GitLab

* beam sample in beam search

* rag now works with the updated beam search

* delete legacy (non-XLA) generation code related to beam sample

* update pyknp to rhoknp

* fix linter

* fix linter

* fix linter

* fix linter

* fix linter

* support rhoknp==1.1.0, fix testcase

* Add generate kwargs to AutomaticSpeechRecognitionPipeline

* Add test for generation kwargs

* Add generate kwargs to AutomaticSpeechRecognitionPipeline

* Add test for generation kwargs

[trainer: distributed_concat] ensure all_gather's input are contiguous

Fixing error message

* Fix error message

* Fix code quality

* add torch_dtype attribute to Pipeline

* Use torch_dtype to cast input tensor type in AutomaticSpeechRecognitionPipeline

* Fix code quality

* Add TextGenerationPipeline fp16 test

* Fix code quality

* Remove useless require in tests
Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com>
Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com>

load the state dict on cpu.

* Remove non-breaking space in comment

It was likely added unintionally.

* Remove remaining non-breaking spaces

correctly detect default max length

* Avoid collisions in writing metrics via 2 APIs - azureml + mlflow

MLflow tracking API is enabled by default in AzureML and HF MLflow integration is more fully featured. I'd remove the AzureML integration but leaving the current behavior for backwards compatibility (though it should really be removed)

* Trigger CI

* torch.jit._state

* Fix past CI

* Fix for perceiver

* Fix REALM

* Fix for Bloom

* Fix for SwinMode

* Fix for TrajectoryTransformerModel

* Fix for test_wav2vec2_with_lm

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

fix Jupyter typos

Update flan-t5.mdx

* fix fp16 loading issue

* add backward compatibility

* better refactor

* better readability

- remove `force_upcast_dtype` as it is used once
- use `inspect`
- add `TODO`

get correct checkpoint

* add japanese translation of template

* fix japanese translation

- fix special cases
- fix typos
- manually translate special cases
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Add script to convert T5X T5 (v1.0 and v1.1) checkpoints to PyTorch

* Remove unnecessary check and update docstring

* Format docstring

* Fix whitespace in docstring

* Supporting `fp16` for asr pipeline

* Adding test.

* Style.

* Oops.

* Flake8 update ?

* Fixing flake8 ?

* Revert "Flake8 update ?"

This reverts commit 0b917fcb520e5f34d1933d9d37d8f32b64553048.

* Style (acctidentally deleted flake8 F401.)

* Move to a bigger test (no small whisper model, and s2t doesn't seem to
accept torch_dtype=fp16).

Also we need to use a GPU to actually compute on fp16.

* Using BatchFeature capability.

poolformer onnx
Co-authored-by: syed <syed.abdul@sandlogic.com>

* fix typo `onnx`

* fix test

* add `get_encoder` and `get_decoder`

* add additional kwargs support

* fix condition

* add better checks

* better checks

* fix embed positions

* better test to consider padding

* fix debug statement

* Apply suggestions from code review
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* add arguments on docstring
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

change strings to f-strings

* fix doctests

* revert unwanted change

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

* add fast not use warning

* update

* fix doc for generation, dinat, nat and prelayernorm

* style

* update

* fix cpies

* use auto config and auto tokenizer
Co-authored-by: sgugger <sylvain.gugger@gmail.com>

* als modify roberta and the depending models
Co-authored-by: sgugger <sylvain.gugger@gmail.com>

* fix_whisper_export

* update input

* update input

* Fix past CI

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

Adding `evaluate` to the list of libraries to be installed for every generated notebook in transformers

Adds visual_prompt argument to CLIPSegProcessor to enable image-guided segmentation

* fix past_key_values in GPTNeoXForCausalLM.prepare_inputs_for_generation

* fix formatting

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

Update big table
Co-authored-by: Niels Rogge <nielsrogge@Nielss-MBP.localdomain>