Your model and response seem fine to me. Dividing the input torque by the moment of inertia is correct.
The acceleration you're providing is always a positive value ranging from 1250 to 1700 N*m, so your speed should keep increasing with a slightly varying positive slope.
In reality, there would be other subtractive terms before the inertia division corresponding to stiffness, damping, friction, etc. Maybe this is what's missing from your model that you would expect to see?