GitLab

Adding gpt-2 large (774M parameters) model

Add a few of typos corrections, bugs fixes and small improvements

Better use of spacy tokenizer in open ai and xlm tokenizers

Re-implemented tokenize() iteratively in PreTrainedTokenizer.

Fix #1015 (tokenizer defaults to use_lower_case=True when loading from trained models)

Fix a path so that a test can run on Windows

Fix typo. configuratoin -> configuration

Fix bug of multi-gpu training in lm finetuning

Swap of optimizer.step and scheduler.step for lm finetuning examples

This reverts commit a1359b97.