GitLab

* 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>

* doc formatting nits

* ignore non-autodocs

* Apply suggestions from code review
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

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

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

* make fixup

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

* adds: extra_repr() to RMSNorm layers in multiple models

* adds: extra_repr for deprecated models as well

* formatting as per style guide

* let's not warn when someone is running a foward without cache + self.training

* more models

* fixup

* 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

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

* tmp commit

* shorter

* nit

* explicit kwargs

* propagate changes

* mass propagation with a few manual touches (let's see how CI behaves)

* fix cacheless case

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

* make fixup

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

* dumb commit

* nit

* update

* something like this

* unpack in modeling utils

* safe import

* oups

* update

* nits

* diff convert gemma

* update

* start propagating

* udpate other modeling code as well

* update for sliding window models

* nits

* more init cleanups

* styling

* fixup

* noice

* pass fixup

* typo typing_extension -> typing_extensions

* torch.nn.functionnal -> torch.nn.functional

* add to import structure

* unpack

* simplify a bit more for this first version

* nut

* update

* update

* nit

* ease the import of `Unpack`

* remove useless `use_sliding_window`

* no qua please

* protect import?

* style

* [run-slow]

* [run slow] llama,gemma,mistral,mixtral

* remove extra kwargs

* fix llama

* address review comments

* apply diff_model_converter to modeling_gemma.py

* remove cache_position 1

* remove cache_position 2

* some cleaning

* refactor gemma2 as well

* apply review comments

* rename file to modeling_flash_attention_utils.py

* siglip refactor

* remove dead code

* is the hub down?

* still down?

* fix siglip

* fix gemma2

* fatal: Could not read from remote repository.

* fix typo in softcap implem

* flacky

* Failed: Timeout >120.0s

---------
Co-authored-by: fxmarty <9808326+fxmarty@users.noreply.github.com>

* fix llama fsdp

* fixup

* adding FSDP tests for CPU offloading

* fixes

* fix tests

* fix tests

* add it for mixtral

* propagate the changes on other models

* Update src/transformers/models/phi/modeling_phi.py

* Delete utils/testing_scripts/fsdp_cpu_offloading.py

Remove script - FSDP + CPU offloading it tested in the test suite

* Delete utils/testing_scripts/dummy_fsdp_config.yml

* Update + add cache_positions docstring

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

* tmp

* update models

* revert utils

* delete

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

* modify warning msg

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

* Update modeling_qwen2.py

* Fix llama

* More fixes

* Fix contrastive_search for new cache structure, and improve performance by removing inneficient torch.stack(torch.split(x, top_k, dim=0))

* Fix _contrastive_search for non-standard cache using ellipsis slicing

* Fix all outputs.logits memory leaks for all decoding strategies!

* Fix small error in _contrastive_search()

* Make all necessary change and revert for the new class

* Apply coding style

* Remove pipes in type hints for compatibility

* correct type hint

* apply style

* Use DynamicCache by default and solve conflicts

* Fix rebase issues

* Add `_supports_dynamic_cache_class` in models for models that support DynamicCache but not other caches to make DynamicCache the default for more models

* Create generation config to return legacy format by default, or to choose not to

* style

* Fix case when use_cache is False

* Remove default DynamicCache in assiste_decoding if assistant_model does not support it + fix _seen_tokens when cropping cache

* Update prepare_inputs_for_generation() for case with empty DynamicCache

* Correct return of args in _assisted_decoding

* Remove EfficientDynamicCache as it is no longer needed

* Correct mistake in generation config

* Move cache logic of assisted decoding to AssistedCandidateGenerator.__init__

* change DynamicCache function names from "split" to "batch_split" for readability + apply coding style

* Remove `_supports_dynamic_cache_class` attribute after rebase

* Correct missing line lost in conflict resolution during rebasing

* Add special case for Jamba

* Fix jamba test

* Coding style

* coding style

* Correct missing import in rebasing

* Simplify _validate_model_kwargs based on removal of _supports_dynamic_cache attribute

