llama.cpp

dcdd65e2 - ggml : optimize ggml_vec_dot_q4_0_q8_0() using vectorized accumulators