Is it desirable to have a physical models for the internal learning activity?

It must be desirable, but definitely not realistic. Only a compromise solution can be found, using the imagined, hypothetical model. Otherwise, people either have their own ambiguous models, abstract models, or no models, no senses of these models.

Eventually, we rely on the ‘schemes/mental models’ guide our learning behavior, make sense of the world of learning: the events of learning, the outcome of learning, the success/failure of learning, the emotional effects of all learning. So, the answer would be: YES, we need a physical models on learning.