Triclinic box simulation losing atom

[stfalxndria]

Hi all,
I have trained an MLIP for my MOF structure filled with water. My model had a pretty decent RMSE for both energy (1.34e-04) and forces (4.47e-02) and I even tested my initial structure using dp test and the RMSE was good (in my opinion). However when I run my MLMD simulation, my KE is extremely high and I lose an atom at step 50. I am wondering if this has anything to do with the fact that the structure is a triclinic (90 122 90) since I have replicated the same steps and method employed for another structure that is orthorombic and everything worked well. I have tried with both LAMMPS and GROMACS and both of them I have this issue for this specific structure and not the orthorombic structure.

I have attached some input and outputs that I thought were relevant.
thank you in advance!
lammps_MLMD.zip