I am trying to customize gradient descent methods in multiple layers with tf.GradientTape. The Tensorflow tutorial provides examples for a regression which does not have multipl