MoZIP
收藏资源简介:
MoZIP是一个多语言知识产权评估基准,旨在评估大型语言模型在知识产权领域的性能。该基准包含三个挑战性任务:知识产权多项选择测验(IPQuiz)、知识产权问答(IPQA)和专利匹配(PatentMatch)。数据集涵盖九种语言,通过收集来自不同国家和语言的在线知识产权知识测试问题构建。MoZIP不仅用于评估模型,还推动了首个面向知识产权的多语言大型语言模型MoZi的开发,该模型基于BLOOMZ,通过监督微调处理多语言知识产权相关文本数据。数据集的应用领域主要集中在知识产权保护和创新激励,旨在解决当前大型模型在特定领域评估中的不足。
MoZIP is a multilingual intellectual property (IP) evaluation benchmark designed to assess the performance of large language models (LLMs) in the intellectual property domain. This benchmark encompasses three challenging tasks: Intellectual Property Quiz (IPQuiz), Intellectual Property Question Answering (IPQA), and Patent Matching (PatentMatch). The dataset spans nine languages and is constructed by gathering online intellectual property knowledge test questions from diverse countries and linguistic backgrounds. MoZIP not only serves as a tool for model evaluation but also facilitates the development of MoZi, the first multilingual large language model tailored for intellectual property. Built on BLOOMZ, MoZi processes multilingual intellectual property-related textual data via supervised fine-tuning. Its application scenarios primarily focus on intellectual property protection and innovation incentives, aiming to address the current shortcomings in domain-specific evaluation of large-scale models.

- 1MoZIP: A Multilingual Benchmark to Evaluate Large Language Models in Intellectual Property中国科学院深圳先进技术研究院 · 2024年



