THU FASTsys Research Group

    

个人简介

  • 胡世鹏，清华大学计算机系高性能所 2021 级博士生，导师为张广艳老师。主要研究方向为机器学习系统，尤其是大模型的 KV Cache 系统优化。本人正在寻找大模型基础设施相关的工作机会，联系方式：hsp21@mails.tsinghua.edu.cn
  • I am Shipeng Hu, a PhD candidate at FastSys Lab, Department of Computer Science and Technology, Tsinghua University. My research interest focuses on machine learning systems, particularly KV cache optimization for LLM serving. I am currently seeking job opportunities in LLM infra. Contact me at: hsp21@mails.tsinghua.edu.cn

教育背景

  • 2021.09 - 至今，清华大学计算机科学与技术系高性能所，博士生，导师：张广艳
  • 2017.09 - 2021.06，华中科技大学计算机学院卓越工程师班（工学学士学位）

研究内容

  • 提出了计算-存储双向感知的大模型KV缓存系统，大幅减小历史KV加载开销，提升大模型服务质量。
  • 提出了大模型键值缓存的无损压缩方法，大幅减小大模型键值缓存的存储开销和加载延迟。
  • 提出了高效联邦树训练框架，在保证精度的前提下大幅减小掉队者本地计算负载，提升同步效率。

论文发表

  • [FAST 2026] Shipeng Hu, Guangyan Zhang, Yuqi Zhou, Yaya Wei, Ziyan Zhong, Jike Chen. Bidaw: Enhancing Key-Value Caching for Interactive LLM Serving via Bidirectional Computation–Storage Awareness. In the Proceedings of the 24th USENIX Conference on File and Storage Technologies (FAST’26), Santa Clara, CA, February 2026. Pages 101-106.

  • Shipeng Hu, Guangyan Zhang, Weimin Zheng. Survey on KV Cache Compression for Large Language Model Inference. Journal of Computer Research and Development (in Chinese), 2026.

  • Yun Teng, Dawei Sun, Shipeng Hu, Zhiyue Li, Guangyan Zhang, Haidong Tian, Rui Chang. FastCheck: Fast Checkpointing and Recovery for DNN Training via Parallel Transmission and Compression. ENGINEERING Information Technology & Electronic Engineering, 2026.

个人荣誉

  • 2023.10 清华大学综合优秀奖学金
  • 2019.10 国家奖学金
  • 2018.10 华中科技大学本科特优生