我在网上翻了好几篇技术文章，试了好几种实现方案，这里给你分享其中一种基于TensorFlow的BiLSTM实现方式：

def biLSTM(inputs, n_layers, seq_len, n_hidden, batch_size):
    output = inputs
    for n in range(n_layers):
        lstm_fw = tf.nn.rnn_cell.LSTMCell(n_hidden, state_is_tuple=True)
        lstm_bw = tf.nn.rnn_cell.LSTMCell(n_hidden, state_is_tuple=True)
        _initial_state_fw = lstm_fw.zero_state(batch_size, tf.float32)
        _initial_state_bw = lstm_bw.zero_state(batch_size, tf.float32)
        output, _states = tf.nn.bidirectional_dynamic_rnn(lstm_fw, ...)

内容的提问来源于stack exchange，提问作者Martin_at_Coventry