169Pi/exambench

ExamBench数据集（约6亿个token，405,000个示例）是专为竞争性考试准备和推理AI设计的大型开源语料库之一。它结合了结构化的思维链推理，并涵盖了超过25个印度和国际考试，包括JEE、NEET、UPSC、银行、GRE和IELTS等多个领域，使其成为训练和评估教育重点、推理能力AI系统的强大基础。The ExamBench Dataset (~600M tokens, 405k examples) is one of the largest open-source corpora designed for competitive exam preparation and reasoning AI. Generated using advanced distillation techniques, it combines structured chain-of-thought reasoning with comprehensive coverage of over 25 Indian and international examinations. From JEE and NEET to UPSC, Banking, GRE, and IELTS, the dataset spans multiple domains like STEM, humanities, current affairs, language, and reasoning, making it a powerful foundation for training and evaluating education-focused, reasoning-capable AI systems.

搜集汇总

背景与挑战

背景概述

该数据集是一个专注于竞争性考试准备的大规模问答数据集，包含约40.6万个示例，覆盖印度和国际考试如JEE、NEET、UPSC等，涉及STEM、人文和推理等多个学科。其特点在于提供详细的链式推理步骤，旨在训练和评估具备教育推理能力的AI系统，适用于AI辅导和自适应学习平台开发。