GitLab

Necessary to run xlnet (at least in squad) with `--fp16 --fp16_opt_level="O2"`, otherwise loss is immediately `NaN` and fine-tuning cannot proceed.

fixed: hard coding for max and min number will out of range in fp16, which will cause nan.

merege from original repo

changes in evaluate function in run_lm_finetuning.py

Fix to prevent crashing on assert len(tokens_b)>=1

Fixing typo in gpt2 for doc site's class link

Trying to fix the head masking test

changes in return statement of evaluate function

changed `results` to `result` and removed `results` dict defined previously

[2.0] Reodering arguments for torch jit #1010 and future TF2.0 compatibility

Better examples

Fix typo

[2.0] - Split configuration and modeling files