Haocheng Xi ☕️
Haocheng Xi

First-year PhD student

Berkeley AI Research

About Me

I am a first-year PhD student at Berkeley AI Research, advised by Prof Kurt Keutzer. I graduated from Yao Class, Tsinghua University in 2024, advised by Prof Jianfei Chen and Prof Jun Zhu. I have been very fortunate to work with Prof Song Han as a research intern at Nvidia, and to work with Prof Sheng Wang as an student intern in University of Washington.

Email: xihc (at) berkeley (dot) edu.

Download CV
Interests
  • Efficient Machine Learning
  • Quantizated LLM training
Education

  • PhD Computer Science

    University of California, Berkeley

  • Bachelor Computer Science

    Yao Class, Tsinghua University

📚 My Research

My research interests is about the efficiency of large language models (LLMs) and quantized training. I aim to push the boundaries of how we can effectively compress and accelerate deep learning models while maintaining their accuracy and robustness.

Email: xihc (at) berkeley (dot) edu. Feel free to reach out 😃

Recent Publications
Haocheng Xi, Yuxiang Chen, Kang Zhao, Kai Jun Teh, Jianfei Chen, Jun Zhu (2024). Jetfire: Efficient and Accurate Transformer Pretraining with INT8 Data Flow and Per-Block Quantization. In ICML 2024 (Spotlight Paper).
PDF Cite Code Poster
Haocheng Xi, Changhao Li, Jianfei Chen, Jun Zhu (2023). Training Transformers with 4-bit Integers. In NeurIPS 2024.
PDF Cite Code

Experience

  1. Research Intern

    Nvidia
    Advised by Prof Song Han. Focus on Low-precision Training.

Education

  1. PhD Computer Science

    University of California, Berkeley
    Focus on LLM efficiency. Advised by Prof Kurt Keutzer.

  2. Bachelor Computer Science

    Yao Class, Tsinghua University
    Graduate from Yao Class, IIIS. Advised by Prof Jianfei Chen and Prof Jun Zhu.