1 paper · 1 filter
Yinmin Zhong, Shengyu Liu, Junda Chen +5
DistServe improves the performance of large language models (LLMs) serving by disaggregating the prefill and decoding computation. Existing LLM serving systems colocate the two pha…