Developed a performance optimization for the ROCm/vllm repository by introducing caching to the content format detection function, targeting faster chat template resolution and improved throughput. The solution leveraged Python programming and focused on performance optimization techniques, specifically caching, to reduce latency in user-facing chat flows and support greater scalability. The implementation maintained compatibility with existing systems and was structured to allow for future caching enhancements across content formatting paths. All changes were traceable through a dedicated pull request, ensuring rollback safety and clear documentation. This work addressed a key business goal of enhancing the responsiveness and scalability of chat-based applications.