* Simplify code paths in _contrastive_search

* coding style

* Update docstrings of cache methods

* Update prepare_inputs_for_generation() -> past_key_values are always Cache objects

* update ruff version

* fix research projects

* Empty

* Fix errors

---------
Co-authored-by: Lysandre <lysandre@huggingface.co>

* first version

* fix sliding window

* fix style

* add sliding window cache

* fix style

* address comments

* fix test

* fix style

* move sliding window check inside cache init

* revert changes on irrelevant files & add comment on SlidingWindowCache

* address comments & fix style

fix style

* update causal mask

* [run-slow] mistral

* [run-slow] mistral

* [run-slow] mistral

* [run-slow] mistral

* [run-slow] mistral

* [run-slow] llama

* [run-slow] mistral

* [run-slow] mistral

* [run-slow] mistral

* revert CI from a10 to t4

* wrap up

* add torch.compile dynamic support

* Add SDPA dynamic shapes compile test & improve SDPA comment

* comment consistency

* Add MistralForTokenClassification

* Add tests and docs

* Add token classification for Mixtral and Qwen2

* Save llma for token classification draft

* Add token classification support for Llama, Gemma, Persimmon, StableLm and StarCoder2

* Formatting

* Add token classification support for Qwen2Moe model

* Add dropout layer to each ForTokenClassification model

* Add copied from in tests

* Update src/transformers/models/llama/modeling_llama.py
Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>

* Propagate suggested changes

* Style

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

* Remove deprecated logic and warnings

* Add back some code that seems to be important...

* Let's just add all he nllb stuff back; removing it is a bit more involved

* Remove kwargs

* Remove more kwargs

* jamba cache

* new flag

* generate exception

* tracing for mistral

* typo

* fix copies

* fix sdpa + sliding window

* give credit
Co-authored-by: ehuaa <ehuamail@163.com>

* remove unnecessary warning

* fix typog

* add test

---------
Co-authored-by: ehuaa <ehuamail@163.com>

* fix mixtral onnx export

* fix qwen model

This commit adds gate jitter to MixtralSparseMoeBlock's input data
before passing it through the MoE layer, if turned on.

* fix stablelm dropout argument type error

* fix docs of _flash_attention_forward

* fix all docs of _flash_attention_forward

* fix docs of _flash_attention_forward in starcoder2

---------
Co-authored-by: oliang <oliang@tencent.com>

Fix mixtral load balancing loss
Co-authored-by: dingkunbo <dingkunbo@baidu.com>

* Set output_router_logits=False in prepare_inputs_for_generation for mixtral

* Add output_router_logits=False to prepare_inputs_for_generation for mixtral

* Fix style

* revert unrelated changes that got in

* style

Co-authored-by: fxmarty <9808326+fxmarty@users.noreply.github.com>
Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

DeepSpeed: hardcode `torch.arange` dtype on `float` usage to avoid incorrect initialization (#28760)

* fix the function load_balancing_loss_func in Mixtral_Moe to include attention_mask

* format code using black and ruff

* skip computing mask if attention_mask=None

* add tests for load balancing loss Mixtral-Moe

* fix assert loss is different in mixtral_test

* fix pad_leng

* use assertNotAlmostEqual and print to debug

* remove print for debug

* minor updates

* reduce rtol and atol

Update warning, a word was missing

* Correct the implementation of auxiliary loss of mixtrtal

* correct the implementation of auxiliary loss of mixtrtal

* Implement a simpler calculation method

---------
Co-authored-by: zhangliangxu3 <zhangliangxu3@jd.com>

* normalize reverse indexing for causal lm sequence classifiers

* normalize reverse indexing for causal lm sequence classifiers

* normalize reverse indexing for causal lm sequence classifiers

* use modulo instead

* unify modulo-based sequence lengths

* some nits

* update test

* add support d\sd[a

* remove some dummy inputs

* all good

* style

* nits

* fixes

* fix more copies

* nits

* styling

* fix

* Update src/transformers/models/mistral/modeling_mistral.py
Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>

* add a slow test just to be sure

* fixup

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