[Question] Clarification on units for QM9 property prediction benchmarks in Uni-Mol2

[YuX-Ren]

Details

Hello,

Thank you for the impressive work on Uni-Mol2 (Exploring Molecular Pretraining Model at Scale).

I am currently studying the benchmark results on the QM9 dataset and would like to double-check the exact units used for the reported MAE (Mean Absolute Error) values to ensure a fair comparison.

Could you please confirm the target units for the following properties?

1. Energy-related tasks (e.g., HOMO, LUMO, Gap, ZPVE, U0, U, H, G):
   • Are the reported MAE values in Hartree (Ha), eV, or meV?
2. Dipole moment ($\mu$): Is this in Debye?
3. Isotropic Polarizability ($\alpha$): Is this in Bohr³?
4. Electronic spatial extent ($R^2$): Is this in Bohr²?

Since different papers/benchmarks sometimes use different units (e.g., meV vs. Hartree), a confirmation would be very helpful.

Thanks!