GitLab

Update apex fp16 implementation

fix: hard coding for max number

fix adding special tokens

Shortcut to special tokens' ids - fix GPT2 & RoBERTa tokenizers - improved testing for GPT/GPT-2

Added cleaned configuration properties for tokenizer with serialization - improve tokenization of XLM

Torch.hub now based on AutoModels - Updating AutoModels with AutoModelWithLMHead, Sequence Classification and Question Answering

Added multiple AutoModel classes: AutoModelWithLMHead, AutoModelForQuestionAnswering and AutoModelForSequenceClassification

regarding #1026 pull request 

change layernorm code to pytorch's native layer norm

loads the tokenizer for each checkpoint, to solve the reproducability…