Benchmarking involves designing scientific test methods, tools, and frameworks to quantitatively and comparably assess specific performance indicators of certain test subjects. With the development of artificial intelligence, AI benchmarking datasets such as ImageNet and DataPerf have gradually become consensus standards in both academic and industrial fields. However, constructing a benchmarking framework remains a significant challenge in the open-source domain due to the diverse range of data types, the wide array of research issues, and the intricate nature of collaboration networks. This paper introduces OpenPerf, a benchmarking framework designed for the sustainable development of the open-source ecosystem. This framework defines 9 task benchmarking tasks in the open-source research, encompassing 3 data types: time series, text, and graphics, and addresses 6 research problems including regression, classification, recommendation, ranking, network building, and anomaly detection. Based on the above tasks, we implemented 3 data science task benchmarks, 2 index-based benchmarks, and 1 standard benchmark. Notably, the index-based benchmarks have been adopted by the China Electronics Standardization Institute as evaluation criteria for open-source community governance. Additionally, we have developed a comprehensive toolkit for OpenPerf, which not only offers robust data management, tool integration, and user interface capabilities but also adopts a Benchmarking-as-a-Service (BaaS) model to serve academic institutions, industries, and foundations. Through its application in renowned companies and institutions such as Alibaba, Ant Group, and East China Normal University, we have validated OpenPerf's pivotal role in the healthy evolution of the open-source ecosystem.

翻译：暂无翻译