NaNs in model output, likely caused by arithmetic underflow

[andriuslauraitis]

Hello,

I'm getting this error in deep neural network training. How can i retune the network to avoid this error?

Best regards,
Andrius L.

[basedrhys]

Hi,

Unfortunately, we haven't found an easy solution - the things you can try is using a different learning rate or pretrained model.

Best,
Rhys