How to slide a set of Conv2d layers over not rectangular data (like a kernel) in pytorch?

So in my batches, I have 30 Items of shape Yx80 (Y >= 80). I want to

1. Expand data with 0s so that batch Y+x%80 == 0
2. take a sequential set of say 5 Conv2d lavers that take 80x80 and output 20x20 and pass them thru my batch Items like so:
3. Apply Conv2d layers so that they would not overlap and be like ooooo
4. get a new item of shape (Y+x)/80*20 by 20

How to do such a thing with pytorch?

from How to slide a set of Conv2d layers over not rectangular data (like a kernel) in pytorch?