GitLab

* Working assymmetrical T5

* rename decoder_layers -> num_decoder_layers

* Fix docstring

* Allow creation of asymmetric t5 students

Co-authored-by: Swetha Mandava <smandava@nvidia.com>

Co-authored-by: Suraj Patil <surajp815@gmail.com>
Co-authored-by: Sam Shleifer <sshleifer@gmail.com>

* fix multi-gpu

* fix longformer

* force to delete unnecessary layers

* fix notifications

* fix warning

* fix roberta

* fix tests

* remove hasattr

* fix tests

* fix roberta

* merge and clean authorized keys

Co-authored-by: Your Name <you@example.com>

* Fix #7371

* Fix training

* Fix test values

* Apply the fix to TF as well

* [code quality] merge style and quality targets

Any reason why we don't run `flake8` in `make style`? I find myself needing to run `make style` and `make quality` all the time, but I need the latter just for the last 2 checks. Since we have no control over the source code why bother with separating checking and fixing - let's just have one target that fixes and then performs the remaining checks, as we know the first two have been done already.

This PR suggests to merge the 2 targets into one efficient target.

I will edit the docs if this change resonates with the team.

* move checks into style, re-use target

* better name

* add fixup target

* document new target

* Fix retrieval offset in RAG's HfIndex

* update slow tests

* style

* fix new test

* style

* add better tests
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* fix_rag_sequence

* add second bug fix

* add docs

* add missing docs and attention_mask in fine-tune

* Check config type instead of instance


Bad merge

* Remove for loops

* Style

* Mbart passing

* boom boom

* cleaner assert

* add assert

* Fix tests

Co-authored-by: Sam Shleifer <sshleifer@gmail.com>

Co-authored-by: Sam Shleifer <sshleifer@gmail.com>

* Remove mentions of  RAG from the docs

* Deactivate check

* Formatter

* Docs

* remote debugging

* remote debugging

* moved _store_flos call

* moved _store_flos call

* moved _store_flos call

* removed debugging artefacts

* Clean RAG docs and template docs

* Fix typo

* Better doc

* Refacto the models to remove their interdependencies

* Fix Flaubert model

* Fix Flaubert

* Fix XLM

* Fix Albert

* Fix Roberta

* Fix Albert

* Fix Flaubert

* Apply style + remove unused imports

* Fix Distilbert

* remove unused import

* fix Distilbert

* Fix Flaubert

* Apply style

* Fix Flaubert

* Add the copy comments for the check_copies script

* Fix MobileBert model name

* Address Morgan's comments

* Fix typo

* Oops typo

Updata tokenization_auto.py to handle Inherited tokenizer

correct a very small mistake

* Check decorator order

* Adapt for parametrized decorators

* Fix typos

Previously, the TFTrainingArguments object did a check to see if XLA was enabled, but did this by referencing `self.args.xla`, when it should be `self.xla`, because it is the args object. This can be verified a few lines above, where the XLA field is set.

* Changed name to all no_... arguments and all references to them, inverting the boolean condition

* Change benchmark tests to use new Benchmark Args

* Update src/transformers/benchmark/benchmark_args_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update src/transformers/benchmark/benchmark.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Fix Style. Add --no options in help

* fix some part of tests

* Update src/transformers/benchmark/benchmark_args_utils.py

* Update src/transformers/benchmark/benchmark_args_utils.py

* Update src/transformers/benchmark/benchmark_args_utils.py

* fix all tests

* make style

* add backwards compability

* make backwards compatible
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: fmcurti <fcurti@DESKTOP-RRQURBM.localdomain>