llama.cpp

Fix kq_scale for the attention layers of PLaMo2

#14892

Merged