arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2403.03329cs.CL
收藏

面向大语言模型遗忘的护栏基线

Guardrail Baselines for Unlearning in LLMs

Carnegie Mellon University(卡内基梅隆大学)

Pratiksha Thaker, Yash Maurya, Shengyuan Hu, Zhiwei Steven Wu, Virginia Smith

AI总结:

本文针对大语言模型遗忘成本高的问题,提出以提示和过滤等轻量护栏作为基线,证明其效果可与微调相当,并指出需改进评估指标以区分二者。

摘要:

近期研究表明,微调是从大语言模型中“遗忘”概念的一种有前景的方法。然而,微调可能代价高昂,因为它既需要生成一组示例,又需要运行多轮微调来更新模型。在本工作中,我们表明,诸如提示和过滤等简单的基于护栏的方法能够取得与微调相当的遗忘结果。我们建议研究人员在评估计算更密集的微调方法的性能时,考察这些轻量级基线。虽然我们并不声称提示或过滤等方法是遗忘问题的通用解决方案,但我们的工作表明,需要能够更好地区分护栏与微调能力的评估指标,并强调了护栏暴露出当前指标和基准中可能存在的非预期行为的情形。

英文摘要:

Recent work has demonstrated that finetuning is a promising approach to 'unlearn' concepts from large language models. However, finetuning can be expensive, as it requires both generating a set of examples and running iterations of finetuning to update the model. In this work, we show that simple guardrail-based approaches such as prompting and filtering can achieve unlearning results comparable to finetuning. We recommend that researchers investigate these lightweight baselines when evaluating the performance of more computationally intensive finetuning methods. While we do not claim that methods such as prompting or filtering are universal solutions to the problem of unlearning, our work suggests the need for evaluation metrics that can better separate the power of guardrails vs. finetuning, and highlights scenarios where guardrails expose possible unintended behavior in existing metrics and benchmarks.

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。

↑