CUDA: attention sinks for mma FlashAttention#15157

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

JohannesGaessler:cuda-fa-mma-sink-3JohannesGaessler/llama.cpp:cuda-fa-mma-sink-3Copy head branch name to clipboard