GitLab

* beit-flax

* updated FLAX_BEIT_MLM_DOCSTRING

* removed bool_masked_pos from classification

* updated Copyright

* code refactoring: x -> embeddings

* updated test: rm from_pt

* Update docs/source/model_doc/beit.rst

* model code dtype updates and
other changes according to review

* relative_position_bias
revert back to pytorch design

* upload

* correct

* correct

* correct

* finish

* up

* up

* up again

the variables of run example is not correct

* Dynamic model

* Use defensive flag

* Style

* Doc and arg rename

* Arg rename

* Add tests

* Apply suggestions from code review
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Apply suggestions from code review
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Address review comments

* Apply suggestions from code review
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* [megatron_gpt2] checkpoint v3

* bug fix

* fixes

* switch to default  from  - which is what the current megatron-lm uses

* cleanup

* back compat

Fixed expand_dims axis

* Fix MT5 documentation

The abstract is incomplete

* MT5 -> mT5

* Init FNet

* Update config

* Fix config

* Update model classes

* Update tokenizers to use sentencepiece

* Fix errors in model

* Fix defaults in config

* Remove position embedding type completely

* Fix typo and take only real numbers

* Fix type vocab size in configuration

* Add projection layer to embeddings

* Fix position ids bug in embeddings

* Add minor changes

* Add conversion script and remove CausalLM vestiges

* Fix conversion script

* Fix conversion script

* Remove CausalLM Test

* Update checkpoint names to dummy checkpoints

* Add tokenizer mapping

* Fix modeling file and corresponding tests

* Add tokenization test file

* Add PreTraining model test

* Make style and quality

* Make tokenization base tests work

* Update docs

* Add FastTokenizer tests

* Fix fast tokenizer special tokens

* Fix style and quality

* Remove load_tf_weights vestiges

* Add FNet to  main README

* Fix configuration example indentation

* Comment tokenization slow test

* Fix style

* Add changes from review

* Fix style

* Remove bos and eos tokens from tokenizers

* Add tokenizer slow test, TPU transforms, NSP

* Add scipy check

* Add scipy availabilty check to test

* Fix tokenizer and use correct inputs

* Remove remaining TODOs

* Fix tests

* Fix tests

* Comment Fourier Test

* Uncomment Fourier Test

* Change to google checkpoint

* Add changes from review

* Fix activation function

* Fix model integration test

* Add more integration tests

* Add comparison steps to MLM integration test

* Fix style

* Add masked tokenization fix

* Improve mask tokenization fix

* Fix index docs

* Add changes from review

* Fix issue

* Fix failing import in test

* some more fixes

* correct fast tokenizer

* finalize

* make style

* Remove additional tokenization logic

* Set do_lower_case to False

* Allow keeping accents

* Fix tokenization test

* Fix FNet Tokenizer Fast

* fix tests

* make style

* Add tips to FNet docs
Co-authored-by: patrickvonplaten <patrick.v.platen@gmail.com>

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

Co-authored-by: Alessandro Suglia <asuglia@fb.com>

* Removed misfiring warnings

* Revert "Removed misfiring warnings"

This reverts commit cea90de325056b9c1cbcda2bd2613a785c1639ce.

* Retain the warning, but only when the user actually overrides things

* Fix accidentally breaking just about every model on the hub simultaneously

* Style pass

* Fix special tokens not correctly tokenized

* Add testing

* Fix

* Fix

* Use user workflows instead of directly assigning variables

* Enable test of fast tokenizers

* Update test of canine tokenizer

* 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