GitLab

* add sdpa

* wip

* cleaning

* add ref

* yet more cleaning

* and more :)

* wip llama

* working llama

* add output_attentions=True support

* bigcode sdpa support

* fixes

* gpt-bigcode support, require torch>=2.1.1

* add falcon support

* fix conflicts falcon

* style

* fix attention_mask definition

* remove output_attentions from attnmaskconverter

* support whisper without removing any Copied from statement

* fix mbart default to eager renaming

* fix typo in falcon

* fix is_causal in SDPA

* check is_flash_attn_2_available in the models init as well in case the model is not initialized through from_pretrained

* add warnings when falling back on the manual implementation

* precise doc

* wip replace _flash_attn_enabled by config.attn_implementation

* fix typo

* add tests

* style

* add a copy.deepcopy on the config in from_pretrained, as we do not want to modify it inplace

* obey to config.attn_implementation if a config is passed in from_pretrained

* fix is_torch_sdpa_available when torch is not installed

* remove dead code

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

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

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

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

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

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

* remove duplicate pretraining_tp code

* add dropout in llama

* precise comment on attn_mask

* add fmt: off for _unmask_unattended docstring

* precise num_masks comment

* nuke pretraining_tp in LlamaSDPAAttention following Arthur's suggestion

* cleanup modeling_utils

* backward compatibility

* fix style as requested

* style

* improve documentation

* test pass

* style

* add _unmask_unattended tests

* skip meaningless tests for idefics

* hard_check SDPA requirements when specifically requested

* standardize the use if XXX_ATTENTION_CLASSES

* fix SDPA bug with mem-efficient backend on CUDA when using fp32

* fix test

* rely on SDPA is_causal parameter to handle the causal mask in some cases

* fix FALCON_ATTENTION_CLASSES

* remove _flash_attn_2_enabled occurences

* fix test

* add OPT to the list of supported flash models

* improve test

* properly test on different SDPA backends, on different dtypes & properly handle separately the pad tokens in the test

* remove remaining _flash_attn_2_enabled occurence

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

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

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

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

* Update docs/source/en/perf_infer_gpu_one.md
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* remove use_attn_implementation

* fix docstring & slight bug

* make attn_implementation internal (_attn_implementation)

* typos

* fix tests

* deprecate use_flash_attention_2=True

* fix test

* add back llama that was removed by mistake

* fix tests

* remove _flash_attn_2_enabled occurences bis

* add check & test that passed attn_implementation is valid

* fix falcon torchscript export

* fix device of mask in tests

* add tip about torch.jit.trace and move bt doc below sdpa

* fix parameterized.expand order

* move tests from test_modeling_attn_mask_utils to test_modeling_utils as a relevant test class is already there

* update sdpaattention class with the new cache

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

* Update src/transformers/models/bark/modeling_bark.py

* address review comments

* WIP torch.jit.trace fix. left: test both eager & sdpa

* add test for torch.jit.trace for both eager/sdpa

* fix falcon with torch==2.0 that needs to use sdpa

* fix doc

* hopefully last fix

* fix key_value_length that has no default now in mask converter

* is it flacky?

* fix speculative decoding bug

* tests do pass

* fix following #27907

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

* added flash attention for opt

* added to list

* fix use cache (#3)

* style fix

* fix text

* test fix2

* reverted until 689f599

* torch fx tests are working now!

* small fix

* added TODO docstring

* changes

* comments and .md file modification

---------
Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>

* first batch of structure improvements for model_docs

* second batch of structure improvements for model_docs

* more structure improvements for model_docs

* more structure improvements for model_docs

* structure improvements for cv model_docs

* more structural refactoring

* addressed feedback about image processors

* Rename index.mdx to index.md

* With saved modifs

* Address review comment

* Treat all files

* .mdx -> .md

* Remove special char

* Update utils/tests_fetcher.py
Co-authored-by: Lysandre Debut <lysandre.debut@reseau.eseo.fr>

---------
Co-authored-by: Lysandre Debut <lysandre.debut@reseau.eseo.fr>

small doc nit

* added resources: links to task guides that support these models

* minor polishing

* conflict resolved

* link fix

* Update docs/source/en/model_doc/vision-encoder-decoder.mdx
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

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

* Add resources to OPT

* Add additional resources for OPT

* Remove -{" "} after <PipelineTag pipeline="question-answering" />

* Change bitsnbytes to bitsandbytes

* Revert formatting

* Revert automatic format changes

* Remove - sign after <PipelineTag pipeline="question-answering" />

* Add `OPTForQuestionAnswering`

- added `OPTForQuestionAnswering` class based on `BloomForQuestionAnswering`
- added `OPTForQuestionAnswering` in common tests
- all common tests pass
- make fixup done

