llama.cpp

66841fdb - ggml : multi-thread mul and diag_mask ops (#1428)