GitLab

Revert "Fix gradient checkpointing + fp16 autocast for most models (#24247)"

This reverts commit 285a4801.

* fix gc bug

* continue PoC on OPT

* fixes

* :exploding_head:

* fix tests

* remove pytest.mark

* fixup

* forward contrib credits from discussions

* forward contrib credits from discussions

* reverting changes on untouched files.

---------
Co-authored-by: zhaoqf123 <zhaoqf123@users.noreply.github.com>
Co-authored-by: 7eu7d7 <7eu7d7@users.noreply.github.com>

use _make_causal_mask in clip models

* Patching clip model to create mask tensor on the device

* Addressing PR's comments

* Addressing PR's comments

* Addressing PR's comments

---------
Co-authored-by: Shanmugam Ramasamy <shanmugamr@shanmugamr-mlt.client.nvidia.com>

* fix

* fix

* make style

* For CLIP

* For OwlViT

* For XCLIP

* For CLIPSeg

* For GroupViT

* fix docstrings

* fix docstrings

* For AltCLIP

* For ChineseCLIP

* For Blip

* For GiT

* make style

* update

* update

* update

* fix

* fix

* fix

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

* Clean all models

* Style

* Last to remove

* address review comments

* Address review comments

* Update docs examples FE -> IP

* Remove _IMAGE_PROCESSOR_FOR_DOC

Use config.num_channels in CLIP-like modeling files
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* [clip] fix error message

* sync

* fix 1

* fix 2

* fix 3

* fix 4
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

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

* allow loading projection in text and vision model

* begin tests

* finish test for CLIPTextModelTest

* style

* add slow tests

* add new classes for projection heads

* remove with_projection

* add in init

* add in doc

* fix tests

* fix some more tests

* fix copies

* fix docs

* remove leftover from fix-copies

* add the head models in IGNORE_NON_AUTO_CONFIGURED

* fix docstr

* fix tests

* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* add docstr for models
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Add first draft

* Update conversion script

* Improve conversion script

* Improve conversion script some more

* Add conditional embeddings

* Add initial decoder

* Fix activation function of decoder

* Make decoder outputs match original implementation

* Make decoder outputs match original implementation

* Add more copied from statements

* Improve model outputs

* Fix auto tokenizer file

* Fix more tests

* Add test

* Improve README and docs, improve conditional embeddings

* Fix more tests

* Remove print statements

* Remove initial embeddings

* Improve conversion script

* Add interpolation of position embeddings

* Finish addition of interpolation of position embeddings

* Add support for refined checkpoint

* Fix refined checkpoint

* Remove unused parameter

* Improve conversion script

* Add support for training

* Fix conversion script

* Add CLIPSegFeatureExtractor

* Fix processor

* Fix CLIPSegProcessor

* Fix conversion script

* Fix most tests

* Fix equivalence test

* Fix README

* Add model to doc tests

* Use better variable name

* Convert other checkpoint as well

* Update config, add link to paper

* Add docs

* Update organization

* Replace base_model_prefix with clip

* Fix base_model_prefix

* Fix checkpoint of config

* Fix config checkpoint

* Remove file

* Use logits for output

* Fix tests
Co-authored-by: Niels Rogge <nielsrogge@Nielss-MacBook-Pro.local>

* create the arange tensor on device for enabling CUDA-Graph at higher-performace for SD

* sync
Co-authored-by: Stas Bekman <stas@stason.org>

* add first generation tutorial

* uP

* [Clip] Add text model to device map

* onnx config for clip

* default opset as 14

* changes from the original repo

* input values order fix

* outputs fix

* remove unused import

* ran make fix-copies

* black format

* review comments: forward ref, import fix, model change revert, .to cleanup

* make style

* formatting fixes

* revert groupvit

* comment for cast to int32

* comment fix

* make .T as .t() for onnx conversion

* ran make fix-copies

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

* fix copies

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

* Use torch.finfo(self.dtype).min

* for GPTNeoX

* for Albert

* For Splinter

* Update src/transformers/models/data2vec/modeling_data2vec_audio.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* fix -inf used in Bart-like models

* Fix a few remaining -inf

* more fix

* clean up

* For CLIP

* For FSMT

* clean up

* fix test

* Add dtype argument and use it for LayoutLMv3

* update FlaxLongT5Attention
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Support for Bart and LayoutLM, and partial support for XLNet

* Support for mbart

* A lot of new models supported

* Support for other models

* LayoutLM fix

* Use strings instead of classes

* Black preview

* Fixup too!

