Towards the Law of Capacity Gap in Distilling Language Models

Open in new window