About
Welcome to Yongchao He's homepage!
I’m a researcher and engineer working on LLM systems and AI infrastructure. I build systems that keep large models alive — and sometimes make them faster, from distributed training to LLM inference and all the little optimizations that help big things scale.
My work sits at the intersection of distributed systems, runtime design, and large-scale LLM training and inference, with a focus on system-level abstractions, performance trade-offs, and system–model co-design.
I earned my Ph.D. from the Institute for Interdisciplinary Information Sciences (IIIS, 交叉信息研究院) at Tsinghua University, where I was advised by Wei Xu and Wenfei Wu.
Occasional cat wrangler, full-time system tinkerer. Just another small node in the AI era.
Selected Publications
View All →LMTracer: Fine-Grained and Real-Time Performance Profiling for Production LLM Systems
Wei Liu,Yongchao He,Bohan Zhao,Hongyi Wang,Zhenhua Li,Junping Zhao
ConfSpec: Efficient Step-Level Speculative Reasoning via Confidence-Gated Verification
Liu Siran,Yongchao He*
Umap: Revisiting Memory-mapped I/O on Distributed File Systems for Efficient Matrix Access
Yongchao He,Guangyan Zhang,Zane Cao,Wenfei Wu
MegatronApp: Efficient and Comprehensive Management on Distributed LLM Training
Bohan Zhao,Guang Yang,Shuo Chen,Ruitao Liu,Tingrui Zhang,Yongchao He,Wei Xu
A Generic Service to Provide In-Network Aggregation for Key-Value Streams
Yongchao He,Wenfei Wu*,Yanfang Le,Ming Liu,ChonLam Lao
News
🎉🎉 LMTracer accepted in SOSP 2026!
🎉🎉 UniSparse accepted in ICML 2026!
🎉🎉 Two papers accepted in ACL 2026!
🎉🎉 Umap accepted in OSDI 2026!
⚙️⚙️ I founded and lead the AI Infrastructure R&D group at UbiQuant.
🏆🏆 Selected as the first member of UbiQuant’s CTO List, a high-potential technical leadership program.
🚀🚀 I started my position at UbiQuant under the Wutong Program (梧桐计划), a premier Technical Talent Program.
🎓🎓 I earned my Ph.D. in Computer Science!
🏆🏆 Our paper titled 'A Generic Service to Provide In-Network Aggregation for Key-Value Streams' received the Distinguished Paper Award at ASPLOS 2023.