The Accelerated Intelligent Systems Lab (AlSys, led by Professor Jin-Ho Lee) of the Electrical and Computer Engineering (ECE) Department has proposed a methodology for accelerating the training of Large Language Models (LLMs) using SmartSSD. Training LLMs typically requires the utilization of numerous GPUs due to the large memory requirements. However, this study suggests a technique that achieves high performance with just one GPU, by utilizing a SmartSSD that can perform computations internally.

This study has been particularly recognized for its utility in situations such as fine-tuning, where training times are relatively short compared to memory requirements. As a result, it was awarded the Best Paper Award – Honorable Mention at the 2024 HPCA (International Symposium on High-Performance Computer Architecture, March 2-6, 2024, Edinburgh, Scotland), one of the top-tier conferences in the field of computer architecture.

[Presented Paper]
Smart-Infinity: Fast Large Language Model Training using Near-Storage Processing on a Real System, Hongsun Jang, Jaeyong Song, Jaewon Jung, Jaeyoung Park, Youngsok Kim, and Jinho Lee, HPCA, 2024

   

(left) HCPA 2024 award session, (right) Award certificate

Source:  https://ece.snu.ac.kr/ece/news?md=v&bbsidx=55465

Translated by: Jiyong Yoo, English Editor of the Department of Electrical and Computer Engineering, cyoo7@snu.ac.kr