llama.cpp

a817a22b - ggml : implement fast walsh-hadamard transform for kv rotation (#21352) (#22631)