$\pi^2$: Structure-Originated Reasoning Data Improves Long-Context Reasoning Ability of Large Language Models

TL;DR

The paper introduces $ ext{ extpi}^2$, a data pipeline that enhances long-context reasoning in large language models by generating high-quality, multi-hop reasoning data from structured sources like Wikipedia.

Contribution

It presents a novel data curation method that creates realistic reasoning questions with verified answers, improving LLM performance on reasoning benchmarks.

Findings

Supervised fine-tuning with $ ext{ extpi}^2$ data improves LLM accuracy by up to 4.3%.

Self-distillation with $ ext{ extpi}^2$ data boosts GPT-OSS-20B performance by 4.4%.

Open-source code, data, and models are available at the provided GitHub link.

Abstract

We study a pipeline that curates reasoning data from initial structured data for improving long-context reasoning in large language models (LLMs). Our approach, $\pi^2$, 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 $\pi^2$ yields consistent improvements across four long-context reasoning benchmarks and our alike $\pi^2$-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 $\pi^2$'s usefulness. Our code, data, and models are open-source at https://github.com/vt-pi-squared/pi-squared.