Multiple node support?

[settle]

@haifeng-jin and others, I read that multiple GPUs are supported via distributed_strategy using MirroredStrategy, but what about multiple node support, i.e., using MultiWorkerMirroredStrategy or something similar? https://www.tensorflow.org/guide/distributed_training

[settle]

@haifeng-jin and @qingquansong, I saw that your "Automated Machine Learning in Action" MEAP book with Manning (link) a section on multi-GPU and a Jupyter notebook example (link). Would one of you please comment regarding support for parallelization on multi-nodes? It would be nice addition to the scaling section in your book too.

[haifeng-jin]

We haven't tested the multi-node support in autokeras. It is now only officially supported in KerasTuner.

[settle]

@haifeng-jin Thank you for pointing that out. I think you're referring to Section 8.2.3 "Parallel tuning" with environment variables for tuner IDs, oracle IPs, and oracle ports, is that right, or something else entirely?