Focusing on What Matters: Saliency-Harnessing Accurate Routing for Diffusion MoE

Hugging Face Daily Papers Papers

Summary

SharpMoE is a post-training framework that improves routing in diffusion mixture-of-experts models by using clean latent features to identify salient tokens and a trajectory routing loss to allocate compute precisely, achieving state-of-the-art visual generation.

Mixture-of-Experts (MoE) architectures have emerged as a powerful paradigm for scaling diffusion models in visual generation. Recent advancements have focused on adaptively allocating computational resources across diverse tokens to improve efficiency and performance. However, we identify a routing assignment problem in existing diffusion MoE frameworks: the router fails to accurately allocate more computational resources to salient tokens. Our analysis attributes this failure to the router's reliance on noise-corrupted latent features throughout the denoising process. Such stochastic noise obscures the critical structural and textural information, thereby preventing the router from effectively distinguishing salient tokens. To address this, we propose SharpMoE, a post-training framework with a saliency-harnessing accurate routing mechanism, which utilizes clean latent features as a noise-free guidance signal for routing. By bypassing the noise-distorted inputs, SharpMoE provides the router with clear saliency guidance, enabling the identification of salient tokens even in high-noise stages. Furthermore, we introduce a trajectory routing loss to constrain the compute allocation throughout the multi-step denoising trajectory, ensuring precise resource allocation along the generation rollout. Extensive experiments demonstrate that SharpMoE serves as a versatile, plug-and-play solution that further enhances the pretrained, converged MoE models, achieving state-of-the-art performance in visual generation.
Original Article
View Cached Full Text

Cached at: 06/30/26, 03:33 AM

Paper page - Focusing on What Matters: Saliency-Harnessing Accurate Routing for Diffusion MoE

Source: https://huggingface.co/papers/2606.26938

Abstract

SharpMoE addresses routing inefficiencies in diffusion models by using clean latent features to guide salient token identification and employs trajectory routing loss for precise compute allocation during multi-step denoising.

Mixture-of-Experts(MoE) architectures have emerged as a powerful paradigm for scalingdiffusion modelsinvisual generation. Recent advancements have focused on adaptively allocating computational resources across diverse tokens to improve efficiency and performance. However, we identify arouting assignmentproblem in existing diffusion MoE frameworks: the router fails to accurately allocate more computational resources tosalient tokens. Our analysis attributes this failure to the router’s reliance on noise-corruptedlatent featuresthroughout thedenoising process. Such stochastic noise obscures the critical structural and textural information, thereby preventing the router from effectively distinguishingsalient tokens. To address this, we propose SharpMoE, apost-training frameworkwith a saliency-harnessing accurate routing mechanism, which utilizes cleanlatent featuresas a noise-free guidance signal for routing. By bypassing the noise-distorted inputs, SharpMoE provides the router with clear saliency guidance, enabling the identification ofsalient tokenseven in high-noise stages. Furthermore, we introduce atrajectory routing lossto constrain thecompute allocationthroughout the multi-step denoising trajectory, ensuring precise resource allocation along the generation rollout. Extensive experiments demonstrate that SharpMoE serves as a versatile, plug-and-play solution that further enhances the pretrained, converged MoE models, achieving state-of-the-art performance invisual generation.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2606\.26938

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.26938 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.26938 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.26938 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Sticky Routing: Training MoE Models for Memory-Efficient Inference

arXiv cs.LG

StickyMoE proposes a differentiable routing consistency loss that encourages adjacent tokens to activate the same experts in MoE models, reducing expert-swapping overhead and cache misses during inference on edge devices by up to 3.92× while improving perplexity.