Xiaohongshu and Peking University open-source UltraEP for real-time MoE load balancing

小红书与北大开源 UltraEP：面向大规模 MoE 训推的实时负载均衡方案

Xiaohongshu and Peking University open-sourced UltraEP, a real-time load balancing method for large MoE models. It dynamically replicates hot experts per microbatch and per layer using exact routing info, hitting 94.6% of ideal training throughput. On Qwen3-235B, training throughput is 42% higher than Megatron-LM and prefill throughput is 1.56× SGLang. The post doesn't disclose the license or deployment requirements.

Why it matters: Joint open-source release from Xiaohongshu and Peking University tackles real MoE idle-GPU pain with strong numbers (94.6% ideal training throughput, 1.56x SGLang inference). Missing license and deployment requirements keep it from scoring higher—those determine real-world ado...