Zelei Shao

MS student in Statistics at the University of Illinois Urbana-Champaign, building practical and efficient machine learning systems.

View Publications Email Me Resume
Portrait of Zelei Shao

About

I'm working at the intersection of RL training infrastructure and LLM inference optimization.

Most recently, I've been working on DAS (Distribution-Aware Speculative Decoding for RL Training), which adapts speculative decoding to the unique distribution shift that arises during on-policy RL rollouts. The work integrates with verl and vLLM, and tackles GPU workload balancing for tree-structured speculation under heterogeneous sequence lengths. I'm currently based in the Bay Area.

Publications

More details on OpenReview profile.

Contact