GitLab

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

- It appears that `tqdm` only introduced `tqdm.auto` in 4.27.
- See https://github.com/tqdm/tqdm/releases/tag/v4.27.0.
- Without the lower bound I received the following stack trace in an environment where I already had tqdm installed:
```
  File "/home/brendanr/anaconda3/envs/allennlp/lib/python3.6/site-packages/transformers/__init__.py", line 20, in <module>
    from .file_utils import (TRANSFORMERS_CACHE, PYTORCH_TRANSFORMERS_CACHE, PYTORCH_PRETRAINED_BERT_CACHE,
  File "/home/brendanr/anaconda3/envs/allennlp/lib/python3.6/site-packages/transformers/file_utils.py", line 24, in <module>
    from tqdm.auto import tqdm
ModuleNotFoundError: No module named 'tqdm.auto'
```

* add failing test

* fix call to _run_split_on_punc

* format with black

Rul -> Run

HANS evaluation

💄 super

[PyTorch 1.4] Fix failing torchscript test for xlnet

model.parameters() order is apparently not stable (only for xlnet, for some reason)

Fix an issue where `prepare_for_model()` gives a `KeyError` when
`return_token_type_ids` is set to `False` and `return_tensors` is
enabled.

Created a link between the linear layer bias and the model attribute bias. This does not change anything for the user nor for the conversion scripts, but allows the `resize_token_embeddings` method to resize the bias as well as the weights of the decoder.

Added a test.