BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:EECS Faculty Candidate Seminar: Dong Li
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260711T173600Z
UID:tag:localist.com\,2008:EventInstance_52834105346011
DTSTART:20260511T150000Z
DTEND:20260511T160000Z
DESCRIPTION:Tensor Offloading: GPU Memory-Efficient Execution Paradigm for 
 Large AI Models\n \n\nAbstract\n\nGiven the increase of AI model size and 
 sacristy of GPU hardware\, training and deploying AI models is often const
 rained by GPU memory capacity. Using CPU memory as an extension to GPU mem
 ory\, tensor offloading to CPU memory provides a cost-effective solution t
 o save GPU memory while enabling larger AI models. However\, this executio
 n paradigm introduces extra data movement\, and faces a series of challeng
 es\, such as tensor-migration promptness and granularity\, load balancing\
 , and tensor coherence. In this talk\, we present our efforts that use ten
 sor/computation co-offloading and a learning-based approach to address tho
 se challenges. Our work enables industry-quality transformer models with t
 ens of billion parameters on a single GPU\, a 10x increase in size compare
 d to popular frameworks such as PyTorch\, and we do so without requiring a
 ny model change from data scientists or sacrificing computational efficien
 cy. Our work has been integrated into Microsoft DeepSpeed\, and it is now 
 being utilized across the industry to democratize the use of large AI mode
 ls.\n \n\nBiography\n\nDong Li\, associate professor at Electrical Enginee
 ring and Computer Science\, works at the University of California\, Merced
 . Previously\, he was a research scientist at the Oak Ridge National Labor
 atory (ORNL)\, studying computer architecture and programming models for n
 ext generation supercomputer systems. Li earned his PhD in computer scienc
 e from Virginia Tech. His research focuses on high performance computing (
 HPC)\, and maintains a strong relevance to computer systems. Li received a
 n ORNL/CSMD Distinguished Contributor Award in 2013\, a CAREER Award from 
 the National Science Foundation in 2016\, Facebook faculty research award 
 in 2021\, Oracle Research Award in 2022\, Amazon Research Award in 2025. H
 is paper in SC'14 was in the best paper final list. His paper in ASPLOS'21
  won the distinguished artifact award. He was also the lead PI for the NVI
 DIA CUDA Research Center at UC Merced. He is an associate editor for IEEE 
 Transactions on Parallel and Distributed Systems (TPDS).
GEO:35.958178;-83.924516
LOCATION:Min H. Kao Electrical Engineering and Computer Science\, 435
SUMMARY:EECS Faculty Candidate Seminar: Dong Li
URL;VALUE=URI:https://calendar.utk.edu/event/eecs-faculty-candidate-seminar
 -dong-li
CATEGORIES:Lectures & Presentations
END:VEVENT
END:VCALENDAR