I am trying to implement a tensorflow model (encoder decoder like) in which I train initially with a small number of layers, and append the model with more layers after trai