As mobile devices increasingly become the primary computing platform for users, deploying large language models directly on these devices has emerged as a critical frontier