llama.cpp

5b351e94 - cuda : remove nchannels_x argument from mul_mat_vec_nc_f16_f32 (#2028)