李翔宇

博士生, 清华大学智能产业研究院

我目前就读于清华大学智能产业研究院（AIR），博士四年级，研究方向为面向端侧智能的高效推理系统（博士阶段工作概述）。目前，我主要关注具身智能的模型系统协同设计，尤其是 VLA 的训练与部署。

最新动态

2026/07 DWI: Efficient On-Device LLM Decoding Through Dynamic Width Selection on Nested Models accepted to IEEE TMC. Congrats to Weining Chen. Paper
2026/06 Vec-LUT received Best Paper Award Runner-Up at ACM MobiSys 2026. Paper Blog Code Slides
2026/06 ActProbe: Action-Space Probe for Early Failure Detection of Generative Robot Policies released. Paper Code Page

代表论文

ActProbe: Action-Space Probe for Early Failure Detection of Generative Robot Policies

Bingjia Huang^*, Xiangyu Li^*, Xiang Wang, Liang Mi, Zixu Hao, Weijun Wang, Hao Wu, Kun Li, Yunxin Liu, Ting Cao

ArXiv preprint, 2026

Paper Code Page

OxyGen: Unified KV Cache Management for VLA Inference under Multi-Task Parallelism

Xiangyu Li, Huaizhi Tang, Xin Ding, Weijun Wang, Ting Cao, Yunxin Liu

ArXiv preprint, 2026

Paper Code

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

ACM MobiSys 2026Best Paper Award Runner-Up, Results Reproduced @AE

Paper Blog Code Slides

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

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

ACM MobiCom 2024Results Replicated @AE

Paper Code Slides

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, 2024Survey & Position, “Efficiency” Section Lead

Paper Code

近期文章

用向量查表真正加速低比特 LLM 的并行推理

2026/07/11

Vec-LUT 将重复的标量查表转化为连续的向量读取，在 x86 和 ARM CPU 上将并行三值 LLM 推理加速至多 4.2 倍。

增强 vLLM 对 GPTQv2 格式的支持：分析与实现

2025/10/12

分析 vLLM 对 GPTQv2 格式支持的局限，以及低比特非对称量化推理所需的 CUDA 内核改动。

视觉—语言—动作（VLA）模型：近期进展综述

2025/09/16

近期 VLA 正从离散控制走向连续控制，也从单系统架构转向双系统架构。