Closed
Description
Description
Numpy, along with derivative tensor libraries like jax, torch, etc, all offer padding functions to insert values to the left and right (and top and bottom, etc) to a tensor. We should probably have this as well. It shouldn't be too difficult (famous last words) -- I see it essentially as some shape computation followed by an AllocSubtensor, although we could certainly try to figure out the best way to do this.