?▼

Yes, this AutoML approach orchestrates foundation fine-tuning runs for FairChem, MatGL, and MACE across materials, chemistry, and drug discovery datasets to evaluate configurations consistently.

Why does my MLIP training validation MAE plateau or degrade during hyperparameter search?▼

Validation MAE plateaus or degrades when the hyperparameter configuration reaches convergence or overfitting, which triggers the early stopping search strategy to halt the trial.

Can I use this hyperparameter tuning workflow without writing custom search scripts?▼

Yes, you can ask your agent to run an LLM-guided hyperparameter search, which automatically reads each run's training_history.json to choose the next configuration from the available CLI space.

Related Skills

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

general-property-units

Define ASE-standard energy, force, and stress units across MLIP and DFT toolchains.

Official
Basic
hung-phanhung-phan
1

ray-tune

Automate distributed hyperparameter tuning with Ray and early stopping.

Community
Advanced
pedapudipedapudi
4

zicato-design-judges

Design outcome expectations and process judges for zicato board evaluations.

Community
Advanced
ovachieverovachiever
52

scikit-learn

Implement, evaluate, and optimize scikit-learn machine learning models with pipelines.

Community
Intermediate