Efficient Architectures for High Resolution Vision-Language Models