micro LMs

Micro LM delivers large‑model quality on device

A new study introduces Micro Language Models (μLMs), ultra-compact models (8M–30M parameters) that can deliver large-model quality on devices. This approach solves the dilemma between responsive first words and complete answers in edge assistants by seeding answers locally, reducing latency caused by cloud round-trips.