Model distillation is a method to create a small, efficient model by training it on the outputs of a larger, more capable model