Nan value in convolution output

I am getting nan value in the convolution output like this [ 6.6468e+00, 7.1471e+00, -7.2259e+00, nan].

I am using efficient net. Please help me.

Hi, Can you please put the code here for reproduction?

It is not the output of the convolution that produces nan but the input to the loss function. It is due to divisibility by negligible amount I think.

log of a negative decimal value is resulting nan value. Do you know how to solve this?. Should take absolute difference or make relu?