Training API

Trainer

ununennium.training.Trainer

Features: - Mixed precision training (AMP) - Distributed Data Parallel (DDP) - Gradle accumulation - Experiment logging

Usage

trainer = Trainer(
    model=model,
    optimizer=optimizer,
    scheduler=scheduler,
    callbacks=[EarlyStopping(), ModelCheckpoint()]
)
trainer.fit(train_loader, val_loader, epochs=100)

Callbacks

EarlyStopping

Stop training when validation metric stops improving.

ModelCheckpoint

Save model weights periodically or on best metric.

LearningRateMonitor

Log learning rate changes.