Chaofan Lin (林超凡)

I'm a third-year Ph.D. student in Institute for Interdisciplinary Information Sciences (IIIS), Tsinghua University advised by Prof. Mingyu Gao. I'm now also a research intern @ DeepSeek-AI Infra Team. Previously I received my my B.Eng. in Computer Science from ACM Honors Class, Shanghai Jiao Tong University. advised by Prof. Yong Yu. During my undergraduate years, I had a wonderful time working with Yuqing Yang @ System Group of Microsoft Research Asia, focusing on LLM serving; and Tianqi Chen @ Apache TVM community, doing research on Deep Learning Compilers. Please check out my CV if interested. I go by `SiriusNEO` on most websites. And I usually use `Chaos` as a nickname (a variant of my real name). (Github | Zhihu | Scholar)

Research

Publications

• Twilight: Adaptive Attention Sparsity with Hierarchical Top-p Pruning
  

  Chaofan Lin, Jiaming Tang, Shuo Yang, Hanshuo Wang, Tian Tang, Boyu Tian, Ion Stoica, Song Han, Mingyu Gao†

  The Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS'25 Spotlight)

  [arXiv] | [code]
• Parrot: Efficient Serving of LLM-based Applications with Semantic Variable
  

  Chaofan Lin, Zhenhua Han†, Chengruidong Zhang, Yuqing Yang†, Fan Yang, Chen Chen†, Lili Qiu

  18th USENIX Symposium on Operating Systems Design and Implementation (OSDI'24)

  [🤗Hugging Face Daily Papers] | [arXiv] | [code] | [conference page]
• Accelerating Large-Scale Reasoning Model Inference with Sparse Self-Speculative Decoding
  

  Yilong Zhao*, Jiaming Tang*, Kan Zhu, Zihao Ye, Chi-Chih Chang, Chaofan Lin, Jongseok Park, Guangxuan Xiao, Mohamed S. Abdelfattah, Mingyu Gao, Baris Kasikci, Song Han, Ion Stoica

  Ninth Annual Conference on Machine Learning and Systems (MLSys'26)

  [arXiv] | [code]
• SparseForge: Efficient Semi-Structured LLM Sparsification via Annealing of Hessian-Guided Soft-Mask
  

  Hanzuo Liu, Chaofan Lin, Weixuan Sun, Yulong Wang, Key, Rayying, Mingyu Gao

  (Under review)

  [arXiv] | [code]
• Pedestal: Fully Disaggregated LLM Serving with Computational Memory Pool
  

  (Under review)
Tech Reports.
• DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression
  

  DeepSeek-AI. (As a contributor)

  [arXiv] | [🤗V41 Flash 1017] | [🤗Hugging Face Repo]
• DeepSeek Elastic Compute (DSec): A Sandbox Infrastructure for Effective Agentic Training at Scale
  

  Jialiang Huang, Hongxuan Tang, Jingchang Chen, ..., Chaofan Lin, ..., Mingxing Zhang, Liyue Zhang, Panpan Huang, Wenfeng Liang

  [arXiv]
• DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence
  

  DeepSeek-AI. (As a contributor)

  [arXiv] | [🤗V4 Flash 0731] | [🤗Hugging Face Repo]
• DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models
  

  DeepSeek-AI. (As a contributor)

  [arXiv] | [🤗Hugging Face Repo]

News

Contact

Miscellaneous