Xusheng Chen's Homepage

prof_pic.jpg

Email:

michael.xschen@gmail.com

I am Xusheng Chen. At Tencent, I lead a team building the inference infrastructure behind the R&D of the Hunyuan (HY) large language models. Broadly, my team owns inference across the model R&D lifecycle, with ongoing threads in model–infra co-design, high-precision and deterministic batch inference, and train–inference consistency and acceleration for RL rollout.

Before Tencent, I was at Huawei Cloud, where I led the team incubating XDeepServe, Huawei Cloud’s Model-as-a-Service (MaaS) platform running on the CloudMatrix SuperPod.

I received my Ph.D. from the University of Hong Kong, supervised by Dr. Heming Cui. My doctoral research was in distributed systems — consensus protocols, fault-tolerance systems, and distributed transaction processing, often built on datacenter networking hardware such as RDMA. My early work also explored systems for accelerating the training and inference of Transformer models.

I received my Bachelor of Engineering degree from the University of Hong Kong with first class honour, and enjoyed a wonderful exchange program at UCSD during my undergraduate study.

I am from Shenyang, a beautiful city in northeast China. See my Google Scholar for a full list of publications.

Selected publications

  1. Tech Report
    xDeepServe: Huawei Cloud Model-as-a-Service on the CloudMatrix384 SuperPod
    XDS Team
    2025
  2. ATC ’25
    DeepServe: Serverless Large Language Model Serving at Scale
    Junhao Hu, Jiang Xu, Zhixia Liu, Yulong He, Yuetao Chen, Hao Xu, Jiang Liu, Jie Meng, Baoquan Zhang, Shining Wan, Gengyuan Dan, Zhiyu Dong, Zhihao Ren, Changhong Liu, Tao Xie, Dayun Lin, Qin Zhang, Yue Yu, Hao Feng,  Xusheng Chen, and Yizhou Shan
    In 2025 USENIX Annual Technical Conference (USENIX ATC 25) 2025
  3. ATC ’25
    Toppings: CPU-Assisted, Rank-Aware Adapter Serving for LLM Inference
    Suyi Li, Hanfeng Lu, Tianyuan Wu, Minchen Yu, Qizhen Weng,  Xusheng Chen, Yizhou Shan, Binhang Yuan, and Wei Wang
    In 2025 USENIX Annual Technical Conference (USENIX ATC 25) 2025
  4. ICML ’25
    EPIC: Efficient Position-Independent Caching for Serving Large Language Models
    Junhao Hu, Wenrui Huang, Weidong Wang, Haoyi Wang, Tiancheng Hu, Qin Zhang, Hao Feng,  Xusheng Chen, Yizhou Shan, and Tao Xie
    In Proceedings of the 42nd International Conference on Machine Learning (ICML 2025) 2025
  5. TACO ’25
    ShuffleInfer: Disaggregate LLM Inference for Mixed Downstream Workloads
    Cunchen Hu, Heyang Huang, Liangliang Xu,  Xusheng Chen, Chenxi Wang, Jiang Xu, Shuang Chen, Hao Feng, Sa Wang, Yungang Bao, Ninghui Sun, and Yizhou Shan
    ACM Transactions on Architecture and Code Optimization 2025
  6. arXiv ’24
    MemServe: Context Caching for Disaggregated LLM Serving with Elastic Memory Pool
    Cunchen Hu, Heyang Huang, Junhao Hu, Jiang Xu,  Xusheng Chen, Tao Xie, Chenxi Wang, Sa Wang, Yungang Bao, Ninghui Sun, and Yizhou Shan
    arXiv preprint arXiv:2406.17565 2024
  7. SOSP ’21
    BIDL: A High-throughput, Low-latency Permissioned Blockchain Framework for Datacenter Networks
    Ji Qi* ,  Xusheng Chen*, Yunpeng Jiang, Jianyu Jiang , Tianxiang Shen , Shixiong Zhao , Sen Wang, Gong Zhang, Li Chen, Man Ho Au , and Heming Cui
    *Parallel First Author, equal contribution.
    In Proceedings of the ACM SIGOPS 28th Symposium on Operating Systems Principles (SOSP) 2021
  8. SoCC ’17
    APUS: Fast and Scalable Paxos on RDMA
    Cheng Wang , Jianyu Jiang ,  Xusheng Chen, Ning Yi, and Heming Cui
    In Proceedings of the 2017 Symposium on Cloud Computing (SoCC) 2017