I built my experimental system and I'm using it. I'm genuinely very happy with it. If I had more compute and could run a real, full 30B model locally, I'd be absolutely thrilled.

https://github.com/8lamaster8/GIN