GitLab

* Remove redundant `nn.log_softmax` in `run_flax_glue.py`

`optax.softmax_cross_entropy` expects unnormalized logits, and so it already calls `nn.log_softmax`, so I believe it is not needed here. `nn.log_softmax` is idempotent so mathematically it shouldn't have made a difference.

* Remove unused 'flax.linen' import

* Added logic to return attention from flax-bert model and added test cases to check that

* Added new line at the end of file to test_modeling_flax_common.py

* fixing code style

* Fixing Roberta and Elextra models too from cpoying bert

* Added temporary hack to not run test_attention_outputs for FlaxGPT2

* Returning attention weights from GPT2 and changed the tests accordingly.

* last fixes

* bump flax dependency
Co-authored-by: jayendra <jayendra@infocusp.in>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* seq classification changes

* fix tests

* Adding new argument `max_new_tokens` for generate.

This is a proposal to add a new argument `max_new_tokens` to `generate`.
This include a `MaxNewTokensCriteria` that enables callers that don't
know about the token length ahead (like pipelines callers) to manage
more easily the length of their generated output.

* Adding a test for the user warning when both`max_length` and
`max_new_tokens` are used together.

* Removed redundant `no_grad`.

* rebuild deepspeed config for hyperparameter search

* reformat code to fix style issues

* fix_torch_device_generate_test

* remove @

* add

* indexing

* correct a couple of tests

* fix tests

* add logits processor

* finish top_k, top_p, temp

* add docs

* correct flax prng key default

* improve generate

* add generation docs

* add docs

* make style

* revert model outputs change

* make style

* correct typo

* fix tests

* fix slow test

* add raise

* finish generation
Co-authored-by: Patrick von Platen <patrick@huggingface.co>

* changing find_batch_size to work with tokenizer outputs

trainer_pt_utils.find_batch_size does not recognize the batch size of BatchEncoding objects. This can cause an error when a trainer relies on find_batch_size to report the number of observed examples in the evaluation loop.

* Trigger CI
Co-authored-by: jrenner <joseph.renner@inria.fr>

* fix_torch_device_generate_test

* remove @

* change dataclasses to flax ones

* fix typo

* fix jitted tests

* fix bert & electra

* Correcting comments to reflect correct tuple order

In order to match the actual order (line 513 and 516, and as accessed in 968), I've changed the order mentioned in comments L962 and L966-967.

* Update modeling_t5.py

Updating another comment as well

* Removing extra space

* Fixing style and quality

* style & quality

* Update src/transformers/models/t5/modeling_t5.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Fix Bart

* Fix Blenderbot{,_small}

* Fix LED

* Fix Marian

* Fix MBart

* Fix Pegasus

* Fix T5

* Add test for generation with head_mask

* Add a common TF test

* Override a test for the LED model as head masking is not yet properly implemented

* Remove all head_masks from input preparation for LED

* Drop masking for T5 as it needs a bit of refactor

The feature extractor does not create tensors on the appropriate device,
so we call `ensure_tensor_on_device` before feeding the processed inputs
to the model.

* create custom model on the flight

* better wording

* add update_from_string

* cleanup

* cleanup

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

* more bool options

* style

* fix logger

* add test

* add the doc

* assert on conflict of options
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* fix overflow in perplexity calc

* use inf

* fix

* first try

* finish

* Add option to long only once in multinode training

* Use an alternate property

* [Trainer] Report both steps and num samples per second

* Fix batch number

* Update src/transformers/trainer_utils.py
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>

* Address review comments
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>

* typo2

* fix typo

* fixing flos bug/typo in non-distributed setting

* storing flos every logging_interval

* Switch mem metrics flag

* Update src/transformers/training_args.py
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>

* fix_torch_device_generate_test

* remove @

* change pytorch import to flax import

get_length_grouped_indices() in LengthGroupedSampler and DistributedLengthGroupedSampler
is prohibitively slow for large number of megabatches (in test case takes hours for ~270k
megabatches with 100 items each) due to slow list concatenation with sum(megabatches, []).

Resolves: #11795
Co-authored-by: ctheodoris <cvtheodo@ds.dfci.harvard.edu>

* fix_torch_device_generate_test

* remove @

* add flax glue link

* support zero.Init in from_config

* no need for eval test

* fix_torch_device_generate_test

* remove @

* correct best seed for flax fine-tuning
Co-authored-by: Patrick von Platen <patrick@huggingface.co>

* speed up flax glue

* remove unnecessary line

* remove folder

* remove run in loop
Co-authored-by: Patrick von Platen <patrick@huggingface.co>