nan found when training with example data given

[zzzh1y]

Hi! I tried to run train.py with example data you given to get a quick understanding of your code. However, it failed to train and I found it got a Nan Loss in every iter. Is that normal for the the example data? And I also train it with my own data (I suppose my data processing was correct) and got the same problem. Do you have any clue? Thanks a lot!