Also very silly prediction given this would dramatically lower the amount of models hosted on HuggingFace and move users elsewhere to platforms that do support quantized models (not what Nvidia wants).