tensor(2.4211, grad_fn=<NllLossBackward0>)
tensor(0.0831, grad_fn=<NllLossBackward0>)