AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

SoftMoE: Soft Differentiable Routing for Mixture-of-Experts in LLMs

Mikołaj Zasada, Łukasz Struski, Jacek Tabor, Marcin Kurdziel

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
use soft top-k routing (SoftMoE*) instead of sparse top-2 routingSparse MoE k=2 · OWT validationloss: 2.79 -> 2.78Improved
use soft top-k routing (SoftMoE*) instead of sparse top-2 routingSparse MoE k=2 · C4 validationloss: 2.24 -> 2.23Improved
use SoftMoE* k=1 alpha=4 instead of sparse top-4Sparse MoE k=4 · OWT validationloss: 2.75 -> 2.70Improved
use SoftMoE* k=1 alpha=4 instead of sparse top-4Sparse MoE k=4 · C4 HellaSwagaccuracy: 47.23 -> 48.48Improved
add learned expert allocation across layers (ablation)SoftMoE* k=1.5 alpha=2 · OWT validationloss: 2.78 -> 2.76Improved
add learned expert allocation across layers (ablation)SoftMoE* k=2 alpha=2 · OWT validationloss: 2.75 -> 2.74Improved
add learned expert allocation across layers (ablation)SoftMoE* k=2 alpha=2 · C4 validationloss: 2.21 -> 2.20Improved
add learned expert allocation across layers (ablation)SoftMoE* k=2 alpha=2 · C4 HellaSwagaccuracy: 48.05 -> 47.48Hurt
lower inference truncation threshold tau from 2.4 to 1.2 (ablation)SoftMoE k=2 · C4 HellaSwagaccuracy: 46.98 -> 47.48Mixed
use truncation tau 1.3 instead of 1.5 (ablation)SoftMoE* k=1.5 alpha=2 · OWT PIQAaccuracy: 62.02 -> 63.87Improved
add routing jitter 0.05 to sparse MoE top-2 (ablation)Sparse MoE k=2 · OWT-style small runloss: 2.99 -> 3.00Hurt
add routing jitter 0.05 to sparse MoE top-1 (ablation)Sparse MoE k=1 · HellaSwagaccuracy: 28.79 -> 29.21No clear effect

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