TensorFlow 使用上個模型輸出的一個向量表示 來給RNN生成一段文本

import tensorflow as tf

from tensorflow.contrib.rnn import LSTMCell

cell = LSTMCell(128,state_is_tuple=False)
init_state = tf.random_normal([16,128]) # 可以是上一模型的輸出

output,new_state = cell(inputs=init_state,state=init_state)

for i in range(20):
    output,new_state = cell(inputs=output,state=init_state) # 這個state=也試試傳output

print()

或者

import tensorflow as tf

from tensorflow.contrib.rnn import LSTMCell

cell = LSTMCell(128,state_is_tuple=False)
init_state = tf.random_normal([16,128]) # 可以是上一模型的輸出

output,new_state = cell(inputs=init_state,state=init_state) # new_state是[batch_size,256]

mlp = tf.keras.layers.Dense(units=128)
new_state = mlp(new_state)

for i in range(20):
    output,new_state = cell(inputs=output,state=new_state) 
    new_state = mlp(new_state)

print()