GitLab

* fixed hybrid cache init, added test

* Fix Test Typo

---------
Co-authored-by: Aaron Haag <aaron.haag@siemens.com>

* Initial implementation of OffloadedCache

* enable usage via cache_implementation

* Address feedback, add tests, remove legacy methods.

* Remove flash-attn, discover synchronization bugs, fix bugs

* Prevent usage in CPU only mode

* Add a section about offloaded KV cache to the docs

* Fix typos in docs

* Clarifications and better explanation of streams

* Fix conflicting key in init kwargs in PreTrainedTokenizerBase

* Update code to check for callable key in save_pretrained

* Apply PR suggestions

* Invoke CI

* Updates based on PR suggestion

* Remove size check between attn_weights and kv_seq_len

* add unit tests

* [whisper] compile compatibility with long-form decoding

* clarify comment

* fix after rebase

* finalise

* fix bsz

* fix cache split

* remove contiguous

* style

* finish

* update doc

* prevent cuda graph trace

* draft

* apply changes to all relevant archs

* rerun ci - check_docstrings.py failing?

* fix docstring

* move 2D->4D mask creation to modeling file

* repo consistency

* fix the batch size = 1 case - calling contiguous is not enough

* nit

* style

* propagate to gemma/gemma-2

* prepare inputs for gemma generation

* implement test and tiny fix in gemma2

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

* fix copies

* ci pass

* fix gemma's test_compile_static_cache tests

* flacky

* retrigger ci

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

* Fix FA2 call for Perciever layer

* [run_slow] idefics2

* [run_slow] idefics2

* [run_slow] idefics2

* Fix up

* [run_slow] idefics2

* [run_slow] idefics2

* [run_slow] idefics2

fix 💩

* enable flash-attn & static cache

* this works, not the prev

* fix for sliding window layers

* not needed anymore

* Remove user-defined tokens which can be obtained through merges

* Remove debug line

* formatting

* Refactor spm slow -> fast converter

* revert unnecessary refactor

* set comprehension

* remove test files

* Use `vocab_scores`

* Always replace spiece underline with space in decode

* we no longer need token filtering

* Add save fast load slow unit test

* Remove tokenizers version check

* Remove duplicate code

* Make `<start_of_turn>` and `<end_of_turn>` special tokens

* Bias merge priority with length if score is the same

* Add unit test for merge priority

* CI

* [pipeline] fix padding for 1-d tensors

* add test

* make style

* Update tests/pipelines/test_pipelines_automatic_speech_recognition.py
Co-authored-by: Kamil Akesbi <45195979+kamilakesbi@users.noreply.github.com>

* Update tests/pipelines/test_pipelines_automatic_speech_recognition.py

---------
Co-authored-by: Kamil Akesbi <45195979+kamilakesbi@users.noreply.github.com>

* fix _fix_key in PreTrainedModel

* fix _find_longest_common_sequence

* add test

* remove result.json

* nit

* update test

* mvp

* added test (a few models need fixes)

* fix a few test cases

* test nits

* harder test 😈

* revert changes in stablelm

* test with improved condition

* add todo

* tmp commit

* merged with main

* nits

* add todo

* final corrections

* add docs for generation compilation

* docs nits

* add  tip

* PR suggestions

* add more details to the compilation docs

* fix cache positions

* cache is now init in generate; update docs

* tag test as flaky

* docs

* post rebase make fixup and other nits

* remove unintended changes

* whisper (encoder-decoder) not supported

* move token default updates to ; add tests for token defaults

* push changes

* manual rebase

* chameleon doesn't support this

* fix test_static_cache_mha_mqa_gqa (broken in another PR)

* docs: dynamic is better with end-to-end compilation

* bloom dynamic cache

* bloom follows standard cache format

* no skips for bloom anymore

* use cache position when possible

* clean up

* codestyle

* Update src/transformers/models/bloom/modeling_bloom.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/bloom/modeling_bloom.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/bloom/modeling_bloom.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* pr comments

* isinstance fix

* address comments

* make musicgen test happy

* [run-slow] bloom

---------
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* fix

* fix prev test (half of failures)

* [run-slow] llama, gemma2

* [run-slow] llama, gemma2

[tests] fix `static` cache implementation is not compatible with `attn_implementation==flash_attention_2` (#32039)

* add flash attention check

* fix

* fix

* Refactored to remove un-necessary object base class.

* small fix.

