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

TL;DR

DECO introduces a sparse Mixture-of-Experts architecture that achieves dense-transformer performance on end-side devices with significantly reduced computational and storage requirements.

Contribution

The paper proposes DECO, a novel sparse MoE architecture with adaptive routing, a new activation function, and simplified expert design, matching dense model performance efficiently.

Findings

DECO activates only 20% of experts while maintaining dense performance.

DECO outperforms existing MoE baselines in experiments.

Achieves a 2.93× speedup on Jetson AGX Orin with specialized kernel.

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.