llama.cpp

e1b1db9f - simple : do not perform tensor data copy if not needed