用结构化数据生成推理题，提升大模型长文本推理能力
$π^2$: Structure-Originated Reasoning Data Improves Long-Context Reasoning Ability of Large Language Models
- 从维基百科表拓展出多跳推理问题，自动验证答案
- 在4个基准上平均准确率提升4.3%至2.7%，支持自蒸馏
- 适合研究长上下文推理或数据构建的学者
本文提出π²，一种从初始结构化数据中构建高质量推理数据的流程，以提升大语言模型的长上下文推理能力。方法包括：1）从维基百科提取并扩展表格；2）基于表格和相关上下文生成真实且需多步推理的问题，通过双路径代码执行自动确定与验证答案；3）将逐步推理过程反向翻译为问答对的解题步骤，置于真实网页搜索上下文中。在π²数据上对GPT-OSS-20B和Qwen3-4B-Instruct-2507进行监督微调，在四个长上下文推理基准及自建的π²-Bench上均实现一致提升，平均准确率分别提高+4.3%和+2.7%。值得注意的是，π²支持自蒸馏，使GPT-OSS-20B仅用自身推理轨迹就实现了+4.4%的性能提升，证明其数据价值。代码、数据与模型均已开源。
原文摘要 · Abstract (English)
We study a pipeline that curates reasoning data from initial structured data for improving long-context reasoning in large language models (LLMs). Our approach, $π^2$, constructs high-quality reasoning data through rigorous QA curation: 1) extracting and expanding tables from Wikipedia, 2) from the collected tables and relevant context, generating realistic and multi-hop analytical reasoning questions whose answers are automatically determined and verified through dual-path code execution, and 3) back-translating step-by-step structured reasoning traces as solutions of QA pairs given realistic web-search context. Supervised fine-tuning with \textsc{\small{gpt-oss-20b}} and \textsc{\small{Qwen3-4B-Instruct-2507}} on $π^2$ yields consistent improvements across four long-context reasoning benchmarks and our alike $π^2$-Bench, with average absolute accuracy gains of +4.3% and +2.7% respectively. Notably, our dataset facilitates self-distillation, where \textsc{\small{gpt-oss-20b}} even improves its average performance by +4.4% with its own reasoning traces, demonstrating $π^2$'s usefulness. Our code, data, and models are open-source at https://github.com/vt-pi-squared/pi-squared.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。