llama.cpp

2a39955a - move mul_mat(k,q) to outside of chunking