tf.distribute.experimental.CollectiveCommunication
Stay organized with collections
Save and categorize content based on your preferences.
Communication choices for CollectiveOps.
AUTO
: Default to runtime's automatic choices.
RING
: TensorFlow's ring algorithms for all-reduce and
all-gather.
NCCL
: Use ncclAllReduce for all-reduce, and ring algorithms for
all-gather.
Class Variables
Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates.
Last updated 2020-10-01 UTC.
[null,null,["Last updated 2020-10-01 UTC."],[],[],null,["# tf.distribute.experimental.CollectiveCommunication\n\n\u003cbr /\u003e\n\n|---------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------------------------|\n| [TensorFlow 2 version](/api_docs/python/tf/distribute/experimental/CollectiveCommunication) | [View source on GitHub](https://github.com/tensorflow/tensorflow/blob/v1.15.0/tensorflow/python/distribute/cross_device_ops.py#L975-L986) |\n\nCommunication choices for CollectiveOps.\n\n#### View aliases\n\n\n**Compat aliases for migration**\n\nSee\n[Migration guide](https://www.tensorflow.org/guide/migrate) for\nmore details.\n\n[`tf.compat.v1.distribute.experimental.CollectiveCommunication`](/api_docs/python/tf/distribute/experimental/CollectiveCommunication), \\`tf.compat.v2.distribute.experimental.CollectiveCommunication\\`\n\n\u003cbr /\u003e\n\n- `AUTO`: Default to runtime's automatic choices.\n- `RING`: TensorFlow's ring algorithms for all-reduce and all-gather.\n- `NCCL`: Use ncclAllReduce for all-reduce, and ring algorithms for all-gather.\n\nClass Variables\n---------------\n\n- `AUTO`\n- `NCCL`\n- `RING`"]]