[blitzar]

I've done a couple side by sides on web chat with the same prompt on local 4b, 14b, 32b open models and the output gets longer/more verbose on version increment.

Its rather frustrating, slower tokens and more tokens.