llama.cpp

fix: correct EXAONE3 FFN_DOWN tensor mapping prefix

#20076

Open