llama.cpp

610394ff - fix supported ops for kompute backend