GitLab

* Fix add-new-model-like command when old checkpoint can't be recovered

* Style

* Fix model templates

* Update paths

* Per-folder tests reorganization

* Review comments
Co-authored-by: sgugger <sylvain.gugger@gmail.com>
Co-authored-by: Stas Bekman <stas@stason.org>

* Tests fetcher

* Review comments
Co-authored-by: sgugger <sylvain.gugger@gmail.com>
Review comments

* Per-folder tests reorganization
Co-authored-by: sgugger <sylvain.gugger@gmail.com>
Co-authored-by: Stas Bekman <stas@stason.org>

* clean commit of changes to NLP tasks

* 🖍 apply feedback

* 📝

 move tf data collator in multiple choice
Co-authored-by: Steven <stevhliu@gmail.com>

Co-authored-by: Pierric Cistac <Pierrci@users.noreply.github.com>

In the code, `do_normalize` defaults to True

* custom_models: tiny doc addition

* mention security feature earlier in the section
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* Enabling Beit SegFormer to `image-segmentation`.

* Fixing the score.

* Fix import ?

* Missing in type hint.

* Multiple test fixes:

- Add `raw_image` support. It should be the default IMHO since in Python
  world it doesn't make any sense to base64 encode the image (Sorry
  @mishig, didn't catch that in my review). I really think we should
  consider breaking BC here.
- Add support for Segformer tiny test (needed
  `SegformerModelTester.get_config` to enable TinyConfig
  @NielsRogge)
- Add the check that `batch_size` works correctly on that pipeline.
  Uncovered that it doesn't for Detr, which IMO is OK since images
  after `feature_extractor` don't have the same size. Comment should
  explain.

* Type hint as a string.

* Make fixup + update black.

* torch+vision protections.

* Don't use torchvision, use F.interpolate instead (no new dep).

* Last fixes for Segformer.

* Update test to reflect new image (which was broken)

* Update tests.

* Major BC modification:

- Removed the string compressed PNG string, that's a job for users
`transformers` stays in python land.
- Removed the `score` for semantic segmentation. It has hardly a meaning
  on its own in this context.
- Don't include the grayscale with logits for now (which could enable
  users to get a sense of confidence). Might be done later.
- Don't include the surface of the mask (could be used for sorting by
  users, to filter out small masks). It's already calculable, and
  it's easier to add later, than to add now and break later if we need.

* `make fixup`.

* Small changes.

* Rebase + doc fixup.

* [ViLT] Fix checkpoint url in config

* Apply suggestions from code review
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

(`hf-internal-testing/tiny-clip` for instance)

* [Proposal] Adding ZeroShotImageClassificationPipeline

- Based on CLIP

* WIP, Resurection in progress.

* Resurrection... achieved.

* Reword handling different `padding_value` for `feature_extractor` and
`tokenizer`.

* Thanks doc-builder !

* Adding docs + global namespace `ZeroShotImageClassificationPipeline`.

* Fixing templates.

* Make the test pass and be robust to floating error.

* Adressing suraj's comments on docs mostly.

* Tf support start.

* TF support.

* Update src/transformers/pipelines/zero_shot_image_classification.py
Co-authored-by: Suraj Patil <surajp815@gmail.com>
Co-authored-by: Suraj Patil <surajp815@gmail.com>

* Fix `HfArgumentParser` when passing a generator

* Add missing import

* Always convert `dataclass_types` into a list

cna -> can

Co-authored-by: Boumadane Abdelmoumene <moumene.boumadane@gmail.com>

* [Wav2Vec2 Time Stamps]

* Add first version

* add word time stamps

* Fix

* save intermediate space

* improve

* [Finish CTC Tokenizer]

* remove @

* remove @

* push

* continue with phonemes

* up

* finish PR

* up

* add example

* rename

* finish

* Apply suggestions from code review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* correct split

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

* Add GeLU10 (clipped version of GeLU) to transformers to improve quantization performances.

* Add unittests.

* Import tensorflow after `is_tf_available` check.

* Fix tensorflow wrong function `tf.tensor` to `tf.constant`

* style.

* use `tf.math.max`

* Fix tf tests.

* style.

* style style style style style style

* style style style style style style

* Address @sgugger comments.

* Fix wrong operator for raising ValueError for ClippedGELUActivation.

* TF train_step docstring

* Working example with to_tf_dataset

* updated text_classification

* more comments

* Add layer_idx to CrossAttention

* Add layer_idx to crossattention of ImageGPT model

* begin script

* update script

* fix features and data args

* main

* add requirements

* add column name args

* fix captions

* don't jit transforms

* fix caption

* fix labels, handle attention mask

* convert pixel values to numpy

* labels => input_ids

* transform images on the fly

* use AutoModel class, create the hybird model outside of the script

* fix version message

* add readme

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

* adderss review comments

* add more comments

* allow freezing vision and text models
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

Remove input and target reset after preprocessing

* Add missing PLBart entry in index

* Fix README

* Fix README

* Fix style

* Change to master model doc

* fix bug in PT speech-encoder-decoder

* add pt test for `inputs is not None`

* fix test

* new pt test

* Update tests/test_modeling_speech_encoder_decoder.py

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

`HfDeepSpeedConfig` accepts a dictionary or path to `.json` file containing DS configurations, not `TrainingArguments`.