The training effect of energy and strength is too different

[lammps-vasp]

As shown in the figure, we can see that the training effect of force is very good, but the energy difference is too far. I want to know what causes this, how to use energy to achieve the training effect of force, and whether the model in this case is available

[njzjz]

You may meet the same situation as #514. The current advice is to increase start_pref_e and limit_pref_e in your parameters.

By the way, I don't think your forces are very good. Did you calculate the root mean squared error (RMSE)?

[lammps-vasp]

Thank you
I don't know how to calculate the root mean squared error (RMSE). In lcurve.out, the order of magnitude of the force reaches 10 ^ - 1. DP users say that they can use the straight line y = x to judge the fitting effect. If this effect is not very good, then what conditions need to be met before the fitting effect can be achieved