nvalchemi.training.TrainingStrategy.restore_checkpoint

TrainingStrategy.restore_checkpoint(root_folder, checkpoint_index=-1, map_location=None, *, validators=None)

Restore checkpoint state into this already-constructed strategy.

Parameters:

• root_folder (Path | str) – Root directory containing checkpoint files.

• checkpoint_index (int, optional) – Checkpoint index to load. -1 loads the latest manifest index.

• map_location (str | torch.device | None, optional) – Device override passed through to torch.load().

• validators (Sequence[CheckpointValidator] | None, optional) – Optional loaded-checkpoint validators forwarded to the lower-level loader.

Returns:

Loaded checkpoint payload from nvalchemi.training.load_checkpoint().

Return type:

Mapping[str, Any]