Hongbo delivered a targeted performance optimization for the HabanaAI/vllm-fork repository by implementing StreamK scheduling for block-quantized CUTLASS kernels. Using C++ and CUDA, he refactored the kernel calling structure and introduced scheduler arguments to enable dimension-aware execution, allowing tensor operations to adapt more flexibly to varying workloads. This work focused on high-performance computing and advanced GPU programming, laying the groundwork for broader kernel-level scheduling strategies. Although the contribution spanned a single feature over one month, the depth of the engineering effort is evident in the careful restructuring and the focus on scalable, efficient tensor operation throughput for future development.