llama.cpp

ggml-backend : only offload from host buffers (fix)

#11124

Merged