GitLab

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

* First draft

* More improvements

* Improve model, add custom CUDA code

* Import torch before

* Add script that imports custom layer

* Add everything in new ops directory

* Import custom layer in modeling file

* Fix ARCHIVE_MAP typo

* Creating the custom kernel on the fly.

* Import custom layer in modeling file

* More improvements

* Fix CUDA loading

* More improvements

* Improve conversion script

* Improve conversion script

* Make it work until encoder_outputs

* Make forward pass work

* More improvements

* Make logits match original implementation

* Make implementation also support single_scale model

* Add support for single_scale and dilation checkpoint

* Add support for with_box_refine model

* Support also two stage model

* Improve tests

* Fix more tests

* Make more tests pass

* Upload all models to the hub

* Clean up some code

* Improve decoder outputs

* Rename intermediate hidden states and reference points

* Improve model outputs

* Move tests to dedicated folder

* Improve model outputs

* Fix retain_grad test

* Improve docs

* Clean up and make test_initialization pass

* Improve variable names

* Add copied from statements

* Improve docs

* Fix style

* Improve docs

* Improve docs, move tests to model folder

* Fix rebase

* Remove DetrForSegmentation from auto mapping

* Apply suggestions from code review

* Improve variable names and docstrings

* Apply some more suggestions from code review

* Apply suggestion from code review

* better docs and variables names

* hint to num_queries and two_stage confusion

* remove asserts and code refactor

* add exception if two_stage is True and with_box_refine is False

* use f-strings

* Improve docs and variable names

* Fix code quality

* Fix rebase

* Add require_torch_gpu decorator

* Add pip install ninja to CI jobs

* Apply suggestion of @sgugger

* Remove DeformableDetrForObjectDetection from auto mapping

* Remove DeformableDetrModel from auto mapping

* Add model to toctree

* Add model back to mappings, skip model in pipeline tests

* Apply @sgugger's suggestion

* Fix imports in the init

* Fix copies

* Add CPU implementation

* Comment out GPU function

* Undo previous change

* Apply more suggestions

* Remove require_torch_gpu annotator

* Fix quality

* Add logger.info

* Fix logger

* Fix variable names

* Fix initializaztion

* Add missing initialization

* Update checkpoint name

* Add model to doc tests

* Add CPU/GPU equivalence test

* Add Deformable DETR to pipeline tests

* Skip model for object detection pipeline
Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com>
Co-authored-by: Nouamane Tazi <nouamane98@gmail.com>
Co-authored-by: Sylvain Gugger <Sylvain.gugger@gmail.com>

* [WIP] Skeleton of VisualQuestionAnweringPipeline extended to support LayoutLM-like models

* Fixup

* Use the full encoding

* Basic refactoring to DocumentQuestionAnsweringPipeline

* Cleanup

* Improve args, docs, and implement preprocessing

* Integrate OCR

* Refactor question_answering pipeline

* Use refactored QA code in the document qa pipeline

* Fix tests

* Some small cleanups

* Use a string type annotation for Image.Image

* Update encoding with image features

* Wire through the basic docs

* Handle invalid response

* Handle empty word_boxes properly

* Docstring fix

* Integrate Donut model

* Fixup

* Incorporate comments

* Address comments

* Initial incorporation of tests

* Address Comments

* Change assert to ValueError

* Comments

* Wrap `score` in float to make it JSON serializable

* Incorporate AutoModeLForDocumentQuestionAnswering changes

* Fixup

* Rename postprocess function

* Fix auto import

* Applying comments

* Improve docs

* Remove extra assets and add copyright

* Address comments
Co-authored-by: Ankur Goyal <ankur@impira.com>

* Further reduce the number of alls to head for cached models/tokenizers/pipelines

* Fix tests

* Address review comments

Co-authored-by: Olivier Dehaene <olivier@huggingface.co>

* Add Image2TextGenerationPipeline to supported pipelines

* Add Flax and Tensorflow support

* Add Flax and Tensorflow small tests

* Add default model for Tensorflow

* Add docstring

* Fix doc style

* Add tiny models for pytorch and flax

* Remove flax from pipeline.
Fix tests

* Use ydshieh/vit-gpt2-coco-en as a default for both PyTorch and Tensorflow

* Fix Tensorflow support
Co-authored-by: Olivier Dehaene <olivier@huggingface.co>

* Use commit hash to look in cache instead of calling head

* Add tests

* Add attr for local configs too

* Stupid typos

* Fix tests

* Update src/transformers/utils/hub.py
Co-authored-by: Julien Chaumond <julien@huggingface.co>

* Address Julien's comments
Co-authored-by: Julien Chaumond <julien@huggingface.co>

* Adding a new `align_to_words` param to qa pipeline.

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

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

* [DX fix] Fixing QA pipeline streaming a dataset.

QuestionAnsweringArgumentHandler would iterate over the whole dataset
effectively killing all properties of the pipeline.
This restores nice properties when using `Dataset` or `Generator` since
those are meant to be consumed lazily.

* Handling TF better.

