Zirui Zhang

LLM Training Engineer — Post-training, Reinforcement Learning, and Training Systems

I work on large-scale language model training, with experience in pre-training, post-training, reinforcement learning, and training infrastructure. My current interests include scalable RL for reasoning and self-improving agents.

Email · GitHub · Hugging Face

Experience

AMD — LLM Training

Large-scale LLM training and training-system optimization on AMD GPUs, including kernel optimization.

01.AI — Pre-training and Post-training

• Worked on Yi-Nano pre-training, including model architecture and training strategy optimization.
• Built and optimized post-training pipelines for Yi-Large and Yi-Lightning, including preference optimization and online/offline reinforcement learning.
• Explored reinforcement learning for reasoning, code generation, and tool use.

Alibaba — LLM Engineer

Worked on continued pre-training, retrieval augmentation, supervised fine-tuning, and alignment for LLM-based e-commerce customer service.

Research Interests

• Scalable post-training and reinforcement learning
• Reasoning, coding, and tool-using models
• Efficient large-scale training systems
• Self-improving agents

Education

M.S. in Computer Science, Chinese Academy of Sciences.