English

ASLoRA: Adaptive Sharing Low-Rank Adaptation Across Layers

Computation and Language 2024-12-17 v2

Abstract

As large language models (LLMs) grow in size, traditional full fine-tuning becomes increasingly impractical due to its high computational and storage costs. Although popular parameter-efficient fine-tuning methods, such as LoRA, have significantly reduced the number of tunable parameters, there is still room for further optimization. In this work, we propose ASLoRA, a cross-layer parameter-sharing strategy combining global sharing with partial adaptive sharing. Specifically, we share the low-rank matrix A across all layers and adaptively merge matrix B during training. This sharing mechanism not only mitigates overfitting effectively but also captures inter-layer dependencies, significantly enhancing the model's representational capability. We conduct extensive experiments on various NLP tasks, showing that ASLoRA outperforms LoRA while using less than 25% of the parameters, highlighting its flexibility and superior parameter efficiency. Furthermore, in-depth analyses of the adaptive sharing strategy confirm its significant advantages in enhancing both model flexibility and task adaptability.

Keywords

Cite

@article{arxiv.2412.10135,
  title  = {ASLoRA: Adaptive Sharing Low-Rank Adaptation Across Layers},
  author = {Junyan Hu and Xue Xiao and Mengqi Zhang and Yao Chen and Zhaochun Ren and Zhumin Chen and Pengjie Ren},
  journal= {arXiv preprint arXiv:2412.10135},
  year   = {2024}
}
R2 v1 2026-07-22T20:52:06.061Z