On March 31, 2025, a team led by Professor Wan Jiguang from the Parallel Data and Storage Laboratory (HUST-PDSL) at Wuhan National Laboratory for Optoelectronics (WNLO) received the Best Paper Award at the Design, Automation and Test in Europe Conference (DATE) 2025. The DATE conference is considered a significant academic event within the field of Electronic Design Automation (EDA), which underpins the chip design industry.
A Novel Mixed-Precision Quantization Strategy
The award-winning paper was co-authored by Dr. Wang Jianzong from Ping An Technology (PAT) and introduced a novel chunk-level mixed-precision quantization strategy named “Cocktail.” This strategy aims to enhance long-text inference for large language models (LLMs) by assessing the similarity between LLM queries and contexts at the chunk level. The researchers reported notable advancements in hardware efficiency of the key-value (KV) cache through this mixed-precision quantization method, which resulted in a 10.4% reduction in video memory usage and a 21.4% decrease in token generation latency relative to current industry benchmarks.
Significance of the DATE Conference
The DATE 2025 conference received over 1,200 submissions, with a selective acceptance rate of 25%. Co-organized by the Institute of Electrical and Electronics Engineers (IEEE) and the Association for Computing Machinery (ACM) since its inception in 1994, the conference attracts leading scholars and professionals in electronic design automation and testing from around the world.
Recognition and Institutional Support
The Best Paper award recognized Huazhong University of Science and Technology (HUST) as the principal institution, with PhD candidate Tao Wei noted as the first author. Professor Wan Jiguang and Dr. Wang Jianzong were identified as co-corresponding authors. The research received support from several initiatives, including the Guangdong Provincial Key R&D Program and the National Key R&D Program.
Future Collaborations and Innovations
Looking ahead, the HUST-PDSL laboratory plans to extend its collaboration with Ping An Technology to explore innovative applications of large models in the optimization of storage systems. The laboratory has a reputation for its contributions to storage systems, computer architecture, and database systems, achieving significant milestones such as the development of a parallel file system that enhanced performance rankings in supercomputer storage.
Link Between Storage Technology and AI
This recognition illustrates the vital link between storage technology and artificial intelligence, emphasizing the collaborative efforts between academic institutions and industry partners to further advance these fields.
(Original source: https://english.hust.edu.cn/info/1104/4067.htm)
