DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices

• URL: http://arxiv.org/abs/2605.10933v3
• Date: Wed, 20 May 2026 17:26:14 GMT
• Title: DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices
• Abstract summary: DECO is a sparse MoE architecture designed to match the performance of dense Transformers under identical total parameter budgets and training tokens.<n>Our specialized acceleration kernel delivers a 2.93$times$ speedup on Jetson AGX Orin compared with dense inference.
• Score: 27.533004971063175
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: While Mixture-of-Experts (MoE) scales model capacity without proportionally increasing computation, its massive total parameter footprint creates significant storage and memory-access bottlenecks, which hinder efficient end-side deployment that simultaneously requires high performance, low computational cost, and small storage overhead. To achieve these properties, we present DECO, a sparse MoE architecture designed to match the performance of dense Transformers under identical total parameter budgets and training tokens. DECO utilizes the differentiable and flexible ReLU-based routing enhanced by learnable expert-wise scaling, which adaptively balances the contributions of routed and shared experts. Furthermore, we introduce NormSiLU, an activation function that normalizes inputs prior to SiLU operators, producing a more stable trend of routed-expert activation ratio and a higher intrinsic sparsity level. We also identify an empirical advantage in using non-gated MLP experts with ReLU-based routing, indicating the possibility of MoE architecture simplification. Experiments demonstrate that DECO, activating only 20% of routed experts, matches dense performance and outperforms established MoE baselines. Our specialized acceleration kernel delivers a 2.93$\times$ speedup on Jetson AGX Orin compared with dense inference. Code and checkpoints are available at https://github.com/thunlp/DECO.

