xlm.utils.checkpoint_with_thinning

ThinningCheckpoint

Bases: ModelCheckpoint

Checkpoint callback that saves after every N steps and keeps checkpoints at K*N step intervals.

Parameters:

Name	Type	Description	Default
keep_multiple	int	Keep checkpoints for every keep_multiple * every_n_train_steps	1
dirpath	Optional[Union[Path, str]]	directory to save the model file	None
filename	Optional[str]	checkpoint filename. Must contain {step} in the pattern	None
every_n_train_steps	Optional[int]	Number of training steps between checkpoints	None
**kwargs		Additional arguments passed to ModelCheckpoint	required

Example:: >>> callback = StepBasedCheckpoint( ... dirpath='checkpoints', ... filename='model-{epoch}-{step}', ... every_n_train_steps=1000, ... keep_multiple=10 ... ) # This will: # - Save a checkpoint every 1000 steps # - Keep checkpoints at steps 10000, 20000, etc. # - Delete intermediate checkpoints