语义相似度模型用第一层的隐藏状态计算相似度？是不是有问题？

[IceFlameWorm]

sr/ss/model.py

    def forward(self, left, right):
        left_vec = self.embedding(left.to(DEVICE)).to(DEVICE)
        #         left_vec = pack_padded_sequence(left_vec, left_sent_lengths)
        right_vec = self.embedding(right.to(DEVICE)).to(DEVICE)
        #         right_vec = pack_padded_sequence(right_vec, right_sent_lengths)

        self.hidden = self.init_hidden(batch_size=left.size(1))

        left_lstm_out, (left_lstm_hidden, _) = self.lstm(left_vec, self.hidden)
        #         left_lstm_out, left_batch_size = pad_packed_sequence(left_lstm_out)
        #         assert torch.equal(left_sent_lengths, left_batch_size.to(DEVICE))

        right_lstm_out, (right_lstm_hidden, _) = self.lstm(right_vec, self.hidden)
        #         right_lstm_out, right_batch_size = pad_packed_sequence(right_lstm_out)
        #         assert torch.equal(right_sent_lengths, right_batch_size.to(DEVICE))

        return self.manhattan_distance(left_lstm_hidden[0], right_lstm_hidden[0])

最后一行计算曼哈顿距离 self.manhattan_distance(left_lstm_hidden[0], right_lstm_hidden[0])， 用的是第一层的隐藏状态，是不是有问题？