llama.cpp

ec7b9329 - gguf-py : use copy-on-write mode for localtensor (#18162)