arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2507.23751cs.AIcs.CL
收藏

CoT-Self-Instruct:为推理与非推理任务构建高质量合成提示

CoT-Self-Instruct: Building high-quality synthetic prompts for reasoning and non-reasoning tasks

FAIR at Meta(Meta的FAIR)

Ping Yu, Jack Lanchantin, Tianlu Wang, Weizhe Yuan, Olga Golovneva, Ilia Kulikov, Sainbayar Sukhbaatar, Jason Weston, Jing Xu

AI总结:

该研究提出CoT-Self-Instruct合成数据生成方法,通过思维链推理规划加自动指标过滤生成高质量数据,在可验证推理与不可验证指令遵循任务的多个基准上均优于现有方案。

摘要:

我们提出CoT-Self-Instruct,这是一种合成数据生成方法,它指示大语言模型(LLM)先基于给定的种子任务,通过思维链(Chain-of-Thought, CoT)进行推理与规划,再生成一个质量和复杂度相近的新合成样本。随后设置过滤步骤,用自动指标筛选高质量数据,再将其用于LLM训练。在可验证推理任务中,我们的合成数据在MATH500、AMC23、AIME24和GPQA-Diamond上评估时,表现显著优于s1k、OpenMathReasoning等现有训练数据集。在不可验证的指令遵循任务中,我们的方法在AlpacaEval 2.0和Arena-Hard基准上的表现超过了人工训练数据与标准Self-Instruct训练数据的性能。

英文摘要:

We propose CoT-Self-Instruct, a synthetic data generation method that instructs LLMs to first reason and plan via Chain-of-Thought (CoT) based on given seed tasks, and then generate a new synthetic example of similar quality and complexity. This is followed by a filtering step to select high-quality data using automatic metrics, which are then used for LLM training. In verifiable reasoning, our synthetic data significantly outperforms existing training datasets, such as s1k and OpenMathReasoning, when evaluated on MATH500, AMC23, AIME24, and GPQA-Diamond. For non-verifiable instruction-following tasks, our method surpasses the performance of both human and standard Self-Instruct training data on the AlpacaEval 2.0 and Arena-Hard 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。

↑