keras can save model with tf.keras.models.save_model(), then restore model from tf.keras.models.load_model()
tf.keras.models.save_model()
tf.keras.models.load_model()
Here is a scenario,