1 paper
Chao Xu, Xu Zhang, Zihang Luo +5
Reducing collective communication latency is a critical goal for large model training and inference in both academia and industry. Many-to-many communications, such as AllGather an…