llama.cpp

efac2d46 - common : don't do warm-up with more than n_batch tokens (close #3058)