llama.cpp

e59fcb2b - Add --n-predict -2 for stopping generation on full context (#2565)