GitLab

* improve handling of short inputs for reformer

* correct typo in assert statement

* fix other tests

* Glue task cleaup

* Enable writing cache to cache_dir in case dataset lives in readOnly
filesystem.
* Differentiate match vs mismatch for MNLI metrics.

* Style

* Fix pytype

* Fix type

* Use cache_dir in mnli mismatch eval dataset

* Small Tweaks
Co-authored-by: Julien Chaumond <chaumond@gmail.com>

* 🐛 Fix model ids for BART and Flaubert

* Kill model archive maps

* Fixup

* Also kill model_archive_map for MaskedBertPreTrainedModel

* Unhook config_archive_map

* Tokenizers: align with model id changes

* make style && make quality

* Fix CI

* allow to not add special tokens

* remove print

Fixes bug reported in https://github.com/huggingface/transformers/issues/4669

See #3967 for context

* pass on tokenizer to pipeline

* order input names when convert to onnx

* update style

* remove unused imports

* make ordered inputs list needs to be mutable

* add test custom bert model

* remove unused imports

Co-authored-by: Julien Chaumond <chaumond@gmail.com>

Co-authored-by: Julien Chaumond <chaumond@gmail.com>

Readme for HooshvareLab/bert-base-parsbert-armananer-uncased

Readme for HooshvareLab/bert-base-parsbert-peymaner-uncased

mBERT results added regarding NER datasets!

* fix bug

* add more tests

* Fix longformer attention mask casting when using apex

* remove extra type casting

* better api

* improve automatic setting of global attention mask

* fix longformer bug

* fix global attention mask in test

* fix global attn mask flatten

* fix slow tests

* update docstring

* update docs and make more robust

* improve attention mask