Worked on enhancing the caching documentation for the ModelTC/LightX2V repository, focusing on TeaCache and TaylorSeer Cache. Developed comprehensive Markdown guides that clarified when caching mechanisms should be applied, with detailed explanations of L1 distance threshold handling and the use of Taylor expansion to balance inference speed and accuracy. The documentation provided actionable strategies to streamline caching decisions, aiming to reduce misconfigurations and support faster, more consistent model deployments. By improving the clarity and maintainability of the caching guidance, the work facilitated smoother onboarding for new contributors and helped ensure reliable model performance across various deployment scenarios.