In this project, contrastive learning techniques are leveraged to improve ophthalmic biomarker identification. Utilizing the EfficientViT_m5.r224_in1k model as the foundation, the baseline accuracy was enhanced from 69% to 73% through the integration of contrastive learning. This study presents a novel approach to apply contrastive learning on multi-label classes of both labeled and unlabeled data, contributing valuable insights into biomarkers associated with eye health.