Lin Sun developed and validated INT8 support for grouped 2D convolution forward operations within the ROCm/composable_kernel and ROCm/MIOpen repositories. Leveraging C++ and CUDA, Lin introduced new int8 instances and configurations across multiple tensor layouts, updating problem-descriptor logic to accurately identify and process INT8 operations. The work included comprehensive unit tests to ensure correctness and performance in mixed-precision inference scenarios, supporting both NCHW and NHWC data formats. By aligning development across repositories, Lin enabled efficient, low-precision inference workflows, laying a foundation for future hardware-accelerated optimizations and improving throughput and energy efficiency for high-performance GPU computing tasks.