arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2307.12950cs.CLcs.AI

RLCD:基于对比蒸馏的强化学习用于语言模型对齐

RLCD: Reinforcement Learning from Contrastive Distillation for Language Model Alignment

Meta AI;UC Berkeley(加州大学伯克利分校);UCLA(加州大学洛杉矶分校)

Kevin Yang, Dan Klein, Asli Celikyilmaz, Nanyun Peng, Yuandong Tian

AI总结:

RLCD通过对比蒸馏生成偏好对训练偏好模型,结合强化学习对齐语言模型,无需人类反馈,在无害性、帮助性和故事大纲生成任务上超越RLAIF和上下文蒸馏基线。

摘要:

我们提出了对比蒸馏强化学习(RLCD),一种无需人类反馈即可使语言模型遵循自然语言表达的原则(例如,更加无害)的对齐方法。RLCD通过两个对比模型输出创建偏好对,一个使用旨在鼓励遵循给定原则的正向提示,另一个使用旨在鼓励违反原则的负向提示。使用两种不同的提示使模型输出平均而言更具区分度,从而在没有人类标注的情况下产生更干净的偏好标签。然后,我们使用这些偏好对训练一个偏好模型,该模型进而通过强化学习用于改进基础未对齐的语言模型。实验上,RLCD在三个不同的对齐任务——无害性、有帮助性和故事大纲生成——中,以及在模拟偏好数据时使用7B和30B两种模型规模,均优于RLAIF(Bai等人,2022b)和上下文蒸馏(Huang等人,2022)基线。

英文摘要:

We propose Reinforcement Learning from Contrastive Distillation (RLCD), a method for aligning language models to follow principles expressed in natural language (e.g., to be more harmless) without using human feedback. RLCD creates preference pairs from two contrasting model outputs, one using a positive prompt designed to encourage following the given principles, and one using a negative prompt designed to encourage violating them. Using two different prompts causes model outputs to be more differentiated on average, resulting in cleaner preference labels in the absence of human annotations. We then use the preference pairs to train a preference model, which is in turn used to improve a base unaligned language model via reinforcement learning. Empirically, RLCD outperforms RLAIF (Bai et al., 2022b) and context distillation (Huang et al., 2022) baselines across three diverse alignment tasks--harmlessness, helpfulness, and story outline generation--and when using both 7B and 30B model scales for simulating preference data.

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。

↑