Greetings! I am a 4th year PhD student from the Chinese University of Hong Kong (CUHK) and I am honored to be supervised by Prof. LAM, Wai. Previously, I received my Master degree from King’s College London.

My previous work generally focus on Parallel Context Encoding, Context Compression, Block Attention, and Internal mechanism of LLM (CoachHooK). Specifically, my recent work explored the generalization of block attention, which constructs a general semantic segmenter and designs a more efficiency and effective training paradigm for block attention. The general block attention could help the KV cache reuse across prompts and has broad application scenarios in Agents. I also studied an elegant way of parallel context encoding and compression (InComeS), which increases considerably the efficiency of ICL.

Feel free to contact (li_shuaiyi2021@outlook) if you want more discussion.

News

• The Huggingface Repo and ModelScope Repo of my recent work generalization of block attention have been released, feel free to check out.
• One paper has been accepted as a Main Conference to EMNLP 2026.

Publications

• Towards Generalization of Block Attention via Automatic Segmentation and Block Distillation
  - Enable extreme KV cache reuse across prompts
  - Has broad application scenarios in Long-horizon agent tasks like coding, multi-turn agentic workflows, etc
  Authors: Shuaiyi Li, Zhisong Zhang, Yan Wang, Lei Zhu, Dongyang Ma, Chenlong Deng, Yang Deng, Lam Wai
  NeurIPS 2026 Avg.4

• InComeS: Integrating Compression and Selection Mechanisms into LLMs for Efficient Model Editing
  - A finer token-level RAG
  - Increase the efficiency of ICL while maintaining its flexibility
  Authors: Shuaiyi Li, Zhisong Zhang, Yang Deng, ChenLong Deng, Tianqing Fang, Hongming Zhang, Haitao mi, Dong Yu, Lam Wai
  EMNLP 2026

• Attention Entropy is a Key Factor: An Analysis of Parallel Context Encoding with Full-attention-based Pre-trained Language Models
  Authors: Zhisong Zhang, Yan Wang, Xinting Huang, Tianqing Fang, Hongming Zhang, Chenlong Deng, Shuaiyi Li, Dong Yu
  ACL 2025 oral

• UniGist: Towards General and Hardware-aligned Sequence-level Long Context Compression
  Authors: Chenlong Deng, Zhisong Zhang, Kelong Mao, Shuaiyi Li, Tianqing Fang, Hongming Zhang, Haitao Mi, Dong Yu, Zhicheng Dou
  NeurIPS 2025

• Consecutive Batch Model Editing with HooK Layers
  Authors: Shuaiyi Li, Yang Deng, Deng Cai, Hongyuan Lu, Liang Chen, Lam Wai
  EMNLP 2024

Check my google scholar for full publication list.

Experience

March 2024 - March 2026, Tencent, Research Intern

• Mentor: Prof. Zhisong Zhang and Dr. Yan Wang
• Complete two projects about LLM. Topic in parallel encoding, block attention, and compression.

Community Service

Reviewer for ICML, ICLR, NeurIPS, ACL, and EMNLP.

Teaching

• Information System Management
  Type: Undergraduate course
  Time: 2025-2026 2nd term
  

• Computer Processing System Concepts
  Type: Undergraduate course
  Time: 2025-2026 1st term
  

• E-Commerce Data Mining Technique
  Type: Postgraduate course
  Time: 2023-2024 1st term, 2024-2025 1st term
  

• Multivariable Calculus for Engineers
  Type: Undergraduate course
  Time: 2022-2023 2nd term, 2023-2024 1st term