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