I’m reading the documentation for distributed training in tensorflow:
Suppose I have three machines, and start two workers on each machine. What port number can I give each worker in TF_CONFIG? Can I choose any arbitrary port number that isn’t already taken by some standard program and the worker will use that port?