* Make ZeroShotImageClassificationPipeline faster
The pipeline makes separate calls to model for each candidate label.
This commit combines all labels into one call.
Original code takes more that 60 seconds to process one image and 1000
candidate labels. Updated code takes less than 2 seconds.
* implement batching
* code formatting
* Creating an even faster zero-shot-image-classifiction.
Unfortunately super tailored towards CLIP.
Co-Authored-By: Yessen Kanapin <yessen@deepinfra.com>
* Quality.
* Cleanup.
* Order different on the CI it seems.
* Cleanup.
* Quality.
---------
Co-authored-by: Yessen Kanapin <yessen@deepinfra.com>
* [WIP] whisper refacto to support language output.
* Handling merges.
* A bit more cleanup and comments.
* Many improvements.
Lots of details everywhere.
* Cleanup old code and tests.
* Handle lone timestamp tokens (just recover when something bad happens).
* Adding return_language example.
* No ffmpeg.
* Hmm.
* Some corrections.
* Both fast and slow.
* New black.
* Update src/transformers/models/whisper/tokenization_whisper.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/whisper/tokenization_whisper.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Remove print.
* Undoing tests modifications.
* Smaller test modifications.
* Rename.
* Remove maxDiff.
---------
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Make schedulers picklable by making lr_lambda fns global
* add unused _get_constant_schedule_lr_lambda arg
* remove unneeded _get_constant_schedule_lr_lamda
* add test
* make style
* rebase, remove torch dep, put lambda back
* repo-consistency and style
* Mark pipeline tests to skip them easily
* Mark the mixin as pipeline test
* Update src/transformers/testing_utils.py
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>
---------
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>
* Use PyAV instead of Decord
* Get frame indices
* Fix number of frames
* Update src/transformers/models/videomae/image_processing_videomae.py
* Fix up
* Fix copies
* Update timesformer doctests
* Update docstrings
Adds the ALIGN model to transformers. ALIGN is introduced in "Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision" by Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, Tom Duerig.
* Change the .view call to .reshape
* Change the .view call to .reshape to all the copies from bart attention
* Fix copies and style
* Fix copies and style
* Fix copies and style
* Fix copies and style
* Fix copies and style
* Revert unneccessary changes
* Revert unneccessary changes
* Revert unneccessary changes
* Revert unneccessary changes
* trying to figure out whether model is NLP
* drop my changes and apply easier fix
* trying to handle all int input types
* fix logic
---------
Co-authored-by: Stas Bekman <stas@stason.org>
* rounding_mode = "floor" instead of // to prevent behavioral change
* add other TODO
* use `torch_int_div` from pytrch_utils
* same for tests
* fix copies
* style
* use relative imports when needed
* Co-authored-by: sgugger <sylvain.gugger@gmail.com>
* First commit for the improved PT-TF weight loading
* Remove workarounds from TFEncoderDecoder tests
* Allow a custom weight renaming function in from_pretrained and use that to clean up EncoderDecoder
* make fixup
* First attempt at visionencoderdecoder
* Disable tensorfloat32 in tests to get consistent outputs
* Quick fix to tf_vision_encoder_decoder tests
* make fixup
* Update Blenderbot tests
* Remove unused arg in modeling_tf_opt
* load_tf_sharded_weights had strict=True! This meant transfer learning was impossible, so I'm setting it to False.
* Support prefixes when loading sharded TF checkpoints
* make fixup
* Add test to load sharded models with a weight prefix
* Fix sharded weight loading test
* Add a test for transfer from a sharded checkpoint
* make fixup
* Add test to check that crossloading from PT with a prefix works
* Refactor from_pretrained in the encoderdecoder classes
* Refactor from_pretrained in the encoderdecoder classes
* missmatched -> mismatched
* Explicitly check for None
* No comments showing my very impressive and attractive knowledge of Py3.9+
* Disable TF32 across all TF tests
* Add loss for BridgeTowerForMaskedLM and BridgeTowerForImageAndTextRetrieval
* minor fix return_dict
* implement test for loss computation
---------
Co-authored-by: Tiep Le <97980157+tileintel@users.noreply.github.com>
Co-authored-by: Tiep Le <tiep.le@intel.com>
* zero shot object detection part 1
* added batch prediction section
* added image guided object detection section
* make style
* added the task guide to the TOC
* minor polishing
* Apply suggestions from code review
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>
Co-authored-by: Alara Dirik <8944735+alaradirik@users.noreply.github.com>
* added embedded owlvit demo
* Apply suggestions from code review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* minor fix
* make style
---------
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>
Co-authored-by: Alara Dirik <8944735+alaradirik@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* If applied, this commit fixes generate bug in gptj
* Remove extra same code block
* formatting and test fix
* Conflict fix and declaration error fix
---------
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Fix the issue of blip model returning loss even when the label is not provoided
* Fix ruff failure
* Incorporate PR feedbacks
* Incorporate PR feedbacks
* Incorporate PR feedbacks
* Incorporate PR feedbacks
