Tag
Researchers introduce 8M-30M parameter micro language models that instantly generate the first few words on-device before cloud models complete responses, enabling responsive AI on ultra-constrained devices like smartwatches.