1 paper
Juntao Zhao, Jiuru Li, Chuan Wu
CPUs are critical for LLM serving due to their availability, cost efficiency, and edge applicability. However, efficient CPU serving is hindered by conflicting prefill/decode resou…