Since you’re reading sequentially through the tensor’s memory, you need contigous tensor. There is a pretty good explanation what contigous does over here https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107
1 Like