Just look at your thermo output.
The problem is not the errors, but it starts earlier when your temperature is “exploding” in just a few steps.
That is usually an indication of a bad initial geometry.
When your system has a temperature of over 8 million Kelvin, you cannot expect much sanity.
It looks to me, you first need to gather some experience in running meaningful MD simulations at all. And that is easier with simpler systems, like the “classic” liquid argon system.