Weiyan Wang (王伟俨)

Weiyan Wang (王伟俨)

Senior R&D Engineer at TEG, Tencent. PhD from CSE, HKUST.

Weiyan Wang (王伟俨)

Nothing is more practical than a good theory. —— Kurt Lewin
What I cannot create, I do not understand. —— Richard Feynman

Hi there! I am a senior R&D engineer at Hunyuan, Tencent, where I fortunately get support from Project Up (青云计划). I focus on the co-design of both systems and algorithms for large-scale machine learning systems. By exploring the large design space combining system and algorithm, I try to find better ways to make big models trained on big data really happen. I advocate the philosophy proposed by Wang Yangming, which regards the knowledge (i.e. abstract theory) and the actions(i.e. concrete practices) as one (知行合一).

Brief Biography

Currently, I work in Tencent’s Large Model Department, focusing on improving the scalability and efficiency of RL training. My work also includes accelerating pretraining phase and improving sample efficienc for the large models, especially the multimodal generative models.

I recieved my Ph.D. from Computer Science and Engineering, Hong Kong University of Science and Technology, advised by Prof. Kai Chen. I obtained my M.Phil. degree in Computer Software and Theory from Institute of Software, Chinese Academy of Sciences, under the supervision of Prof. Yunquan Zhang and Dr. Guoping Long. My B.Eng. degree in Computer Science and Technology is from Huazhong University of Science and Technology.

Research Interests

Co-design in both System and Algorithm especially for:

  • Efficient large-scale training
  • Efficient model inference for real-time requirements
  • Real big data without perfect supervision (e.g. sparse/scarce, noisy, and out-of-distribution labels/rewards)

Selected Publications

HY Multi-Modal Seriers

  • HunyuanVideo-1.5: A Leading Lightweight Video Generation Model, Technical Report 2025
    Core Contributor: Weiyan Wang | GitHub
    To the best of my knowledge, the first large diffusion model pretrained by the Muon optimizer.
  • HunyuanImage-3.0: A Powerful Native Multimodal Model for Image Generation, Technical Report 2025
    Contributor: Weiyan Wang | GitHub
  • HunyuanVideo: A Systematic Framework For Large Video Generative Models, Technical Report 2024
    Core Contributor: Weiyan Wang | GitHub
  • Hunyuan-DiT: A Powerful Multi-Resolution Diffusion Transformer with Fine-Grained Chinese Understanding, Technical Report 2024
    Core Contributor: Weiyan Wang | GitHub

the Selected Works for Efficient Training

  • Surge Phenomenon in Optimal Learning Rate and Batch Size Scaling, NeurIPS 2024(CCF-A)
    Shuaipeng Li, Penghao Zhao, Hailin Zhang, Samm Sun, Hao Wu, Dian Jiao, Weiyan Wang, Chengjun Liu, Zheng Fang, Jinbao Xue, Yangyu Tao, Bin CUI, Di Wang
  • Addressing Network Bottlenecks with Divide-and-Shuffle Synchronization for Distributed DNN Training, Infocom 2022 (CCF-A, co-design in non-convex optimization and network)
    Weiyan Wang, Cengguang Zhang, Liu Yang, Kai Chen, Kun Tan
  • Rethinking transport layer design for distributed machine learning, APNet 2019（CCF-C）
    Jiacheng Xia, Gaoxiong Zeng, Junxue Zhang, Weiyan Wang, Wei Bai, Junchen Jiang, Kai Chen
  • Quantifying the Performance of Federated Transfer Learning, IJCAI FL Workshop 2019 (Best Student Paper)
    Qinghe Jing, Weiyan Wang, Junxue Zhang, Han Tian, Kai Chen

the Selected Works for Efficient Inference

  • Exploiting Student Parallelism for Low-latency GPU Inference of BERT-like Models in Online Services, KDD 2025 (CCF-A, co-design in model and system)
    Weiyan Wang, Yilun Jin, Yiming Zhang, Victor Junqiu Wei, Han Tian, Li Chen, Jinbao Xue, Yangyu Tao, Di Wang, Kai Chen
  • Enabling Edge-Cloud Video Analytics for Robotics Applications, Infocom 2021 (CCF-A)
    Yiding Wang, Weiyan Wang, Duowen Liu, Xin Jin, Junchen Jiang, Kai Chen
  • CLSIFT: An Optimization Study of the Scale Invariance Feature Transform on GPUs, HPCC 2013 (CCF-C, GPU kernel)
    Weiyan Wang, Yunquan Zhang, Guoping Long, Shengen Yan, Haipeng Jia
  • Parallelization and Performance Optimization on Face Detection Algorithm with OpenCL: A Case Study, Tsinghua Science and Technology (SCI Journal, GPU kernel)
    Weiyan Wang, Yunquan Zhang, Shengen Yan, Ying Zhang, Haipeng Jia

the Selected Works about Imperfect Supervision

  • BeamVQ: Metric-Guided Latent Search for Physical Spatiotemporal Forecasting, Technical Report 2026
    Hao Wu, Xingjian Shi, Weiyan Wang (Corresponding)
  • RoboAlign-R1: Distilled Multimodal Reward Alignment for Robot Video World Models, NeurIPS 2026 (CCF-A, To appear)
    Hao Wu, Yuqi Li, Yuan Gao, Fan Xu, Fan Zhang, Kun Wang, Penghao Zhao, Qiufeng Wang, Yizhou Zhao, Weiyan Wang(Corresponding), Yingli Tian, Xian Wu, Xiaomeng Huang
  • Prometheus: Out-of-distribution Fluid Dynamics Modeling with Disentangled Graph ODE, ICML 2024（CCF-A）
    Hao Wu, Huiyuan Wang, Kun Wang, Weiyan Wang, ChanganYe, Yangyu Tao, Chong Chen, Xian-Sheng Hua, Xiao Luo
  • Multi-task Learning Based Keywords Weighted Siamese Model for Semantic Retrieval, PAKDD 2023 (CCF-C)
    Mengmeng Kuang, Zhenhong Chen, Weiyan Wang (Corresponding), Lie Kang, Qiang Yan, Min Tang, Penghui Hao
  • Efficient Two-stage Label Noise Reduction for Retrieval-based Tasks, WSDM 2022 (CCF-B)
    Mengmeng Kuang, Weiyan Wang, Zhenhong Chen, Lie Kang, Qiang Yan

See the full Publications page for all 23 papers.

Professional Services

Reviewer

  • 2026    KDD Reviewer
  • 2025    NeurIPS Reviewer
  • 2025    Journal Reviewer: Performance Evaluation
  • 2025    KDD Commendable Reviewer
  • 2023    Globalcom （AINextGenWN Workshop） Reviewer
  • 2020    Sigcomm Artifact Reviewer

Open Source Code Contributor

CSDN Forum Moderator

Contact

Miscellaneous

I am a badminton fan and amateur player, especially for men’s doubles. I prefer combining drop shots with smashes, using an even-balance long-size Duora racket from Yonex. If I have time, I also like reading, hiking, and watching football.