The code snipped provided in the Fnetdata section suggests that energy targets are expected in eV. Although this isn't an issue for this particular example, it would not be compatible with the internal force calculation and ASE calculator (that expects energy predictions in Hartree).
There should either be a clear explanation of energy units or change towards more consistent behavior, i.e. expecting "ASE-units" for every property.
The code snipped provided in the Fnetdata section suggests that energy targets are expected in eV. Although this isn't an issue for this particular example, it would not be compatible with the internal force calculation and ASE calculator (that expects energy predictions in Hartree).
There should either be a clear explanation of energy units or change towards more consistent behavior, i.e. expecting "ASE-units" for every property.