A large error of fitting a Lennard-Jones potential solid

[phys-chem]

I am doing a test by using Lennard-Jones to sample some structures. Then, I fit these structures with deepmd for the test, but the RMSE is very high. I try to adjust my parameters and still cannot solve it. Could you give me some suggestions? The attachment are my data and input,json Thank you very much!

Here is the fitting error:
(Units are reduce units LJ in LAMMPS, not eV)
DEEPMD INFO # number of test data : 61
DEEPMD INFO Energy MAE : 2.294327e-01 eV
DEEPMD INFO Energy RMSE : 4.881255e-01 eV
DEEPMD INFO Energy MAE/Natoms : 1.911939e-02 eV
DEEPMD INFO Energy RMSE/Natoms : 4.067713e-02 eV
DEEPMD INFO Force MAE : 2.293103e-02 eV/A
DEEPMD INFO Force RMSE : 7.531464e-02 eV/A
DEEPMD INFO Virial MAE : 2.434960e+00 eV
DEEPMD INFO Virial RMSE : 7.304477e+00 eV
DEEPMD INFO Virial MAE/Natoms : 2.029133e-01 eV
DEEPMD INFO Virial RMSE/Natoms : 6.087064e-01 eV

fit.zip

[njzjz]

What's the conversion between LJ length and Angstrom? I see you set rcut to 2.

[phys-chem]

Thank you for the reply! The LJ cutoff is two, and I use the LJ unit. I just test the fitting and it is not a real system.