Dayou Du

Master of Data Science and Analytic

HKUST (GZ)

Biography

My name is Dayou Du (杜大猷). I am currently a second-year Mphil student at the Data Science and Analytics Thrust, Information Hub of HKUST(GZ) under the guidance of Prof. Xiaowen Chu. I received my B.Eng. from South China University of Technology. During the summer of 2023 and 2024, I interned at Microsoft Research Asia with Dr. Shijie Cao.

Interests

• Machine Learning Systems
• Model Compression
• Hardware Accelerators

Education

• MPhil in Data Science and Analytics, 2022

  The Hong Kong University of Science and Technology

• BE in Intelligent Science and Technology, 2018

  South China University of Technology

Experience

 

 

 

 

 

Research Intern

Microsoft Research Asia

July 2024 – December 2024 Beijing

Research on efficient inference of Large Language Models.

 

 

 

 

 

Research Intern

Microsoft Research Asia

July 2023 – December 2023 Beijing

Research on model quantization of Large Language Model.

 

 

 

 

 

Engineering Intern

DMAI

June 2021 – October 2021 GuangZhou

Designed a framework for Nature Language Understanding tasks.

Projects

CuTe

GEMM optimization using CuTe and Cutlass.

Code Blogs

Awesome Vit Quantization and Acceleration

List of papers on Vision Transformers (ViT) quantization and hardware acceleration, featured in top-tier AI conferences and journals.

PDF Code

BitDistiller

A novel QAT with Self-Distillation framework to enhance ultra low-bit LLMs.

PDF Code

LabelTrack

A graphical automated annotation platform for multi-object tracking.

Code Video

Publications

Model Quantization and Hardware Acceleration for Vision Transformers: A Comprehensive Survey

[Arxiv] This article surveys ViTs quantization and hardware acceleration, covering their architectural attributes, model quantization principles, state-of-the-art techniques, and the importance of hardware-friendly algorithm design.

Dayou Du, Gu Gong, Xiaowen Chu

PDF Cite Code Project

BitDistiller: Unleashing the Potential of Sub-4-Bit LLMs via Self-Distillation

[ACL 2024] BitDistiller synergizes Quantization-Aware Training and Knowledge Distillation, significantly enhancing sub-4-bit LLMs’ performance on complex tasks with reduced data and training resource demands.

Dayou Du, Yijia Zhang, Shijie Cao, Jiaqi Guo, Ting Cao, Xiaowen Chu, Ningyi Xu

PDF Cite Code Project

AFPQ: Asymmetric Floating Point Quantization for LLMs

[ACL 2024 Findings] This study introduces asymmetric FP quantization for LLMs to address poor performance with small sizes or low bits, improving accuracy without extra storage compared to asymmetric INT quantization.

Yijia Zhang, Sicheng Zhang, Shijie Cao, Dayou Du, Jianyu Wei, Ting Cao, Ningyi Xu

PDF Cite Code

Benchmarking and Dissecting the Nvidia Hopper GPU Architecture

[IPDPS 2024] This research benchmarks Hopper GPUs, analyzing new features like FP8 tensor cores and DPX, to aid in software optimization and understanding of their microarchitectural and programming nuances.

Weile Luo, Ruibo Fan, Zeyu Li, Dayou Du, Qiang Wang, Xiaowen Chu

PDF Cite