During February 2026, Xuejian Qian developed dynamic LoRA weights management for runtime deployment in the ModelTC/LightX2V repository. He engineered a system using Python that enables on-the-fly loading and removal of LoRA adapters, allowing model behavior to be reconfigured in production without requiring a full model reload. By implementing robust handling for LoRA parameters and file paths, he improved deployment agility and reduced operational downtime. His work focused on machine learning model deployment, decoupling LoRA changes from the core model to support safer and more flexible updates. This feature addressed the need for rapid, low-risk configuration changes in production environments.