* llava w/o images

* tests

* fix

* move changes to prompt lookup

* add test

* set eos in assistant model

* style

* fix flakiness

* changes for new `main`

* Update tests/generation/test_utils.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update tests/generation/test_utils.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* add comment to explain

---------
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* fix

* [test_all] trigger full CI

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

fix E721 warnings

* [whisper] fix short-form output type

* add test

* make style

* update long-form tests

* fixes

* last fix

* finalise test

Replaced deprecated unittest method with the correct one.

* No more default chat templates

* Add the template to the GPT-SW3 tests since it's not available by default now

* Fix GPT2 test

* Fix Bloom test

* Fix Bloom test

* Remove default templates again

* support gguf fp16

* support gguf bf16 with pytorch

* add gguf f16 test

* remove bf16

* relaxed rope check

* lets also accept rope_type=None, defaulting to the original implementation

* type and rope_type can coexist

Remove conversation pipeline tests

* Updated ruff version and fixed the required code accorindg to the latest version.

* Updated ruff version and fixed the required code accorindg to the latest version.

* Added noqa directive to ignore 1 error shown by ruff

* add DataCollatorBatchFlattening

* Update data_collator.py

* change name

* new FA2 flow if position_ids is provided

* add comments

* minor fix

* minor fix data collator

* add test cases for models

* add test case for data collator

* remove extra code

* formating for ruff check and check_repo.py

* ruff format

ruff format tests src utils

* custom_init_isort.py

Revert "Incorrect Whisper long-form decoding timestamps  (#32003)"

This reverts commit cd48553f.

* renamed phi3 rope_scaling type

* fixed trailing whitespaces

* fixed test

* added warning

* fixed format

---------
Co-authored-by: Merve Noyan <mervenoyan@Merve-MacBook-Pro.local>

* gguf conversion forces add_prefix_space=False for llama3, this is not required and forces from_slow, which fails. changing to None + test

* typo

* clean test

Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Change resize_token_embeddings to make it return same Class that is passed to it

* Add explanatory comment as requested in review

* Add explanatory comments for add resizing function in lxmert

* Add comment for padding_idx and moving _resize_bias in lxmert to LxmertForPreTraining

---------
Co-authored-by: Prashanth Sateesh <prasatee@Prashanths-MBP.attlocal.net>
Co-authored-by: Prashanth Sateesh <prasatee@Prashanths-MacBook-Pro.local>

* Add YaRN and Dynamic-YaRN RoPE Scaling Methods

YaRN (Yet another RoPE extension method) combines the NTK-By-Parts
Interpolation and Attention Scaling methods, improving upon existing
RoPE interpolation methods for longer context window sizes.

Fine-tuned models maintain their original performance across benchmarks
while enabling efficient extrapolation and transfer learning for
quicker convergence, especially in compute-limited environments.

We implement YaRN and Dynamic-YaRN for the following list of models:

 - LLaMA
 - Falcon
 - GPT-NeoX
 - Olmo
 - Persimmon
 - Phi
 - StableLM
 - OpenLLaMA

New unit tests are added to assert YaRN's correct behavior on both
short and long sequence inputs.

For more details, please refer to https://arxiv.org/abs/2309.00071

.
Co-authored-by: Miguel Almeida <miguel.pessanha.almeida@tecnico.ulisboa.pt>

* Refactor YaRN implementation for LLaMA

Iterate on YaRN implementation for LLaMA and remove diff from remaining
models for increased PR modularity.

This commit includes the following changes:
- Merge 'yarn_rope_scaling' and 'rope_scaling' dictionaries
- Remove unnecessary attributes ('extrapolation_factor' and 'finetuned')
  from YaRN classes
- Inherit 'forward' method in YaRN classes from superclass
- Rename 'yarn' method to 'compute_yarn_scaling'
- Extend YaRN tests with further assertions
- Fix style inconsistencies
Co-authored-by: Miguel Monte e Freitas <miguelmontefreitas@tecnico.ulisboa.pt>

* Refactor Tensor Building Logic for YaRN

- Comply with the the tensor building logic introduced in #30743
- Add referencing to the optimized Attention Factor equation
- Remove Dynamic YaRN for a more agile deployment
Co-authored-by: mig-mfreitas <mig-mfreitas@users.noreply.github.com>

* remove unwanted file

---------
Co-authored-by: Miguel Almeida <miguel.pessanha.almeida@tecnico.ulisboa.pt>
Co-authored-by: mig-mfreitas <mig-mfreitas@users.noreply.github.com>
Co-authored-by: Joao Gante <joao@huggingface.co>

* fix mask creation of gpt2 and gpt_neox caused by me

* forgot the reshape of masks when shape > 2

* add tests for gpt neox and gpt2

* nit on a comment

* [whisper integration] use parquet dataset for testing

* propagate to others

* more propagation

* last one