llama.cpp

974e3cad - ggml : try another fix