CUDA: fix kernel selection logic for tile FA - #19686

CUDA: fix kernel selection logic for tile FA#19686

JohannesGaessler merged 2 commits into

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

JohannesGaessler:cuda-fa-tile-fix-dkq576JohannesGaessler/llama.cpp:cuda-fa-tile-fix-dkq576Copy head branch name to clipboard