Xiangyu Li

Ph.D. Student
Institute for AI Industry Research (AIR), Tsinghua University

About Me

I am a 4th year Ph.D. candidate at Institute for AI Industry Research (AIR), Tsinghua University, advised by Dr. Yunxin Liu. Prior to this, I received my B.E. degree from Department of Electronic Engineering, Tsinghua Univeristy in 2022.

My research is focused on on-device and physical AI. Main topics:

Efficient inference system (main line during Ph.D.):
- FlexNN (MobiCom 2024): efficient and adaptive memory management.
- Vec-LUT (MobiSys 2026): GeMM optimization for ultra-low-bit LLMs.
- KV cache management for VLA: comming soon.
Autonomous embodied agent (recent interest):
- Pre-training and model architecture.
- Post-training self-evolvement.

Welcome to checkout Xyu’s Blog for more! ✍️

News

[Mar. 2026] Paper “Vec-LUT: Vector Table Lookup for Parallel Ultra-Low-Bit LLM Inference on Edge Devices” is (conditionally) accepted to MobiSys 2026. Thanks to all coauthors! [Paper] [Code] [Model]
[Aug. 2025] Paper “An Empirical Study of LLM Reasoning Ability Under Strict Output Length Constraint” is accepted to EMNLP 2025 main conference. Congrats to Yi Sun! [Paper] [Page]
[Jul. 2025] Paper “Squeezer: Efficient Multi-DNN Inference for Edge Video Analytics via Cross-Model Scheduling” is accepted to IEEE Transactions on Mobile Computing (TMC). Congrats to Xiang Wang! [Paper]
[Nov. 2024] I presented FlexNN at MobiCom 2024, Washington, D.C., USA. [Paper] [Code] [Slides]

See More News

Publications

Vec-LUT: Vector Table Lookup for Parallel Ultra-Low-Bit LLM Inference on Edge Devices

Xiangyu Li*, Chengyu Yin*, Weijun Wang, Jianyu Wei, Ting Cao, Yunxin Liu.

ArXiv preprint, 2025.

Paper Code Model conditionally accepted to MobiSys 2026

MobiCom

FlexNN: Efficient and Adaptive DNN Inference on Memory-Constrained Edge Devices

Xiangyu Li, Yuanchun Li, Yuanzhe Li, Ting Cao, Yunxin Liu.

The 30th Annual International Conference on Mobile Computing and Networking (MobiCom), 2024.

Paper Slides Code Oral Presentation, Artifact Evaluated

Personal LLM Agents: Insights and Survey about the Capability, Efficiency and Security

Yuanchun Li, Hao Wen, Weijun Wang, Xiangyu Li, Yizhen Yuan, Guohong Liu, Jiacheng Liu, Wenxing Xu, Xiang Wang, Yi Sun, Rui Kong, Yile Wang, Hanfei Geng, Jian Luan, Xuefeng Jin, Zilong Ye, Guanjing Xiong, Fan Zhang, Xiang Li, Mengwei Xu, Zhijun Li, Peng Li, Yang Liu, Ya-Qin Zhang, Yunxin Liu.

ArXiv preprint, 2024.

Paper Code Survey, "Efficiency" Section Lead

See More Publications

Experiences

Awards

清华之友-智能产业研究院清智奖学金, 2025.12.
清华之友-济宁英才奖学金, 2024.11.

Internship

ByteDance, Big Data (Serverless/FaaS) Infrastructure Development, 2021.06-2021.09.

Teaching Assistant

Tsinghua University, Computer Program Design (undergraduate), 2025 Spring & Summer.
Tsinghua University, Basic Music Theory and Vocal Practice, 2023 Fall.

Others

EESAST (清华大学电子工程系学生科协), 2019.07-2022.06.

Skills

Programming Languages

C/C++, Python — primary languages for research/projects.
Shell, HTML/CSS/JavaScript — frequently used with AI.
C#, Go — was familiar years ago.
Java/Kotlin, MATLAB, Verilog, MIPS asm.

Frameworks

vLLM, llama.cpp, NCNN, openpi — most familiar and customized/contributed.
PyTorch, Transformers — freqently used for research.
SGLang, JAX, TensorRT, TF Lite, LlamaIndex — ever used.
Electron, React, Unity 3D.

Powered by Jekyll and Minimal Light theme.