tf.contrib.nccl.all_sum

tf.contrib.nccl.all_sum(tensors)

Returns a list of tensors with the all-reduce sum across tensors.

The computation is done with an all-reduce operation, so if only some of the returned tensors are evaluated then the computation will hang.

tensors: The input tensors across which to sum; must be assigned to GPU devices.

List of tensors, each with the sum of the input tensors, where tensor i has the same device as tensors[i].

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.