llama.cpp

convert : use n_groups instead of hardcoded values in reshape

#18929

Merged