GitLab

* Add return lengths

* make pad a bit more flexible so it can be used as collate_fn

* check all kwargs sent to encoding method are known

* fixing kwargs in encodings

* New AddedToken class in python

This class let you specify specifique tokenization behaviors for some special tokens. Used in particular for GPT2 and Roberta, to control how white spaces are stripped around special tokens.

* style and quality

* switched to hugginface tokenizers library for AddedTokens

* up to tokenizer 0.8.0-rc3 - update API to use AddedToken state

* style and quality

* do not raise an error on additional or unused kwargs for tokenize() but only a warning

* transfo-xl pretrained model requires torch

* Update src/transformers/tokenization_utils.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* improve mem handling

* improve mem for pos ax encodings

* [Modelcard] bart-squadv2

* Update README.md

* Update README.md

* Create README.md

* changes in model usage section

* minor changes in output visualization

* minor errata in readme

* Create README.md

* Update README.md

* Create README.md

* Apply suggestions from code review
Co-authored-by: Julien Chaumond <chaumond@gmail.com>

* Add link to new comunity notebook (optimization)

related to https://github.com/huggingface/transformers/issues/4842#event-3469184635

This notebook is about benchmarking model training with/without dynamic padding optimization. 
https://github.com/ELS-RD/transformers-notebook 

Using dynamic padding on MNLI provides a **4.7 times training time reduction**, with max pad length set to 512. The effect is strong because few examples are >> 400 tokens in this dataset. IRL, it will depend of the dataset, but it always bring improvement and, after more than 20 experiments listed in this [article](https://towardsdatascience.com/divide-hugging-face-transformers-training-time-by-2-or-more-21bf7129db9q-21bf7129db9e?source=friends_link&sk=10a45a0ace94b3255643d81b6475f409

), it seems to not hurt performance.

Following advice from @patrickvonplaten I do the PR myself :-)

* Update notebooks/README.md
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* nyu-mll: roberta on smaller datasets

* Update README.md

* Update README.md
Co-authored-by: Alex Warstadt <alexwarstadt@gmail.com>

* Quicktour part 1

* Update

* All done

* Typos
Co-authored-by: Thomas Wolf <thomwolf@users.noreply.github.com>

* Address comments in quick tour

* Update docs/source/quicktour.rst
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Update from feedback
Co-authored-by: Thomas Wolf <thomwolf@users.noreply.github.com>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Cleaner warning when loading pretrained models

This make more explicit logging messages when using the various `from_pretrained` methods. It also make these messages as `logging.warning` because it's a common source of silent mistakes.

* Update src/transformers/modeling_utils.py
Co-authored-by: Julien Chaumond <chaumond@gmail.com>

* Update src/transformers/modeling_utils.py
Co-authored-by: Julien Chaumond <chaumond@gmail.com>

* style and quality
Co-authored-by: Julien Chaumond <chaumond@gmail.com>

* Have documentation fail on warning

* Force ci failure

* Revert "Force ci failure"

This reverts commit f0a4666ec2eb4cd00a4da48af3357defc63324a0.

* Create README.md

@julien-c check out that dataset meta tag is right

* Fix typo
Co-authored-by: Julien Chaumond <chaumond@gmail.com>

* fix #5081 and improve backward compatibility (slightly)

* add nlp to setup.cfg - style and quality

* align default to previous default

* remove test that doesn't generalize

Fix for https://github.com/huggingface/transformers/issues/3809

* add support for gradient checkpointing in BERT

* fix unit tests

* isort

* black

* workaround for `torch.utils.checkpoint.checkpoint` not accepting bool

* Revert "workaround for `torch.utils.checkpoint.checkpoint` not accepting bool"

This reverts commit 5eb68bb804f5ffbfc7ba13c45a47717f72d04574.

* workaround for `torch.utils.checkpoint.checkpoint` not accepting bool
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Configure all models to use output_hidden_states as argument passed to foward()

* Pass all tests

* Remove cast_bool_to_primitive in TF Flaubert model

* correct tf xlnet

* add pytorch test

* add tf test

* Fix broken tests

* Configure all models to use output_hidden_states as argument passed to foward()

* Pass all tests

* Remove cast_bool_to_primitive in TF Flaubert model

* correct tf xlnet

* add pytorch test

* add tf test

* Fix broken tests

* Refactor output_hidden_states for mobilebert

* Reset and remerge to master
Co-authored-by: Joseph Liu <joseph.liu@coinflex.com>
Co-authored-by: patrickvonplaten <patrick.v.platen@gmail.com>

* Add model cards for Microsoft's MiniLM

* XLMRobertaTokenizer

* format

* Add thumbnail

* finishing up

* Fixed resize_token_embeddings for transfo_xl model

* Fixed resize_token_embeddings for transfo_xl.

Added custom methods to TransfoXLPreTrainedModel for resizing layers of
the AdaptiveEmbedding.

* Updated docstring

* Fixed resizinhg cutoffs; added check for new size of embedding layer.

* Added test for resize_token_embeddings

* Fixed code quality

* Fixed unchanged cutoffs in model.config

* Added feature to move added tokens in tokenizer.

* Fixed code quality

* Added feature to move added tokens in tokenizer.

* Fixed code quality

* Fixed docstring, renamed sym to 	oken.
Co-authored-by: Rafael Weingartner <rweingartner.its-b2015@fh-salzburg.ac.at>

* Update glossary

* Update docs/source/glossary.rst
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* finish benchmark

* fix isort

* fix setup cfg

* retab

* fix time measuring of tf graph mode

* fix tf cuda

* clean code

* better error message

fix bart doc

Authored-by: i.boytsov <i.boytsov@MAC867.local>