Worked on the StreamHPC/rocm-libraries repository to address a critical issue in memory usage reporting for high-performance computing workloads. Focused on C++ development, the work involved correcting the calculation of memory requirements for Key and Value by replacing the use of nhead with nhead_k, ensuring that reported metrics accurately reflected actual resource consumption. This targeted bug fix eliminated inflated memory metrics, which previously led to misreported resource needs and unreliable performance data. By improving memory management and performance optimization, the changes enabled more accurate capacity planning and data-driven tuning, directly supporting the reliability and efficiency of the ROCm libraries ecosystem.