[youarenotyu]

current LLMs are too large or efficient enough for local use needs more breakthroughs on both chip side and model side