ashwath-vaithina/toy-shapes-dataset

This dataset is used in the work "Do VLMs Have Bad Eyes? Diagnosing Compositional Failures via Mechanistic Interpretability" Paper For more details, please refer to the GitHub repository.