hrantzsch opened this issue 9 years ago · 1 comments
Overflows occur when using SGD. E.g. [...]batch_normalization.py:96: RuntimeWarning: overflow encountered in multiply gy - (self.x_hat * ggamma[expander] + gbeta[expander]) / m) Warning is only given when using CPU.
[...]batch_normalization.py:96: RuntimeWarning: overflow encountered in multiply gy - (self.x_hat * ggamma[expander] + gbeta[expander]) / m)