I am using a UNet-like model (actually Attentional UNet) with tensorflow version 2.3.0 and tensorflow-gpu 1.14.0. I run my model in a GPU. It takes in a tensor, pads it and even