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 2407.19262cs.CLcs.LG
收藏

理解大型语言模型中的记忆化:动态、影响因素与启示

Understanding Memorisation in LLMs: Dynamics, Influencing Factors, and Implications

MPI-SWS(马克斯·普朗克软件系统研究所);IIIT Delhi(德里印度信息技术学院);Boston University(波士顿大学)

Till Speicher, Mohammad Aflah Khan, Qinyuan Wu, Vedant Nanda, Soumi Das, Bishwamittra Ghosh, Krishna P. Gummadi, Evimaria Terzi

AI总结:

本文提出基于重复暴露随机字符串的实验框架,研究大型语言模型记忆化动态,发现跨模型家族的一致阶段、影响记忆的因素及顺序暴露的显著影响,为LLMs的可靠性与隐私提供重要启示。

摘要:

了解大型语言模型(LLMs)是否以及在多大程度上记忆了训练数据,对于其输出的可靠性和训练数据的隐私具有重要启示。为了干净地测量并将记忆化与其他现象(例如上下文学习)解耦,我们创建了一个基于向LLMs重复暴露随机字符串的实验框架。我们的框架使我们能够更好地理解动态,即在向模型重复暴露随机字符串时模型的行为。使用我们的框架,我们得出了几个惊人的观察结果:(a)我们发现了跨模型家族(Pythia、Phi和Llama2)动态的一致阶段,(b)我们识别了使某些字符串比其他字符串更容易记忆的因素,以及(c)我们识别了局部前缀和全局上下文在记忆化中的作用。我们还展示了,对不同随机字符串的顺序暴露对记忆化有显著影响。我们的结果往往出人意料,在LLMs的研究和使用中具有重要的下游启示。

英文摘要:

Understanding whether and to what extent large language models (LLMs) have memorised training data has important implications for the reliability of their output and the privacy of their training data. In order to cleanly measure and disentangle memorisation from other phenomena (e.g. in-context learning), we create an experimental framework that is based on repeatedly exposing LLMs to random strings. Our framework allows us to better understand the dynamics, i.e., the behaviour of the model, when repeatedly exposing it to random strings. Using our framework, we make several striking observations: (a) we find consistent phases of the dynamics across families of models (Pythia, Phi and Llama2), (b) we identify factors that make some strings easier to memorise than others, and (c) we identify the role of local prefixes and global context in memorisation. We also show that sequential exposition to different random strings has a significant effect on memorisation. Our results, often surprising, have significant downstream implications in the study and usage of LLMs.

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。

↑