TRINS: Towards Multimodal Language Models that Can Read (Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2024)