NaNs During Training

[ctralie]

Hello,
Thank you for making this very cool work publicly accessible!
I'm running into an issue where I get NaNs in training. I noticed you created a catch for that. I thought it might be coming from taking logs of very small numbers in the MSS loss, but that doesn't seem to be the issue. Since you put a catch there, I'm assuming you've dealt with this before, so do you have any idea what's causing it and how to avoid it?
Thanks,
Chris Tralie