Why a full-1 tensor on GPU returns a value smaller than 1 when we take product of it?

1

I think the answer should be 1? However, when I switch the tensor to CPU, it returns 1. Why is the difference so significant?
Thank you.

2

Wow, that’s weird. It’s a bug in the C backend, the same thing happens in Lua Torch. We’ll fix that right away.

1 Like