Training datasets for AIMNet2(2025) machine-learned neural network potential

The datasets contain molecular structures and the properties computed with B97-3c (GGA DFT) (or wB97M-def2-TZVPP (range-separated hybrid DFT) soon). The data file contains about 3.8M structures. DFT calculation performed with ORCA 6.0.1 software. Properties include energy, forces, atomic charges(Hirshfeld), and total atomic dipole moment.This dataset supports the study Benchmarking Universal Machine-learned Interatomic Potentials for Intermolecular and Noncovalent Interactions. The primary motivation of this work is to systematically assess and improve the description of intermolecular interactions and long-range effects relative to the original AIMNet2 framework introduced in Chemical Science (2025) https://pubs.rsc.org/en/content/articlelanding/2025/sc/d4sc08572hThe work is described in:Kamal Singh Nayal, Ilkwon Cho, and Olexandr Isayev, ChemRxiv, 2026, Benchmarking Universal Machine-learned Interatomic Potentials for Intermolecular and Noncovalent Interactions | ChemRxivFor the main codebase, pretrained models, and reproducibility resources, see the AIMNet2 central repository:https://github.com/isayevlab/aimnetcentral