This repository provides a PyTorch implementation of DSMIL, a dual-stream multiple instance learning network designed for whole slide image (WSI) classification and tumor detection. The model employs vision transformers and self-supervised contrastive learning to handle weakly supervised learning where only bag-level (slide-level) labels are available. It processes histopathology images by extracting features from image patches and aggregating them using attention-based pooling, enabling tumor classification without pixel-level annotations.