llama.cpp

84e72365 - speculative: add --n-gpu-layers-draft option (#3063)