GitLab

* finish

* add test

* push

* remove unnecessary code

* up

* correct test

* Update src/transformers/training_args.py

* Optimize Token Classification models for TPU

As per the XLA document XLA cannot handle masked indexing well. So token classification
models for BERT and others use an implementation based on `torch.where`. This implementation
works well on TPU. 

ALBERT token classification model uses the masked indexing which causes performance issues
on TPU. This PR fixes this issue by following the BERT implementation.

* Same fix for ELECTRA

* Same fix for LayoutLM

* made tokenizer fully picklable

* remove whitespace

* added testcase

* Properly use test_fetcher for examples

* Fake example modification

* Fake modeling file modification

* Clean fake modifications

* Run example tests for any modification.

* [deepspeed] replaced deprecated init arg

* Trigger CI

* correct

* add tests

* Update src/transformers/feature_extraction_sequence_utils.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Fix issue when labels are supplied as Numpy array instead of list

* Fix issue when labels are supplied as Numpy array instead of list

* Fix same issue in the `TokenClassification` data collator

* Style pass

* Fix issue when labels are supplied as Numpy array instead of list

* Fix issue when labels are supplied as Numpy array instead of list

* finish

* delete bogus file

* correct some stuff

* finish

* finish

Update GPT Neo ONNX config to match the changes implied by the simplification of the local attention
Co-authored-by: Michael Benayoun <michael@huggingface.co>

* Fix test_fetcher when setup is updated

* Remove example

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Add checks to build cleaner model cards

* Address review comments

* added initial files

* fixes pipeline

* fixes style and quality

* fixes doc issue and positional encoding

* fixes layer norm and test

* fixes quality issue

* fixes code quality

* removed extra layer norm

* added layer norm back in encoder and decoder

* added more code copy quality checks

* update tests

* Apply suggestions from code review

* fix import

* fix test
Co-authored-by: patil-suraj <surajp815@gmail.com>

* Push to hub when saving checkpoints

* Add model card

* Revert partial model card

* Small fix for checkpoint

* Add tests

* Add documentation

* Fix tests

* Bump huggingface_hub

* Fix test

* Add long-overdue link to the Google TRC project

* Apply suggestions from code review
Co-authored-by: Suraj Patil <surajp815@gmail.com>
Co-authored-by: Stefan Schweter <stefan@schweter.it>

* Nightly CI torch

* Version

* Reformat

* Only subset
Fix

* Revert

* Better formatting

* New channel

* add imports

* Update docs/source/perplexity.rst
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* [tokenizer] use use_auth_token for config

* args order

apparently).

fix use_cache value assign

* simplify local attention

* update tests

* add a comment and use torch.bitwise_xor

* fix

* 2nd fix

Fix issues with `TFMultipleChoiceLoss` if the choices dimension is None when `build()` is called.

* finalize

* Apply suggestions from code review

* finish cleaner implementation

* more tests

* small fix

* finish

* up