Train and valid loss are nan

[ztw33]

Hi shido,
Recently I've been trying to reproduce your work. I download the dataset, parse them using parser.jar. However, when I'm training the multiway model using command python train.py -m multiway -r 0.001 -b 80 -e 1 -g 0, both the train and valid loss are nan. I tried other models such as codenn, deepcom and childsum, things all seem to be fine. So I wonder where I'm wrong. If you can offer me some help, I will be appreiciated a lot!