About
Building better foundation models through better data.
I am currently a principal researcher on the HY Foundation Model Team at Tencent, where I lead the overall mid-training efforts for both vision-language models (VLMs) and large language models (LLMs), including CT, CPT, and long-context training. I have published more than 40 papers at leading conferences, including more than 20 papers as first or co-first author. My research focuses on how synthetic data can be used effectively to scale the capabilities of foundation models.
Previously, I contributed to the Qwen2.5 and Qwen3 model series. Graph-Reasoner is a collection of projects on graph reasoning; please feel free to explore our work.
Collaborate
Research & Hiring
We are looking for full-time researchers and self-motivated interns with strong hands-on skills and an interest in LLM training, data-centric AI, and large-scale experimentation.
If you are interested in data scaling laws, synthetic data generation, or mid-training strategies for LLMs, please get in touch.
Start a conversation- 01Data quality
Improving the quality and diversity of synthetic data.
- 02Data mixtures
Understanding the optimal mix of synthetic and real data.
- 03Scaling strategies
Developing effective scaling strategies for synthetic data.
- 04Data substitution
Exploring when synthetic data can replace real data.
Updates
News
- Our paper LoopRPT was accepted to NeurIPS 2026.
- Two papers were accepted to ICML 2026 and ACL 2026.
- Two papers were accepted to ICLR 2026.
- One paper was accepted to NeurIPS 2025.
- I presented a tutorial on role-playing at IJCAI 2025.
- One paper was accepted to COLM 2025.
- One paper was accepted to KDD 2025, and three papers were accepted to ACL 2025.
- I received the 2024 Baidu Scholarship.
- One paper was accepted to COLING 2025 as an oral presentation.
- Two papers were accepted to EMNLP 2024.
- I interned with the Qwen Foundation Model Team, focusing on pre-training methods for improving LLM reasoning.
- We released our survey on role-playing with language models: The Oscars of AI Theater.
- Served as a reviewer for EMNLP 2024.
- One paper was accepted to KDD 2024.
- We released two papers on arXiv: MathOctopus and LLaMA-Probing.
- Served as a reviewer for NAACL 2024 and ACL 2024.
- Three papers were accepted to EMNLP 2023.
- Served as a reviewer for IEEE TPAMI.
- Two papers were accepted to ACL 2023.
Research output
Publications
Selected work and a complete list synchronized with Google Scholar.
Recognition
Awards
Community
Academic Service
- Invited talk at Shanghai University of Engineering Science.
- Reviewer for ACM MM, ACL, EMNLP, AAAI, NeurIPS, NAACL, ICLR, IEEE TPAMI, and COLM.
Background
Education
- 2022–PresentPh.D. Candidate, DSA, HKUST
- 2019–2022Master’s degree in Computer Science, Peking University
- 2015–2019Bachelor’s degree, Beijing University of Posts and Telecommunications
Get in touch
Let’s explore ideas together.
I welcome research collaborations and inquiries from prospective researchers and interns.