Developed and validated INT8 support for grouped 2D convolution forward operations within the ROCm/composable_kernel and ROCm/MIOpen repositories, focusing on enabling efficient low-precision inference. Leveraged C++ and CUDA, including advanced template metaprogramming, to introduce new int8 instances and configurations across multiple tensor layouts and optimization strategies. Enhanced the problem-descriptor logic to accurately identify INT8 operations and implemented comprehensive unit tests to ensure correctness and performance in mixed-precision scenarios. This work established a robust foundation for high-throughput, energy-efficient inference on supported GPU hardware, aligning cross-repository development for consistent INT8 support throughout the ROCm deep learning stack.