User Guide

• Introduction
  • What is MACE ?
• Installation
  • pypi installation
  • pip installation from source
• Troubleshooting and Q&A Guide
  • Data Loading Issues
  • Energy reference (E0s) Problems
  • Initial Metrics Interpretation
  • Dataset Quality Issues
  • Multi-head Finetuning Issues
  • Cutoff Radius Selection
  • Memory Issues
  • Getting More Help
• Training
  • Script
  • Training files
  • Model
  • Reference energies
  • SWA and EMA
  • Data keys
  • Float precision
  • Set batch size
  • Validation parameters
  • Heterogeneous labels
  • Devices
  • Checkpoints
• Evaluation
• Heterogeneous Data Training
  • Supported File Formats
  • Configuration
• ASE calculator
  • Running MD simulations
• MACE descriptors
• Analytical Hessians
• Dipole Moments and Polarizabilities with MACE
  • Using MACE-MDP: Pretrained Model for Dipole Moments and Polarizabilities
  • Training Your Own Model
• CUDA Acceleration with cuEquivariance Library
  • Installation
  • Usage
• OpenMM Interface
  • Installation
  • Pure MD simulations
  • Hybrid ML/MM simulations
  • Alchemical simulations with MACE
• MACE in LAMMPS
  • Preparing your model
  • Instructions for GPU
  • Instructions for CPU
• MACE in LAMMPS with ML-IAP
  • Introduction
  • Preparing Your Model for ML-IAP
  • Installation Requirements
  • Using MACE with ML-IAP in LAMMPS
  • Performance Considerations
  • Limitations and Caveats
  • Debugging and Environment Variables
  • Example LAMMPS Script
• Foundation models
  • Pretrained MACE-MP models
  • MACE-OFF23: Transferable Organic Force Fields
  • MACE-MDP: Pretrained Model for Dipole Moments and Polarizabilities
  • MACE-ANI-CC: Coupled cluster Accurate Pretrained Model for H, C, N, O elements
• Electrostatic MACE
  • Architecture Summary
  • Installation
  • Available Checkpoints
  • Basic Inference
  • Reading Dipole and Charge Density
  • Fine-tuning from Polar Foundation Checkpoints
• Fine-tuning Foundation Models
  • Naive Fine-tuning
  • Multihead Replay Fine-tuning
  • LoRA Fine-tuning
• Multihead Replay Finetuning
  • Introduction
  • Replay Dataset Options
  • Method 1: Preprocessing via CLI
  • Method 2: MP Shortcut for Compatible Models
  • Method 3: Direct Multihead Finetuning with run_train
  • Monitoring and Evaluating
  • Tips for Successful Multihead Finetuning
• LoRA Fine-tuning
  • Introduction
  • How It Works
  • Usage
  • Key Parameters
  • Weight Merging
  • Python API
  • Tips for Successful LoRA Fine-tuning
• Multihead Training for MACE
• Large Dataset Pre-processing
  • HDF5 files preprocessing
  • LMDB files preprocessing
• Multi-GPUs Training
  • Multi-nodes training
  • Single-node multi-GPU training