Academician Zheng Weimin delivers CityU lecture on designing computer systems for large model training
Professor Zheng Weimin, an expert in supercomputing, an academician of the Chinese Academy of Engineering, and a professor and Ph.D. supervisor in the Department of Computer Science and Technology at Tsinghua University, visited City University of Macau (CityU) today. He was warmly received by Professor Zhou Wanlei, vice rector and dean of the Faculty of Data Science (FDS). Academician Zheng Weimin delivered an engaging lecture titled “Key Considerations in Designing Computer Systems for Model Training” as a part of the Doctor honoris causa Lecture Series. He delved into cutting-edge research in artificial intelligence (AI), covering areas such as industry computing power, large model training, and software ecosystems.
Academician Zheng began by discussing the demand for computing power, emphasizing its crucial role in every stage in the development of a large model. He then highlighted the software ecosystem challenges associated with three types of systems: those based on NVIDIA GPUs (Graphic Processing Units), those based on domestically developed AI chips, and those based on supercomputers. He concluded by advocating for intensified research efforts on large model infrastructure based on domestically developed systems.
Furthermore, Academician Zheng Weimin introduced a parallel training system designed for next-generation home-produced supercomputers. He emphasized the importance of coordinated hardware and software design, arguing that supercomputers can not only handle complex algorithms but also facilitate large model training, thereby enhancing overall efficiency. Academician Zheng Weimin also analyzed five key challenges facing large model infrastructure and proposed corresponding solutions. He further emphasized that utilizing supercomputers for large model training and building a domestically developed Wanka system are both necessary and challenging, requiring collaborative efforts and continuous innovation from all parties involved.
The lecture sparked lively interaction and discussion between the audience and the speaker. Students and faculty engaged in in-depth exchanges with Academician Zheng Weimin on topics such as the development of domestic GPUs, medical image applications, algorithms for intermediate system, underlying hardware, as well as GPUs and related unification issues.
The attendees included Dean José Alves of the Faculty of Business, FDS Associate Deana Zhu Tianqing and Liu Wenjian, faculty members, and students.
[foogallery id="64379"]
