Zongqian Li

😃 About Me

I am Zongqian Li, a PhD student in Natural Language Processing at Cambridge. I am interested in large language model, efficiency, agent, and multimodality. I have an interdisciplinary background in computer science, physics, and materials science. I also have internships in research institutes and finance companies.

 

📕 Education

• [2024.01-Now] Univeristy of Cambridge
  PhD in Natural Language Processing
  Supervisor: Prof. Nigel Collier
  Fully funded by Cambridge International Scholarship, Cambridge Trust Scholar

• [2023.10-2024.01] University of Cambridge
  MPhil in Physics
  Supervisor: Prof. Jacqueline Cole
  Member of St John's College

• [2018.09-2022.06] Queen Mary University of London
  BEng with Hons in Materials Science and Engineering
  First Class with Honours, Rank 1/234 in school
  Distinguished Student Scholar (Top 0.03%)

 

🏔️ Experiences

• [2025.06-Now] Microsoft Research Asia
  Research Intern
  Supervisor: Dr. Shaohan Huang, Dr. Furu Wei

• [2024.09-2024.10] ELE, Alibaba Group
  Delivery Driver

• [2023.07-2024.01] ZhenFund
  Top 2 Early-stage VC in China
  Investment Intern
  Mentor: Partner Yuan Liu

• [2023.04-2023.07] Deloitte
  NLP Developer
  Mentor: Associate Chenglong Li

• [2022.05-2022.10] Tsinghua University
  Department of Computer Science and Technology
  Research Intern
  Supervisor: Prof. Jie Tang, Prof. Juanzi Li, Dr. Jifan Yu

 

📝 Publications

Topics: Efficiency ^E, Model Training ^T, Data ^D, Reasoning ^R, Survey ^S, Evaluation ^V

• Breaking Training Bottlenecks: Effective Reinforcement Learning for Modern Coding Models ^TR
  Zongqian Li, Shaohan Huang, Zewen Chi, Yixuan Su, Lexin Zhou, Li Dong, Nigel Collier, Furu Wei
  Under Review

• MicroCoder-Dataset: A Latest and Challenging Dataset for Training Modern Coding Models
  Zongqian Li, Shaohan Huang, Tengchao Lv, Lei Cui, Nigel Collier, Furu Wei
  Under Review

• Flexi-LoRA: Efficient LoRA Finetuning with Input-Aware Dynamic Ranks ^ET
  Zongqian Li, Yixuan Su, Han Zhou, Zihao Fu, Nigel Collier
  Under Review

• PT-MoE: An Efficient Finetuning Framework for Integrating Mixture-of-Experts into Prompt Tuning ^ET [Paper] [Page] [Github]
  Zongqian Li, Yixuan Su, Nigel Collier
  NeurIPS 2025 Main

• 500xCompressor: Generalized Prompt Compression for Large Language Models ^ET [Paper] [Page] [Github]
  Zongqian Li, Yixuan Su, Nigel Collier
  ACL 2025 Main Long

• Prompt Compression for Large Language Models: A Survey ^ES [Paper] [Page] [Github]
  Zongqian Li, Yinhong Liu, Yixuan Su, Nigel Collier
  NAACL 2025 Main Long (Selected Oral)

• Auto-generating Question-answering Datasets with Domain-specific Knowledge for Language Models in Scientific Tasks ^ETD [Paper]
  Zongqian Li, Jacquiline Cole
  Digital Discovery (Q1)

• ReasonGraph: Visualization of Reasoning Methods and Extended Inference Paths ^R [Paper] [Demo] [Page] [Github]
  Zongqian Li, Ehsan Shareghi, Nigel Collier
  ACL 2025 Demo (Github 500+ Stars)

• A Survey on Prompt Tuning ^ES [Paper] [Page] [Github]
  Zongqian Li, Yixuan Su, Nigel Collier
  ICML 2025 Workshop

• General Scales Unlock AI Evaluation with Explanatory and Predictive Power ^V [Paper] [Page]
  Lexin Zhou, Lorenzo Pacchiardi, Fernando Martínez-Plumed, Katherine M. Collins, Yael Moros-Daval, ..., Zongqian Li, ..., Peter Henderson, Sherry Tongshuang Wu, Patrick C. Kyllonen, Lucy Cheke, Xing Xie, José Hernández-Orallo
  Nature