rnn.py 文件源码

def forward_extended(self, input, weight, hx):
        assert cudnn.is_acceptable(input)

        output = input.new()

        if torch.is_tensor(hx):
            hy = hx.new()
        else:
            hy = tuple(h.new() for h in hx)

        cudnn.rnn.forward(self, input, hx, weight, output, hy)

        self.save_for_backward(input, hx, weight, output)
        return output, hy