llama.cpp

cd4a7c4c - Make quantize_row_iq4_nl do the same thing is quantization on CUDA