Training module# Training checkpoints Manual save and restart Training functions Restartable hook state Periodic checkpoint hook Model reconstruction specs Serialization scope MACE checkpoints and cuEquivariance Distributed training Lower-level loader API reference Training strategy API Strategies Optimizer helpers Serialization and checkpoints Fine-tuning API Strategy Hooks Training update hooks TrainingStage Distributed data parallel PyTorch profiler traces Mixed precision Autocast scope Gradient accumulation Validation Stage constraints Composition rules Checkpointing EMA model averaging Restartable update hooks API reference Losses — Training Terms Dtype alignment Leaf and composition Concrete losses Weight schedules Reduction helpers Validation Training vs validation ValidationConfig Standalone validation API reference
