Comment by stackedinserter

The main gotcha for local models is insane hardware requirements.

Even for $10K you get mediocre performance.