At 9:00 am, researchers from different centers were convened to discuss system architectures; at 10:00 am, he discussed some technical details with external experts; at noon, he picked up where they left off in the morning; after lunch, he agreed on test plans with business partners; at 3:00 pm, he talked about project planning with the Research Center for Space Computing System; in the evening, a regular meeting of the project team was held to review the day's work... As the head of ZJ Lab's Intelligent Computing Infrastructure Innovation Center, ZHANG Ruyun performed well-defined and fast-paced tasks.

Long accustomed to such conditions, he works nearly 12 hours a day and remains motivated and fully engaged. Having long been engaged in scientific research and management in the fields of AI and computer architecture, ZHANG Ruyun has undertaken six national major projects such as those supported by the National Key R&D Program of China, key foundations, and innovation-driven development projects of the Ministry of Industry and Information Technology. Additionally, he has won one First Prize of the National Science and Technology Progress Award, four first prizes and two second prizes of provincial and ministerial science and technology progress awards. Recently, he has been honored as a Model Worker of Zhejiang Province in 2024.

ZHANG Ruyun is Honored at the Award Ceremony

At present, ZHANG Ruyun is leading the team to promote the building of a 10,000 GPU Intelligent Computing Cluster, aiming to establish a high-efficiency intelligent computing cluster that supports AI foundation model training. The 10,000 GPU Intelligent Computing Cluster lays a foundation for enhancing ZJ Lab's strengths in intelligent computing, which requires close collaboration between research centers and external companies.

"It's a structured and highly coupled task where individuals work together towards a shared purpose," ZHANG Ruyun said. "We need to be clear about our cluster development goals. System design varies with goals. Perhaps, some systems aim towards stability, while others prioritize accuracy. However, we'll identify elements based on goals, and design the total system and subsystems with interfaces between them, and finally carry out system integration."

In view of limited time, heavy workload and high standards, the whole team dared not relax for a moment in the face of this complex and difficult task. Under the leadership of ZHANG Ruyun, the team has always worked together with a selfless attitude.

ZHANG Ruyun is Discussing with Team Members. (Photo by ZHOU Zhihao)

During the Spring Festival 2024, foundation model training was affected due to network flashing, causing significant stress among members. "There are many factors that affect the operation of intelligent computing clusters, including server, cable fault, and switch, all of which may lead to network flashing, making it challenging to pinpoint the problem." Recalled GAO Xiang, one of the key team members. "The problem was not solved until late at night on the day before the New Year's Eve. Throughout the Spring Festival, team members were put on duty at the lab every day to ensure that clusters were functioning properly."

In the months following cluster development, GAO Xiang and his team members often worked until three or four o'clock in the morning. "We often need to go the whole hog. Maybe, if this problem were not solved today, we could not work it out at all tomorrow. Dr. ZHANG Ruyun often joins us and provides strong support and guidance," said GAO Xiang.

We insist on solving problems once identified to maximize work continuity and efficiency. Under the leadership of ZHANG Ruyun, the team has completed the most critical part of the task of constructing a 10,000 GPU intelligent computing cluster.

At the Intelligent Computing Infrastructure Innovation Center, ZHANG Ruyun's office is situated amidst a row of "research and discussion rooms". Whenever problems arise, he promptly studies and solves them. As the head of the center, ZHANG Ruyun hopes that all team members possess strong execution capabilities and embrace "altruistic thinking", fostering an "open, transparent, positive and public" culture. "When individuals find their niche and area of expertise within a complex system, and fully utilize their abilities within that framework while collaborating as much as possible, the whole system gains cohesive strength."