如何在TF1.15中创建有状态TensorFlowLite RNN模型

import tensorflow as tf
import numpy as np

num_units = 128  # RNN单元的数量
num_classes = 10  # 分类的数量

input_data = tf.placeholder(tf.float32, [None, num_steps, num_input])
output_data = tf.placeholder(tf.float32, [None, num_classes])

cell = tf.nn.rnn_cell.BasicRNNCell(num_units)
outputs, states = tf.nn.dynamic_rnn(cell, input_data, dtype=tf.float32)

output = tf.layers.dense(states, num_classes)
prediction = tf.nn.softmax(output)

loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=output, labels=output_data))
optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss)

init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)

for epoch in range(num_epochs):
    # 执行训练操作
    sess.run(optimizer, feed_dict={input_data: input_data_batch, output_data: output_data_batch})
    # 计算损失
    curr_loss = sess.run(loss, feed_dict={input_data: input_data_batch, output_data: output_data_batch})
    # 打印损失
    print("Epoch: {} Loss: {}".format(epoch, curr_loss))

saver = tf.train.Saver()
saver.save(sess, "model.ckpt")