Ray
Restores training state from a checkpoint object.
These checkpoints are returned from calls to save_to_object().