llama.cpp

39bb8e7d - ggml : 2x faster scalar implementations