During the month, contributed to the ROCm/aiter repository by developing a feature that enhances multi-head attention efficiency through FP8 block-scale quantization parameters. This work introduced stride, start pointer, and block size controls to optimize memory usage and improve compute throughput for deep learning workloads on AMD GPUs. The implementation involved C++ and CUDA, focusing on GPU programming and performance optimization. The update included integration notes, dependency management, and code formatting, laying the groundwork for further performance tuning. No major bugs were addressed during this period, as efforts centered on feature delivery and collaborative development with attention to maintainability and scalability.