4 ms·TPI-LLM: Serving 70B-Scale LLMs Efficiently on Low-Resource Edge Devices1 points by zoobab 2y ago