目录 最保险的办法: 1 tensor.clone() 2 tensor.detach() 问题来源: a = torch.zeros(2, 3) for i in range(2): for j in range(<