CUDA: faster Deepseek FA, add Turing support#13435

JohannesGaessler merged 1 commit intoggml-org:masterggml-org/llama.cpp:masterfrom

JohannesGaessler:cuda-fa-opt-2JohannesGaessler/llama.cpp:cuda-fa-opt-2Copy head branch name to clipboard