get NAN even for max sequence length of 10(without any modification)

Hi, I'm studying NTM in recent days, and I'm trying to learn your code. 
- But when I just downloaded your code and run "python train.py" without any modification, I  got NAN gradient after 6000 iterations. 
- Do you have the same experience?