Comment by slim

use llama.cpp with cuda