Language Model Distillation

Knowledge distillation is a technique for training smaller neural networks to perform like larger ones. The basic idea is simple: train a small "student" model to copy the behavior of a large "teacher