About Me

Hi, I’m Lily (Xiaoxuan) Liu. I’m currently a researcher at OpenAI. Before joining OpenAI, I completed my PhD in Computer Science at UC Berkeley, where I was part of Sky Lab (formerly RISE/AMP Lab) and was fortunate to be advised by Professor Alvin Cheung and Professor Ion Stoica. I received my undergraduate degree from Peking University and my master’s degree from Carnegie Mellon University, where I had the pleasure of working with Andy and Huanchen.

Research Interest: My research is driven by curiosity. I currently work on pretraining algorithms, where the process of training models remains both mysterious and endlessly fascinating. Previously, I worked on inference for gpt-oss and post-training infra for gpt-5. During my PhD, I built systems for efficient LLM inference and contributed to vLLM as a member of the team.

Selected Manuscripts (full list)

• Speculative Decoding: Performance or Illusion?, MLsys 2026. Best Research Paper Honorable Mention.
  Xiaoxuan Liu^*, Jiaxiang Yu^*, Jongseok Park, Ion Stoica, Alvin Cheung
• TurboSpec: Closed-loop Speculation Control System for Optimizing LLM Serving Goodput, arxiv, 2024.
  Xiaoxuan Liu, Jongseok Park, Langxiang Hu, Woosuk Kwon, Zhuohan Li, Chen Zhang, Kuntai Du, Xiangxi Mo, Kaichao You, Alvin Cheung, Zhijie Deng, Ion Stoica, Hao Zhang
• Online Speculative Decoding, ICML 2024. [code][blog]
  Xiaoxuan Liu, Lanxiang Hu, Peter Bailis, Alvin Cheung, Zhijie Deng, Ion Stoica, Hao Zhang