COSMOSDataset

Description

COSMOS Dataset (Combined Organic, Surface and Materials Open Source Dataset) aggregates 15 publicly available ab initio databases spanning molecules, inorganic crystals, metal and oxide surfaces, and metal-organic frameworks, covering levels of theory from generalized gradient approximation to hybrid functionals. It also includes a domain-bridging set, sampled from OC20, OC22, MatPES, ODAC23, OMol25, and QCML dataset and recomputed with MPtrj-consistent computational settings, which facilitates cross-domain knowledge transfer in a multi-task training framework. For detailed information about the dataset composition, see Table 1 from the paper.