View source on GitHub
|
A distributed computation library for TF.
See tensorflow/contrib/distribute/README.md for overview and examples.
Classes
class AllReduceCrossDeviceOps: Reduction using all-reduce.
class CollectiveAllReduceStrategy: Distribution strategy that uses collective ops for all-reduce.
class CrossDeviceOps: Base class for cross-device reduction and broadcasting algorithms.
class DistributeConfig: A config tuple for distribution strategies.
class DistributionStrategy: A list of devices with a state & compute distribution policy.
class MirroredStrategy: Mirrors vars to distribute across multiple devices and machines.
class Monitor: Executes training steps, recovers and checkpoints.
class MultiWorkerAllReduce: All-reduce algorithms for distributed TensorFlow.
class OneDeviceStrategy: A distribution strategy for running on a single device.
class ParameterServerStrategy: A parameter server DistributionStrategy.
class ReplicaContext: tf.distribute.Strategy API when in a replica context.
class StandardInputStep: Step with a standard implementation of input handling.
class StandardSingleLossStep: A step function that implements a training step for a feed forward network.
class Step: Interface for performing each step of a training algorithm.
class TPUStrategy: TPU distribution strategy implementation.
class UpdateContext: Context manager when you are in update() or update_non_slot().
Functions
get_cross_replica_context(...): Returns the current tf.distribute.Strategy if in a cross-replica context.
get_distribution_strategy(...): Returns the current tf.distribute.Strategy object.
get_loss_reduction(...): tf.distribute.ReduceOp corresponding to the last loss reduction.
get_replica_context(...): Returns the current tf.distribute.ReplicaContext or None.
get_strategy(...): Returns the current tf.distribute.Strategy object.
has_distribution_strategy(...): Return if there is a current non-default tf.distribute.Strategy.
has_strategy(...): Return if there is a current non-default tf.distribute.Strategy.
in_cross_replica_context(...): Returns True if in a cross-replica context.
initialize_tpu_system(...): Initialize the TPU devices.
require_replica_context(...): Verify in replica_ctx replica context.
run_standard_tensorflow_server(...): Starts a standard TensorFlow server.