SpikeDetection

class lightning.pytorch.callbacks.SpikeDetection(mode='min', window=10, warmup=1, atol=None, rtol=2.0, exclude_batches_path=None, finite_only=True)[source]

Bases: SpikeDetection, Callback

on_train_batch_end(trainer, pl_module, outputs, batch, batch_idx)[source]

Checks if we currently have a loss-spike.

Return type:

None