Aritra Roy Gosthipaty

Hey @NiklasRosenstein I made a unit test for the PR Feel free to review. I think I need to look into the enhancement part for the other processors. But if...

Hey @NiklasRosenstein [Here is what I woud really like to see.](https://github.com/NiklasRosenstein/pydoc-markdown/commit/10848424ada53e742f1a29e7e18eff9279d4ac18#r44610820)

Hey @AntivistRock Thanks for the ticket! We will track it internally. The colab notebook will help a lot in the process 😄

Tagging @NielsRogge @Rocketknight1 @gante for the PR review!

Hey @gante thanks for the insights. > The other errors is because the attribute base_model_prefix does not match the name of the attribute that is used to hold the models....

Fixes #18543 CC: @NielsRogge

References - hard_softmax: https://gist.github.com/ariG23498/08cdae21637b8b61bdd6d21d11719fb3 - resize_attention_map: https://gist.github.com/ariG23498/3777f8d9be25de8ae782256f5aacb2c5

@amyeroberts I have added the `shape_list` back to places which are copied from other parts of the repository. This was done to pass the `make fixup`.

Hey @amyeroberts and @gante I had to swap back in the `shape_list` and also the `axis=range(len(shape_list(logits)))[dim]` to pass the tests.

> Perhaps relevant to the TF-PT mismatch: [#18555 (comment)](https://github.com/huggingface/transformers/pull/18555#issuecomment-1230066514) Hey @gante I do not think this might be the source of the problem. `nn.functional.interpolate` has been used twice in the...