llama.cpp

speculative: add --n-gpu-layers-draft option

#3063

Merged