Distributed Training with TensorFlow Java

Please make sure that this is a feature request. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:feature_template

System information

• TensorFlow version (you are using): 2.X
• Are you willing to contribute it (Yes/No): Yes, when able and available

Describe the feature and the current behavior/state.
Tensorflow on Python has tf.distribute.Strategy API to distribute training across multiple GPUs or multiple machines.

Will this change the current api? How?
Yes, it will add a new awesome feature

Who will benefit with this feature?

• Anyone that requires to speed up training a DL model
• Anyone that requires to train a DL model with big data
• Anyone who wants to create or add Java support for APIs that leverages tf.distribute.Strategy such as TensorflowOnSpark, Spark Tensorflow Distributor or Horovod

Any Other info.
https://www.tensorflow.org/guide/distributed_training

Start by reading the TensorFlow distributed training guide and comparing its tf.distribute.Strategy API with the Java bindings in this repository. Determine which distributed-training capabilities and Java API surface are in scope, then define tests or examples that demonstrate multi-GPU or multi-machine training before implementation.