Welcome to modelforge’s documentation!

Modelforge is a Python package to build and train machine learned interatomic potentials using PyTorch and Lightning. It is designed to be modular and flexible, allowing for easy extension and customization. It provides access to popular datasets and models, making it trivial to get started with training and evaluation.

The best way to get started is to read the Getting Started guide, which oultines how to

Contents:

• Getting Started
  • Installation
  • Use Cases for Modelforge
  • How to Use Modelforge
• Potentials
  • Potentials: Overview
  • Using TOML files to configure potentials
  • Use cases of the factory class
  • Example
  • AimNet2: how to define the postprocessing operations
• Dataset Module
  • Dataset Configuration TOML file
  • Processing of dataset entries
  • Interacting with the Dataset Module
  • yaml Metadata File Structure
  • Available Datasets and Versions
• Training
  • Training: Overview
  • Training Configuration
  • Train a Model
• Inference Mode
  • Neighborlists
• For Developers
  • How to deal with units
  • Base structure of machine learned potentials
  • Contributing to the modelforge package
• tuning module
  • RayTuner
  • tune_model()
• modelforge.curate
  • Basic Usage
  • Examples
• API Documentation
  • modelforge.dataset
  • modelforge.utils

Indices and tables

• Index

• Module Index

• Search Page