transformer_tutorial.py mentions log softmax even though it's not used.

[Fraser-Greenlee]

After this merge https://github.com/pytorch/tutorials/pull/695/files
LogSoftmax is not used even though the tutorial text says it is.
I thought the code was wrong and added the old softmax in which made it preform worse.

[Fraser-Greenlee]

It would be great to update the text & explain why a softmax is not used.

[Fraser-Greenlee]

Why the softmax was removed #682

[Fraser-Greenlee]

Where softmax is still mentioned https://github.com/pytorch/tutorials/blob/master/beginner_source/transformer_tutorial.py#L44