pytorch loss.backward()核心理解 和lstm中hidden的典型错误

    X = torch.ones(2, 2, requires_grad=False)
    w = torch.ones(2, 2, requires_grad=True)
    c1 = X*w
    c1.backward()
    print(w.grad)

    # 会报错
    c2 = c1 * X
    c2.backward()

   #重新建立c1的链子
    c1=X*w

    c2 = c1 * X
    c2.backward()