@inproceedings{qing-etal-2024-alphalora,
    title = "{A}lpha{L}o{RA}: Assigning {L}o{RA} Experts Based on Layer Training Quality",
    author = "Qing, Peijun  and
      Gao, Chongyang  and
      Zhou, Yefan  and
      Diao, Xingjian  and
      Yang, Yaoqing  and
      Vosoughi, Soroush",
    editor = "Al-Onaizan, Yaser  and
      Bansal, Mohit  and
      Chen, Yun-Nung",
    booktitle = "Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2024",
    address = "Miami, Florida, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/author-page-patrick-haller/2024.emnlp-main.1141/",
    doi = "10.18653/v1/2024.emnlp-main.1141",
    pages = "20511--20523",
    abstract = "Parameter-efficient fine-tuning methods, such as Low-Rank Adaptation (LoRA), are known to enhance training efficiency in Large Language Models (LLMs). Due to the limited parameters of LoRA, recent studies seek to combine LoRA with Mixture-of-Experts (MoE) to boost performance across various tasks. However, inspired by the observed redundancy in traditional MoE structures, prior studies find that LoRA experts within the MoE architecture also exhibit redundancy, suggesting a need to vary the allocation of LoRA experts across different layers. In this paper, we leverage Heavy-Tailed Self-Regularization (HT-SR) Theory to design a fine-grained allocation strategy. Our analysis reveals that the number of experts per layer correlates with layer training quality, which exhibits significant variability across layers. Based on this, we introduce AlphaLoRA, a theoretically principled and training-free method for allocating LoRA experts to reduce redundancy further. Experiments on three models across ten language processing and reasoning benchmarks demonstrate that AlphaLoRA achieves comparable or superior performance over all baselines. Our code is available at https://github.com/morelife2017/alphalora."
}