AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

Sem-MoE: Semantic-aware Model-Data Collaborative Scheduling for Efficient MoE Inference

Yan Li, Zhenyu Zhang, Zhengang Wang, Pengfei chen, Pengfei Zheng

ICLR 2026 · Architectures and efficiency › Mixture of experts · approach: Efficient MoE language model systems

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Sem-MoE semantic scheduling vs SGLang with DeepEPSGLang, Attention-DP · DeepSeek-V2-Litethroughput under TTFT and E2E SLO: 31% and 221%Improved
add Sem-MoE vs MoETunerMoETuner, Attention-DP · DeepSeek-V2-Litethroughput under TTFT and E2E SLO: 32% and 278%Improved
add Sem-MoE vs SGLangSGLang, Attention-DP · Qwen3-30B-A3Bthroughput under SLO: 98% and 11%Improved
add Sem-MoE vs MoETunerMoETuner, Attention-DP · Qwen3-30B-A3Bthroughput under SLO: 35% and 32%Improved
add Sem-MoE token reschedulingSGLang, Attention-TP · DeepSeek-V2-LiteTTFT: 12.21%, 10.60%, 18.89%Improved
add Sem-MoE token reschedulingSGLang, Attention-TP · Qwen3-30B-A3BTTFT: 17.16%, 24.90%, 3.80%Improved
use MoETuner expert placement instead of vanillaSGLang, Attention-TP · DeepSeek-V2-Lite TTFT, input 1024p99 TTFT (ms): 111.19 -> 119.76Hurt
add Sem-MoE collocation (ablation)vanilla placement · single MoE layerlocal activation rate: 0.24 -> 0.61Improved
add Sem-MoE collocation (ablation)vanilla placement · single MoE layer, all2all latencylatency: -41.8% and -46.6%Improved
use Sem-MoE placement vs best baseline (ablation)MoETuner/vanilla · DeepSeek-V2-Lite EP8local activation rate: 15.4% and 36.7%Improved
zero-shot transfer of predictor trained on ShareGPT to Lmsys-Chat-1M (ablation)Sem-MoE · DeepSeek-V2-LiteLAR (p50): 46.49% -> 41.25% (vs SGLang 25.01%)No clear effect
zero-shot transfer of ShareGPT predictor to MMLU (ablation)Sem-MoE · Qwen3-30B-A3BLAR (p50): 35.30% vs in-domain 36.57%, SGLang 25.00%Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy