GitLab

supprot fa-2 + right padding + forward

* Pin Keras for now out of paranoia

* Add the keras pin to _tests_requirements.txt too

* Make sure the Keras version matches the TF one

* make fixup

* remove docstrings CodeGen from objects_to_ignore

* autofix codegen docstrings

* fill in the missing types and docstrings

* fixup

* change descriptions to be in a separate line

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

* update n_ctx description in CodeGenConfig

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

* Remove ChineseCLIPImageProcessor, ChineseCLIPTextConfig, ChineseCLIPVisionConfig from check_docstrings

* Run fix_and_overwrite for ChineseCLIPImageProcessor, ChineseCLIPTextConfig, ChineseCLIPVisionConfig

* Replace <fill_type> and <fill_docstring> in configuration_chinese_clip.py, image_processing_chinese_clip.py with type and docstring values

---------
Co-authored-by: vignesh-raghunathan <vignesh_raghunathan@intuit.com>

revert

* initial commit

* add processor, add fuyu naming

* add draft processor

* fix processor

* remove dropout to fix loading of weights

* add image processing fixes from Pedro

* fix

* fix processor

* add basic processing fuyu test

* add documentation and TODO

* address comments, add tests, add doc

* replace assert with torch asserts

* add Mixins and fix tests

* clean imports

* add model tester, clean imports

* fix embedding test

* add updated tests from pre-release model

* Processor: return input_ids used for inference

* separate processing and model tests

* relax test tolerance for embeddings

* add test for logit comparison

* make sure fuyu image processor is imported in the init

* fix formattingh

* more formatting issues

* and more

* fixups

* remove some stuff

* nits

* update init

* remove the fuyu file

* Update integration test with release model

* Update conversion script.

The projection is not used, as confirmed by the authors.

* improve geenration

* Remove duplicate function

* Trickle down patches to model call

* processing fuyu updates

* remove things

* fix prepare_inputs_for_generation to fix generate()

* remove model_input

* update

* add generation tests

* nits

* draft leverage automodel and autoconfig

* nits

* fix dtype patch

* address comments, update READMEs and doc, include tests

* add working processing test, remove refs to subsequences

* add tests, remove Sequence classification

* processing

* update

* update the conversion script

* more processing cleanup

* safe import

* take out ModelTesterMixin for early release

* more cl;eanup

* more cleanup

* more cleanup

* and more

* register a buffer

* nits

* add postprocessing of generate output

* nits

* updates

* add one working test

* fix test

* make fixup works

* fixup

* Arthur's updates

* nits

* update

* update

* fix processor

* update tests

* passe more fixups

* fix

* nits

* don't import torch

* skip fuyu config for now

* fixup done

* fixup

* update

* oups

* nits

* Use input embeddings

* no buffer

* update

* styling processing fuyu

* fix test

* update licence

* protect torch import

* fixup and update not doctested

* kwargs should be passed

* udpates

* update the impofixuprts in the test

* protect import

* protecting imports

* protect imports in type checking

* add testing decorators

* protect top level import structure

* fix typo

* fix check init

* move requires_backend to functions

* Imports

* Protect types

---------
Co-authored-by: Pedro Cuenca <pedro@huggingface.co>
Co-authored-by: ArthurZucker <arthur.zucker@gmail.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: Lysandre <lysandre@huggingface.co>

* final fix for FA2 dtype

* try

* oops

* Update src/transformers/models/falcon/modeling_falcon.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* apply fix everywhere

---------
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Add default template warnings

* make fixup

* Move warnings to FutureWarning

* Move warnings to FutureWarning

* fix make fixup

* Remove futurewarning

* fix

* last attempt

* current work

* fix forward compatibility

* save all special tokens

* current state

* revert additional changes

* updates

* remove tokenizer.model

* add a test and the fix

* nit

* revert one more break

* fix typefield issue

* quality

* more tests

* fix fields for FC

* more nits?

* new additional changes

* how

* some updates

* simplify all

* more nits

* revert some things to original

* nice

* nits

* a small hack

* more nits

* ahhaha

* fixup

* update

* make test run on ci

* use subtesting

* update

* Update .circleci/create_circleci_config.py

* updates

* fixup

* nits

* replace typo

* fix the test

* nits

* update

* None max dif pls

* a partial fix

* had to revert one thing

* test the fast

* updates

* fixup

* and more nits

* more fixes

* update

* Oupsy 👁



* nits

* fix marian

* on our way to heaven

* Update src/transformers/models/t5/tokenization_t5.py
Co-authored-by: Lysandre Debut <hi@lysand.re>

* fixup

* Update src/transformers/tokenization_utils_fast.py
Co-authored-by: Leo Tronchon <leo.tronchon@gmail.com>

* Update src/transformers/tokenization_utils_base.py
Co-authored-by: Leo Tronchon <leo.tronchon@gmail.com>

* fix phobert

* skip some things, test more

* nits

* fixup

* fix deberta

* update

* update

* more updates

* skip one test

* more updates

* fix camembert

* can't test this one

* more good fixes

* kind of a major update

- seperate what is only done in fast in fast init and refactor
- add_token(AddedToken(..., speicla = True)) ignores it in fast
- better loading

* fixup

* more fixups

* fix pegasus and mpnet

* remove skipped tests

* fix phoneme tokenizer if self.verbose

* fix individual models

* update common tests

* update testing files

* all over again

* nits

* skip test for markup lm

* fixups

* fix order of addition in fast by sorting the added tokens decoder

* proper defaults for deberta

* correct default for fnet

* nits on add tokens, string initialized to special if special

* skip irrelevant herbert tests

* main fixes

* update test added_tokens_serialization

* the fix for bart like models and class instanciating

* update bart

* nit!

