"metaevo_data"
收藏DataCite Commons2026-05-09 更新2026-05-19 收录
下载链接：
https://ieee-dataport.org/documents/metaevodata
下载链接
链接失效反馈官方服务：
资源简介：
"Our experiments are conducted on two widely used multi-domain reasoning benchmarks: MMLU and C-Eval. For MMLU, we group its 57 subjects into seven major domains: Natural Sciences, Engineering & Technology, Social Sciences, Humanities & History, Law & Public Affairs, Ethics & Morality, and Business & Management. For C-Eval, we organize its 52 subjects into five domains: Natural Sciences, Engineering, Social Sciences & Humanities, Vocational Qualifications, and Medicine & Life Sciences. Additionally, we construct C-Eval Hard by extracting three challenging science subdomains\u2014Mathematics, Chemistry, and Physics\u2014to test reasoning under domain variation. Following official splits, we use training sets for reference graph construction and test sets for evaluation, covering diverse question types from factual recall to complex quantitative reasoning across all domains."
提供机构：
IEEE DataPort
创建时间：
2026-05-09
