llama.cpp

Move GLM4 f32 attention fix to the correct function

#13750

Merged