llama.cpp

hexagon: allow dflash lm-head offload experiment

#25166

Open