LMSYS：Blog（Chatbot Arena 团队）·· 13 小时前AI 评分52
Novita AI 分享 GLM4-MoE 生产部署优化：基于 SGLang 将 TTFT 最高降低 65%
Blog Optimizing GLM4-MoE for Production: 65% Faster TTFT with SGLang A suite of production-tested, high-impact optimizations has been developed by Novita AI for deploying GLM4-MOE models based on SGLANG. We introduce an end-to-end performance optimization strategy that... Novita AI January 21, 2026
AI 导读
Novita AI 发布基于 SGLang 部署 GLM4-MoE 模型的端到端推理优化方案，TTFT 最高降低 65%，agentic coding 负载下 TPOT 提升 22%，结果在 H200 集群 TP8、FP8 配置下验证。
来源：LMSYS：Blog（Chatbot Arena 团队） · lmsys.org