Creates an optimizer training op.
(loss, init_lr, num_train_steps, num_warmup_steps, use_tpu)
source not stored for this graph (policy: none)
nothing calls this directly
no test coverage detected