TIGER-Lab/WebInstruct-verified

WebInstruct-Verified是一个多样化的高质量数据集，旨在促进大型语言模型在广泛领域的稳健推理能力，包括但不限于物理、化学、金融、人文学科等。数据集由超过230,000个经过人类验证的、具有明确可验证答案的问题组成，这些问题经过精心筛选以确保质量和挑战性。

WebInstruct-Verified is a diverse high-quality dataset designed to enhance the robust reasoning capabilities of large language models (LLMs) across a wide range of domains, including but not limited to physics, chemistry, finance, humanities, and more. The dataset consists of over 230,000 human-verified questions with clear verifiable answers, carefully selected to ensure quality and challenge.