使用 ModuleDict,我有: 输入类型 (torch.cuda.FloatTensor) 和权重类型 (torch.FloatTensor) 应该相同

Using ModuleDict, I have: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

我正在尝试 __init__ 函数:


        self.downscale_time_conv = np.empty(8, dtype=object)
        for i in range(8):
            self.downscale_time_conv[i] = torch.nn.ModuleDict({})

但在我的 forward 中,我有:

        down_out = False
        for i in range(8):
            if not down_out:
                down_out = self.downscale_time_conv[i][side](inputs)
            else:
                down_out += self.downscale_time_conv[i][side](inputs)

我得到:

RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same
        self.downscale_time_conv = torch.nn.ModuleList()
        for i in range(8):
            self.downscale_time_conv.append(torch.nn.ModuleDict({}))

这解决了它。显然我需要使用 ModuleList