2025/02/18 by M. Lv, Zhenpeng Su, Lv, Minxuan +23 · 1 citation
Computer Science · #Computation and Language (cs.CL) #Data Mining Algorithms and Applications #FOS: Computer and information sciences #Machine Learning and Algorithms #Machine Learning and Data Classification
paper · pdf · doi:10.48550/arxiv.2502.12455
openalex publication_date 2025/02/18 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
As large language models continue to scale, computational costs and resource consumption have emerged as significant challenges. While existing sparsification methods like pruning reduce computational overhead, they risk losing model knowledge through parameter removal. This paper proposes DSMoE (Dynamic Sparse Mixture-of-Experts), a novel approach that achieves sparsification by partitioning pre-trained FFN layers into computational blocks. We implement adaptive expert routing using sigmoid activation and straight-through estimators, enabling tokens to flexibly access different aspects of model knowledge based on input complexity. Additionally, we introduce a sparsity loss term to balance performance and computational efficiency. Extensive experiments on LLaMA models demonstrate that under equivalent computational constraints, DSMoE achieves superior performance compared to existing pruning and MoE approaches across language modeling and downstream tasks, particularly excelling in generation tasks. Analysis reveals that DSMoE learns distinctive layerwise activation patterns, providing new insights for future MoE architecture design.