Experimental Distribution Strategy library.
Modules
coordinator
module: Public API for tf.distribute.experimental.coordinator namespace.
partitioners
module: Public API for tf.distribute.experimental.partitioners namespace.
rpc
module: Public API for tf.distribute.experimental.rpc namespace.
Classes
class CentralStorageStrategy
: A one-machine strategy that puts all variables on a single device.
class CollectiveCommunication
: Cross device communication implementation.
class CollectiveHints
: Hints for collective operations like AllReduce.
class CommunicationImplementation
: Cross device communication implementation.
class CommunicationOptions
: Options for cross device communications like All-reduce.
class MultiWorkerMirroredStrategy
: A distribution strategy for synchronous training on multiple workers.
class ParameterServerStrategy
: An multi-worker tf.distribute strategy with parameter servers.
class PreemptionCheckpointHandler
: Preemption and error handler for synchronous training.
class PreemptionWatcher
: Watch preemption signal and store it.
class TPUStrategy
: Synchronous training on TPUs and TPU Pods.
class TerminationConfig
: Customization of PreemptionCheckpointHandler
for various platforms.
class ValueContext
: A class wrapping information needed by a distribute function.