Training function utilities

Call these ray.train APIs from inside your training function to report metrics and checkpoints, read the training context, and get dataset shards. The collective operations live in ray.train.collective.

Classes

Checkpoint	A reference to data persisted as a directory in local or remote storage.
CheckpointUploadMode	The manner in which we want to upload the checkpoint.
CheckpointConsistencyMode	Read semantics for checkpoint retrieval during an ongoing run.
TrainContext	Abstract interface for training context.
ValidationFn	Protocol for a function that validates a checkpoint.
ValidationTaskConfig	Configuration for a specific validation task, passed to report().

PreemptionInfo

Information about an imminent preemption event.

Functions

get_all_reported_checkpoints	Get all the reported checkpoints so far.
get_checkpoint	Access the latest reported checkpoint to resume from if one exists.
get_context	Get or create a singleton training context.
get_dataset_shard	Returns the ray.data.DataIterator shard for this worker.
get_preemption_info	Return the imminent preemption info for the current worker, or None.
report	Report metrics and optionally save a checkpoint.

Collective operations

barrier	Create a barrier across all workers.
broadcast_from_rank_zero	Broadcast small (<1kb) data from the rank 0 worker to all other workers.