tf.VariableAggregation

Indicates how a distributed variable will be aggregated.

tf.distribute.Strategy distributes a model by making multiple copies (called "replicas") acting data-parallel on different elements of the input batch. When performing some variable-update operation, say var.assign_add(x), in a model, we need to resolve how to combine the different values for x computed in the different replicas.

  • NONE: This is the default, giving an error if you use a variable-update operation with multiple replicas.
  • SUM: Add the updates across replicas.
  • MEAN: Take the arithmetic mean ("average") of the updates across replicas.
  • ONLY_FIRST_REPLICA: This is for when every replica is performing the same update, but we only want to perform the update once. Used, e.g., for the global step counter.
Class Variables
MEAN tf.VariableAggregation
NONE tf.VariableAggregation
ONLY_FIRST_REPLICA tf.VariableAggregation
SUM tf.VariableAggregation

© 2020 The TensorFlow Authors. All rights reserved.
Licensed under the Creative Commons Attribution License 3.0.
Code samples licensed under the Apache 2.0 License.
https://www.tensorflow.org/versions/r2.4/api_docs/python/tf/VariableAggregation