llama.cpp

context : simplify output counting logic during decode

#14142

Merged