Relaxation in lmp

[alib-l]

Hello,
I have trained a model for SiO2 and frozen it (the loss function converged and the predicted values were accurate). Then I run four lmp simulations to relax it with "fix NVT", but I got different results each time. In 3 of them, the structure was destroyed and the atoms dispersed, but in one of them, the cell was relaxed. Could you please help me with this issue?
Thanks,

[njzjz]

but I got different results each time.

Can you post the lammps command you used?

Also, you may need to add more data into your training set.