llama.cpp

sync : ggml (ggml-backend)

#3548

Merged