Contents Menu Expand Light mode Dark mode Auto light/dark, in light mode Auto light/dark, in dark mode Skip to content
metatrain 2025.11.dev64+ge8d22bebf documentation
Light Logo Dark Logo
  • Installation
  • Getting started
    • Quickstart
    • Basic Usage
    • Customize a Dataset Configuration
    • Advanced Base Configuration
    • Training YAML Reference
    • Override Architecture’s Default Parameters
    • Restarting and Checkpoints
    • Finetuning example
  • Configuration and Units
    • Units
  • Available Architectures
    • FlashMD
    • GAP
    • LLPR
    • NanoPET (deprecated)
    • PET
    • SOAP-BPNN
  • Tutorials
    • Beginner Tutorials
      • How to prepare data for training
      • Fine-tuning
      • Training a model from scratch
      • Running molecular dynamics with ASE
      • Model validation with parity plots for energies and forces
    • Advanced Tutorials
      • Transfer Learning (experimental)
      • Computing LLPR uncertainties
      • Training a model with ZBL corrections
      • Fitting generic targets
      • Training a FlashMD model
      • Multi-GPU training
  • Concepts and Design
    • Output naming
    • Loss functions
    • Auxiliary outputs
  • Frequently Asked Questions
  • Citing Metatrain
  • Developer documentation
    • Contributing
    • Life Cycle of an Architecture
    • Adding a new architecture
    • Dataset Information
    • Adding a new loss function
    • CLI API
      • Train
      • Eval
      • Export
      • Formatter
    • Utility API
      • Additive models
        • Removing additive contributions
        • Composition model
        • ZBL short-range potential
      • Additive models
        • Scaler model
        • Removing the scale from targets
      • Data
        • Combining dataloaders
        • Dataset
        • Reading a dataset
        • Readers
        • Target data Writers
        • Converting Systems to ASE
      • Architectures
      • Augmentation
      • Device
      • Dtype
      • Errors
      • Evaluating a model
      • External Naming
      • IO
      • Jsonschema
      • Logging
      • Long-range
      • Loss
      • Metrics
      • Neighbor lists
      • Custom omegaconf functions
      • Output gradient
      • Averaging predictions per atom
      • Summing over atoms
      • Data type and device transfers
      • Unit handling
    • Changelog
Back to top
View this page

Concepts and Design¶

This section covers advanced concepts of the metatrain library, such as output naming, auxiliary outputs, and wrapper models.

  • Output naming
  • Loss functions
  • Auxiliary outputs
Next
Output naming
Previous
Multi-GPU training
Copyright © 2025, metatrain developers
Made with Sphinx and @pradyunsg's Furo