arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.04937cs.LGcs.AIcs.CL
收藏

线性模型合并解锁了简单且可扩展的多模态数据混合优化

Linear Model Merging Unlocks Simple and Scalable Multimodal Data Mixture Optimization

University of Trento(特伦托大学);Fondazione Bruno Kessler(布鲁诺·凯斯勒基金会)

Davide Berasi, Matteo Farina, Massimiliano Mancini, Elisa Ricci

AI总结:

本文提出通过模型合并高效估算多模态数据混合效果,结合领域专家训练与参数空间组合,实现可扩展的混合优化方法。

摘要:

选择最佳的数据混合对于成功进行多模态大语言模型的监督微调(SFT)至关重要。然而,确定在多个领域特定数据集上最佳的混合权重仍是一个重大瓶颈,因为组合搜索空间和单次训练运行的高成本。这就是所谓的数据混合优化(DMO)问题。另一方面,模型合并通过参数插值将领域特定专家统一起来。这种策略是高效的,因为它只需要每个领域一次训练运行,但往往导致次优模型。在本工作中,我们结合了两者的优势,研究模型合并作为估算不同数据混合性能的高效策略。我们训练领域特定的多模态专家,并评估其加权参数空间组合以估计相应数据混合的有效性。我们在14个多模态基准上进行了广泛的实验,并实证表明,合并的代理模型与实际数据混合训练的模型具有高排名相关性。这将最佳混合的搜索与资源密集型的训练过程解耦,从而提供了一种可扩展且高效的策略,以导航复杂的混合权重景观。代码可在https://github.com/BerasiDavide/mLLMs_merging_4_DMO上公开获取。

英文摘要:

Selecting the best data mixture is critical for successful Supervised Fine-Tuning (SFT) of Multimodal Large Language Models. However, determining the optimal mixture weights across multiple domain-specific datasets remains a significant bottleneck due to the combinatorial search space and the high cost associated with even a single training run. This is the so-called Data Mixture Optimization (DMO) problem. On the other hand, model merging unifies domain-specific experts through parameter interpolation. This strategy is efficient, as it requires only a single training run per domain, yet it often leads to suboptimal models. In this work, we take the best of both worlds, studying model merging as an efficient strategy for estimating the performance of different data mixtures. We train domain-specific multimodal experts and evaluate their weighted parameter-space combinations to estimate the efficacy of corresponding data mixtures. We conduct extensive experiments on 14 multimodal benchmarks, and empirically demonstrate that the merged proxy models consistently select near-optimal mixtures. This decouples the search for optimal mixtures from the resource-intensive training process, thereby providing a scalable and efficient strategy for navigating the complex landscape of mixture weights. Code is publicly available at https://github.com/BerasiDavide/mLLMs_merging_4_DMO.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