@inproceedings{tian-etal-2026-grass,
    title = "{GRASS}: Gradient-based Adaptive Layer-wise Importance Sampling for Memory-efficient Large Language Model Fine-tuning",
    author = "Tian, Kaiyuan  and
      Qiao, Linbo  and
      Tang, Yu  and
      Jiang, Gongqingjian  and
      Liu, Baihui  and
      Gao, Yifu  and
      Su, Xialin  and
      Li, Dongsheng",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.findings-acl.475/",
    pages = "9777--9790",
    ISBN = "979-8-89176-395-1",
    abstract = "Full-parameter fine-tuning of large language models is constrained by substantial GPU memory demands. Low-rank adaptation methods mitigate this challenge by updating only a subset of parameters. However, these approaches often limit model expressiveness and yield lower performance than full-parameter fine-tuning. Layer-wise fine-tuning methods have emerged as an alternative, enabling memory-efficient training through static layer importance sampling strategies. However, these methods overlook variations in layer importance across tasks and training stages, resulting in suboptimal performance on downstream tasks. To address these limitations, we propose GRASS, a gradient-based adaptive layer-wise importance sampling framework. GRASS utilizes mean gradient norms as a task-aware and training-stage-aware metric for estimating layer importance. Furthermore, GRASS adaptively adjusts layer sampling probabilities through an adaptive training strategy. We also introduce a layer-wise optimizer state offloading mechanism to further reduce memory usage while maintaining comparable training throughput. Extensive experiments across multiple models and benchmarks demonstrate that GRASS consistently outperforms state-of-the-art methods, achieving an average accuracy improvement of up to 4.38 points and reducing memory usage by up to 19.97{\%}."
}