llama.cpp

c12b14b7 - benchmark : fix result validation in benchmark-q4_0-matmult (#987)