Energy offset at the end of TorchANI training

[WardLT]

I get very high $R^2$ scores for training an surrogate model with TorchANI, but the energies are off by a constant factor. Maybe I can just adjust the weights manually after training to remove the offset.