llama.cpp

fa6383ca - CUDA : conditionally add cuda architectures (ggml/1341)