GitLab

The "default=True" makes args.do_lower_case always True.

```python
parser.add_argument("--do_lower_case",
                        default=True,
                        action='store_true')
```

add do_lower_case arg and adjust model saving for lm finetuning.

Added Squad 2.0

Fix it to run properly even if without `--do_train` param.

It was modified similar to `run_classifier.py`, and Fixed to run properly even if without `--do_train` param.

Never split some texts.

LayerNorm initialization

Fix error when `bert_model` param is path or url.

Allow do_eval to be used without do_train and to use the pretrained model in the output folder

Adding new pretrained model to the help of the `bert_model` argument.

Correct the  wrong note

loading saved model when n_classes != 2

Fixing various class documentations.

Add example for fine tuning BERT language model

The LayerNorm gamma and beta should be initialized by .fill_(1.0) and .zero_().

reference links:

https://github.com/tensorflow/tensorflow/blob/989e78c412a7e0f5361d4d7dfdfb230c8136e749/tensorflow/contrib/layers/python/layers/layers.py#L2298

https://github.com/tensorflow/tensorflow/blob/989e78c412a7e0f5361d4d7dfdfb230c8136e749/tensorflow/contrib/layers/python/layers/layers.py#L2308

Error occurs when `bert_model` param is path or url. Therefore, if it is path, specify the last path to prevent error.

Required to for: Assertion `t >= 0 && t < n_classes` failed,  if your default number of classes is not 2.

add exemplary training data. update to nvidia apex. refactor 'item -> line in doc' mapping. add warning for unknown word.

raises value error for bert tokenizer for long sequences

Use entry-points instead of scripts

Add license to source distribution

The recommended approach to create launch scripts is to use entry_points
and console_scripts.

xref: https://packaging.python.org/guides/distributing-packages-using-setuptools/#scripts