About me

Ph.D. Candidate @ University of Science and Technology of China · Advisor: Prof. Xiang Wang [Google Scholar]

Hi! I am Jie Sun, a fourth-year Ph.D. candidate at USTC, advised by Prof. Xiang Wang and jointly trained at Shanghai Innovation Institute. My research focuses on LLM post-training, spanning agent trajectory and environment curation, and on-policy distillation (including Model Merging). I am also interested in auto-research / AI4AI and benchmarks. I have authored 8 first-author papers and conducted research internships at Tencent and Ant Group.

I am happy to chat about research and collaborations, and always welcome in-person coffee chats! Feel free to reach out at sunjie2019@mail.ustc.edu.cn.

📖 Education

📰 News

📝 Distill What You Trust introduces reliability-aware multi-teacher on-policy distillation.

🎉 SimCT is accepted to NeurIPS 2026!

🎉 Guided-OPD is accepted to EMNLP 2026!

💻 EasyOPD: a unified framework for on-policy distillation.

🎉 SepSeq is accepted to ACL 2026!

💼 I joined Tencent as an LLM Algorithm Intern in the Foundation Model Department.

🎓 I joined Shanghai Innovation Institute for joint Ph.D. training with USTC.

📝 Publications

First- and co-first-author papers are listed first, followed by co-authored work. Click a topic to filter. My name is bold; * denotes equal contribution. See also Google Scholar.

💼 Experience

Apr 2026 — Present

Tencent / Foundation Model Department

LLM Algorithm Intern · Project Up (Qingyun Program)

  • On-policy distillation for large-to-small capability transfer and model merging: SimCT recovers cross-tokenizer supervision; TrustMOPD allocates label-free multi-teacher supervision.
  • Developed EasyOPD, a modular verl-based framework supporting cross-tokenizer, self-, and step-wise distillation for in-house OPD research.
Sep 2024 — Jul 2025

Ant Group

LLM Algorithm Intern · Post-training

  • Developed γ-PO with instance-adaptive target margins and a label-smoothing interpretation, adding little training overhead.
  • Studied training-free long numerical sequence processing in SepSeq through attention analysis and separator insertion, improving accuracy and generation efficiency.
Jul 2023 — Sep 2023

Lingjun Invest

Quantitative Algorithm Intern

High-frequency alpha factors, feature engineering, and factor evaluation.

Apr 2023 — Jul 2023

X Asset Management

Quantitative Research Intern

Robust stock prediction under distribution shift and out-of-distribution generalization.

🏆 Awards

  • USTC First-Prize Academic Scholarship
  • Kaggle Optiver — Trading at the Close · Silver Medal, Top 3% Globally
  • 2nd USTC Quantitative Investment Competition · 2nd Place / ~100 Teams
  • USTC Outstanding Graduate
  • USTC Outstanding Student Scholarship · Silver Award, Top 10%
  • National Middle School Physics Olympiad · First Prize
  • Zhejiang Provincial Middle School Mathematics Competition · First Prize
  • Pearl River Delta Physics Competition · First Prize