Taiqiang Wu

Biography

I am now a PH.D. student from the NGai Lab in HKU. Before that, I received my master degree from Tsinghua University in 2023. For master, I studied in IIGroup in Tsinghua University, supervised by Prof. Yujiu Yang. In 2020, I received my bachelor degree in Department of Automation from Tsinghua University. My major research interests lie in Effcient Model methods for Large Language Models. Recently, I focus on the efficient reasoning via RL.

Recent News

• 2026.02 - Release one report on LLM efficient reasoning via reward shaping. We losslessly compress CoT of Qwen3 models. [Project]
• 2025.11 - Present a talk about "Rethinking LLM Post-train at Weight Level". [TechBeat] [Youtube] [Slide]
• 2025.03 - The paper A Survey on the Honesty of Large Language Models is accepted by TMLR!
• 2025.02 - Release one blog about the KL divergence in RL algorithm. [Blog(中)]
• 2025.01 - Release one report for the unified view of Attention and MoE. [Report][Blog(中)]
• 2024.11 - Two papers are accepted by COLING 2025.
• 2024.11 - Release one report for modeling Binary Quantization via Convex Optimization Methods (Project for HKU COMP9602, 2023). [Report]
• 2024.09 - One paper is accepted by EMNLP 2024.
• 2024.06 - One blog about improving LoRA is released. [Eng | 中]
• 2024.04 - One blog about rethinking KL divergence in KD for LLMs is released. [Eng | 中]
• 2024.03 - One paper is accepted by NAACL 2024 findings.
• 2023.02 - Two papers are accepted by ICASSP 2023.
• 2022.10 - One research long paper is accepted by WSDM 2023.

Publications & Preprints

• 🔥 The Art of Efficient Reasoning: Data, Reward, and Optimization

  • Taiqiang Wu, Zenan Xu, Bo Zhou, Ngai Wong
  • Arxiv Preprint [Paper] [Weights] [Blog]

• 🔥 Revisiting Model Interpolation for Efficient Reasoning

  • Taiqiang Wu, Runming Yang, Tao Liu, Jiahao Wang, Ngai Wong
  • Arxiv Preprint [Paper] [Code]

• Timber: Training-free Instruct Model Refining with Base via Effective Rank

  • Taiqiang Wu, Runming Yang, Tao Liu, Jiahao Wang, Zenan Xu, Ngai Wong
  • Arxiv Preprint [Paper] [Code] [Weight]

• Shadow-FT: Tuning Instruct Model via Training on Paired Base Model

  • Taiqiang Wu^*, Runming Yang^*, Jiayi Li, Pengfei Hu, Yik Chung WU, Ngai Wong, Yujiu Yang
  • Arxiv Preprint [Paper] [Code] [Weight]

• A Survey on the Honesty of Large Language Models

  • Siheng Li^*, Cheng Yang^*, Taiqiang Wu^*, Chufan Shi, Yuji Zhang, Xinyu Zhu, Zesen Cheng, Deng Cai, Mo Yu, Lemao Liu, Jie Zhou, Yujiu Yang, Ngai Wong, Xixin Wu, Wai Lam
  • TMLR 2025 Journal [Paper] [Code]

• Mixture-of-Subspaces in Low-Rank Adaptation

  • Taiqiang Wu, Jiahao Wang, Zhe Zhao, Ngai Wong
  • EMNLP 2024 Conference [Paper] [Code]

• Rethinking Kullback-Leibler Divergence in Knowledge Distillation for Large Language Models

  • Taiqiang Wu, Chaofan Tao, Jiahao Wang, Runming Yang, Zhe Zhao, Ngai Wong
  • CoLING 2025 Conference [Paper] [Code] [Video]

• Weight-Inherited Distillation for Task-Agnostic BERT Compression

  • Taiqiang Wu ^*, Cheng Hou ^*, Shanshan Lao, Jiayi Li, Ngai Wong, Zhe Zhao, Yujiu Yang.
  • NAACL 2024 findings Conference [Paper] [Code] [Poster]

• Edge-free but Structure-aware: Prototype-Guided Knowledge Distillation from GNNs to MLPs

  • Taiqiang Wu, Zhe Zhao, Jiahao Wang, Xingyu Bai, Lei Wang, Ngai Wong, Yujiu Yang
  • CoLING 2025 Conference [Paper]

Internship

• 2021.3~2022.5, Tencent
• 2022.5~2023.5, Tencent Rahio Research Plan

Services

• ARR2022, Reviewer
• ARR2023, Reviewer
• ARR2024, Reviewer
• ARR2025, Reviewer
• ARR2026, Reviewer

(Last updated on Mar, 2026)