llama.cpp

ggml : adjust logic for offloading ops to weight's backend

#25832

Merged