CUDA: broadcasting for FlashAttention mask#14500

JohannesGaessler merged 1 commit intoggml-org:gg/ggml-batch-soft-max-opsggml-org/llama.cpp:gg/ggml-batch-soft-max-opsfrom

JohannesGaessler:cuda-fa-mask-broadcastJohannesGaessler/llama.cpp:cuda-fa-mask-broadcastCopy head branch name to clipboard