llama.cpp

finetune: rename feed-forward tensors (w1/w2/w3)

#4839

Merged