Tensor([1]),requires_grad=True) >> y=torch.sqrt(torch.pow(x - 1,0)) >> y.backward() >> x.grad Variable containing: nan [torch.FloatTensor of size 1].
確定！ 回上一頁
