[segmondy]

Curious why you went this route, don't you think you could have achieved near this performance 80%+ or more within llama.cpp?