Weird energies compared to FHI-aims

[EYErbil]

Hello, I am comparing the calculations with aims and my graph .pb ,
this is my ML model :
relaxation.log
Here is FHI-aims DFT output:
relaxation.log
Energies are really different, what I first thought is, I trained my model with CP2K outputs, so the energies should be Hartree, thus, I multiply the energy with 27, but still, it is 1/2 of what I get, any ideas?

[njzjz]

Are you sure CP2K and FHI-aims have the same output? They may have different energy shifts.