olm.train.callbacks.lr_monitor_cb¶
Learning rate monitoring callback.
Classes¶
LRMonitorCallback([log_every]) |
Callback to monitor and log learning rate. |
|---|---|
class olm.train.callbacks.lr_monitor_cb.LRMonitorCallback(log_every: int = 100)¶
Bases: TrainerCallback
Callback to monitor and log learning rate.
- Parameters: log_every – Log learning rate every N steps.
on_step_end(trainer, step: int, loss: float) → None¶
Log learning rate after each optimization step if needed.
class olm.train.callbacks.lr_monitor_cb.TrainerCallback¶
Bases: object
Base class for trainer callbacks.
on_batch_begin(trainer: Trainer, batch_idx: int) → None¶
Called at the beginning of each batch.
on_batch_end(trainer: Trainer, batch_idx: int, loss: float) → None¶
Called at the end of each batch.
on_epoch_begin(trainer: Trainer, epoch: int) → None¶
Called at the beginning of each epoch.
on_epoch_end(trainer: Trainer, epoch: int) → None¶
Called at the end of each epoch.
on_step_begin(trainer: Trainer, step: int) → None¶
Called at the beginning of each optimization step (after gradient accumulation).
on_step_end(trainer: Trainer, step: int, loss: float) → None¶
Called at the end of each optimization step.
on_train_begin(trainer: Trainer) → None¶
Called at the beginning of training.
on_train_end(trainer: Trainer) → None¶
Called at the end of training.