Portrait
Jinyang Du
Ph.D. Student
Beihang University · Computer Science and Technology
Advisors
About

I am a Ph.D. student in Computer Science and Technology at Beihang University, advised by Prof. Xianglong Liu and Prof. Ruihao Gong. My research focuses on efficient training technologies for world models, with interests in scalable training, model compression, and system-algorithm co-design for generative AI.

I received my B.S. in Computer Science and Technology from Beihang University, ranking in the top 10% of my cohort. My recent work includes a survey on low-bit LLMs, Triton-based low-bit FlashAttention operators, and near-lossless 4-bit LLM training.

Research
World Models Generative dynamics and simulation
Efficient Training Scalable and stable optimization
System Co-design Compression, kernels, and data pipelines

My work studies efficient training technologies for world models, spanning scalable optimization, data and systems efficiency, compression-aware training, and hardware-conscious implementation.

Current topics: world model training, efficient generative modeling, training system optimization, and model compression.

Education
  • Beihang University
    Beihang University
    School of Computer Science and Engineering
    Ph.D. Student, Computer Science and Technology
    Sep. 2025 - Present
  • Beihang University
    Beihang University
    School of Computer Science and Engineering
    B.S. in Computer Science and Technology · GPA 3.83/4 · Top 10%
    Sep. 2021 - Jun. 2025
Experience
  • Beihang University
    Beihang University
    Efficient LLMs and Model Quantization
    Worked on block-search quantization experiments for LLaMA-based models and surveyed low-bit LLM frameworks for a review paper.
    2023 - Present
  • Beihang University
    Beihang University
    Long-context Inference and Low-bit Operators
    Participated in a Baidu long-context project and designed low-bit mixed-precision operators for quantized inference acceleration.
    2024 - 2025
  • Huawei 2012 Laboratories
    Huawei 2012 Laboratories
    Research Internship
    Worked on photovoltaic AI development, including Pangu data processing and model training.
    2024
  • Beihang University
    Beihang University
    Low-precision Training and Video Segmentation Quantization
    Contributed to a Huawei low-precision training project, the Hif8 numerical format specification, and simulated quantization for SAM2 video segmentation.
    2024 - Present
Honors & Awards
  • LPCV 2026 Track 1 & Track 2 Champion
    2026
  • Graduate Academic Scholarship, ranked 1st
    2025
  • Academic Excellence Scholarship, Discipline Competition Scholarship, Social Work Scholarship, Social Practice Scholarship
    Undergraduate
  • Qian-Shen Scholarship and Lenovo Scholarship
    Undergraduate
  • First Prize, Chinese Mathematics Competition; Beijing First Prize, Contemporary Undergraduate Mathematical Contest in Modeling
    Undergraduate
  • Fengru Cup Main Track, Third Prize as first author and Second Prize as second author
    Undergraduate
  • University-level Merit Student, Outstanding Student Cadre, Outstanding Student, and Outstanding Party School Trainee
    Undergraduate
News
2026
Half-S was accepted to ACL Findings 2026, focusing on near-lossless 4-bit LLM training.
Mar 20
2025
Low-bit FlashAttention Accelerated Operator Design Based on Triton appeared at ICCVW 2025.
Oct 01
The low-bit LLM survey was published in Neural Networks, covering basics, systems, and algorithms for efficient LLM quantization.
Jul 10
Selected Publications (view all )
Half-S: Halving the Scale for Near-Lossless 4-Bit LLM Training
Half-S: Halving the Scale for Near-Lossless 4-Bit LLM Training

Jinyang Du, Ruihao Gong#, Linghan Ai, Zining Wang, Yunke Peng, Yao Wang, Lei Yan, Xuefei Wang, Yaoyuan Wang, Jinyang Guo, Dahua Lin, Xianglong Liu (# corresponding author)

Findings of the Association for Computational Linguistics (ACL Findings) 2026 First Author ACL Findings

Half-S revisits FP4 scaling for heavy-tailed LLM tensors and proposes a minimal scale correction that improves quantization grid utilization for practical near-lossless 4-bit training.

[OpenReview]

Half-S: Halving the Scale for Near-Lossless 4-Bit LLM Training

Jinyang Du, Ruihao Gong#, Linghan Ai, Zining Wang, Yunke Peng, Yao Wang, Lei Yan, Xuefei Wang, Yaoyuan Wang, Jinyang Guo, Dahua Lin, Xianglong Liu (# corresponding author)

Findings of the Association for Computational Linguistics (ACL Findings) 2026 First Author ACL Findings

Half-S revisits FP4 scaling for heavy-tailed LLM tensors and proposes a minimal scale correction that improves quantization grid utilization for practical near-lossless 4-bit training.

[OpenReview]

Low-bit FlashAttention Accelerated Operator Design Based on Triton
Low-bit FlashAttention Accelerated Operator Design Based on Triton

Jinyang Du, Jinyang Guo, Yifu Ding, Xianglong Liu

IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) 2025 First Author

This work designs low-bit FlashAttention operators with Triton, using operator fusion and mixed-precision execution to improve long-context quantized inference efficiency.

[DOI]

Low-bit FlashAttention Accelerated Operator Design Based on Triton

Jinyang Du, Jinyang Guo, Yifu Ding, Xianglong Liu

IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) 2025 First Author

This work designs low-bit FlashAttention operators with Triton, using operator fusion and mixed-precision execution to improve long-context quantized inference efficiency.

[DOI]

A Survey of Low-bit Large Language Models: Basics, Systems, and Algorithms
A Survey of Low-bit Large Language Models: Basics, Systems, and Algorithms

Ruihao Gong, Yifu Ding, Zining Wang, Chengtao Lv, Xingyu Zheng, Jinyang Du, Yong Yang, Shiqiao Gu, Haotong Qin, Jinyang Guo, Dahua Lin, Michele Magno, Xianglong Liu

Neural Networks 2025 Journal

This survey reviews low-bit quantization for large language models from basic formats, system support, and algorithmic strategies, connecting practical toolchains with future efficient LLM deployment.

[DOI] [arXiv]

A Survey of Low-bit Large Language Models: Basics, Systems, and Algorithms

Ruihao Gong, Yifu Ding, Zining Wang, Chengtao Lv, Xingyu Zheng, Jinyang Du, Yong Yang, Shiqiao Gu, Haotong Qin, Jinyang Guo, Dahua Lin, Michele Magno, Xianglong Liu

Neural Networks 2025 Journal

This survey reviews low-bit quantization for large language models from basic formats, system support, and algorithmic strategies, connecting practical toolchains with future efficient LLM deployment.

[DOI] [arXiv]

All publications
Leadership & Campus Service
Leader, Dandelion Rural Practice Team

Led field practice trips to Zhongyang, Shanxi and Yu County, Hebei.

Student Union Activity Practice Department

Helped organize more than 10 activities and completed over 250 hours of volunteer service.

College and Residential College Service

Served as organization committee member, cohort academic representative, and class coordinator.