DeepJSONEval
收藏资源简介:
DeepJSONEval是一个多语言深度嵌套JSON评估基准和框架,旨在全面评估大型语言模型(LLMs)将原始文本映射到给定JSON模式并返回语法和语义正确的多层嵌套JSON对象的能力。该基准包含2100个高质量数据实例,涵盖十个多样化的网络应用领域,如旅游景点推广、电子产品介绍、患者信息等。数据集根据嵌套深度进行系统性难度分级,为模型能力提供渐进式评估。
DeepJSONEval is a multilingual deep nested JSON evaluation benchmark and framework aimed at comprehensively evaluating the ability of large language models (LLMs) to map raw text to a specified JSON schema and return multi-layer nested JSON objects that are syntactically and semantically correct. This benchmark includes 2100 high-quality data instances spanning ten diverse web application domains, such as tourist attraction promotion, electronic product introductions, patient information, and more. The dataset is systematically graded based on nesting depth to deliver progressive assessments of model capabilities.




