CVPR2026AlgoFine-Grained Post-Training Quantization for Large Vision Language Models with Quantization-Aware Integrated Gradients