1 paper
Guanbin Xu, ZhenGuo Xu, Yuzhe Li +4
Overlapping communication with computation is crucial for distributed large-model training, yet optimizing it - especially when computation becomes the bottleneck-remains challengi…