llama.cpp

b1b993c1 - (test) add n_decode_total on master