PAAI Distinguished Lecture: Heterogeneous and Disaggregated PD Separation for Large Model Inference Systems
Date: 2 June 2026 | Time: 9:30 am - 11:00 am | Venue: AG206, PolyU
The PolyU Academy for Artificial Intelligence (PAAI) successfully hosted a PAAI Distinguished Lecture on the morning of 2 June 2026, featuring Academician ZHENG Weimin, Professor at Tsinghua University and Member of the Chinese Academy of Engineering.
The lecture, entitled "Heterogeneous and Disaggregated PD Separation for Large Model Inference Systems," attracted a packed audience of faculty, researchers, and students. Academician Zheng's passionate and engaging delivery sparked continuous discussion, enthusiastic questions, and rounds of applause—making it one of the most energetic sessions at PolyU.
Drawing from his pioneering work in high-performance computing, Prof. Zheng introduced the Mooncake disaggregated architecture (KVCache-centric) and the Prefill as a Service (PaaS) design, which together deliver 54% higher throughput and 32% lower latency, while helping overcome domestic computing hardware bottlenecks. He laid out a phased roadmap to deploy the technology across same-room, same-campus, and cross-region computing networks.
The overwhelmingly positive audience response proved one thing: great research, when shared with heart, inspires action.
| Topics | Event |
|---|---|
| Research Units | PolyU Academy for Artificial Intelligence |
