AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

MoSE: Mixture of Slimmable Experts for Efficient and Adaptive Language Models

Nurbek Tastan, Stefanos Laskaridis, Karthik Nandakumar, Samuel Horváth

ICML 2026 · Architectures and efficiency › Mixture of experts · approach: Efficient MoE language model systems · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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
replace MoE with MoSE at full width (w=1.0)GPT2-Small 3B tokens · OpenWebTextperplexity: 41.82 -> 39.02Improved
replace MoE with MoSE at full width (w=1.0)GPT2-Small 15B tokens · OpenWebTextperplexity: 31.03 -> 31.26No clear effect
add test-time width identification (TTT) (ablation)MoSE GPT2-Standard 15B · OpenWebTextperplexity: 20.65 -> 20.38Improved
add test-time width identification (TTT) (ablation)MoSE GPT2-Standard 15B · LAMBADAaccuracy: 0.336 -> 0.338Improved
use MoSE (TTT) instead of MoEGPT2-Medium · HellaSwagaccuracy: 0.335 -> 0.385Improved
use MoSE (TTT) instead of MoEGPT2-Medium · PIQAaccuracy: 0.675 -> 0.710Improved
use MoSE (TTT) instead of MoEGPT2-Medium · SIQAaccuracy: 0.290 -> 0.305Improved
router-conditioned normalized-probability widths instead of uniform widths (ablation)MoSE inference · OpenWebText Pareto frontierperplexity at fixed FLOPsImproved
learn sharpness gamma by test-time training instead of fixed gamma=1.0 (ablation)MoSE inference · OpenWebText Pareto frontierperplexity at fixed FLOPsImproved
layer-wise gamma instead of shared gamma (ablation)MoSE TTT · OpenWebText Pareto frontierperplexityImproved
scale to GPT2-Large 2.5B with MoSEMoSE · OpenWebTextFLOPs at matched perplexity: 32.8% lower FLOPsImproved
continual pretraining slimmability adaptationDeepSeek-V2-Lite 16B · validation perplexityFLOPs at comparable perplexity: 38.1% lower FLOPsImproved

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