On December 12, the signing ceremony of the Peking University-ByteDance" Doubao Large Model System Software Joint Laboratory" and the academic seminar on" Intelligent Software Technology and Ecology for Large Models" were held at the Yingjie Exchange Center, Peking University. The attendees included Zhenjiang Hu, Dean of the School of Computer Science, Peking University; Peng Wei, Deputy Party Secretary of the School of Computer Science, Peking University; Ruqing Zheng, Deputy Director of the Office of Science and Technology Development, Peking University; Xiaowei Jin, Head of ByteDance Research Strategy and Academic Cooperation; and Liang Xiang, Head of Doubao Foundation Model Team. The event was moderated by Xuanzhe Liu, who is the Director of the Joint Laboratory and Professor at the School of Computer Science, Peking University.

Professor Zhenjiang Hu, Dean of the School of Computer Science, Peking University, said in his speech: "The software research team at the School of Computer Science, Peking University, is recognized as one of the leading teams in the world and a top performer in China. The team has won the first prize of the State Technology Invention Award in the field of computer science and is the only recipient of the IEEE TCSE Software Synergy Award in Asia. In response to the challenges posed by large models to system software, Professor Xuanzhe Liu's team and ByteDance Doubao (Seed) team have collaborated effectively, and published many high-quality papers published. Specifically, they have publicly introduced the large model training system design of GPUs of MegaScale (More than 10,000 GPUs). As the earliest known research in such field, their work has had a significant impact on both academia and industry. The establishment of the joint laboratory marks a new stage of deeper, more specialized and closer cooperation between the two sides. In terms of basic research, we will focus on the basic science and key technical issues of intelligent software for large models, especially in the context of real-world enterprise scenarios, while also advancing original, leading, and critical scientific research. Regarding talent training, we will establish a high-level innovative talent training platform through the joint laboratory, develop a new IUR education model, and cultivate complex talents with solid theoretical skills and practical innovation capabilities."

Speech by Zhenjiang Hu, Dean of the School of Computer Science, Peking University

Ruqing Zheng, Deputy Director of the Office of Science and Technology Development at Peking University, conveyed the warm congratulations from Director Weihao Yao on the establishment of the Joint Laboratory. She said: "System software provides fundamental support for the development of artificial intelligence. We are pleased to see that Peking University and ByteDance have established a joint laboratory based on long-term cooperation, which will further promote the application and transformation of research results from Peking University's system software technology in significant AI scenarios, including large models. We look forward to the joint efforts of both sides to effectively study real-world problems and achieve more impactful collaborative outcomes. The Office of Science and Technology Development will continue to serve as a bridge by actively exploring new patterns of university-enterprise cooperation. This aims to promote the establishment of an innovation ecosystem that deeply integrates ITU (Industry-University-Research) research and accelerates technological transformation and industrialization. We also look forward to collaborating with ByteDance to create a model for school-enterprise cooperation and promote more scientific and technological innovations to serve economic and social development."

Speech by Ruqing Zheng, Deputy Director of the Office of Science and Technology Development, Peking University

Xiaowei Jin, Head of ByteDance Research Strategy and Academic Cooperation, said: "Peking University and ByteDance have established a long-term cooperation, particularly yielding significant research outcomes in recent years regarding the management of ultra-large-scale cluster resources. This research has been applied in the production environment of the Doubao large model. The establishment of the joint laboratory signifies a new phase in collaboration between the two sides. In the future, we will continue to explore innovations in model architecture, training frameworks, and inference optimization, aiming to break through key core technologies in R&D. Regarding industrial practice, the Doubao large model offers a variety of application scenarios and practical experiences, providing valuable real-world feedback for basic research. The Doubao (Seed) Team will collaborate closely with experts from Peking University to achieve breakthroughs in key technologies related to large model systems and advance the development of these technologies."

Speech from Xiaowei Jin, Head of ByteDance Research Strategy and Academic Cooperation

Professor Xuanzhe Liu, the Director of the Joint Laboratory, reviewed the context and process of the establisment of the joint laboratory and pointed out that system software is the fundamental support for efficient AI development. The joint laboratory will fully leverage the advantages of Peking University Software Team's basic research and those of the Doubao large model in infrastructure and application scenarios to carry out joint research and address fundamental technical challenges in large model system software. It aims at creating outcomes that have a significant impact on both academia and industry. At the same time, in the process of university-enterprise cooperation, the laboratory will also focus on training high-end software talents and on the integration of production with education.

Introduction of the laboratory addressed by Xuanzhe Liu, the Director of the Joint Laboratory and Professor at the School of Computer Science, Peking University

Peng Wei, Deputy Party Secretary of the School of Computer Science, Peking University, presided over the signing and opening ceremonies for the joint laboratory. Xuanzhe Liu and Liang Xiang signed the cooperation agreement on behalf of both sides, while Dean Zhenjiang Hu, Deputy Director Ruqing Zheng, and Mr. Jin Xiaowei jointly declared the opening of the joint laboratory.

Opening Ceremony of the Joint Laboratory

The subsequent academic seminar on 'Intelligent Software Technology and Ecology for Large Models' was hosted by Researcher Xin Jin from the School of Computer Science at Peking University. Academic reports were presented by Researcher Tong Yang and graduate students Yihao Zhao, Bingyang Wu, and Shimao Chen.

Academic Seminar Hosted by Researcher Xin Jin, School of Computer Science, Peking University

Academic Report Addressed by Researcher Tong Yang, School of Computer Science, Peking University

The experts agreed that advancing the deep integration of ITU is of great significance for promoting breakthroughs in core AI technology. The establishment of the joint laboratory marks an important milestone in collaborative innovation between universities and enterprises. It will leverage the strengths of both sides to advance technological innovation and the application development of large model system software.

About Peking University-ByteDance "Doubao Large Model System Software Joint Laboratory"

Based on research conducted by the School of Computer Science at Peking University in the field of system software, the joint laboratory focuses on developing system software for AI large models, leveraging ByteDance's practical advantages in applying these models and systems. Many members of the team have published their outcomes at top international conferences such as SIGCOMM and NSDI, or have a wealth of experience in successfully deploying large-scale production environments in the field of training and inference systems, laying a solid foundation for the team's technological innovation.

About Doubao (Seed) Team

The Doubao (Seed) team was established by ByteDance in 2023, dedicated to developing the industry-leading AI large language model technology. The team aims to become a world-class research team with contributions to technological and social advancements. The Doubao (Seed) Team has a long-term vision and determination in AI, with research covering disciplines including deep learning, reinforcement learning, LLM, speech, vision, AI Infra, AI Safety, etc. Leveraging the resources on the platform, the team continues to invest in relevant fields. It has launched a self-developed general-purpose large model that provides multimodal capabilities, supporting over 50 downstream businesses, such as Doubao, Coze, and Dreamina, all accessible to enterprise customers through VolcEngine. To date, the Doubao App has become the largest AIGC application in China by user base.