February 20, 2026 · papers
APEX accepted at IPDPS 2026
APEX runs online LLM inference on GPUs too small to hold the model, overlapping CPU and GPU execution so neither waits on the other. The paper appears at the 40th IEEE International Parallel and Distributed Processing Symposium in New Orleans. Congratulations to Jiakun Fan, Yiyuan Zhang and Xiangchen Li.