Function init

a07_Transformer/a2_decoder.py:100–126

()

    return result

def init():
    d_model = 512
    d_k = 64
    d_v = 64
    sequence_length =6 #5
    decoder_sent_length=6
    h = 8
    batch_size = 4*32
    num_layer=6
    # 2.set Q,K,V
    vocab_size = 1000
    embed_size = d_model
    initializer = tf.random_normal_initializer(stddev=0.1)
    Embedding = tf.get_variable("Embedding_d", shape=[vocab_size, embed_size], initializer=initializer)
    decoder_input_x = tf.placeholder(tf.int32, [batch_size, decoder_sent_length], name="input_x")  # [4,10]
    print("1.decoder_input_x:", decoder_input_x)
    decoder_input_embedding = tf.nn.embedding_lookup(Embedding, decoder_input_x)  # [batch_size*sequence_length,embed_size]
    #Q = embedded_words  # [batch_size*sequence_length,embed_size]
    #K_s = embedded_words  # [batch_size*sequence_length,embed_size]
    #K_v_encoder = tf.placeholder(tf.float32, [batch_size,decoder_sent_length, d_model], name="input_x") #sequence_length
    Q = tf.placeholder(tf.float32, [batch_size,sequence_length, d_model], name="input_x")
    K_s=decoder_input_embedding
    K_v_encoder= tf.get_variable("v_variable",shape=[batch_size,decoder_sent_length, d_model],initializer=initializer) #tf.float32,
    print("2.output from encoder:",K_v_encoder)
    mask = get_mask(decoder_sent_length) #sequence_length
    decoder = Decoder(d_model, d_k, d_v, sequence_length, h, batch_size, Q, K_s, K_v_encoder,decoder_sent_length,mask=mask,num_layer=num_layer)
    return decoder,Q, K_s

decoder,Q, K_s=init()