CUDA: stream-k decomposition for MMQ#8018

JohannesGaessler merged 2 commits intoggml-org:masterggml-org/llama.cpp:masterfrom

JohannesGaessler:cuda-mmq-stream-k-2JohannesGaessler/llama.cpp:cuda-mmq-stream-k-2Copy head branch name to clipboard