Qunhong Zeng

profile.jpg

Beijing, China

qunhongzeng@gmail.com

I am a second-year graduate student at the School of Computer Science and Technology at Beijing Institute of Technology, advised by Prof. Yuxia Zhang. Prior to this, I received my B.S. degree in Computer Science from the same university in 2023.

My research primarily focuses on the application of LLMs to SE fields. Recently, I have concentrated on reinforcement learning (RL) approaches to enhance LLMs’ reasoning capabilities, with the goal of improving their performance on automatic SE tasks such as code generation.

I am also interested in MLSys and enjoy building scalable systems. I had several MLSys internships and recently I’ve been actively contributing to the verl community. I believe RL represents a critical pathway toward AGI and it’s an art of combining research and engineering.

Please feel free to reach out if you’re interested in having a discussion :)

Publications

  1. A First Look at Conventional Commits Classification
    Qunhong Zeng, Yuxia Zhang, Zhiqing Qiu, and Hui Liu
    In Proceedings of the IEEE/ACM 47th International Conference on Software Engineering, 2025
  2. COLARE: Commit Classification via Fine-grained Context-aware Representation of Code Changes
    Qunhong Zeng, Yuxia Zhang, Zeyu Sun, Yujie Guo, and Hui Liu
    In 2024 IEEE International Conference on Software Analysis, Evolution and Reengineering (SANER), 2024

Experiences

  1. ByteDance, MarsCode CodeAI Intern, 2024.10 - now
    • Conducted research on commit message generation and commit message quality evaluation.

    • Currently researching reinforcement learning to enhance LLM’s reasoning capabilities, such as mathematical reasoning, competitive programming contest test generation, and tool-interative reasoning. Collaborating with Zexiong Ma and mentored by Bo Jiang.

  2. Oneflow, Framework Intern, 2024.3 - 2024.9
    • Extended deep learning framework oneflow compatibility beyond CUDA to diverse hardware accelerators (NPU/XPU), implementing Ascend CANN/AscendC-based operators that enabled end-to-end ResNet, GPT2, and Llama model inference/training on Ascend chips.

    • Managed framework infrastructure including Docker containerization, CI/CD, cross-platform compilation, and Manylinux-compliant Python/C++ extension packaging.

  3. Momenta, HD-Map Backend Intern, 2022.9 - 2023.1
    • Engineered high-performance Python/C++ extensions for HD maps algorithm library, delivering 20x performance improvement.

  4. ByteDance, AI-Lab Intern, 2021.12 - 2022.6
    • Contributed to ByteDance’s machine learning platform development as part of the engineering team.