llama.cpp

017cc5f4 - ggml-backend : only offload from host buffers (fix) (#11124)