GitLab

* add Bigbird ONNX config

* [FlaxWav2Vec2Model] Fix bug in attention mask

* more fixes

* add (Flax)SpeechEncoderDecoderModel PT-FX cross-test

* [FlaxSpeechEncoderDecoder] Fix input shape bug in weights init

* make style

* start working

* fix: ReformerForQA doctest

* fix: ReformerModelWithLMHead doctest

* fix: ReformerModelForSC doctest

* fix: ReformerModelForMLM doctest

* add: documentation_tests.txt

* make fixup

* change: ReformerModelForSC doctest

* change: checkpoint

* Moved functions to pytorch_utils.py

* isort formatting

* Reverted tf changes

* isort, make fix-copies

* documentation fix

* Fixed Conv1D import

* Reverted research examples file

* backward compatibility for pytorch_utils

* missing import

* isort fix

* Change the chunk_iter function to handle

the subtle cases where the last chunk gets ignored since all the
data is in the `left_strided` data.

We need to remove the right striding on the previous item.

* Remove commented line.

* Updated assertions to exceptions

* updated assertions to exceptions

* bug fixes

* fix-copies

* Update modeling_ctrl.py

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

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

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

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

* Update modeling_led.py

* Update modeling_led.py

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

* add ort-trt benchmark script

* Update README.md

* ort version can be newer

* formatting

* specify ORT version

args.model_name_or_path -> args.config_name
fix it

This avoids an unnecessary call and avoids problems during
initialization of class hierarchies.
Co-authored-by: Samuel Melm <samuel.melm@stud.uni-heidelberg.de>

* First Pass All Tests Pass

* WIP

* Adding file to documentation tests

* Change the base model for the example in the doc test.

* Fix Code Styling by running
make fixup

* Called Style

* Reverted to gpt2 model rather than distill gpt2
Then used a token classification model over a sequence model for an example.

* Fix Styling Issue

* Hopefully ignores the formatting issue.
Co-authored-by: ArEnSc <xx.mike.chung.xx@gmail.com>

Move declaration of log streams to before tests, so that results won't get compounded on top of each other

* add error message

* Use names in the error message

* allow ModelOutput

* rename to check_pt_tf_outputs and move outside

* fix style

* skip past_key_values in a better way

* Add comments

* improve code for label/loss

* make the logic clear by moving the ignore keys out

* fix _postprocessing_to_ignore

* fix _postprocessing_to_ignore: create new outputs from the remaining fields

* ignore past_key_values in TFGPT2 models for now

* make check_pt_tf_outputs better regarding names

* move check_pt_tf_models outside

* rename methods

* remove test_pt_tf_model_equivalence in TFCLIPModelTest

* Reduce TFViTMAEModelTest.test_pt_tf_model_equivalence

* move prepare_pt_inputs_from_tf_inputs outside check_pt_tf_models

* Fix quality

* Clean-up TFLxmertModelTester.test_pt_tf_model_equivalence

* Fix quality

* fix

* fix style

* Clean-up TFLEDModelTest.test_pt_tf_model_equivalence

* Fix quality

* add docstring

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

* update

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

* private repo argument to trainer

* format
Co-authored-by: Nicholas Broad <nicholas@nmbroad.com>

Adds checkpoint prefixes to the gitignore if `push_to_hub` is used along with `checkpointint_steps`

* update _create_and_check_torchscript

* Enable test_torchscript

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

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

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

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

* Fix t5 shard on TPU Pods

The current script doesn't work properly on a TPU pod because the global batch is not divided correctly per host.
This pull request fixes this issue by dividing the global batch to each host before it is shared on each host.

* fix style
Co-authored-by: ahmed-elnaggar <ahmed.elnaggar@allianz.com>

* Add doctest BERT

* make fixup

* fix typo

* change checkpoints

* make fixup

* define doctest output value, update doctest for mobilebert

* solve fix-copies

* update QA target start index and end index

* change checkpoint for docs and reuse defined variable

* Update src/transformers/models/bert/modeling_tf_bert.py
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

* Apply suggestions from code review
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

* Apply suggestions from code review
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

* make fixup
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

I create an archive of older checkpoints during training the checkpoint has a  name with `f"{checkpoint_prefix}-*.zip/.tar ` 
previously `glob(f"{checkpoint_prefix}-*")` takes all files/folders starting with the name checkpoint, and later `shutil.rmtree(checkpoint)` takes a folder name; since at some point it my get a zip file; it crashes training; adding this `if os.path.isdir(x)` allows only folders on `glob_checkpoints`

* min length must be smaller than max length

* Update min_length in tests

* add simple multi gpu complet

* add human_eval_multi_gpu

* use copy strategy to distribute across gpu, to avoid padding

* add doc string

* update code style

* use task id to arrange output

* truncate input to avoid zero pad

* Stop the copy mechanism

* update style

* restore copies to scale better in distributed mode

* update style

* replace human eval

* Apply suggestions from code review

1. Tokenize all input at the same time
2. use attention_mask to get the input length
3. other small fixes
Co-authored-by: Leandro von Werra <lvwerra@users.noreply.github.com>

* correct typo and update docstring

* update code style

* remove num sample division constraint

* remove max len calculation

* use accelerator.gather once to speed up

* use accelerate set_seed; update accelerate version

* correct gather bug
Co-authored-by: Leandro von Werra <lvwerra@users.noreply.github.com>

* Fix some doc examples
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* update proto sentencepiece model

* Revert "update proto sentencepiece model"

This reverts commit b07f671747fec35773d0b3d4788b8b15aefa0229.

* add check

* add test

* Revert "Revert "update proto sentencepiece model""

This reverts commit 46108257b8927b73627ec8f4f3eed53a95fc700d.

* test for log level

* test for log level 2

* warning at the warning level

* clean

* format

* add explanation in docstring

* ✨ update audio examples with minds dataset

* 🖍 make style

* 🖍 minor fixes for doctests

* [Trainer] tf32 arg doc

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

* Fixed some bugs involving saving during epochs
* Added tests mimicking the existing examples tests
* Added in json exporting to all `no_trainer` examples for consistency 

* fix QA sample

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

The defalut value of `padding` in `DataCollatorWithPadding` is `True`, not `False`.