llama.cpp

30eef31b - Make quantize_row_iq4_nl do the same thing is quantization on CUDA