@inproceedings{zhu-etal-2025-dynamic,
    title = "Dynamic Data Mixing Maximizes Instruction Tuning for Mixture-of-Experts",
    author = "Zhu, Tong  and
      Dong, Daize  and
      Qu, Xiaoye  and
      Ruan, Jiacheng  and
      Chen, Wenliang  and
      Cheng, Yu",
    editor = "Chiruzzo, Luis  and
      Ritter, Alan  and
      Wang, Lu",
    booktitle = "Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)",
    month = apr,
    year = "2025",
    address = "Albuquerque, New Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/moar-dois/2025.naacl-long.80/",
    doi = "10.18653/v1/2025.naacl-long.80",
    pages = "1663--1677",
    ISBN = "979-8-89176-189-6",
    abstract = "Mixture-of-Experts (MoE) models have shown remarkable capability in instruction tuning, especially when the number of tasks scales. However, previous methods simply merge all training tasks (e.g. creative writing, coding, and mathematics) and apply fixed sampling weights, without considering the importance of different tasks as the model training state changes. In this way, the most helpful data cannot be effectively distinguished, leading to suboptimal model performance. To reduce the potential redundancies of datasets, we make the first attempt and propose a novel dynamic data mixture for MoE instruction tuning. Specifically, inspired by MoE{'}s token routing preference, we build dataset-level representations and then capture the subtle differences among datasets. Finally, we propose to dynamically adjust the sampling weight of datasets by their inter-redundancies, thus maximizing global performance under a limited training budget. The experimental results on two MoE models demonstrate the effectiveness of our approach on both downstream knowledge {\&} reasoning tasks and open-ended queries."
}