Dear Colleagues and Students,

The State Key Laboratory of Internet of Things for Smart City would like to invite you to join our “IOTSC Talk Series” on 24/08/2023 (Thursday). We are pleased to invite Prof. Jingwen LENG from Shanghai Jiao Tong University as the speaker.

Low-bit Adaptive Numerical Data Type for Large Language Models
Speaker: Prof. Jingwen LENG, Professor, Department of Computer Science and Engineering, Shanghai Jiao Tong University
Date: 24/08/2023 (Thursday)
Time: 16:00 – 17:00
Language: English
Venue: N21-5/F Exhibition Hall

Abstract:

The computational power required for large language models has increased 240 times in two years, far exceeding the improvements brought by the chip manufacturing process in accordance with Moore’s Law. Therefore, the evolution of computing architecture and the innovation of computational numerical data types have become the key to improving computational efficiency. In this talk, we have deeply analyzed the numerical distribution characteristics in large language models, and designed a system that can automatically adjust the numerical type to fit the different tensor distributions in large models, thereby reducing the computational numerical bit width of large models to 4 bits. On this basis, we have also designed a storage subsystem-friendly encoding format for adaptive numerical types, which can exploit the computational efficiency advantages of adaptive numerical types with very low hardware overhead.

Speaker’s Bio:
Jingwen LENG is a professor at the Department of Computer Science and Engineering at Shanghai Jiao Tong University. His research direction is the intelligent computer system design for the artificial intelligence, with the focus on performance, energy efficiency, and reliability. He has received multiple grants from National Science Foundation of China and top industrial companies. He has published more than 40 papers in top tier computer architecture conferences and more than 10 domestic/international patents. His work has received best paper award or nomination at venues/conferences including IEEE Micro Top Picks, DAC, and PACT. He was also awarded the DAMO Young Fellow from Alibaba.

For enquiries: Tel: 8822 9141
Email: shirleyfong@um.edu.mo

Best Regards,
State Key Laboratory of Internet of Things for Smart City