Fixed typing exception throwing issues with JIT

[frgfm]

As per #3027, some JIT issues were introduced in #2860 with typing, that weren't caught initially by tests.
This PR addresses the issues related to the densenet module.

This PR should close #3027

[frgfm]

I'm still getting the following error with mypy:

torchvision/models/densenet.py:74: error: Name 'forward' already defined on line 70  [no-redef]
        @torch.jit._overload_method  # noqa: F811
         ^
torchvision/models/densenet.py:80: error: Name 'forward' already defined on line 70  [no-redef]
        def forward(self, input: Tensor) -> Tensor:  # noqa: F811
        ^

But I'm not very knowledgeable on JIT overlead decorator. Should we accept that some mypy tests fail? Or is there a way around this without having JIT issues? @fmassa @pmeier

[pmeier]

Should we accept that some mypy tests fail?

Sort of. I suggest we simply exclude densenet.py from the type check for now and only re-enable it if the issue with torchscript is fixed.

You can include

[mypy-torchvision.models.densenet.*]
ignore_errors=True

into the mypy.ini file.

[frgfm]

Alright, it should be good to go now 👌

[fmassa]

Thanks!

[pmeier]

@frgfm Do you want to have another go at this as pytorch/pytorch#51675 fixed the behavior of type: ignore comments for torch.jit?

[frgfm]

@pmeier Thanks for the mention! Yes, it would be bad karma for me to leave this unfinished haha

[frgfm]

@pmeier Here it is #3395 !