Determine what backends are going to be loaded at runtime

[CommanderLake]

I have a situation where the backend may or may not be CUDA and the host OS may not have CUDA available so when my C++ DLL which uses llama.cpp gets loaded the program may crash so i need to somehow determine what backend is about to be loaded and stop it loading if it cant be used, how do i do solve this?

[slaren]

You can build with GGML_BACKEND_DL to make the backends dynamically loadable. Use ggml_backend_load or ggml_backend_load_all to load the backends at runtime.