GitLab

Adding the state-of-the-art contrastive search decoding methods for the codebase of generation_utils.py (#19477)

* add: the contrastive search for generaton_utils

* add: testing scripts for contrastive search under examples/text-generation

* update the quality of codes

* revise the docstring; make the generation_contrastive_search.py scripts;

* revise the examples/pytorch/text-generation/run_generation_contrastive_search.py to the auto-APIs format

* revise the necessary documents

* fix: revise the docstring of generation_contrastive_search.py

* Fix the code indentation

* fix: revise the nits and examples in contrastive_search docstring.

* fix the copyright

* delete generation_contrastive_search.py

* revise the logic in contrastive_search

* update the intergration test and the docstring

* run the tests over

* add the slow decorate to the contrastive_search intergrate test

* add more test

* do the style, quality, consistency checks

* [Examples] make default preprocessing_num_workers=1

* [Examples] revert changes in research projects

* fixed typo for SQuAD

* Fixed the preprocess_validation_function function for the labels to reflect the remaining truncated instances

* Rolled back the trainer_seq2seq_qa.py for UnboundLocalError: local variable 'metrics' referenced before assignment
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Add multi-node conditions in trainer_qa.py and trainer_seq2seq.py

* Code improvement

Fix link typo of the following content.
PyTorch version, Trainer
PyTorch version, no Trainer

* edit: casting attention_mask to long in DataCollatorCTCWithPadding

* edit: casting attention_mask to long in DataCollatorCTCWithPadding

* Update no_trainer script for summarization

* removed unnecessary import

* fixes notation mistake

* removed: unused variable

* fix args option

* fix trainer eval log

* fix out of memory qa script

* do isort, black, flake

* fix tokenize target

* take it back.

* fix: comment

* Use repo_type instead of deprecated datasets repo IDs

* Add missing one in doc

The overwrite_cache parameter is declared twice.

* fixed bug which caused overwrite_cache to always be True (#18967).

* reformatting changes

* NeptuneCallback improvements

* After review suggestions and deduplication of initial run

* Added volatile checkpoints support due to missing post-rebase commit

* Update README per review comments

- Remove list formatting
- Correct Neptune docs link
Co-authored-by: Sabine <sabine.nyholm@neptune.ai>

* add accelerator.end_training()

Some trackers need this to end their runs.

* fixup and quality

* add space

* add space again ?!?

Co-authored-by: Arun Rajaram <arunrajaram@Aruns-MacBook-Pro.local>

* fixed incorrect param to hasattr

* simplified condition checks

* code cleanup

* `model.tie_weights()` should be applied after `accelerator.prepare`

Weight tying should be done after the model has been moved to XLA device as mentioned on PyTorch/XLA Troubleshooting guide [here](https://github.com/pytorch/xla/blob/master/TROUBLESHOOTING.md#xla-tensor-quirks)

* format code

* examples: add Bloom support for token classification (FLAX, PyTorch and TensorFlow)

* examples: remove support for Bloom in token classication (FLAX and TensorFlow currently have no support for it)

* Update run_translation_no_trainer.py

found an error in selecting `no_decay` parameters and some small modifications when the user continues to train from a checkpoint

* fixs `no_decay` and `resume_step` issue

1. change `no_decay` list
2. if use continue to train their model from provided checkpoint, the `resume_step` will not be initialized properly if `args.gradient_accumulation_steps != 1`

* Added accelerate gradient accumulation wrapper to run_image_classification_no_trainer.py example script

* make fixup changes

* PR comments

* changed input to Acceletor based on PR comment, ran make fixup

* Added comment explaining the sync_gradients statement

* Fixed lr scheduler max steps

* Changed run_clm_no_trainer.py script to use accelerate gradient accum wrapper

* Fixed all scripts except wav2vec2 pretraining to use accelerate gradient accum wrapper

* Added accelerate gradient accum wrapper for wav2vec2_pretraining_no_trainer.py script

* make fixup and lr_scheduler step inserted back into run_qa_beam_search_no_trainer.py

* removed changes to run_wav2vec2_pretraining_no_trainer.py script and fixed using wrong constant in qa_beam_search_no_trainer.py script

* Fix compatibility with 1.12

* Remove pin from examples requirements

* Update torch scatter version

* Fix compatibility with 1.12

* Remove pin from examples requirements

* Update torch scatter version

* fix torch.onnx.symbolic_opset12 import

* Reject bad version
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* zero chance anyone's using that constant no?

* `transformers-cli login` => `huggingface-cli login`

* `transformers-cli repo create` => `huggingface-cli repo create`

* `make style`

* Delete valohai.yaml

* NLP => ML

* typo

* website supports https

* datasets

* 60k + modalities

* unrelated link fixing for accelerate

* Ok those links were actually broken

* Fix link

* Make `AutoTokenizer` auto-link

* wording tweak

* add at least one non-nlp task

* swag_no_trainer updated for with gather_metrics

* Removed unused variable samples_seen

* updated examples with gather_for_metrics

* swag_no_trainer updated for with gather_metrics

* Removed unused variable samples_seen