Related papers

• Distribution-Consistent Inference for Dynamic Sparse Mixture-of-Experts [48.294824672909876]
  Inference-time dynamic top-$k$ routing can reduce computation without retraining.<n>We show that reducing the number of activated experts consistently increases the RMS scale and variance of SMoE outputs.<n>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.
  arXiv  Detail & Related papers  (2026-09-08T06:23:07Z)
• Accelerating Unified Multimodal Models with Core-Expansion Routing and Unified Computation Scheduling [61.28599393707943]
  Unified multimodal models jointly support understanding and generation, but incur substantial redundant computation across tokens, layers, and generation timesteps.<n>We propose a task-shared core scorer and progress-conditioned generation expansions, optimized through generation decomposition and cross-task core alignment.<n> Experiments on two representative UMM architectures demonstrate consistent quality--efficiency improvements across both tasks, retaining 98.03% of dense understanding performance with a 1.93times$ end-to-end inference speedup.
  arXiv  Detail & Related papers  (2026-08-29T14:27:55Z)
• Tying the Loop -- Tied Expert Layers in Mixture-of-Experts Language Models [50.34585122347149]
  We introduce Expert Tying, an architectural modification that shares expert parameters across consecutive transformer layers.<n>We evaluate this approach across common, state-of-the-art architectures, including OLMoE, Qwen3, and DeepSeek-style MoEs.
  arXiv  Detail & Related papers  (2026-06-15T15:08:09Z)
• OScaR: The Occam's Razor for Extreme KV Cache Quantization in LLMs and Beyond [50.440302567029654]
  Multi-modal intelligence has made the Key-Value cache a dominant memory bottleneck for efficient deployment.<n>In this work, we revisit the inherent limitations of the per-channel quantization paradigm.<n>We propose OScaR, an accurate and lightweight KV cache compression framework for X-LLMs.
  arXiv  Detail & Related papers  (2026-05-19T10:53:03Z)
• BEAM: Binary Expert Activation Masking for Dynamic Routing in MoE [18.898045833395095]
  Mixture-of-Experts (MoE) architectures activate only a subset of experts per token.<n>Existing acceleration methods require costly retraining with architectural changes or suffer from severe performance drop at high sparsity.<n>We propose BEAM, a novel method that learns token-adaptive expert selection via trainable binary masks.
  arXiv  Detail & Related papers  (2026-05-14T06:33:41Z)
• Algorithm and Hardware Co-Design for Efficient Complex-Valued Uncertainty Estimation [10.567064384827196]
  Complex-Valued Neural Networks (CVNNs) have advantages in handling tasks that involve complex numbers.<n>Existing CVNNs are unable to quantify predictive uncertainty.<n>We propose, for the first time, dropout-based Bayesian Complex-Valued Neural Networks (BayesCVNNs) to enable uncertainty quantification.
  arXiv  Detail & Related papers  (2026-04-21T21:06:00Z)
• Ge$^\text{2}$mS-T: Multi-Dimensional Grouping for Ultra-High Energy Efficiency in Spiking Transformer [84.8831358775386]
  Spiking Neural Networks (SNNs) offer superior energy efficiency over Artificial Neural Networks (ANNs)<n>Existing paradigms including ANN-SNN Conversion and Spatial-Temporal Backpropagation (STBP) suffer from inherent limitations.<n>We propose Ge$text2$mS-T, a novel architecture implementing grouped computation across temporal, spatial and network structure dimensions.
  arXiv  Detail & Related papers  (2026-04-10T02:58:46Z)
• AdaFuse: Accelerating Dynamic Adapter Inference via Token-Level Pre-Gating and Fused Kernel Optimization [84.25316984309725]
  Integration of dynamic, sparse structures with parameter-efficient adapters (e.g., LoRA) is a powerful technique for enhancing Large Language Models (LLMs)<n>Despite minimal increases in computational load, the computation latency often skyrockets, leading to decoding speeds slowing by over 2.5 times.<n>AdaFuse is a framework built on a tight co-design between the algorithm and the underlying hardware system to enable efficient dynamic adapter execution.
  arXiv  Detail & Related papers  (2026-03-12T12:46:42Z)
• MoE-SpAc: Efficient MoE Inference Based on Speculative Activation Utility in Heterogeneous Edge Scenarios [12.409831039709886]
  Mixture-of-Experts (MoE) models enable scalable performance but face severe memory constraints on edge devices.<n>We introduce MoE-SpAc, an MoE inference framework that integrates a Speculative Utility Estor to track expert demand.<n>Experiments on seven benchmarks demonstrate that MoE-SpAc achieves a 42% improvement in TPS over the SOTA SD-based baseline.
  arXiv  Detail & Related papers  (2026-02-12T04:28:00Z)
• SERE: Similarity-based Expert Re-routing for Efficient Batch Decoding in MoE Models [19.56443760368644]
  We present SERE, a Similarity-based Expert Re-routing method for Efficient batch decoding in MoE models.<n>SERE dynamically reduces the number of active experts by re-routing tokens from secondary experts to their most similar primary counterparts.<n>SERE achieves up to 2.0x speedup with minimal quality loss, providing a practical solution for cost-efficient and latency-sensitive large-scale MoE deployment.
  arXiv  Detail & Related papers  (2026-02-07T16:51:16Z)
• HOBBIT: A Mixed Precision Expert Offloading System for Fast MoE Inference [54.40808356999408]
  We present HOBBIT, a mixed precision expert offloading system to enable flexible and efficient MoE inference. Our key insight is that dynamically replacing less critical cache-miss experts with low precision versions can substantially reduce expert-loading latency. HOBBIT achieves up to a 9.93x speedup in decoding compared to state-of-the-art MoE offloading systems.
  arXiv  Detail & Related papers  (2024-11-03T04:25:46Z)
• Read-ME: Refactorizing LLMs as Router-Decoupled Mixture of Experts with System Co-Design [59.00758127310582]
  We propose a novel framework Read-ME that transforms pre-trained dense LLMs into smaller MoE models. Our approach employs activation sparsity to extract experts. Read-ME outperforms other popular open-source dense models of similar scales.
  arXiv  Detail & Related papers  (2024-10-24T19:48:51Z)
• MoE++: Accelerating Mixture-of-Experts Methods with Zero-Computation Experts [63.67734699877724]
  MoE++ is a general and heterogeneous MoE framework that integrates both Feed-Forward Network(FFN) and zero-computation experts. MoE++ achieves better performance while delivering 1.1-2.1x expert forward throughput compared to a vanilla MoE model of the same size.
  arXiv  Detail & Related papers  (2024-10-09T18:01:27Z)
• Layerwise Recurrent Router for Mixture-of-Experts [42.36093735411238]
  Mixture-of-Experts (MoE) architecture stands out for its ability to scale model size without significantly increasing training costs.<n>Current MoE models often display parameter inefficiency.<n>We introduce the Layerwise Recurrent Router for Mixture-of-Experts (RMoE)
  arXiv  Detail & Related papers  (2024-08-13T10:25:13Z)
• Efficient Heterogeneous Large Language Model Decoding with Model-Attention Disaggregation [15.35494431928751]
  Transformer-based large language models (LLMs) exhibit impressive performance in generative tasks but also introduce significant challenges in real-world serving.<n>We introduce model-attention disaggregation to enhance the efficiency of LLM decoding.<n>We develop and deploy Lamina, an LLM inference system that incorporates model-attention disaggregation in a distributed heterogeneous cluster.
  arXiv  Detail & Related papers  (2024-05-03T02:15:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.