CUDA/HIP: Fix kernel slection for mmvq mmid kernel to align host selection with device launch bounds#21238

IMbackK merged 1 commit into

ggml-org:masterggml-org/llama.cpp:masterfrom

IMbackK:mmvq_kernel_selection_fixIMbackK/llama.cpp:mmvq_kernel_selection_fixCopy head branch name to clipboard