Developed an enhancement for the ultralytics/ultralytics repository, focusing on dynamic TensorRT export shape handling within the YOLO-World module. The work involved refining the guide tensor’s shape adjustment logic to support varying input sizes, thereby improving the reliability and scalability of model exports for deployment scenarios. Leveraging expertise in Python, PyTorch, and deep learning, the solution introduced explicit shape management to ensure compatibility with dynamic input dimensions during TensorRT export. This targeted feature addressed a key challenge in computer vision workflows, enabling more flexible and robust model deployment pipelines without introducing regressions or requiring changes to existing inference logic.