Diabetic Retinopathy (DR) is a severe eye disease affecting diabetics. Early detection is crucial to prevent vision loss. In this paper, we adopt and fine-tune transformer-based learning models to capture the crucial features of retinal images for a more nuanced understanding of DR severity. To improve early detection, we convert a 5-class dataset into 3 clinically relevant stages (no DR, early DR, advanced DR). Our methodology follows a three-stage process: initial preprocessing of images to improve image quality, then feature extraction utilizing the vision transformer model (ViT), and finally classification employing multi layer process to categorize them into three stages of DR. Our experimentation on the APTOS dataset showcases that our improved classification approach employing a ViT model achieves highlevels of accuracy (0.9297) and precision (0.98).