* Fix check copies

* Use the same version as the CI

* Bump black

* Updated assertions to exceptions

* updated assertions to exceptions

* bug fixes

* fix-copies

* Update modeling_ctrl.py

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

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

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

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

* Update modeling_led.py

* Update modeling_led.py

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

* Use CLIP model's config for some fields (if specified) instead of those of vision & text components.
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* Big file_utils cleanup

* This one still needs to be treated separately

* Split file_utils in several submodules

* Fixes

* Add back more objects

* More fixes

* Who exactly decided to import that from there?

* Second suggestion to code with code review

* Revert wront move

* Fix imports

* Adapt all imports

* Adapt all imports everywhere

* Revert this import, will fix in a separate commit

* clip typhinting #16059

* removed optional type annotations for dataclass in CLIPOutput

* type annotation fixes per Rocket - Clip Torch

* Upgrade black to version ~=22.0

* Check copies

* Fix code

* Fix bad examples

* Add black formatting to style_doc

* Use first nonempty line

* Put it at the right place

* Don't add spaces to empty lines

* Better templates

* Deal with triple quotes in docstrings

* Result of style_doc

* Enable mdx treatment and fix code examples in MDXs

* Result of doc styler on doc source files

* Last fixes

* Break copy from

* New doc styler

* Fix issue with args at the start

* Code sample fixes

* Style code examples in MDX

* Fix more patterns

* Typo

* Typo

* More patterns

* Do without black for now

* Get more info in error

* Docstring style

* Re-enable check

* Quality

* Fix add_end_docstring decorator

* Fix docstring

* Convert docstrings of all configurations and tokenizers

* Processors and fixes

* Last modeling files and fixes to models

* Pipeline modules

* Utils files

* Data submodule

* All the other files

* Style

* Missing examples

* Style again

* Fix copies

* Say bye bye to rst docstrings forever

* Convert file_utils docstrings to Markdown

* Test on BERT

* Return block indent

* Temporarily disable doc styler

* Remove from quality checks as well

* Remove doc styler mess

* Remove check from circleCI

* Fix typo

* Convert file_utils docstrings to Markdown

* Test on BERT

* Return block indent

* Temporarily disable doc styler

* Remove from quality checks as well

* Remove doc styler mess

* Remove check from circleCI

* Fix typo

* Let's go on all other model files

* Add templates too

* Styling and quality

* Add a main_input_name attribute to all models

* Fix tests

* Wtf Vs Code?

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

* Style

* Fix copies
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

The model outputs a pooler_output whereas the doctype examples were using a pooled_output.

* Add a post init method to all models

* Fix tests

* Fix last tests

* Fix templates

* Add comment

* Forgot to save

* Remove n_ctx from configs

* Fix GPTJ and OpenAIGPT, both are acceptable breaking changes as there are no configs such that it breaks

* Remove unecessary n_positions from TFOpenAIGPT

* Make gradient_checkpointing a training argument

* Update src/transformers/modeling_utils.py
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>

* Update src/transformers/configuration_utils.py
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>

* Fix tests

* Style

* document Gradient Checkpointing as a performance feature

* Small rename

* PoC for not using the config

* Adapt BC to new PoC

* Forgot to save

* Rollout changes to all other models

* Fix typo
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>
Co-authored-by: Stas Bekman <stas@stason.org>

* fix logit_scale init

* add logit_scale_init_value as config param

* Update clip loss calculation

Hello, I'm the author of the blog you took the snippet from. I think this way of calculating is possibly slightly more accurate for calculation.

* Apply suggestions from code review
Co-authored-by: Suraj Patil <surajp815@gmail.com>