llama.cpp

ggml-cuda: Use same regex for GGML_NATIVE=OFF

#18407

Merged