GitLab

* Compile gelu_new with torchscript

* Compile _gelu_python with torchscript

* Wrap gelu_new with torch.jit for torch>=1.4

* Electra wip

* helpers

* Electra wip

* Electra v1

* ELECTRA may be saved/loaded

* Generator & Discriminator

* Embedding size instead of halving the hidden size

* ELECTRA Tokenizer

* Revert BERT helpers

* ELECTRA Conversion script

* Archive maps

* PyTorch tests

* Start fixing tests

* Tests pass

* Same configuration for both models

* Compatible with base + large

* Simplification + weight tying

* Archives

* Auto + Renaming to standard names

* ELECTRA is uncased

* Tests

* Slight API changes

* Update tests

* wip

* ElectraForTokenClassification

* temp

* Simpler arch + tests

Removed ElectraForPreTraining which will be in a script

* Conversion script

* Auto model

* Update links to S3

* Split ElectraForPreTraining and ElectraForTokenClassification

* Actually test PreTraining model

* Remove num_labels from configuration

* wip

* wip

* From discriminator and generator to electra

* Slight API changes

* Better naming

* TensorFlow ELECTRA tests

* Accurate conversion script

* Added to conversion script

* Fast ELECTRA tokenizer

* Style

* Add ELECTRA to README

* Modeling Pytorch Doc + Real style

* TF Docs

* Docs

* Correct links

* Correct model intialized

* random fixes

* style

* Addressing Patrick's and Sam's comments

* Correct links in docs

* BertJapaneseTokenizer accept options for mecab

* black

* fix mecab_option to Option[str]

* solve conflicts

* improve comments

* change tf t5 argument naming for TF 2.2

* correct bug in testing

[T5, Testst] Add extensive hard-coded integration tests and make sure PT and TF give equal results (#3550)

* add some t5 integration tests

* finish summarization and translation integration tests for T5 - results loook good

* add tf test

* fix == vs is bug

* fix tf beam search error and make tf t5 tests pass

* add bad words list

* make style

* add bad_words_tokens

* make style

* better naming

* make style

* fix typo

Co-authored-by: Ioannis Douratsos <ioannisd@amazon.com>

* Update the NER TF script to remove the softmax and make the pad token label id to -1

* Reformat the quality and style
Co-authored-by: Julien Plu <julien.plu@adevinta.com>

* make decoder input ids optional for t5 training

* lm_lables should not be shifted in t5

* add tests

* finish shift right functionality for PT T5

* move shift right to correct class

* cleaner code

* replace -100 values with pad token id

* add assert statement

* remove unnecessary for loop

* make style

* Add clear description of how to train T5

* correct docstring in T5

* correct typo

* correct docstring format

* update t5 model docs

* implement collins feedback

* fix typo and add more explanation for sentinal tokens

* delete unnecessary todos

* add t5 docs basis

* improve docs

* add t5 docs

* improve t5 docstring

* add t5 tokenizer docstring

* finish docstring

* make style

* add pretrained models

* correct typo

* make examples work

* finalize docs

For some reason Sphinx extremely dislikes this and crashes.

* trim seq_len below 1024 if there are columns full of pad_token_id
* Centralize trim_batch so SummarizationDataset can use it too

* Dummy inputs to model.device

* Move self.device to ModuleUtilsMixin

* delete lm_head, skips weight tying
* Fixed s3

* Add the missing token classification for XLM

* fix styling

* Add XLMForTokenClassification to AutoModelForTokenClassification class

* Fix docstring typo for non-existing class

* Add the missing token classification for XLM

* fix styling

* fix styling

* Add XLMForTokenClassification to AutoModelForTokenClassification class

* Fix docstring typo for non-existing class

* Add missing description for AlbertForTokenClassification

* fix styling

* Add missing docstring for AlBert

* Slow tests should be slow
Co-authored-by: Sakares Saengkaew <s.sakares@gmail.com>
Co-authored-by: LysandreJik <lysandre.debut@reseau.eseo.fr>

* fix merge conflicts

* add t5 summarization example

* change parameters for t5 summarization

* make style

* add first code snippet for translation

* only add prefixes

* add prefix patterns

* make style

* renaming

* fix conflicts

* remove unused patterns

* solve conflicts

* fix merge conflicts

* remove translation example

* remove summarization example

* make sure tensors are in numpy for float comparsion

* re-add t5 config

* fix t5 import config typo

* make style

* remove unused numpy statements

* update doctstring

* import translation pipeline

* solve conflicts

* move warnings below

* incorporate changes

* add pad_to_max_length to pipelines

* add bug fix for T5 beam search

* add prefix patterns

* make style

* fix conflicts

* adapt pipelines for task specific parameters

* improve docstring

* remove unused patterns

* add new default configs

* change prefix default to None

* [ci] Also run test_examples in py37

(will revert at the end of the experiment)

* InputExample: use immutable dataclass

* [deps] Install dataclasses for Py<3.7

* [skip ci] Revert "[ci] Also run test_examples in py37"

This reverts commit d29afd9959786b77759b0b8fa4e6b4335b952015.

see #3359

cc @lysandrejik

* make style

* fix conflicts