example in BertForSequenceClassification() conflicts with the api

[labixiaoK]

Hi, firstly, admire u for the great job. but I encounter 2 problems when i use it:
1. UnicodeDecodeError: 'gbk' codec can't decode byte 0x85 in position 4527: illegal multibyte sequence,
same problem as ISSUE 52 when I excute the BertTokenizer.from_pretrained('bert-base-uncased'), but I successfully excute BertForNextSentencePrediction.from_pretrained('bert-base-uncased'), >.<
2. in the pytorch-pretrained-BERT/pytorch_pretrained_bert/modeling.py,
line 761 --> ```
token_type_ids: an optional torch.LongTensor of shape [batch_size, sequence_length] with the token
types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` and type 1 corresponds to
a `sentence B` token (see BERT paper for more details).

but in the following example,  in **line 784**-->     `token_type_ids = torch.LongTensor([[0, 0, 1], [0, **2**, 0]])`, why the '2' appears?  I am confused.  Otherwise, is the situation similar to '0, 1, 0 ' correct ? Or it should be similar to [000000111111] , that is continuous '0' and continuous '1' ?
ty.

[thomwolf]

Hi,
(1) is solved on master. I will release a new release soon with the fixes on pip. In the mean time you can install from sources if you want.
I fixed the typo in the docstring you mention in (2), thanks, it should be a 1 instead of a 2.