Skip to content
AI Atlas
PaperActive

Distribution-Consistent Inference for Dynamic Sparse Mixture-of-Experts

arxiv.org/abs/2609.09241

Updated 50 min ago · first seen 11 Sept 2026

paper_01M294GKVFFQD5FBD6BBBM0E2Y

Published
11 Sept 2026
T1 · 50 min ago
arXiv
2609.09241
T1 · 50 min ago
Category
cs.LG
T1 · 50 min ago

Abstract

Mixture-of-Experts (MoE) architectures have emerged as a powerful paradigm for scaling model capacity while preserving efficient inference in large foundation models. However, most MoE models use a fixed top-$k$ expert selection policy, assigning the same expert budget to every token even when fewer experts may be sufficient. Inference-time dynamic top-$k$ routing can reduce computation without retraining, but existing methods often overlook the distributional shift caused by deviating from the training-time routing configuration. We show that reducing the number of activated experts consistently increases the RMS scale and variance of SMoE outputs, inducing a representation mismatch that contributes to downstream performance degradation in addition to the loss of expert capacity. To address this correctable component, we propose Layer-wise Distribution Alignment (LDA), a lightweight inference-time correction that uses layer-wise calibration statistics to align reduced-routing representations with the default configuration. Across multiple SMoE LLMs, benchmarks, and routing strategies, LDA recovers much of the performance lost induced by the distributional shift under reduced routing while preserving sparse-inference efficiency with negligible overhead.

Authors 3

Dohyeon Kim, Bedionita Soro, Sung Ju Hwang

Specification

Official page

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

Arxiv announce type
cross

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

arXiv id
2609.09241

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

Categories
cs.LG, cs.AI, cs.CL

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

PDF

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

Primary category
cs.LG

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

Published
11 Sept 2026

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

Each value shows its source, tier and observation time. Conflicting claims are kept side by side and flagged — never averaged. How AI Atlas records facts →

Provenance

Attributed facts

9

Source tiers

T19

Freshest observation

50 min ago

Conflicts

None