* added docstrings for OPTForQuestionAnswering

* Fix docstrings for OPTForQuestionAnswering

* Adding OPTForSeqClassification class

* Fix import issues

* Add documentation for optforseqclassification

* Remove checkout

* fix failing tests

* fix typo

* Fix code formatting

* Incorporating the PR feedbacks

* Incorporate PR Feedbacks

* Fix failing test and add new test for multi label setup

* Fix formatting issue

* Fix failing tests

* Fix formatting issues

* Fix failing tests

* Fix failing tests

* Fix failing tests

* Fix failing tests

* PR feedback

* initial commit

* add init file

* update globakl init

* update index and dummy objects

* style

* update modelling auto

* fix initi typo in src/transformers

* fix typo in modeling tf auto, opt was in wrong mapping name

* fixed a slow test : saved_model

* style

* fix positionnal embedding if no position id is provided

* update tf test

* update test flax requirements

* fixed serialization

* update

* update tf name to allow smooth convertion

* update flax tests

* style

* fix test typo

* fix tf typo test

* add xla for generate support in causal LM

* fixed bug

* cleaned tf tests

* style

* removed from PT for slow tests

* fix typp

* opt test as slow

* trying to fix GPT2 undefined

* correct documentation and add to test doc

* update tf doc

* fix doc

* fake commit

* Apply suggestions from code review
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* update test based on review

* merged main layer for functionning test

* fixup + quality

* Apply suggestions from code review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* update long comment

* make fix copies
Co-authored-by: Arthur <arthur@huggingface.co>
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* First version - OPT model

* Final changes

- putting use cache to False

* few changes

- remove commented block

* few changes

- remove unecessary files

* fix style issues

* few changes

- remove a test file
- added the logits test

* Update src/transformers/models/auto/tokenization_auto.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* add gen tests

* few changes

- rm mask filling example on docstring

* few changes

- remove useless args

* some changes

- more tests should pass now
- needs to clean more
- documentation still needs to be done

* fix code quality

* major changes

- change attention architecture to BART-like
- modify some tests
- style fix

* rm useless classes

- remove opt for:
- QA
- cond generation
- seq classif

* Removed autodoc calls to non-existant classes

TOkenizers are not implemented

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

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

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

* Replaced OPTTokeniser with GPT2 tokenizer

* added GPT2Tokenizer.from_pretrained("patrickvonplaten/opt_gpt2_tokenizer")

* Removed OPTTokenizer

* make style

* Make style replaces

``` ...).unsqueeze(```
by
``` >>>).unsqueeze(```

* make repo consistency

* Removed PretrainedOPTModel

* fix opt.mdx removed other heads

* fix init, removed 3 heads

* removed heads

* finished cleaning head

* removed seauence classif and question answering

* removed unused imports

* removed useless dummy object for QA, SC and CG

* removed tests for removed useless dummy object for QA, SC and CG

* Removed head_mask using encoder layers which don't exist

* fixed test

* fix line

* added OPT to toctree

* Updated model path with pushed weigths

* fix model path

* fixed code quality

* fixed embeddings and generation tests

* update paths

* clean comments

* removed OPTClassificationHead for sentence classification

* renamed hidden layer

* renamed num layers to standard num_hidden_layers

* num_attention_heads fix

* changes for 125m

* add first version for 125m

* add first version - flax

* add new version

* causal LM output

* replace output type with BaseModelOutputWithPastAndCrossAttentions

* revert working config from 150m to 350m

* clean

* removed decoder input ids

* fixed embed dim

* more embed_dim issues

* make style + removed enc_dec test

* update falx model

* removed troublesome copy

* added is_encoder_decoder=False to config

* added set_input emb fuinction to model class

* requires torch on embed test

* use head mask instead of decoder head mask input param solves a test

* 8 test remaining, update

* Updated create_and_check_decoder_model_past_large_inputs

* Make style

* update op tokenizer with condition

* make style

* See if I can push

* some clean up

* remove linear head hack

* save intermediate

* save correct attention

* add copied from from bart

* Update src/transformers/models/opt/modeling_opt.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* fix part of the reviewss
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* same changes in naming / conversion

* correct mask

* more fixes

* delete FlaxOPT and TfOPT

* clean traces of Flax and Tf

* fix mask

* fixed positionnal embedding length when past key value is provoded

* get 125m, 6.7b to work

* Added do_layer_norm

* solved mismatch in load dictionnary

* clean up preapre opt input dict

* fixed past key value as bool

* fix previus

* fixed return dict False tuple issue

* All tests are passing

* Make style

* Ignore OPTDecoder non tested

* make fix-copies

* make repo consistency

* small fix

* removed uselss @torch.no_grad decorator

* make styl;e

* fix previous opt test

* style

* make style

* added opt documentation

* update OPT_PRETRAINED_MODEL_ARCHIVE_LIST

* up

* more fixes

* model & config work

* Update src/transformers/models/opt/modeling_opt.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update src/transformers/models/opt/modeling_opt.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update src/transformers/models/opt/modeling_opt.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* added comment on padding hack (+2)

* cleaup

* review update

* docstring for missing arg

* Update docs/source/en/model_doc/opt.mdx
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update docs/source/en/model_doc/opt.mdx
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update docs/source/en/model_doc/opt.mdx
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update src/transformers/models/opt/__init__.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* update pretrained map

* update path and tests

* make style

* styling

* make consistency

* add gpt2 tok new

* more tok fixes

* Update src/transformers/models/auto/tokenization_auto.py

* Update docs/source/en/model_doc/opt.mdx
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update docs/source/en/model_doc/opt.mdx
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update docs/source/en/model_doc/opt.mdx
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

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

* Update tests/models/opt/test_modeling_opt.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

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

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

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

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

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

* Update based on reviews

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

* make style

* make tokenizer auto tests pass

* apply Lysandre suggestion

* finish tests

* add some good tokenizer tests

* improve docs slighly
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: ArthurZucker <arthur.zucker@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>