Mote-Carlo Dropout from loaded model in tf.keras

Question

I am using tf.keras (not Keras itself, but the tensorflow implementation of keras). I would like to use dropout during prediction, as done in Monte-Carlo Dropout Tutorial, b