CUBLAS Grouped GEMM doesnt support bias addition within the kernel. We need additional kernels for Bias/DBias