all 10

The 'all 10' dataset/benchmark consists of 2 million molecule images used to evaluate performance across 10 downstream tasks related to molecular property prediction, including physical-property regression, drug-discovery classification, and quantum-chemistry prediction.