Comment by pich

Its not quite that simple here. The iMatrix-guided hybrid uses different quantization levels per tensor/layer, so there isnt one honest Q4/Q5/NVFP4 label I can put in the title