llama.cpp

e773d226 - add optional activation scales to mul_mat and mul_mat_id tensors