GitLab

* Initial commit to get BERT + run_glue.py on TPU

* Add README section for TPU and address comments.

* Cleanup TPU bits from run_glue.py (#3)

TPU runner is currently implemented in:
https://github.com/pytorch-tpu/transformers/blob/tpu/examples/run_glue_tpu.py.

We plan to upstream this directly into `huggingface/transformers`
(either `master` or `tpu`) branch once it's been more thoroughly tested.

* Cleanup TPU bits from run_glue.py

TPU runner is currently implemented in:
https://github.com/pytorch-tpu/transformers/blob/tpu/examples/run_glue_tpu.py

.

We plan to upstream this directly into `huggingface/transformers`
(either `master` or `tpu`) branch once it's been more thoroughly tested.

* No need to call `xm.mark_step()` explicitly (#4)

Since for gradient accumulation we're accumulating on batches from
`ParallelLoader` instance which on next() marks the step itself.

* Resolve R/W conflicts from multiprocessing (#5)

* Add XLNet in list of models for `run_glue_tpu.py` (#6)

* Add RoBERTa to list of models in TPU GLUE (#7)

* Add RoBERTa and DistilBert to list of models in TPU GLUE (#8)

* Use barriers to reduce duplicate work/resources (#9)

* Shard eval dataset and aggregate eval metrics (#10)

* Shard eval dataset and aggregate eval metrics

Also, instead of calling `eval_loss.item()` every time do summation with
tensors on device.

* Change defaultdict to float

* Reduce the pred, label tensors instead of metrics

As brought up during review some metrics like f1 cannot be aggregated
via averaging. GLUE task metrics depends largely on the dataset, so
instead we sync the prediction and label tensors so that the metrics can
be computed accurately on those instead.

* Only use tb_writer from master (#11)

* Apply huggingface black code formatting

* Style

* Remove `--do_lower_case` as example uses cased

* Add option to specify tensorboard logdir

This is needed for our testing framework which checks regressions
against key metrics writtern by the summary writer.

* Using configuration for `xla_device`

* Prefix TPU specific comments.

* num_cores clarification and namespace eval metrics

* Cache features file under `args.cache_dir`

Instead of under `args.data_dir`. This is needed as our test infra uses
data_dir with a read-only filesystem.

* Rename `run_glue_tpu` to `run_tpu_glue`
Co-authored-by: LysandreJik <lysandre.debut@reseau.eseo.fr>

* [examples] Generate argparsers from type hints on dataclasses

* [HfArgumentParser] way simpler API

* Restore run_language_modeling.py for easier diff

* [HfArgumentParser] final tweaks from code review

* Big cleanup of `glue_convert_examples_to_features`

* Use batch_encode_plus

* Cleaner wrapping of glue_convert_examples_to_features for TF

@lysandrejik

* Cleanup syntax, thanks to @mfuntowicz

* Raise explicit error in case of user error

* improve and add features to benchmark utils

* update benchmark style

* remove output files

* Fix RoBERTa/XLNet Pad Token in run_multiple_choice.py

`convert_examples_to_fes atures` sets `pad_token=0` by default, which is correct for BERT but incorrect for RoBERTa (`pad_token=1`) and XLNet (`pad_token=5`). I think the other arguments to `convert_examples_to_features` are correct, but it might be helpful if someone checked who is more familiar with this part of the codebase.

* Simplifying change to match recent commits

* Resizing embedding matrix after sending it to the optimizer prevents from updating the newly resized matrix.

* Remove space for style matter

* replace heavy t5 models with tiny random models as was done by sshleifer

* fix isort

* Start cleaning examples

* Fixup

* fix conflicts

* add model size argument to summarization

* correct wrong import

* fix isort

* correct imports

* other isort make style

* make style

* Using loaded checkpoint with --do_predict

Without this fix, I'm getting near-random validation performance for a trained model, and the validation performance differs per validation run. I think this happens since the `model` variable isn't set with the loaded checkpoint, so I'm using a randomly initialized model. Looking at the model activations, they differ each time I run evaluation (but they don't with this fix).

* Update checkpoint loading

* Fixing model loading

* Update the NER TF script to remove the softmax and make the pad token label id to -1

* Reformat the quality and style
Co-authored-by: Julien Plu <julien.plu@adevinta.com>

* force bleu

* fix wrong file name

* rename file

* different filenames for each example test

* test files should clean up after themselves

* test files should clean up after themselves

* do not force bleu

* correct typo

* fix isort

* Use tokenizer.num_added_tokens to count number of added special_tokens instead of hardcoded numbers.
Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

* run_ner.py - Do not add a label to the labels_ids if word_tokens is empty.

This can happen when using bert-base-multilingual-cased with an input containing an unique space.
In this case, the tokenizer will output just an empty word_tokens thus leading to an non-consistent behavior
over the labels_ids tokens adding one more tokens than tokens vector.
Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

T5-small in test


isort

* add translation example

* make style

* adapt docstring

* add gpu device as input for example

* small renaming

* better README

* rebase to master

* change tf to pytorch

* change to pytorch

* small fix

* renaming

* add gpu training possibility

* renaming

* improve README

* incoorporate collins feedback

* better Readme

* better README.md

For more details, see https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate

* make style

* fix conflicts

torch.cuda.empty_cache() was being called from a TF function (even when torch is unavailable)
not sure any replacement is needed if TF OOMs

* solving bug where for small epochs and large gradient_accumulation_steps we never train

* black formatting

* no need to change these files

* Update examples/ner/run_ner.py to use AutoModel

* Fix missing code and apply `make style` command

* ✨ Alter base pl transformer to use automodels

* 🐛 Add batch size env variable to function call

* 💄 Apply black code style from Makefile

* 🚚 Move lightning base out of ner directory

* ✨ Add lightning glue example

* 💄 self

* move _feature_file to base class

* ✨ Move eval logging to custom callback

* 💄 Apply black code style

* 🐛 Add parent to pythonpath, remove copy command

* 🐛 Add missing max_length kwarg

* change do_samples back

* None better default as boolean

* adapt do_sample to True in test example

* make style