sniklaus/softmax-splatting

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation

hityzy1122 opened this issue · 3 comments

hityzy1122 commented

Hi thanks for your code!
I got an error as the title when call loss.backward, I fixed it when I changed self.save_for_backward(tenIn, tenFlow) into self.save_for_backward(tenIn.clone(), tenFlow.clone()), I don't know if it is right?

sniklaus commented

What is the layer before and the layer after the softmax splatting in your network?

sniklaus commented

Closing due to inactivity, please feel to reopen if this issue still persists. Thanks!

hityzy1122 commented