llama.cpp

llama : gemma3 : use output tensor if it exists in model weight

#12506

Merged