Although there are better ways to test, I used a 3B model to speed up replies from my local AI server when testing out an application I was developing. Yes I could have mocked up HTTP replies etc., but in this case the small model let me just plug in and go.