3 ms·

llama.cpp can do exactly that