metal : use F32 attention accumulators in FA kernels#13975

ggerganov merged 1 commit intomasterggml-org/llama.cpp:masterfrom

gg/metal-fa-acc-f32ggml-org/llama.cpp:gg/metal-fa-acc-f32Copy head branch name to clipboard