GSCBench provides exact graph edit distance (GED) supervision for studying in-collection, cross-collection, zero-shot and few-shot graph similarity models. The GSCBench benchmark data comprise 11 collections, 5,523 graphs and 1,903,452 exact graph pairs. The repository additionally includes a version of PyG LINUX with isomorphic graphs removed (89 graphs and 3,916 pairs), used as auxiliary data for model pretraining. If you use GSCBench in your research, please cite our paper: The tu prefix identifies TU collections, and ogbogbg-code2 identifies the OGB code2 collection. This PyG LINUX subset has isomorphic graphs removed and is provided to reproduce auxiliary model pretraining.…