chore(docs): add usage in supervised_training_steps

[ydcjeff]

Follow up of #1589

Description: add usage examples in supervised_training_steps

Check list:

• New tests are added (if a new feature is added)
• New doc strings: description and/or example code are in RST format
• Documentation is updated (if required)

[ydcjeff]

Should we remove non_blocking argument in supervised_training_step_tpu
as per

non_blocking (bool, optional): if True and this copy is between CPU and GPU, the copy may occur asynchronously
    with respect to the host. For other cases, this argument has no effect.

or leave as it is for simplicity?

[vfdev-5]

Thanks a lot @ydcjeff !
Left a nit comment to add import.
As for non_blocking for TPU, we can keep the docstring, just adapt the text for TPU. Do you know if it works for TPU this argument ?

[ydcjeff]

Do you know if it works for TPU this argument?

Asked in pytorch/xla#2791

[vfdev-5]

Do you know if it works for TPU this argument?

Asked in pytorch/xla#2791

Thanks @ydcjeff ! I would expect that everything is non-blocking in xla until an op that should sync and output a result. My intuition is that this parameter is not taken into account...

[ydcjeff]

Do you know if it works for TPU this argument?

Asked in pytorch/xla#2791

Thanks @ydcjeff ! I would expect that everything is non-blocking in xla until an op that should sync and output a result. My intuition is that this parameter is not taken into account...

Yes, I also think like that, since xla docs doesn't mention about non_blocking.
But, since I am not sure, I filed a question just to be sure. Let's wait for the answer.

[ydcjeff]

So non_blocking is already taken care by xla, current docstring seems ok.
ready to go?