Non-deterministic behavior of Pytorch upsample/interpolate

Same question, my post: Reproduce Training Question