arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2407.06483cs.LGcs.CL
收藏

面向语言模型的可组合干预

Composable Interventions for Language Models

University of Virginia(弗吉尼亚大学);EleutherAI;University of Exeter(埃克塞特大学);Harvard Medical School(哈佛医学院);University of Oxford(牛津大学);Thomson-Reuters Foundational Research(汤森路透基础研究部);Mass General Brigham(麻省总医院布里格姆医疗系统);UNC Chapel Hill(北卡罗来纳大学教堂山分校)
Arinbjorn Kolbeinsson, Kyle O'Brien, Tianjin Huang, Shanghua Gao, Shiwei Liu, Jonathan Richard Schwarz, Anurag Vaidya, Faisal Mahmood, Marinka Zitnik, Tianlong … 展开作者

Arinbjorn Kolbeinsson, Kyle O'Brien, Tianjin Huang, Shanghua Gao, Shiwei Liu, Jonathan Richard Schwarz, Anurag Vaidya, Faisal Mahmood, Marinka Zitnik, Tianlong Chen, Thomas Hartvigsen

AI总结:

本文提出可组合干预框架与统一代码库,通过310种组合实验分析知识编辑、模型压缩和机器遗忘的交互,揭示顺序效应与现有指标不足。

摘要:

语言模型的测试时干预可以提高事实准确性、减少有害输出并提升模型效率,而无需付出高昂的再训练成本。但是,尽管涌现出大量新方法,不同类型的干预在很大程度上仍在独立发展。在实践中,必须对同一模型依次应用多种干预,然而我们缺乏研究干预之间如何相互作用的标准化方法。我们通过引入可组合干预来填补这一空白,这是一个研究在同一语言模型上使用多种干预所产生效果的框架,具备新的指标和统一的代码库。利用我们的框架,我们开展了大量实验,并组合了来自三个新兴干预类别——知识编辑、模型压缩和机器遗忘——的流行方法。我们从310种不同组合中得到的结果揭示了有意义的相互作用:压缩会妨碍编辑和遗忘,组合干预取决于其应用顺序,而流行的通用指标不足以评估可组合性。总体而言,我们的发现展示了可组合性方面的明显差距,表明需要新的多目标干预。我们的所有代码均已公开:https://github.com/hartvigsen-group/composable-interventions。

英文摘要:

Test-time interventions for language models can enhance factual accuracy, mitigate harmful outputs, and improve model efficiency without costly retraining. But despite a flood of new methods, different types of interventions are largely developing independently. In practice, multiple interventions must be applied sequentially to the same model, yet we lack standardized ways to study how interventions interact. We fill this gap by introducing composable interventions, a framework to study the effects of using multiple interventions on the same language models, featuring new metrics and a unified codebase. Using our framework, we conduct extensive experiments and compose popular methods from three emerging intervention categories -- Knowledge Editing, Model Compression, and Machine Unlearning. Our results from 310 different compositions uncover meaningful interactions: compression hinders editing and unlearning, composing interventions hinges on their order of application, and popular general-purpose metrics are inadequate for assessing composability. Taken together, our findings showcase clear gaps in composability, suggesting a need for new multi-objective interventions. All of our code is public: https://github.com/hartvigsen-group/composable-interventions.

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。

↑