* update idefix test

* fix whisper!

* some fixup

* fixups

* revert some of the wrong chanegs

* fixup

* fixup

* skip marian

* skip the correct tests

* skip for tf and flax as well

---------
Co-authored-by: Lysandre Debut <hi@lysand.re>
Co-authored-by: Leo Tronchon <leo.tronchon@gmail.com>

Don't drop decoder_input_ids without also dropping decoder_attention_mask

improve docstrings

Add tf-nightly-rocm to _is_tf_available check

* Deleted LukeConfig and ran check_docstrings.py

* Filled docstring information

---------
Co-authored-by: louie <louisparizeau@Chicken.local>

* add warning when no speaker embeddings in speecht5._generate_speech

* modify warning to error

* adapt generation test

fix fa2 + dropout issue

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Update logits_process.py docstrings + match arg fields to __init__'s

* Ran `make style`

* First step

* fix

* add adjustements for gptq

* change to `_pre_quantization_dtype`

* Update src/transformers/modeling_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* fix serialization

* Apply suggestions from code review
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* fixup

---------
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Adjust length limits and allow naked conversation list inputs

* Adjust length limits and allow naked conversation list inputs

* Maybe use a slightly more reasonable limit than 1024

* Skip tests for old models that never supported this anyway

* Cleanup input docstrings

* More docstring cleanup + skip failing TF test

* Make fixup

* Remove BertGenerationTokenizer from objects to ignore

The file BertGenerationTokenizer is removed from
objects to ignore as a first step to fix docstring.

* Docstrings fix for BertGenerationTokenizer

Docstring fix is generated for BertGenerationTokenizer
by using check_docstrings.py.

* Fix docstring for BertGenerationTokenizer

Added sep_token type and docstring in BertGenerationTokenizer.

* Remove space in template comment

I think the space between the eos and bos tokens is not present in the actual template output. I'm using this documentation as a reference for everyone asking about prompting, so would like to clarify whether there's a space or not :)

* Update fast tokenizer too

* Apply to Code Llama

* Link to original code snippet.

* fix resume_from_checkpoint bug

* update code

make fsdp ram efficient loading optional

* remove from OBJECTS_TO_IGNORE

* run check_docstrings.py

* fill in information

* ignore CodeLlamaTokenizer

* Remove CanineConfig from check_docstrings

* Run fix_and_overwrite for CanineConfig

* Replace <fill_type> and <fill_docstring> in configuration_canine.py with type and docstring values

---------
Co-authored-by: vignesh-raghunathan <vignesh_raghunathan@intuit.com>

* fix flava doctest

* add shape

* adapt

* Fixed KeyError for Mistral

* Removed try block

* Removed whitespace

* First draft

* Update conversion script

* Update copied from statements

* Fix style

* Add copied from to config

* Add copied from to processor

* Run make fixup

* Add docstring

* Update docstrings

* Add method

* Improve docstrings

* Fix docstrings

* Improve docstrings

* Remove onnx

* Add flag

* Address comments

* Add copied from to model tests

* Add flag to conversion script

* Add code snippet

* Address more comments

* Address comment

* Improve conversion script

* More improvements

* Add expected objectness logits

* Skip test

* Improve conversion script

* Extend conversion script

* Convert large checkpoint

* Fix doc tests

* Convert all checkpoints, update integration tests

* Add checkpoint_path arg

* Fix repo_id

* Disable default system prompt for LLaMA

* Update test to not expect default prompt

* fix fa-2 import

* nit

* fix docstring dpr config

* fix style

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

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

* fix

* [skip-ci] fix

* [skip-ci] fix

* [skip-ci] fix

* [skip-ci] fix

* fix

* fix

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

* update check_docstrings

* update docstring

* fix

* fix

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

* [docstring] Remove 'BertGenerationConfig' from OBJECTS_TO_IGNORE

* [docstring] Fix docstring for 'BertGenerationConfig' (#26638)

* [DOCS] Update docstrings for  and  tokenizer

* [DOCS] add pad_token argument to whisper tokenizer docstring

* [FIX] Reword pad_token description

* [CHORE] Apply style formatting

---------
Co-authored-by: jmcdonnell <jmcdonnell@fieldbox.ai>

* Remove UniSpeechConfig

* Remove , at the end otherwise check_docstring changes order

* Auto add new docstring

* Update docstring for UniSpeechConfig

* Remove from check_docstrings

* Remove UniSpeechSatConfig and UniSpeechSatForCTC from check_docstrings

* Remove , at the end

* Fix docstring

* Update docstring for Wav2Vec2ForCTC

* Update Wav2Vec2ForCTC docstring
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

* fix style

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

* Fix backward compatibility of Conversation

I ran into a case where an external library was depending on the `new_user_input` field of Conversation. https://github.com/SeldonIO/MLServer/blob/release/1.4.x/runtimes/huggingface/mlserver_huggingface/codecs/utils.py#L37

 

This field was deprecated as part of the refactor, but if `transformers` wants to maintain backwards compatibility for now (which is mentioned in a few comments) then there's a good argument for supporting it. Some comments referred to it as an "internal" property, but it didn't start with `_` as is Python convention, so I think it's reasonable that other libraries were referencing it directly.

It's not difficult to add it to the other supported backwards-compatible properties. In addition, the implementation of `past_user_inputs` didn't actually match the past behavior (it would contain the most recent message as well) so I updated that as well.

* make style

---------
Co-authored-by: Matt <rocketknight1@gmail.com>