GitLab

* Start the work on FlaxVisionEncoderDecoderModel

* Add FlaxVisionEncoderDecoderModel

* Add VisionEncoderDecoderConfig

* Make FlaxVisionEncoderDecoderModel visible to transformers

* Add test

* Fix wrong getattr usage

* Fix tests

* Add FlaxAutoModelForVision2Seq

* Expose FLAX_MODEL_FOR_VISION_2_SEQ_MAPPING

* clean-up

* add integration test

* update expected logits

* update expected scores

* Add ViT2GPT2ModelIntegrationTest + some cleaning

* Add projection layer + PT/Flax equivalence tests

* Fix import

* minor changes

* make test slow again

* Apply suggestions

* Add modeling_flax_vision_encoder_decoder to _ignore_modules in get_model_modules()

* fix copies

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

* split long strings in multiple lines

* decoder_input_ids can't be None

* Add back test_configuration_tie

* Remove attention_mask parameter

* fix test - encoder_last_hidden_state should be encoder_outputs.last_hidden_state instead of the projected vector

* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Remove more encoder_attention_mask

* remove encoder_attention_mask when calling self.decode (in FlaxVisionEncoderDecoderModule)

* Fix style + pass 1s instead of None as encoder_attention_mask

* fix init_weights

* pass None for encoder_attention_mask

* pass 1s instead of None as encoder_attention_mask

* Fix doc style
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
Co-authored-by: Suraj Patil <surajp815@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* minor modification to the wav2vec2 modeling file to support tensor-parallelism with DeepSpeed on this HuggingFace model

* refine the comments

* synch changes

* fix comments

* refine comments

* fix format

* defer to DS_TEST_PORT if set

* style
Co-authored-by: Stas Bekman <stas@stason.org>

* Dynamic configs

* Add config test

* Better tests

* Add tokenizer and test

* Add to from_config

* With save

* Changed relative imports to absolute to allow convert_graph_to_onnx.py to be run as a script

* isorted code

* Fixing mutable default argument.

* XX.

* Revert "XX."

This reverts commit 61d4bb333f6d39a7fbe31d161b8bd14787ceec2e.

* Fixing tests on master.

* Better fix.

* Lxmert doesn't have feature extractor but is bimodal.

* Add paddings to input shapes

* Add padding comment

`start_` and `end_` tensors now contain a batch_size at this point.

* Pin TF until tests are fixed

* Also pin TF CPU

* Removing Keras version pinning

* make fixup

Fixes #14272

* Fiixng slow pipeline tests

* Remove the image-segmentaiton override.

* Fixing clamping only in training.

* Wav2vec2.

* Remove last mention of `no_grad`.

* Fixing copies.

* Rename.

* Add more instructions to the release guide

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

* Address review comment
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Start PR doc

* Cleanup the quality checks and document them

* Add reference in the contributing guide

* Apply suggestions from code review
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>

* Rename file as per review suggestion
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>

* Pin Keras cause they messed their release

* Put != instead of <

* Try this way

* Back to the beginning but more agressive

* Fix of issue #13327: Wrong weight initialization for TF t5 model

* run black formatter

* fix typo

* remove my name tag from comments
Co-authored-by: Shirron <dan.shirron@intel.com>

* Adding support for `truncation` parameter on `feature-extraction`
pipeline.

Fixes #14183

* Fixing tests on ibert, longformer, and roberta.

* Rebase fix.

minimal fixes to run DataCollatorForWholeWordMask with return_tensors="np" and return_tensors="tf" (#13891)

* minimal fixes to run DataCollatorForWholeWordMask with return_tensors="np" and return_tensors="tf"

* more consinstent implementation for numpy_mask_tokens

* Fix img load rotation

* Add `load_image` to `image_utils.py`

* Implement LoadImageTester

* Use hf-internal-testing dataset

* Add img utils comments

* Refactor LoadImageTester

* Import load_image under is_vision_available

* Add cross attentions to TFGPT2Model

* change to is_pt_tf_cross_test

* A minor correction to a comment

* Remove n_ctx when creating self.crossattention
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Add LayoutXLMTokenizer and LayoutXLMTokenizerFast

* Fix styling issues

* Fix more styling issues

* Fix more styling issues

* Fix docstring

* Fix unit tests

* Fix docs

* Fix unit tests

* Fix typos and styling issues

* Fix styling issues

* Fix docstring

* Make all tests of test_tokenization_layoutxlm pass

* Add LayoutXLMProcessor

* Make fixup

* Make all LayoutXLMProcessor tests pass

* Minor fixes

* Leave LayoutLMv2Processor tests unchanged

* Fix code quality

* Move LayoutXLM tokenizers and processor to separate folder

* Fix code quality

* Apply suggestions from code review

* Replace assertions by value errors

* Remove methods from fast tokenizer
Co-authored-by: King Yiu Suen <kingyiusuen@gmail.com>

* Update Transformers to huggingface_hub >= 0.1.0

* Forgot to save...

* Style

* Fix test

* add Beit model ouput class

* inherting from BaseModelOuputWithPooling

* updated docs if use_mean_pooling is False

* added beit specific outputs in model docs

* changed the import path

* Fix docs
Co-authored-by: Niels Rogge <niels.rogge1@gmail.com>

* Add audio-classification benchmarking results

* fix distilhubert path

* check test_configuration_tie

* Fix test_configuration_tie

* make test slow again

* Remove property and use model.module.bind

* revert to slow test
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* Fix generation docstring

* Style

* Add first draft

* Make forward pass work

* Improve conversion script

* Add notebook that checks if it works

* Add BeitForSemanticSegmentation to the tests

* More improvements

* Make BeitForSemanticSegmentation consistent with Segformer

* Small bug fix

* Add BeitForSemanticSegmentation to docs

* Make sure model doesn't output hidden states when the user doesn't want to

* Make it possible to convert the large model

* Fix issue

* Fix conversion script for large model

* Add auxiliary_head option to semantic segmentation model

* Apply suggestions from @sgugger's review

* Apply suggestions from code review

* Fix failing test
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>