SOCRATIC-PRMBENCH
收藏资源简介:
SOCRATIC-PRMBENCH是一个针对过程奖励模型(PRMs)的系统评估数据集,由中国科学院自动化研究所、中国科学院大学人工智能学院和阿里巴巴集团通义实验室的研究人员创建。该数据集包含2995条推理路径,涵盖了六种推理模式:转换、分解、重新收集、推理、验证和集成,每种模式又细分为20种错误类型。数据集旨在为PRMs提供一个全面的评估框架,帮助研究人员识别PRMs在不同推理模式下的潜在缺陷,并推动PRMs在未来发展中的应用。
SOCRATIC-PRMBENCH is a systematic evaluation dataset for Process Reward Models (PRMs), created by researchers from the Institute of Automation, Chinese Academy of Sciences, School of Artificial Intelligence, University of Chinese Academy of Sciences, and Tongyi Lab of Alibaba Group. This dataset contains 2995 reasoning paths covering six reasoning modes: transformation, decomposition, recollection, reasoning, verification, and integration, each of which is further subdivided into 20 error types. The dataset aims to provide a comprehensive evaluation framework for PRMs, assisting researchers in identifying potential flaws of PRMs across different reasoning modes and advancing the future development and application of PRMs.
Socratic-PRMBench 数据集概述
基本信息
- 数据集名称:Socratic-PRMBench
- 托管平台:GitHub
- 托管地址:https://github.com/Xiang-Li-oss/Socratic-PRMBench
数据集描述
(注:根据提供的README内容,该数据集详情页面未提供具体描述信息)

- 1Socratic-PRMBench: Benchmarking Process Reward Models with Systematic Reasoning Patterns中国科学院自动化研究所, 中国科学院大学人工智能学院, 阿里巴巴集团通义实验室 · 2025年



