通过谱截断与重缩放，解决多模型融合性能下降问题。
STAR: Spectral Truncation and Rescale for Model Merging
- 在谱空间截断小分量，减少融合冲突
- 12个Flan-T5模型融合时提升4.2%性能
- 无需额外训练数据，对超参不敏感
模型融合可在不进行微调的情况下，从多个预训练模型中高效构建多任务模型，已在自然语言处理等领域受到关注。尽管效率高，但随着融合模型数量增加，任务性能通常会下降。本文提出谱截断与重缩放（STAR），通过截断各模型谱空间中的小成分来缓解融合冲突，并采用自动参数重缩放方案保持原矩阵的核范数。STAR无需在原始训练数据上进行额外推理，对超参数选择具有鲁棒性。我们在多种NLP任务上进行了广泛的模型融合实验，结果表明，STAR在不同模型规模下均表现稳健，在将12个Flan-T5模型融合时，性能优于基线4.2%。代码已公开于https://github.com/IBM/STAR。
原文摘要 · Abstract (English)
Model merging is an efficient way of obtaining a multi-task model from several pretrained models without further fine-tuning, and it has gained attention in various domains, including natural language processing (NLP). Despite the efficiency, a key challenge in model merging is the seemingly inevitable decrease in task performance as the number of models increases. In this paper, we propose $\mathbf{S}$pectral $\mathbf{T}$runcation $\mathbf{A}$nd $\mathbf{R}$escale (STAR) that aims at mitigating ``merging conflicts'' by truncating small components in the respective spectral spaces, which is followed by an automatic parameter rescaling scheme to retain the nuclear norm of the original matrix. STAR requires no additional inference on original training data and is robust to hyperparamater choice. We demonstrate the effectiveness of STAR through extensive model merging cases on diverse NLP tasks. Specifically, STAR works robustly across varying model sizes, and can outperform baselines by 4.2$\%$ when merging 12 models on Flan-T5. Our code is publicly available at https://github.com/IBM/STAR.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。