GitLab

* Un-skip tests

* Add aliasing support to tf_to_pt_weight_rename

* Refactor tf-to-pt weight rename for simplicity

* Patch mobilebert

* Let us pray that the transfo-xl one works

* Add XGLM rename

* Expand the test to see if we can get more models to break

* Expand the test to see if we can get more models to break

* Fix MPNet (it was actually an unrelated bug)

* Fix MPNet (it was actually an unrelated bug)

* Add speech2text fix

* Update src/transformers/modeling_tf_pytorch_utils.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/mobilebert/modeling_tf_mobilebert.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update to always return a tuple from tf_to_pt_weight_rename

* reformat

* Add a couple of missing tuples

* Remove the extra test for tie_word_embeddings since it didn't cause any unexpected failures anyway

* Revert changes to modeling_tf_mpnet.py

* Skip MPNet test and add explanation

* Add weight link for BART

* Add TODO to clean this up a bit

---------
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>