license: cc-by-nc-sa-4.0
Introduction
ArcMMLU is a Chinese benchmark specifically designed for evaluating LLMs on Library & Information Science (LIS). It aims to evaluate the knowledge and reasoning capabilities of LLMs in the LIS academic field, which covers four key sub-areas: Archival Science, Data Science, Library Science, and Information Science. Please refer to our paper for more information ArcMMLU: A Library and Information Science Benchmark for Large Language Models
It is important to note that the name ArcMMLU is derived from our previous large language model research project—ArcGPT, which was primarily focused on Archival Science. Later, our research scope expanded from Archival Science to a broader field of information management, but we retained the name ArcMMLU. Therefore, ArcMMLU is not just an evaluation benchmark for Archival Science; it is a comprehensive evaluation dataset for the entire LIS discipline.
For the sake of convenience, ArcMMLU adopts the same data format as CMMLU. Furthermore, based on the CMMLU project, we provide evaluation code. For models that have been evaluated on CMMLU, conducting an evaluation on ArcMMLU will be pretty straightforward. Special thanks to the CMMLU---Chinese Multi-Task Language Understanding Evaluation project for its contribution to the evaluation of Chinese LLMs. We hope that ArcMMLU can serve as a powerful supplement in specialized fields, bringing more detail and depth to the evaluation of Chinese LLMs.