作者位次 #2/17OSDI 2026Oral· Operating Systems & Distributed Systems
Safeguarding LLM Training at Scale: Online SDC Detection and Insights from 35 Million GPU Hours
Kinman Lei · Liyan Zheng · Xiang Li · Hongmin Chen · Yun Zhang · Gaohong Liu · Zuquan Song · Zixuan Ma · Zhiyu Xue · Minghui Yu · Shuguang Wang · Wencong Xiao · Haibin Lin · Yuyang Jin · Jidong Zhai · Bo Liu · Xin Liu
摘要
Silent Data Corruption (SDC) poses a critical threat to large-scale LLM training. Existing offline tests and online detection methods provide solutions for large-scale systems, yet they suffer from high overhead or low detection accuracy in LLM training. This paper presents AEGIS, an online SDC d...