ValueError: You cannot perform fine-tuning on purely quantized models.