▲ 2 pointsTPI-LLM: Serving 70B-Scale LLMs Efficiently on Low-Resource Edge Devicesarxiv.orgby CrypticShift·1y ago·0 comments·view on hn ↗