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

* Fix pipeline tests

* Make sure all pipelines tests run with init changes

* Adding a better error message when the model is improperly configured

within transformers.

* Update src/transformers/pipelines/__init__.py

* Black version.

* Overriding task aliases so that tokenizer+feature_extractor

values are correct.

* Fixing task aliases by overriding their names early

* X.

* Fixing feature-extraction.

* black again.

* Normalizing `translation` too.

* Fixing last few corner cases.

translation need to use its non normalized name (translation_XX_to_YY,
so that the task_specific_params are correctly overloaded).
This can be removed and cleaned up in a later PR.

`speech-encode-decoder` actually REQUIRES to pass a `tokenizer` manually
so the error needs to be discarded when the `tokenizer` is already
there.

* doc-builder fix.

* Fixing the real issue.

* Removing dead code.

* Do not import the actual config classes.

* Update pipeline word heuristic to work with whitespace in token offsets

This change checks for whitespace in the input string at either the
character preceding the token or in the first character of the token.
This works with tokenizers that return offsets excluding whitespace
between words or with offsets including whitespace.

fixes #18111

starting

* Use smaller model, ensure expected tokenization

* Re-run CI (please squash)

* Initial work

* More work

* Add tests for custom pipelines on the Hub

* Protect import

* Make the test work for TF as well

* Last PyTorch specific bit

* Add documentation

* Style

* Title in toc

* Bad names!

* Update docs/source/en/add_new_pipeline.mdx
Co-authored-by: Lysandre Debut <lysandre.debut@reseau.eseo.fr>

* Auto stash before merge of "custom_pipeline" and "origin/custom_pipeline"

* Address review comments

* Address more review comments

* Update src/transformers/pipelines/__init__.py
Co-authored-by: Lysandre Debut <lysandre.debut@reseau.eseo.fr>
Co-authored-by: Lysandre Debut <lysandre.debut@reseau.eseo.fr>

* Adding support for `device_map` directly in `pipeline(..)` function.

* Updating the docstring.

* Adding a better docstring

* Put back type hints.

* Blacked. (`make fixup` didn't work ??!!)

* Fix some typos.
Signed-off-by: Yulv-git <yulvchi@qq.com>

* Fix typo.
Signed-off-by: Yulv-git <yulvchi@qq.com>

* make fixup.

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

* feat: add pipeline registry abstraction

- added `PipelineRegistry` abstraction
- updates `add_new_pipeline.mdx` (english docs) to reflect the api addition
- migrate `check_task` and `get_supported_tasks` from
  transformers/pipelines/__init__.py to
  transformers/pipelines/base.py#PipelineRegistry.{check_task,get_supported_tasks}
Signed-off-by: Aaron Pham <29749331+aarnphm@users.noreply.github.com>

* fix: update with upstream/main

chore: Apply suggestions from sgugger's code review
Signed-off-by: Aaron Pham <29749331+aarnphm@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* chore: PR updates

- revert src/transformers/dependency_versions_table.py from upstream/main
- updates pipeline registry to use global variables
Signed-off-by: Aaron Pham <29749331+aarnphm@users.noreply.github.com>

* tests: add tests for pipeline registry
Signed-off-by: Aaron Pham <29749331+aarnphm@users.noreply.github.com>

* tests: add test for output warning.
Signed-off-by: Aaron Pham <29749331+aarnphm@users.noreply.github.com>

* chore: fmt and cleanup unused imports
Signed-off-by: Aaron Pham <29749331+aarnphm@users.noreply.github.com>

* fix: change imports to top of the file and address comments
Signed-off-by: Aaron Pham <29749331+aarnphm@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* trigger test failure

* upload revision poc

* Update src/transformers/pipelines/base.py
Co-authored-by: Julien Chaumond <julien@huggingface.co>

* up

* add test

* correct some stuff

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

* correct require flag
Co-authored-by: Julien Chaumond <julien@huggingface.co>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Revert "Skip failing test until they are fixed."

This reverts commit 8f400775.

* Use `tiny-detr` checkpts from `hf-internal-testing`

Fixing a regression with `return_all_scores` introduced in #17606

- The legacy test actually tested `return_all_scores=False` (the actual
  default) instead of `return_all_scores=True` (the actual weird case).

This commit adds the correct legacy test and fixes it.

Tmp legacy tests.

Actually fix the regression (also contains lists)

Less diffed code.

* Initial commit

* Make some fixes

* Make PT model full forward pass

* Drop TF & Flax implementation, fix copies etc

* Add Flax model and update some corresponding stuff

* Drop some TF things

* Update config and flax local attn

* Add encoder_attention_type to config

* .

* Update docs

* Do some cleansing

* Fix some issues -> make style; add some docs

* Fix position_bias + mask addition + Update tests

* Fix repo consistency

* Fix model consistency by removing flax operation over attn_mask

* [WIP] Add PT TGlobal LongT5

* .

