University of Michigan · CSE

Hi, I’m Runyu Lu.

I am a second-year PhD student in Computer Science and Engineering at the University of Michigan, co-advised by Professors Mosharaf Chowdhury and Ang Chen. I am also doing research in NVIDIA GEAR Lab. I am interested in Robotics (Embodied AI) and ML systems.

Email Me

Research Overview

Robotics (Embodied AI)
I am currently working on VLA post-training and physical agents. While today’s LLMs, VLMs, and AI agents excel at reasoning and diverse virtual tasks, I’m excited to learn and contribute to extending the intelligence into the physical world.
ML System
I am interested in many aspects of ML systems. I worked on DiT-based Image/Video Generation, Multimodal Model Training, GPU Sharing for LLM Serving, and LLM-enabled Compiler Fuzzing.

Publications

Papers

2026

ML System ASPLOS

TetriServe: Efficiently Serving Mixed DiT Workloads

Runyu Lu*, Shiqi He*, Wenxuan Tan, Shenggui Li, Ruofan Wu, Jeff J. Ma, Ang Chen, Mosharaf Chowdhury

Step-level sequence parallelism and round-based scheduling improve SLO attainment by up to 32% across heterogeneous DiT workloads.

Paper DOI Code

BibTeX

@inproceedings{lu2026tetriserve,
  title     = {TetriServe: Efficiently Serving Mixed DiT Workloads},
  author    = {Lu, Runyu and He, Shiqi and Tan, Wenxuan and Li, Shenggui and Wu, Ruofan and Ma, Jeff J. and Chen, Ang and Chowdhury, Mosharaf},
  booktitle = {Proceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2},
  pages     = {1982--1997},
  year      = {2026},
  doi       = {10.1145/3779212.3790233}
}

Robotics arXiv

ASPIRE: Agentic /Skills Discovery for Robotics

Runyu Lu*†, Yubo Wu*, Ethan Kou*, Max Fu, Wenli Xiao, Ajay Mandlekar, Yinzhen Xu, Guanya Shi, Ken Goldberg, Ang Chen, Mosharaf Chowdhury, Yuke Zhu†, Linxi "Jim" Fan†, Guanzhi Wang†

Coding agents turn multimodal execution traces into validated robot repairs and reusable skills that transfer across tasks, embodiments, and simulation-to-real settings.

Paper Website

BibTeX

@article{lu2026aspire,
  title         = {{ASPIRE}: Agentic /{Skills} Discovery for Robotics},
  author        = {Lu, Runyu and Wu, Yubo and Kou, Ethan and Fu, Letian and Xiao, Wenli and Mandlekar, Ajay and Xu, Yinzhen and Shi, Guanya and Goldberg, Ken and Chen, Ang and Chowdhury, Mosharaf and Zhu, Yuke and Fan, Linxi {Jim} and Wang, Guanzhi},
  journal       = {arXiv preprint arXiv:2607.00272},
  year          = {2026},
  url           = {https://arxiv.org/abs/2607.00272}
}

ML System ICML

Efficient Distributed MLLM Training with Cornstarch

Insu Jang, Runyu Lu, Nikhil Bansal, Ang Chen, Mosharaf Chowdhury

Frozen-aware pipeline parallelism and workload-balanced context parallelism deliver 2.26× average MLLM training throughput.

Paper Code

BibTeX

@inproceedings{jang2026cornstarch,
  title     = {Efficient Distributed {MLLM} Training with {Cornstarch}},
  author    = {Jang, Insu and Lu, Runyu and Bansal, Nikhil and Chen, Ang and Chowdhury, Mosharaf},
  booktitle = {Proceedings of the 43rd International Conference on Machine Learning},
  year      = {2026},
  url       = {https://arxiv.org/abs/2503.11367}
}

ML System ICLR

DSA: Efficient Inference For Video Generation Models via Distributed Sparse Attention

Shenggui Li, Runyu Lu, Qiaoling Chen, Haiyan Yin, Yueming Lyu, Yonggang Wen, Ivor Tsang, Tianwei Zhang

Training-free distributed sparse attention reaches up to 10.79× faster video-generation inference than a single GPU.

Paper

BibTeX

@inproceedings{li2026dsa,
  title     = {{DSA}: Efficient Inference for Video Generation Models via Distributed Sparse Attention},
  author    = {Li, Shenggui and Lu, Runyu and Chen, Qiaoling and Yin, Haiyan and Lyu, Yueming and Wen, Yonggang and Tsang, Ivor and Zhang, Tianwei},
  booktitle = {The Fourteenth International Conference on Learning Representations},
  year      = {2026},
  url       = {https://openreview.net/forum?id=1ZmdfDzGE1}
}

AI for Science arXiv

The Last Human-Written Paper: Agent-Native Research Artifacts

Jiachen Liu, Jiaxin Pei, Jintao Huang, Chenglei Si, Ao Qu, Xiangru Tang, Runyu Lu, et al.

A machine-executable research format preserving scientific logic, code, exploration traces, and evidence for AI-native research.

