* 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>
* move generation_*.py src files into generation/*.py
* populate generation.__init__ with lazy loading
* move imports and references from generation.xxx.object to generation.object
* Support for Bart and LayoutLM, and partial support for XLNet
* Support for mbart
* A lot of new models supported
* Support for other models
* LayoutLM fix
* Use strings instead of classes
