如何解决使用tenssorflow实现贝叶斯神经网络时的获取参数错误

import tensorflow as tf
import tensorflow_probability as tfp

tfd = tfp.distributions

class BayesianDenseLayer(tf.keras.layers.Layer):
    def __init__(self, units, **kwargs):
        super(BayesianDenseLayer, self).__init__(**kwargs)
        self.units = units

    def build(self, input_shape):
        self.kernel = tfd.Independent(
            tfd.Normal(loc=tf.Variable(tf.random.normal([input_shape[-1], self.units])),
                       scale=tf.Variable(tf.random.uniform([input_shape[-1], self.units], minval=0, maxval=1)))
        )
        self.bias = tfd.Independent(
            tfd.Normal(loc=tf.Variable(tf.random.normal([self.units])),
                       scale=tf.Variable(tf.random.uniform([self.units], minval=0, maxval=1)))
        )
        super(BayesianDenseLayer, self).build(input_shape)

    def call(self, inputs):
        kernel = self.kernel.sample()
        bias = self.bias.sample()
        return tf.matmul(inputs, kernel) + bias

class BayesianModel(tf.keras.Model):
    def __init__(self, **kwargs):
        super(BayesianModel, self).__init__(**kwargs)
        self.dense1 = BayesianDenseLayer(64)
        self.dense2 = BayesianDenseLayer(10)

    def call(self, inputs):
        x = tf.nn.relu(self.dense1(inputs))
        return self.dense2(x)

model = BayesianModel()

for var in model.trainable_variables:
    print(var.name, var.shape)

for layer in model.layers:
    for var in layer.trainable_variables:
        print(var.name, var.shape)