Paper Code

BibTeX

@article{liu2026agentnative,
  title         = {The Last Human-Written Paper: Agent-Native Research Artifacts},
  author        = {Liu, Jiachen and Pei, Jiaxin and Huang, Jintao and Si, Chenglei and Qu, Ao and Tang, Xiangru and Lu, Runyu and Chen, Lichang and Bai, Xiaoyan and Zheng, Haizhong and Chen, Carl and Chen, Zhiyang and Ye, Haojie and Fu, Yujuan and He, Zexue and Jin, Zijian and Zhang, Zhenyu and Sun, Shangquan and Harmon, Maestro and Wang, John Dianzhuo and Zeng, Jianqiao and Sun, Jiachen and Wu, Mingyuan and Zhou, Baoyu and You, Chenyu and Lu, Shijian and Qiu, Yiming and Lai, Fan and Yuan, Yuan and Li, Yao and Hong, Junyuan and Zhu, Ruihao and Chen, Beidi and Pentland, Alex and Chen, Ang and Chowdhury, Mosharaf and Zhang, Zechen},
  journal       = {arXiv preprint arXiv:2604.24658},
  year          = {2026},
  url           = {https://arxiv.org/abs/2604.24658}
}

2025

ML System arXiv

Cornfigurator: Automated Planning for Any-to-Any Multimodal Model Serving

Jeff J. Ma, Jae-Won Chung, Jisang Ahn, Yizhuo Liang, Runyu Lu, Akshay Jajoo, Myungjin Lee, Mosharaf Chowdhury

Automated coarse-to-fine deployment planning finds colocated, disaggregated, and mixed configurations with up to 6.32× higher goodput.

Paper Code

BibTeX

@article{ma2025cornfigurator,
  title         = {Cornfigurator: Automated Planning for Any-to-Any Multimodal Model Serving},
  author        = {Ma, Jeff J. and Chung, Jae-Won and Ahn, Jisang and Liang, Yizhuo and Lu, Runyu and Jajoo, Akshay and Lee, Myungjin and Chowdhury, Mosharaf},
  journal       = {arXiv preprint arXiv:2512.14098},
  year          = {2025},
  url           = {https://arxiv.org/abs/2512.14098}
}

2024

Software OOPSLA

WhiteFox: White-Box Compiler Fuzzing Empowered by Large Language Models

Chenyuan Yang, Yinlin Deng, Runyu Lu, Jiayi Yao, Jiawei Liu, Reyhaneh Jabbarvand, Lingming Zhang

LLM-guided white-box compiler fuzzing found 101 bugs, including 92 previously unknown defects across major DL compilers.

Paper DOI Code

BibTeX

@article{yang2024whitefox,
  title   = {WhiteFox: White-Box Compiler Fuzzing Empowered by Large Language Models},
  author  = {Yang, Chenyuan and Deng, Yinlin and Lu, Runyu and Yao, Jiayi and Liu, Jiawei and Jabbarvand, Reyhaneh and Zhang, Lingming},
  journal = {Proceedings of the ACM on Programming Languages},
  volume  = {8},
  number  = {OOPSLA2},
  pages   = {709--735},
  year    = {2024},
  doi     = {10.1145/3689736}
}

Abstract

Compiler correctness is crucial, as miscompilation can falsify program behaviors, leading to serious consequences. Fuzzing has been studied to uncover compiler defects. However, compiler fuzzing remains challenging: Existing arts focus on black- and grey-box fuzzing, which generates tests without sufficient understanding of internal compiler behaviors. Meanwhile, traditional white-box techniques, like symbolic execution, are computationally inapplicable to the giant codebase of compilers. Recent advances demonstrate that Large Language Models (LLMs) excel in code generation/understanding tasks. Nonetheless, guiding LLMs with compiler source-code information remains a missing piece of research in compiler testing. To this end, we propose WhiteFox, the first white-box compiler fuzzer using LLMs with source-code information to test compiler optimization, with a spotlight on detecting deep logic bugs in the deep learning (DL) compilers. WhiteFox adopts a multi-agent framework: an LLM-based analysis agent examines the low-level optimization source code and produces requirements on the high-level test programs that can trigger the optimization; an LLM-based generation agent produces test programs based on the summarized requirements. Additionally, optimization-triggering tests are used as feedback to enhance the generation on the fly. Our evaluation on the three most popular DL compilers (i.e., PyTorch Inductor, TensorFlow-XLA, and TensorFlow Lite) shows WhiteFox can generate high-quality test programs to exercise deep optimizations, practicing up to 8× more than state-of-the-art fuzzers. WhiteFox has found 101 bugs for the DL compilers, with 92 confirmed as previously unknown and 70 fixed. WhiteFox has been acknowledged by the PyTorch team and is being incorporated into its development workflow. Beyond DL compilers, WhiteFox can also be adapted for compilers in different domains.

ML System ICML

MuxServe: Flexible Spatial-Temporal Multiplexing for Multiple LLM Serving

Jiangfei Duan, Runyu Lu, Haojie Duanmu, Xiuhong Li, Xingcheng Zhang, Dahua Lin, Ion Stoica, Hao Zhang

Flexible model colocation improves throughput by up to 1.8× or serves 2.9× more requests while meeting latency objectives.

Paper arXiv Code

BibTeX

@inproceedings{pmlr-v235-duan24a,
  title     = {{MuxServe}: Flexible Spatial-Temporal Multiplexing for Multiple {LLM} Serving},
  author    = {Duan, Jiangfei and Lu, Runyu and Duanmu, Haojie and Li, Xiuhong and Zhang, Xingcheng and Lin, Dahua and Stoica, Ion and Zhang, Hao},
  booktitle = {Proceedings of the 41st International Conference on Machine Learning},
  pages     = {11905--11917},
  year      = {2024},
  volume    = {235},
  series    = {Proceedings of Machine Learning Research},
  publisher = {PMLR},
  url       = {https://proceedings.mlr.press/v235/duan24a.html}
}

* Equal contribution · † Project lead