Computing intermediate gradients using backward method in Pytorch

Question

I am having trouble in understanding backward method in pytorch

x1 = tensor(2.).requires_grad_()
x2 = tensor(3.).requires_grad_() # or x2 = tenso