User Guide

This guide is the in-depth reference for Enerzyme. It explains configuration schemas, model design, workflow mechanics, and operational practices. For a linear first-run tutorial, start with Getting Started.

Concepts

• Overview
  • Workflow map
  • CLI command map
  • Configuration-driven design
  • Reading path
• Units and Standard Fields
  • Standard fields
  • Energy and length units
  • Forces vs gradients
  • Loss and metric weights
  • Time units in MD
  • Uncertainty fields
• Configuration System
  • Section index
  • Training artifacts
  • Override rules at inference
  • Legacy vs current schema

Models

• Architecture Catalog
  • Internal architectures
  • External wrappers
  • Selection guidelines
  • Spin and charge
  • Reference configs
• Modelhub Reference
  • Structure
  • Key fields
  • Output layout
  • Multiple models
• Layer Stack
  • Registered layers
  • Typical charge-aware stack
  • Layer ordering matters
  • Shared build_params
  • Monitoring energy terms
• Loss, Metrics, and Uncertainty
  • Loss types
  • Metrics and early stopping
  • Shallow ensemble uncertainty
  • Committee training
  • UDD at simulation time
  • Unit weights
• Checkpoints and Resume
  • Checkpoint layout
  • Resume modes
  • Pretraining
  • EMA
  • Lightning multi-GPU
  • Logs and config snapshot
  • Internal AL checkpoint

Workflows

• Prediction and Evaluation
  • Command
  • Config overrides
  • Test set sources
  • Outputs
  • Simple predict mode
  • Multi-model predict
• Simulation Tasks
  • Command
  • Task matrix
  • Shared Simulation keys
  • Constraints
  • Optimizers
  • Integrators
  • NEB inputs
  • Calculator integration
• Enhanced Sampling and PLUMED
  • Static PLUMED input
  • PLUMED units
  • Dynamic generators (-pp)
  • plumed vs plumed_scan
  • Hybrid calculators and UDD
  • OPES / proton transfer
• Active Learning Workflows
  • Decision guide
  • Enerzyme internal dataset AL
  • Enerzymette-driven external loop
• Fragment Extraction
  • Command
  • Configuration
  • Parameters
  • Prerequisites
  • Pipeline position
  • Outputs
• QM Annotation
  • Command
  • Configuration
  • Suppliers
  • QMDriver options
  • Output schema
  • Merging into training
  • Environment
  • Related integrations

Integrations

• Enerzymette Integration
  • Install
  • Main launchers
  • PLUMED CV plugins
  • Utility commands
  • Active-learning launcher flags (summary)
  • Boundary with Enerzyme
  • Recommended task directory
• Server Mode
  • Commands
  • Implementation
  • Request format
  • Response
  • Multi-model serving
  • Server config
  • Deployment notes
• Bond Assignment
  • Command
  • Arguments
  • Implementation
  • Use cases
  • Requirements
• ORCA and TeraChem Bridges
  • Batch labeling (Enerzyme annotate)
  • ORCA ExtOpt + TeraChem (Enerzymette)
  • Setup checklist
  • TeraChem timing QC
  • Relationship to training

Operations

• Troubleshooting
  • Installation
  • Data
  • Training
  • Simulation
  • QM annotation
  • Active learning
  • Documentation build
• Performance Tuning
  • Data loading
  • Training
  • Simulation
  • Neighbor list memory
  • When to use server mode
• Reproducibility and Archiving
  • Seeds
  • Archive with each model
  • Document environment
  • Enerzymette task archives
  • Pretrain chains
  • Config drift
  • QM data provenance