On July 20th, reports surfaced indicating that the China Academy of Information and Communications Technology (CAICT) has officially launched the 'AISHPerf Research: Benchmarking Performance in Large-Scale Model Training and Inference Engineering'. This research endeavor will concentrate on three pivotal aspects: firstly, it aims to refine the indicator system by establishing standardized definitions for performance indicators, measurement methodologies, and reporting criteria. This will involve clarifying the applicable scenarios and calculation methods for each indicator, thereby minimizing discrepancies in cross-comparative analyses. Secondly, the research will focus on aggregating industry data by extensively gathering real-world performance data from model trainers, chip manufacturers, and computing power providers. The objective is to create anonymized industry benchmark reference ranges that will be regularly updated, providing a valuable benchmarking resource for the industry. Thirdly, the research will endeavor to promote engineering best practices by compiling and disseminating mature experiences in areas such as communication optimization, memory management, and parallel strategy scheduling. The goal is to formulate reusable optimization path recommendations that will aid computing power providers in making informed, scientific selections and assist model trainers in achieving precise tuning.