这种情况是模型或者数据有的成功to(cuda)了，而有的隐形地还没有成功.cuda()

    大概率是因为model = model.to(device)只对model中实例化在init()中的函数有效，如果在forward中实例化并直接使用则不会将model放置到cuda中，因此很多是不正确地写法，没有成功让layer在init内实例化，导致模型没有成功完全迁移到GPU，但这种情况一般不会直接在该位置报错，因此需要仔细排查。
    错误代码实例


      1. class StdConv1d(nn.Module):
      2.  def __init__(self, in_channels, out_channels, kernel_size, input_shape1d):
      3.      super(StdConv1d, self).__init__()
      4.      self.conv1 = nn.Conv1d(in_channels,
      5.                             out_channels,
      6.                             kernel_size = kernel_size,
      7.                             stride = 1,
      8.                             padding='same')
      9.      #self.bn1 = nn.BatchNorm1d(out_channels)
      10.      self.ln = nn.LayerNorm
      11.      self.relu = nn.ReLU()
      13.  def forward(self, x):
      14.      x = self.conv1(x)
      15.      #x = self.bn1(x)
      16.      x = self.ln(x)
      17.      x = self.relu(x)
      19.      return x


      1. class StdConv1d(nn.Module):
      2.  def __init__(self, in_channels, out_channels, kernel_size, input_shape1d):
      3.      super(StdConv1d, self).__init__()
      4.      self.conv1 = nn.Conv1d(in_channels,
      5.                             out_channels,
      6.                             kernel_size = kernel_size,
      7.                             stride = 1,
      8.                             padding='same')
      9.      #self.bn1 = nn.BatchNorm1d(out_channels)
      10.      #self.ln = nn.LayerNorm
      11.      self.relu = nn.ReLU()
      13.  def forward(self, x):
      14.      x = self.conv1(x)
      15.      #x = self.bn1(x)
      16.      x = nn.LayerNorm(normalize_shape=x.shape[-1])(x)
      17.      x = self.relu(x)
      19.      return x