Imagine this thing for autocomplete.

I'm not sure how good llama 3.1 8b is for that, but it should work, right?

Autocomplete models don't have to be very big, but they gotta be fast.