1 paper
ChonLam Lao, Jiaqi Gao, Jiamin Cao +13
Large-scale ML training jobs are frequently interrupted by hardware and software anomalies, failures, and management events. Existing solutions like checkpoint-restart or runtime r…