User Guide

This guide covers using NequIP, from initial setup to advanced optimization techniques.

Getting Started provides framework-level information for new users, including installation, basic workflow, file types, and the pretrained foundation potentials.

Configuration covers setting up training runs, including config files, data preparation, and model configuration.

Training Techniques explores advanced training methods like fine-tuning (e.g. of the foundation potentials) and multitask learning.

Accelerations details performance optimization techniques for faster training and inference.

Reference contains conventions, troubleshooting guides, and frequently asked questions.

• Getting Started
  • Installation
  • The NequIP Workflow
  • File Types
  • Foundation Potentials
• Configuration
  • Config File
  • Data Configuration
  • Model Hyperparameters
  • Loss and Metrics
• Training Techniques
  • Fine-Tuning
  • Multitask Training
  • Weights and Biases Sweeps
• Accelerations
  • PyTorch 2.0 Compiled Training
  • Distributed Data Parallel Training
  • GPU Kernel Modifiers
  • Precision Settings
• Reference
  • Conventions
  • FAQs
  • PyTorch Version Compatibility
  • Troubleshooting