Model Distillation

Also: knowledge distillation

Training a smaller 'student' model to replicate the behavior of a larger 'teacher' model — producing smaller faster models that retain most of the original's capability.