Computer Science · Huazhong University of Science and Technology (HUST)
Deputy Team Leader · Unique Studio
I'm ZiLang, a Computer Science student at Huazhong University of Science and Technology and Deputy Team Leader of Unique Studio, a student-led technology and innovation team at HUST.
My projects explore large language models, AI agents, and action-conditioned world models. I learn by reading implementations, running experiments, and documenting what works—and what doesn't.
- LLMs & agents — Transformers, parameter-efficient fine-tuning, and ReAct-style tool use.
- World models — Action-conditioned video prediction and the structure of action responses.
- Learning through implementation — Reproducing ideas, examining training behavior, and sharing findings.
| Project | Focus |
|---|---|
| Action-Response Structure in ACWM-Phys | Investigating when action-conditioned world models justify specialist models; documenting experiments and negative results. |
| ReAct Fine-Tuning Pipeline | Exploring ReAct-format data generation and LoRA / QLoRA fine-tuning for tool use. |
| MiniMind Learning Notes | Studying Transformer and LLM implementations, with experiments on batch size, loss, and learning rate. |
At Unique Studio, I'm part of a community built around curiosity, hands-on learning, and collaboration.
Explore my repositories below, or find out more about our team at hustunique.com.