* [WIP] Add flax tglobal model

* [WIP] Update flax model to use the right attention type in the encoder

* Fix flax tglobal model forward pass

* Make the use of global_relative_attention_bias

* Add test suites for TGlobal model

* Fix minor bugs, clean code

* Fix pt-flax equivalence though not convinced with correctness

* Fix LocalAttn implementation to match the original impl. + update READMEs

* Few updates

* Update: [Flax] improve large model init and loading #16148

* Add ckpt conversion script accoring to #16853 + handle torch device placement

* Minor updates to conversion script.

* Typo: AutoModelForSeq2SeqLM -> FlaxAutoModelForSeq2SeqLM

* gpu support + dtype fix

* Apply some suggestions from code review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* * Remove (de)parallelize stuff
* Edit shape comments
* Update README.md
* make fix-copies

* Remove caching logic for local & tglobal attention

* Apply another batch of suggestions from code review

* Add missing checkpoints
* Format converting scripts
* Drop (de)parallelize links from longT5 mdx

* Fix converting script + revert config file change

* Revert "Remove caching logic for local & tglobal attention"

This reverts commit 2a619828f6ddc3e65bd9bb1725a12b77fa883a46.

* Stash caching logic in Flax model

* Make side relative bias used always

* Drop caching logic in PT model

* Return side bias as it was

* Drop all remaining model parallel logic

* Remove clamp statements

* Move test files to the proper place

* Update docs with new version of hf-doc-builder

* Fix test imports

* Make some minor improvements

* Add missing checkpoints to docs
* Make TGlobal model compatible with torch.onnx.export
* Replace some np.ndarray with jnp.ndarray

* Fix TGlobal for ONNX conversion + update docs

* fix _make_global_fixed_block_ids and masked neg  value

* update flax model

* style and quality

* fix imports

* remove load_tf_weights_in_longt5 from init and fix copies

* add slow test for TGlobal model

* typo fix

* Drop obsolete is_parallelizable and one warning

* Update __init__ files to fix repo-consistency

* fix pipeline test

* Fix some device placements

* [wip]: Update tests -- need to generate summaries to update expected_summary

* Fix quality

* Update LongT5 model card

* Update (slow) summarization tests

* make style

* rename checkpoitns

* finish

* fix flax tests
Co-authored-by: phungvanduy <pvduy23@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: patil-suraj <surajp815@gmail.com>

* wip

* rebase

* all tests pass

* rebase

* ready for PR

* address comments

* fix styles

* add require_torch to pipeline test

* remove remote image to improve CI consistency

* address comments; fix tf/flax tests

* address comments; fix tf/flax tests

* fix tests; add alias

* repo consistency tests

* Update src/transformers/pipelines/visual_question_answering.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* address comments

* Update src/transformers/pipelines/visual_question_answering.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* merge

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

* merge
Co-authored-by: Sijun He <sijunhe@Sijuns-MacBook-Pro.local>
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

When we're preparing the tensors for CPU for postprocessing, we need
to upgrade the `float16` to `float32` since CPUs don't have instructions
for `[b]float16`.

* Adding `top_k` and `sort` arguments to `text-classification` pipeline.

- Deprecate `return_all_scores` as `top_k` is more uniform with other
  pipelines, and a superset of what `return_all_scores` can do.
  BC is maintained though.
  `return_all_scores=True` -> `top_k=None`
  `return_all_scores=False` -> `top_k=1`

- Using `top_k` will imply sorting the results, but using no argument
  will keep the results unsorted for backward compatibility.

* Remove `sort`.

* Fixing the test.

* Remove bad doc.

* [BC] Fixing usage of text pairs

The BC is actually preventing users from misusing the pipeline since
users could have been willing to send text pairs and the pipeline would
instead understand the thing as a batch returning bogus results.

The correct usage of text pairs is preserved in this PR even when that
makes the code clunky.

Adds support for {"text":..,, "text_pair": ...} inputs for both dataset
iteration and more explicit usage to pairs.

* Updating the doc.

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

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

* Update tests/pipelines/test_pipelines_text_classification.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* quality.
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Accepting real pytorch device as arguments.

* is_torch_available.

* Black preview

* Fixup too!

* Fix check copies

* Use the same version as the CI

* Bump black

* LogSumExp trick `question_answering` pipeline.

* Adding a failing test.

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

* Temporary commit witht the long QA fix.

* Adding slow tests covering this fix.

* Removing fast test as it doesn't fail anyway.

* Fixing return type tensor with `num_return_sequences>1`.

* Nit.

* Enabling `Tapex` in table question answering pipeline.

* Questions are independant for Tapex, making the test respect that.

* Missing extra space.

* Change the chunk_iter function to handle

the subtle cases where the last chunk gets ignored since all the
data is in the `left_strided` data.

We need to remove the right striding on the previous item.

* Remove commented line.

* Attention mask is important in the case of batching...

* Improve the fix.

* Making the sentence different enough that they exhibit different
predictions.

* Adding ForInstanceSegmentation to pipelines.

* Last fix `category_id` renamed to `label_id`.

* Can't be none no more.

* No `is_thing_map` anymore.