GitLab

* This will reduce "Already borrowed error":

Original issue https://github.com/huggingface/tokenizers/issues/537



The original issue is caused by transformers calling many times
mutable functions on the rust tokenizers.
Rust needs to guarantee that only 1 agent has a mutable reference
to memory at a given time (for many reasons which don't need explaining
here). Usually, the rust compiler can guarantee that this property is
true at compile time.

Unfortunately, this is impossible for Python to do that, so PyO3, the
bridge between rust and python used by `tokenizers`, will change the
compile guarantee for a dynamic guarantee, so if multiple agents try
to have multiple mutable borrows at the same time, then the runtime will
yell with "Already borrowed".

The proposed fix here in transformers, is simply to reduce the actual
number of calls that really need mutable borrows. By reducing them,
we reduce the risk of running into "Already borrowed" error.
The caveat is now we add a call to read the current configuration of the
`_tokenizer`, so worst case we have 2 calls instead of 1, and best case
we simply have 1 + a Python comparison of a dict (should be negligible).

* Adding a test.

* trivial error :(.

* Update tests/test_tokenization_fast.py
Co-authored-by: SaulLu <55560583+SaulLu@users.noreply.github.com>

* Adding reference to original issues in the tests.

* Update the tests with fast tokenizer.
Co-authored-by: SaulLu <55560583+SaulLu@users.noreply.github.com>

* [model.from_pretrained] raise exception early on failed load

Currently if `load` pretrained weights fails in `from_pretrained`, we first print a whole bunch of successful messages and then fail - this PR puts the exception first to avoid all the misleading messages.

* style
Co-authored-by: Suraj Patil <surajp815@gmail.com>

* Fixing the pipeline optimization by rescaling the logits first.

* Add test for target equivalence
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>

* Laying down building stone for more flexible ONNX export capabilities

* Ability to provide a map of config key to override before exporting.

* Makes it possible to export BART with/without past keys.

* Supports simple mathematical syntax for OnnxVariable.repeated

* Effectively apply value override from onnx config for model

* Supports export with additional features such as with-past for seq2seq

* Store the output path directly in the args for uniform usage across.

* Make BART_ONNX_CONFIG_* constants and fix imports.

* Support BERT model.

* Use tokenizer for more flexibility in defining the inputs of a model.

* Add TODO as remainder to provide the batch/sequence_length as CLI args

* Enable optimizations to be done on the model.

* Enable GPT2 + past

* Improve model validation with outputs containing nested structures

* Enable Roberta

* Enable Albert

* Albert requires opset >= 12

* BERT-like models requires opset >= 12

* Remove double printing.

* Enable XLM-Roberta

* Enable DistilBERT

* Disable optimization by default

* Fix missing setattr when applying optimizer_features

* Add value field to OnnxVariable to define constant input (not from tokenizers)

* Add T5 support.

* Simplify model type retrieval

* Example exporting token_classification pipeline for DistilBERT.

* Refactoring to package `transformers.onnx`

* Solve circular dependency & __main__

* Remove unnecessary imports in `__init__`

* Licences

* Use @Narsil's suggestion to forward the model's configuration to the ONNXConfig to avoid interpolation.

* Onnx export v2 fixes (#12388)

* Tiny fixes
Remove `convert_pytorch` from onnxruntime-less runtimes
Correct reference to model

* Style

* Fix Copied from

* LongFormer ONNX config.

* Removed optimizations

* Remvoe bad merge relicas.

* Remove unused constants.

* Remove some deleted constants from imports.

* Fix unittest to remove usage of PyTorch model for onnx.utils.

* Fix distilbert export

* Enable ONNX export test for supported model.

* Style.

* Fix lint.

* Enable all supported default models.

* GPT2 only has one output

* Fix bad property name when overriding config.

* Added unittests and docstrings.

* Disable with_past tests for now.

* Enable outputs validation for default export.

* Remove graph opt lvls.

* Last commit with on-going past commented.

* Style.

* Disabled `with_past` for now

* Remove unused imports.

