Session

· Contributors · Organizations · Search Program · Flagged · Happening Now

Research Manuscript: Transformers: Rise of the Optimized Large Language Models

Session Chairs

DescriptionThis session is about transformers and large language model optimizations. As LLMs are essentially made of transformers due to their superior effectiveness to understand context and relationships within sequences by using self-attention mechanism, their hardware optimizations are quite important. We have a best paper award candidate presentation for DRAFT, which is about a clever hardware technique that approximates the backpropagation hardware for great energy efficiency improvement. Our other papers cover hardware/algorithm co-design, retrieval-in-memory techniques, attention, long-context generation, and an overview of the pareto-frontier for low-precision data formats.

Event TypeResearch Manuscript

TimeMonday, June 231:30pm - 3:00pm PDT

Location3001, Level 3

Topics

Tracks

Presentations

1:30pm - 1:45pm PDT	SSFT: Algorithm and Hardware Co-design for Structured Sparse Fine-Tuning of Large Language Models Authors Miao Yu Trevor E. Carlson
1:45pm - 2:00pm PDT	DRAFT: Decoupling Backpropagation from Pre-trained Backbone for Efficient Transformer Fine-Tuning on Edge Authors Zhirui Huang Shiwei Liu Haozhe Zhu Qi Liu Chixiao Chen
2:00pm - 2:15pm PDT	Move Less, Retrieve Fast: A Retrieval-in-Memory Architecture for Language Models Authors Jiaxian Chen Yuxuan Qi Jianan Yuan Kaoyi Sun Tianyu Wang Chenlin Ma Yi Wang
2:15pm - 2:30pm PDT	KVO-LLM: Boosting Long-Context Generation Throughput for Batched LLM Inference Authors Zhenyu Li Dongxu Lyu Gang Wang Yuzhou Chen Liyan Chen Wenjie Li Jianfei Jiang Yanan Sun Guanghui He
2:30pm - 2:45pm PDT	An Energy-Efficient High-Utilization Hardware Architecture for Attention Mechanism in Transformer using Balanced Systolic Array and Multi-Row Interleaved Operation Ordering Authors Haiyang Zhou Hongyang Hu Jinshan Yue Hanghang Gao Yuanlu Xie Xiaoxin Xu Chunmeng Dou Ming Liu
2:45pm - 3:00pm PDT	Finding the Pareto Frontier of Low-Precision Data Formats and MAC Architecture for LLM Inference Authors Brian Crafton Xiaochen Peng Xiaoyu Sun Ashwin Lele Bo Zhang Win-San Khwa Kerem Akarvardar

62dac.conference-program.com