LMSYS:Blog(Chatbot Arena 团队)·· 13 小时前AI 评分41
SGLang 推出基于 RDMA 的 P2P 权重更新机制,1T 参数 Kimi-K2 传输从 53 秒降至 7.2 秒
Blog Updating 1T parameters in seconds — P2P weight transfer in Large Scale Distributed RL We introduced a RDMA-based, Peer to Peer weight update mechanism for RL workloads in SGLang as a supplement to traditional NCCL broadcast methods, compatible with all major open source models. By util... Jiadong Guo, Xin Ji, Letian Ruan, Teng Ma, Chenyang Zhao, Yueming Yuan, Zhichen Zeng April 29, 2026
AI 导读
SGLang 为 RL 工作负载引入基于 RDMA 的 P2P 权重更新机制,作为传统 NCCL broadcast 的补充,兼容所有主流开源模型。
来源:LMSYS:Blog(Chatbot Arena 团队) · lmsys.org