Maybe its because there's literally no point in using a local llm for code-completion. You'd be spending 90% of your time correcting it. Its barely worth it to use co-pilot.