如何在Tensorflow中使RNN单元的权重不可训练？

import tensorflow as tf

rnn_cell = tf.nn.rnn_cell.BasicRNNCell(num_units=hidden_size)

inputs = tf.placeholder(shape=(batch_size, sequence_length, input_size), dtype=tf.float32)
initial_state = rnn_cell.zero_state(batch_size, dtype=tf.float32)

outputs, final_state = tf.nn.dynamic_rnn(rnn_cell, inputs, initial_state=initial_state)
final_state = tf.stop_gradient(final_state)