GitLab

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

* move deepspeed to `lib_integrations.deepspeed`

* more refactor

* oops

* fix slow tests

* Fix docs

* fix docs

* addess feedback

* address feedback

* final modifs for PEFT

* fixup

* ok now

* trigger CI

* trigger CI again

* Update docs/source/en/main_classes/deepspeed.md
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* import from `integrations`

* address feedback

* revert removal of `deepspeed` module

* revert removal of `deepspeed` module

* fix conflicts

* ooops

* oops

* add deprecation warning

* place it on the top

* put `FutureWarning`

* fix conflicts with not_doctested.txt

* add back `bitsandbytes` module with a depr warning

* fix

* fix

* fixup

* oops

* fix doctests

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

* 🐛

 Handle empty gen_kwargs for seq2seq trainer prediction_step fn
Signed-off-by: gkumbhat <kumbhat.gaurav@gmail.com>

* Update src/transformers/trainer_seq2seq.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

---------
Signed-off-by: gkumbhat <kumbhat.gaurav@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

missing None check

* seq2seq trainer and training arguments accepting GenerationConfig arg

* seq2seq Trainer and training arguments docstring fixes

* Update training_args_seq2seq.py docstring
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Fixing trainer_seq2seq.py docstring
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* seq2seq trainer: legacy gen args back & GenerationConfig created at init

* Seq2seq trainer: fix in case gen_config.max_new_tokens is None
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* seq2seq trainer: adding legacy arg retrocompatibility

* seq2seq trainer and training arguments accepting GenerationConfig arg

* seq2seq Trainer and training arguments docstring fixes

* Update training_args_seq2seq.py docstring
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Fixing trainer_seq2seq.py docstring
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* seq2seq trainer: legacy gen args back & GenerationConfig created at init

* Seq2seq trainer: fix in case gen_config.max_new_tokens is None
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* seq2seq trainer: adding legacy arg retrocompatibility

* seq2seq trainer: evaluate and predict untouched

* Apply suggestions from code review
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* seq2seq trainer: adding init args, keeping IDEs hints

---------
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Let generate pick its inputs

* fix squad seq2seq example

* Result of black 23.1

* Update target to Python 3.7

* Switch flake8 to ruff

* Configure isort

* Configure isort

* Apply isort with line limit

* Put the right black version

* adapt black in check copies

* Fix copies

Do not log the generation config for each iteration

* reflect max_new_tokens in gen_kwargs to `trainer.generate()`

* reflect max_new_tokens in `Seq2SeqTrainer`

* remove unnecessary variable

* Trigger CI

* fix style

* Link to the Datasets doc

* Remove unwanted file

Add logits_processor parameter, used by `generate`, to `Seq2SeqTrainer` methods `evaluate` and `predict` (#17805)

* Add logits_processor parameter, used by `generate`, to `Seq2SeqTrainer` methods `evaluate` and `predict`

* Add all generate parameters to `Seq2SeqTrainer`, and also to `QuestionAnsweringSeq2SeqTrainer` which overrides it

* Remove `self._num_beams` from trainer classes

* - Run fixup
- Fix "Constraint" not exposed
- Fix synced_gpus to actually read from param

* Use kwargs

* Copy kwargs before making changes to it

* Fix style issues unused imports

* Support compilation via Torchdynamo, AOT Autograd, NVFuser

* Address comments

* Lint

* Stas comments - missing quality test

* Lintere

* Quality test

* Doc lint

* Reset CUDA peak mem

* Add CustomTrainer

* require a single gpu
Co-authored-by: Stas Bekman <stas@stason.org>

If global_attention_mask is found in the models inputs (used by certain
models, like LED) in the prediction_step method of Seq2SeqTrainer,
it is added to the gen_kwargs, which are passed to model.decode().
This allows us to properly set the global attention when decoding.

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

It solves the problem that metric_key_prefix is different from trainer.

* redo sans examples

* style

* New doc styler

* Fix issue with args at the start

* Code sample fixes

* Style code examples in MDX

* Fix more patterns

* Typo

* Typo

* More patterns

* Do without black for now

* Get more info in error

* Docstring style

* Re-enable check

* Quality

* Fix add_end_docstring decorator

* Fix docstring

* up

* save

* correct

* up

* correct more

* up

* up

* up

* up

* up

* correct

* fix tf

* fix

* remove tokenizer

* Convert docstrings of all configurations and tokenizers

* Processors and fixes

* Last modeling files and fixes to models

* Pipeline modules

* Utils files

* Data submodule

* All the other files

* Style

* Missing examples

* Style again

* Fix copies

* Say bye bye to rst docstrings forever

* [Seq2SeqTrainer] Remove model input name hack

* Update src/transformers/trainer_seq2seq.py

* make style

* finish

* [trainer] conditional ctx managers into one wrapper

* workaround for contextlib.nullcontext for py<3.7

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

* one more autocast

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

Fix a Bug, trainer_seq2seq.py, in the else branch at Line 172, generation_inputs should be a dict (#14546)

* fix bug, trainer_seq2seq.py, Line 172, generation_inputs must be a dict before feeding into self.model.generation()

* fix bug, trainer_seq2seq.py, Line 172, generation_inputs must be a dict before feeding into self.model.generation()

* First draft

* Make tuple output more readable

* Replace assertions by value errors

* Make it possible to predict_with_generate for vision and speech models

* Adapt Seq2SeqTrainer to work with VisionEncoderDecoder/SpeechEncoderDecoder

* Add deprecation warning

* Add copied from statements to vision and speech encoder decoders

* Fix failing test

* Apply @patrickvonplaten's suggestion

* Use reshape instead of view for consistency

* Add generate kwargs to Seq2SeqTrainingArguments

* typo

* Address review comments + doc

* Style

* Update torch.utils.data namespaces to the latest.

* Format

* Update Dataloader.

* Style

* set max_length and num_beams only when non None

* fix instance variables

* fix code style

* move code and docs

* style

* moved

* restore

* synced gpus

* fix

* fix

* need to use t5-small for quality tests

* notes

* complete merge

* fix a disappearing std stream problem

* start zero3 tests

* wip

* tune params

* sorting out the pre-trained model loading

* reworking generate loop wip

* wip

* style

* fix tests

* split the tests

* refactor tests

* wip

* parameterized

* fix

* workout the resume from non-ds checkpoint pass + test

* cleanup

* remove no longer needed code

* split getter/setter functions

* complete the docs

* suggestions

* gpus and their compute capabilities link

* Apply suggestions from code review
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* style

* remove invalid paramgd

* automatically configure zero3 params that rely on hidden size

* make _get_resized_embeddings zero3-aware

* add test exercising resize_token_embeddings()

* add docstring
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Remove subclass for sortish sampler

* Use old Seq2SeqTrainer in script

* Styling

* Add label smoothing in Trainer

* Add options for scheduler and Adafactor in Trainer

* Put Seq2SeqTrainer in the main lib

* Apply suggestions from code review
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Address review comments and adapt scripts

* Documentation

* Move test not using script to tests folder
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>