RegexPSPACE
收藏资源简介:
RegexPSPACE数据集是针对PSPACE完全正则表达式(regex)问题构建的,用于评估大型语言模型(LLM)和大型推理模型(LRM)的空间计算能力。数据集包括超过一百万个正则表达式实例,涵盖了正则表达式等价性判定(RegexEQ)和最小化(RegexMin)两个PSPACE完全问题。该数据集通过双指数空间探索构建,并经过严格的过滤过程,旨在揭示LLM和LRM在空间复杂性约束下的计算局限性,并为评估它们的先进推理能力提供一个新框架。
The RegexPSPACE dataset is constructed for PSPACE-complete regular expression (regex) problems, aiming to evaluate the spatial computational capabilities of Large Language Models (LLMs) and Large Reasoning Models (LRMs). The dataset contains over one million regular expression instances, covering two PSPACE-complete problems: regular expression equivalence checking (RegexEQ) and minimization (RegexMin). This dataset is built via double-exponential space exploration and has undergone a rigorous filtering process, designed to reveal the computational limitations of LLMs and LRMs under spatial complexity constraints and provide a novel framework for evaluating their advanced reasoning capabilities.

- 1通过韩国首尔延世大学计算机科学系 · 2025年



