GitLab

* [JAX] Replace all jax.tree_* calls with jax.tree_util.tree_*

* fix double tree_util

* [Flax] Add remat (gradient checkpointing)

* fix variable naming in test

* flip: checkpoint using a method

* fix naming

* fix class naming

* apply PVP's suggestions from code review

* add gradient_checkpointing to examples

* Add gradient_checkpointing to run_mlm_flax

* Add remat to longt5

* Add gradient checkpointing test longt5

* Fix args errors

* Fix remaining tests

* Make fixup & quality fixes

* replace kwargs

* remove unecessary kwargs

* Make fixup changes

* revert long_t5_flax changes

* Remove return_dict and copy to LongT5

* Remove test_gradient_checkpointing
Co-authored-by: sanchit-gandhi <sanchit@huggingface.co>

* zero chance anyone's using that constant no?

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

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

* `make style`

* add bart pretraining flax script

* fixup

* add bart pretraining flax script

* add BART to README

* add BART to README

* add BART to README

* add BART to README

* add BART to README

* add bos eos document

* Update README.md

* Update README.md

* Update examples/flax/language-modeling/run_bart_dlm_flax.py
Co-authored-by: Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>

* final

* final

* final

* remove use_auth_token ing from_config
Co-authored-by: Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>

* Fix ROUGE add example check and update README

* Stay consistent in values

* Preliminary work on tokenizers

* Quality + fix tests

* Treat processors

* Fix pad

* Remove all uses of  in tests, docs and examples

* Replace all as_target_tokenizer

* Fix tests

* Fix quality

* Update examples/flax/image-captioning/run_image_captioning_flax.py
Co-authored-by: amyeroberts <amy@huggingface.co>

* Style
Co-authored-by: amyeroberts <amy@huggingface.co>

Currently, tensorflow examples use the `load_metric` function from
Datasets library, commit migrates function call to `load` function
from Evaluate library.

* [Flax] Fix incomplete batches in example scripts

* fix dataloader batching

* convert jnp batch idxs to np array

* add missing `pad_shard_unpad` to final prediction generate step

* only `pad_shard_unpad` at inference time

* merge conflicts

* remove incomplete batch step from eval

* fix run_qa.py

* add `pad_shard_unpad` to run_flax_ner.py

* add `pad_shard_unpad` to run_flax_glue.py

* add `pad_shard_unpad` to run_image_classification.py

* make style

* fix mlm flax eval batches

* remove redundant imports

* generalize decay_mask_fn to find all layernorm params

* fixup

* generalising decay_mask_fn

* remove use_auth_token from from_config

* restore use_auth_token from_pretrained run_t5_mlm_flax

* Fix RESOURCE_EXHAUSTED error for large datasets on Flax example scripts

* using np.permutation for creating batch_idx

* train_samples_idx -> training_samples_idx

* fix type hints

* Add examples telemetry

* Alternative approach

* Add to all other examples

* Add to templates as well

* Put framework separately

* Same for TensorFlow

* Black preview

* Fixup too!

* Fix check copies

* Use the same version as the CI

* Bump black

1. Fixes evaluation errors popping up when you train/eval on squad v2 (one was newly encountered and one that was previously reported Running SQuAD 1.0 sample command raises IndexError #15401 but not completely fixed).
2. Removes boolean arguments that don't use store_true. Please, don't use these: *ANY non-empty string is being converted to True in this case and this clearly is not the desired behavior (and it creates a LOT of confusion).
3. All no-trainer test scripts are now saving metric values in the same way (with the right prefix eval_), which is consistent with the trainer-based versions.
4. Adds forgotten model.eval() in the no-trainer versions. This improved some results, but not everything (see the discussion in the end). Please, see the F1 scores and the discussion below.

* Correct Logging of Eval metric to Tensorboard

An empty dictionary ``eval_metrics`` was being logged, is replaced by ``eval_metric`` which is the output dictionary of ``metric.compute()``.

* Remove unused variable

Move declaration of log streams to before tests, so that results won't get compounded on top of each other

* Fix t5 shard on TPU Pods

The current script doesn't work properly on a TPU pod because the global batch is not divided correctly per host.
This pull request fixes this issue by dividing the global batch to each host before it is shared on each host.

* fix style
Co-authored-by: ahmed-elnaggar <ahmed.elnaggar@allianz.com>

* fix formatting and remove use_auth

* Add use_auth_token to Flax examples

* [examples] max samples can't be bigger than then len of dataset

* do tf and flax

Fix typo in comment: proprocessed -> preprocessed

* Split file_utils in several submodules

* Fixes

* Add back more objects

* More fixes

* Who exactly decided to import that from there?

* Second suggestion to code with code review

* Revert wront move

* Fix imports

* Adapt all imports

* Adapt all imports everywhere

* Revert this import, will fix in a separate commit

* Updates the default branch from master to main

* Links from `master` to `main`

* Typo

* Update examples/flax/README.md
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

Speedup training by using numpy instead of jnp for batch shuffling
Co-authored-by: Yeb Havinga <y.t.havinga@mgrid.net>

* Per-folder tests reorganization
Co-authored-by: sgugger <sylvain.gugger@gmail.com>
Co-authored-by: Stas Bekman <stas@stason.org>

* Harder check for IndexErrors in QA scripts

* Make test stronger

fix typo

* add a section about GPUs

* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* refactor run_flax_glue.py

* updated readme

* rm unused import and args typo fix

* refactor

* make consistent arg name across task

* has_tensorboard check

* argparse -> argument dataclasses

* refactor according to review

* fix