SonicBoom – llama.cpp's own kernels, 14% faster decode via a graph compiler

SonicBoom is llama.cpp’s own kernels that speed up decoding by 14 % using a graph compiler. Adopt it by first ensuring compatibility with your workflow, then ma