我想实现一个具有简单自我注意的rnn模型：步骤i的输出是隐藏状态i和以前的隐藏状态（内存）的注意向量的线性组合。下面的代码有意义吗？
rnn_cell = tf.contrib.rnn.BasicLSTMCell(num_units)
h_states, final_state = tf.nn.dynamic_rnn(
rnn_cell, input_tensor, dtype=tf.float32, time_major=False)
attention_mechanism = tf.contrib.seq2seq.LuongAttention(
num_units, h_states,
memory_sequence_length=source_sequence_length)
rnn_cell_with_attention = tf.contrib.seq2seq.AttentionWrapper(
rnn_cell, attention_mechanism,
attention_layer_size=num_units)