@inproceedings{lv-etal-2025-dsmoe,
    title = "{DSM}o{E}: Matrix-Partitioned Experts with Dynamic Routing for Computation-Efficient Dense {LLM}s",
    author = "Lv, Minxuan  and
      Su, Zhenpeng  and
      Pan, Leiyu  and
      Xiong, Yizhe  and
      Lin, Zijia  and
      Chen, Hui  and
      Zhou, Wei  and
      Han, Jungong  and
      Ding, Guiguang  and
      Ou, Wenwu  and
      Zhang, Di  and
      Gai, Kun  and
      Hu, Songlin",
    editor = "Christodoulopoulos, Christos  and
      Chakraborty, Tanmoy  and
      Rose, Carolyn  and
      Peng, Violet",
    booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2025",
    address = "Suzhou, China",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-emnlp/2025.emnlp-main.997/",
    pages = "19722--19733",
    ISBN = "979-8-89176-332-6",
    abstract = "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."
}