P-MMEVAL
收藏资源简介:
P-MMEVAL是由阿里巴巴集团通义实验室创建的一个大规模多语言多任务基准数据集,旨在全面评估大型语言模型(LLMs)的多语言能力。该数据集包含三个基础自然语言处理(NLP)数据集和五个高级能力专项数据集,涵盖了代码生成、知识理解、数学推理、逻辑推理和指令跟随等任务。数据集通过专家翻译审查,确保了10种语言的一致覆盖,提供了跨语言的平行样本。P-MMEVAL的应用领域主要集中在LLMs的多语言能力评估和跨语言迁移能力的研究,旨在解决现有评估方法在多语言覆盖和任务多样性上的不足。
P-MMEVAL is a large-scale multilingual multi-task benchmark dataset developed by Tongyi Lab of Alibaba Group, designed to comprehensively evaluate the multilingual capabilities of Large Language Models (LLMs). It consists of three foundational Natural Language Processing (NLP) datasets and five specialized advanced capability datasets, covering a range of tasks including code generation, knowledge understanding, mathematical reasoning, logical reasoning, and instruction following. The dataset has been subjected to expert translation and review to ensure consistent coverage across 10 languages, providing cross-lingual parallel samples. Primarily applied to the evaluation of LLMs' multilingual capabilities and research on cross-lingual transfer abilities, P-MMEVAL aims to address the shortcomings of existing evaluation methods in terms of multilingual coverage and task diversity.




