vLLM 2025
Contributor Fixed rope scaling override order so Hugging Face config conversions stay correct after legacy overrides.
LLM serving inference bugfix
Projects & Open Source
Open Source Contributions
VERL (Volcano Engine RL for LLMs) 2025
Contributor Extended-context fixes and truncation improvements for RL pipelines (YaRN), with tests across rollout workers.
RL long context tooling
RULER / InfiniteBench / Retrieval Head 2024
Contributor Long-context evaluation and analysis: reproducible data synthesis, 128k LLaMA3-7B single‑GPU inference, and Phi3 retrieval-head support.
evaluation long context LLM
Competitions & Demos
Kaggle LMSYS – Chatbot Arena Preference Prediction Aug 2024
Team Lead & Algorithm Designer Built LoRA‑fine‑tuned models with pseudo‑labeling and ensembles; earned Silver Medal (39/1,849 teams).
Kaggle preference modeling LoRA
LLM‑MCTS Reasoning from Scratch Jan 2025
Lead & Algorithm Designer Lightweight Monte‑Carlo Tree Search pipeline for LLM agents; supports OpenAI/DeepSeek APIs with optimized rollout policy.
LLM MCTS agents
Architect & Lead Programmer
2.5D narrative adventure; produced teaser and achieved charity certification from Beijing New Sunshine Charity Foundation.
game dev narrative unity