llama.cpp

convert: remove most of the n_mult usage in convert.py

#3098

Merged