Hi there,

I don’t know whether my question will make sense, but I’m trying to construct a complicated tensor involving a large array and functions like swapaxes, reshape, sum etc… The problem is that I’d like to allow a case in which that tensor is repeated several times into "sub"tensors (typically: different slices for sum) and that there’s quickly a memory issue. So, my question would be this one: is there any method / good practice to “share” a tensor variable that won’t be put into memory for each "sub"tensor calculation? Is this the idea between pytensor.shared?

Cheers,

Vian