GitLab

* See if we can get tests to pass with the fixed weights

* See if we can get tests to pass with the fixed weights

* Replace the revisions now that we don't need them anymore

fix copies

* Add datasets.Dataset to Trainer's train_dataset and eval_dataset type hints

* Add is_datasets_available check for importing datasets under TYPE_CHECKING guard

https://github.com/huggingface/transformers/pull/30077/files#r1555939352

* fix sequence length errors

* fix label column name error for vit

* fix the lm_head embedding!=linear layer mismatches for Seq2Seq models

* init: add StableLm 2 support

* add integration test for parallel residual and qk layernorm

* update(modeling): match qk norm naming for consistency with phi/persimmon

* fix(tests): run fwd/bwd on random init test model to jitter norm weights off identity

* `use_parallel_residual`: add copy pointer to `GPTNeoXLayer.forward`

* refactor: rename head states var in `StableLmLayerNormPerHead`

* tests: update test model and add generate check

* adding env variable for mps and is_torch_mps_available for Pipeline

* fix linting errors

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

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

fix typo at ImportError

* remove controlflows

* style

* rename patch_ to padded_ following review comment

* style

* Update trainer.py

* fix copies

* fix falcon without attention_mask & alibi

* add test

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

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

fix learning rate display issue in galore optimizer

* pass token to trainer.push_to_hub

* fmt

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

* pass token to create_repo, update_folder

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

* ImportError: Trainer with PyTorch requires accelerate>=0.20.1 Fix

Adding the evaluate and accelerate installs at the beginning of the cell to fix the issue

* ImportError Fix: Trainer with PyTorch requires accelerate>=0.20.1

* Import Error Fix

* Update installation.md

* Update quicktour.md

* rollback other lang changes

* Update _config.py

* updates for other languages

* fixing error

* Tutorial Update

* Update tokenization_utils_base.py

* Just use an optimizer string to pass the doctest?

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

* Patch fix - don't use safetensors for TF models

* Skip test for TF for now

* Update for another test

* add bnb flag

* move maker

* add accelerator maker

updated examples/pytorch/language-modeling scripts and requirements.txt to require datasets>=2.14.0 (#30120)

updated requirements.txt and require_version() calls in examples/pytorch/language-modeling to require datasets>=2.14.0

* Make MLFlow version detection more robust and handles mlflow-skinny

* Make function name more clear and refactor the logic

* Further refactor

* add _torch_extract_fbank_features_batch function in feature_extractor_whisper

* reformat feature_extraction_whisper.py file

* handle batching in single function

* add gpu test & doc

* add batch test & device in each __call__

* add device arg in doc string

---------
Co-authored-by: vaibhav.aggarwal <vaibhav.aggarwal@sprinklr.com>

* clean-up whisper kwargs

* failing test

* fix

* fix

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

* Add docstrings and types for MambaCache

* Update src/transformers/models/mamba/modeling_mamba.py

* Update src/transformers/models/mamba/modeling_mamba.py

* Update src/transformers/models/mamba/modeling_mamba.py

* make fixup

* import copy in generation_whisper

* ruff

* Revert "make fixup"

This reverts commit c4fedd6f60e3b0f11974a11433bc130478829a5c.

* separate jobs

* separate jobs

* use channel name directly instead of ID

* use channel name directly instead of ID

* use channel name directly instead of ID

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

* [WIP] fix fx

* [WIP] fix fx

* [WIP] fix fx

* [WIP] fix fx

* [WIP] fix fx

* Apply changes to other models

* fix

* fix

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

Update quantizer_bnb_4bit.py: In the ValueError string there should be "....you need to set `llm_int8_enable_fp32_cpu_offload=True`...." instead of "`load_in_8bit_fp32_cpu_offload=True`". (#30013)

* Update quantizer_bnb_4bit.py

There is an mistake in ValueError on line 86 of quantizer_bnb_4bit.py. In the error string there should be "....you need to set `llm_int8_enable_fp32_cpu_offload=True`...." instead of "load_in_8bit_fp32_cpu_offload=True". I think you updated the BitsAndBytesConfig() arguments, but forgot to change the ValueError in quantizer_bnb_4bit.py.

* Update quantizer_bnb_4bit.py

Changed ValueError string "...you need to set load_in_8bit_fp32_cpu_offload=True..." to "....you need to set llm_int8_enable_fp32_cpu_offload=True...."

fix bnb test

* Add image processor to trainer

* Replace tokenizer=image_processor everywhere

* fix mixtral onnx export

* fix qwen model

* if output is tuple like facebook/hf-seamless-m4t-medium, waveform is the first element
Signed-off-by: Wang, Yi <yi.a.wang@intel.com>

* add test and fix batch issue
Signed-off-by: Wang, Yi <yi.a.wang@intel.com>

* add dict output support for seamless_m4t
Signed-off-by: Wang, Yi <yi.a.wang@intel.com>

---------
Signed-off-by: Wang, Yi <yi.a.wang@intel.com>

skip test_encode_decode_fast_slow_all_tokens for now
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

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

* changes

* addressing comments

* smol fix

* Defaulted IdeficsProcessor padding to 'longest', removed manual padding

* make fixup

* Defaulted processor call to padding=False

* Add padding to processor call in IdeficsModelIntegrationTest as well

* Defaulted IdeficsProcessor padding to 'longest', removed manual padding

* make fixup

* Defaulted processor call to padding=False

* Add padding to processor call in IdeficsModelIntegrationTest as well

* redefaulted padding=longest again

* fixup/doc

* implement convert_mamba_ssm_checkpoint_to_pytorch

* Add test test_model_from_mamba_ssm_conversion

* moved convert_ssm_config_to_hf_config to inside mamba_ssm_available check

* fix skipif clause

* moved skips to inside test since skipif decorator isn't working for some reason

* Added validation

* removed test

* fixup

* only compare logits

* remove weight rename

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

* nits

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

feat: enable mult-idevice for efficientnet

* Docstring to note about zero init

* Check for accelerate

* Change conditional return

* Tweak

* Add new accelerate-specific zero3 check

* Fix import

* Revert to RTFM

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

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

* fix

* sort imports