Can a model that small dynamically grow? In other words, can it train itself AS it progresses through the network?