Are you sure the energies are in Hartree?

[UnixJunkie]

model((species, coordinates)).energies.item()

I wonder if some scaling factor of like 1/100_000 is not missing somewhere.

But, it might also be that my conformers should first be minimized in your ~QM FF before I look at the scores they get.

[yueyericardo]

Hi Francois, I confirm that the energy output from the model is indeed in hartree. To clarify in the document Computing Energy and Force Using Models Inside Model Zoo — TorchANI 2.2.3 documentation, you'll observe that the energy for CH4 is approximately -40.45979 Hartree.

I would suggest verifying the unit of your coordinates, they should be in Angstrom.

[UnixJunkie]

I can reproduce your CH4 energy, thanks.