First, making sure to offer an upvote here. I happen to be VERY enthusiastic about local models, but I've found them to be incredibly hard to host, incredibly hard to harness, and, despite everything, remarkably powerful if you are willing to suffer really poor token/second performance...