Remove softmax from output

[rasbt]

I don't think there should be a softmax in the output layer if you use CrossEntropyLoss since CrossEntropyLoss takes logits as inputs (and performs a LogSoftmax internally): https://pytorch.org/docs/stable/generated/torch.nn.CrossEntropyLoss.html.

[Yard1]

Have you tried running the code?

[rasbt]

No, the code generally doesn't work on my computer (getting import errors with the logger). Looking at it carefully, I can't see an issue with the fix. It should run ok.

[rasbt]

Fixed it in the example code so that it can be tested by the CI workflow

[Yard1]

Thanks!