#gpu-cpu

is it okay to calculate loss in cpu and backpropagate in torch?