Jun 23 Adversarial Pre-training and Instruction Tuning for Improving Vision-Language Model Robustness