* Remove framework argument

* Remove TFPreTrainedModel reference

* Add documentation

* Add onnxruntime tests to CircleCI

* Add test

* Rename `convert_pytorch` to `export`

* Use OrderedDict for dummy inputs

* WIP Wav2Vec2

* Revert "WIP Wav2Vec2"

This reverts commit f665efb04c92525c3530e589029f0ae7afdf603e.

* Style

* Use OrderedDict for I/O

* Style.

* Specify OrderedDict documentation.

* Style :)
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Try to pickle transformers

* Deal with special objs better

* Make picklable

* raise exception when arguments are incomplete

* change exception to runtime error

Successfully logging this requires Pytorch. For the purposes of this script we are not using Pytorch.

* Double check for attribute

* Use right name

It was used in shift_right.
After this change TF code is more similar to Pytorch implementations
Also, TF graphs are optimized (one node less)

Adding prepare_decoder_input_ids_from_labels methods to all ConditionalGeneration TF models (#12560)

* Adding support for `pipeline("automatic-speech-recognition")`.

- Ugly `"config"` choice for AutoModel. It would be great to have the
possibility to have something like `AutoModelFor` that would implement
the same logic (Load the config, check Architectures and load the first
one)

* Remove `model_id` was not needed in the end.

* Rebased !

* Remove old code.

* Rename `nlp`.

* fix_torch_device_generate_test

* remove @

* finish

* Validation split percentage to be used for custom data files also

Issue same as https://github.com/huggingface/transformers/issues/12406 fixed for pytorch branch run_mlm.py

* Validation split added in the right place

* Update run_clm.py

* validation split added for custom files

* Validation split added for custom files

* Update run_plm.py

* fixed validation split for custom files as input for pytorch examples in lm

* Update run_clm_no_trainer.py

* args modified

* Copy BART to MBart and rename some stuff

* Add copy statements pointing to FlaxBart

* Update/add some common files

* Update shift_tokens_rigth + fix imports

* Fix shift_tokens_right method according to MBart implementation

* Update shift_tokens_right in tests accordingly

* Fix the import issue and update docs file
* make style quality

* Do some minor changes according to patil-suraj suggestions

* Change the order of normalization layer and attention

* Add some copu statementes

* Update generate method and add integration test for mBart

* Make a few updates after a review

Besides, add `lang_code_to_id` to MBartTokenizeFast

* fix-copies; make style quality

* Apply suggestions from code review

* Apply suggestions from code review

* Apply suggestions from code review

* fix output type, style

* add copied from

* resolve conflicts
Co-authored-by: Suraj Patil <surajp815@gmail.com>

* add adafactor

* Update examples/flax/language-modeling/run_mlm_flax.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* fix_torch_device_generate_test

* remove @

* up

* up

* correct

* upload
Co-authored-by: Patrick von Platen <patrick@huggingface.co>

* implementing tflxmertmodel integration test

* move import

* revert and fix

* flax gpt neo

* fix query scaling

* update generation test

* use flax model for test

* add RoFormerTokenizerFast into AutoTokenizer

* fix typo in roformer docs

* make onnx export happy

* update RoFormerConfig embedding_size

* use jieba not rjieba

* fix 12244 and make test_alignement passed

* update ARCHIVE_MAP

* make style & quality & fixup

* update

* make style & quality & fixup

* make style quality fixup

* update

* suggestion from LysandreJik
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* make style

* use rjieba
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* fix saving and loading

* update readme

* fix_torch_device_generate_test

* remove @

* adapt flax pretrain script

* fix_torch_device_generate_test

* remove @

* up

* fix_torch_device_generate_test

* remove @

* push

* fix_torch_device_generate_test

* remove @

* add logging steps

* correct training scripts

* correct readme

* correct

* NER example for Tensorflow

* Style pass

* Style pass

* Added metric computation on the evaluation set

* Style pass

* Fixed label masking

* Style pass

* Style pass

* fix_torch_device_generate_test

* remove @

* upload

* finish dataset streaming

* adapt readme

* finish

* up

* up

* up

* up

* Apply suggestions from code review

* finish

* make style

* make style2

* finish
Co-authored-by: Patrick von Platen <patrick@huggingface.co>

* begin script

* clean example, add readme

* update readme

* remove decay mask

* remove masking

* update readme & make flake happy