Zhoujun (Jorge) Cheng     承洲骏

Glad to meet you here! I am Zhoujun. I am now a Ph.D. student at UC San Diego, advised by Prof. Zhiting Hu.

Prior to this, I received my B.E. and M.E. degrees in Computer Science (IEEE class) from Shanghai Jiao Tong University. I also spent great times in HKUNLP with Prof. Tao Yu and Microsoft Research Asia with Haoyu Dong.

 /   /   /   / 

My research focuses on scaling RL and agents along multiple axes:

• Compute (Compute-Optimal LLM RL): how to optimally allocate RL rollout compute as it scales;
• Data (Guru): scaling RL training data across domains and sizes for general reasoning;
• Environments (NanoRollout): scaling parallel environments for digital agents (coding & computer-use) for large-scale agent RL and distillation.

I am also a core developer of foundation models like Mocha-Coder-32B and K2-V2-70B, which apply these insights at production scale. I also contribute to multiple real-world agent evaluations, including CocoaBench, OSWorld, and BigCodeBench. I believe good evaluation shapes model capabilities.