6fe058037c6126c5806c34debf86b3cf302219be,basic/model.py,,bi_attention,#Any#Any#Any#Any#Any#Any#Any#Any#,17

Before Change

    :return: [N, M, d], [N, M, JX, d]
    
    with tf.variable_scope(scope or "bi_attention"):
        N, M, JX, JQ, d = config.batch_size, config.max_num_sents, config.max_sent_size, config.max_ques_size, config.hidden_size
        JX = tf.shape(h)[2]
        M = tf.shape(h)[1]
        JQ = tf.shape(u)[1]

After Change

        if tensor_dict is not None:
            a_u = tf.nn.softmax(u_logits)  // [N, M, JX, JQ]
            a_h = tf.nn.softmax(tf.reduce_max(u_logits, 3))
            tensor_dict["a_u"] = a_u
            tensor_dict["a_h"] = a_h

        return u_a, h_a

Instances