@inproceedings{fu-etal-2024-disperse,
    title = "Disperse-Then-Merge: Pushing the Limits of Instruction Tuning via Alignment Tax Reduction",
    author = "Fu, Tingchen  and
      Cai, Deng  and
      Liu, Lemao  and
      Shi, Shuming  and
      Yan, Rui",
    editor = "Ku, Lun-Wei  and
      Martins, Andre  and
      Srikumar, Vivek",
    booktitle = "Findings of the Association for Computational Linguistics: ACL 2024",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-emnlp/2024.findings-acl.175/",
    doi = "10.18653/v1/2024.findings-acl.175",
    pages = "2967--2985",
    abstract = "Supervised fine-tuning (SFT) on instruction-following corpus is a crucial approach toward the alignment of large language models (LLMs). However, the performance of LLMs on standard knowledge and reasoning benchmarks tends to suffer from deterioration at the latter stage of the SFT process, echoing the phenomenon of alignment tax. Through our pilot study, we put a hypothesis that the data biases are probably one cause behind the phenomenon. To address the issue, we introduce a simple disperse-then-merge framework. To be concrete, we disperse the instruction-following data into portions and then train multiple sub-models using different data portions. Lastly, we merge multiple models into a single one via model merging techniques. Despite its simplicity, our framework outperforms various sophisticated methods such as data curation and training regularization on a series of standard knowledge and reasoning benchmarks."
}