Stanford Alpaca, and the acceleration of on-device large language model