Hey,
This is what I have been working on recently. I am a deep learning engineer / researcher - I guess and after exiting my last startup, I decided I wanted to try self funding / bootstrapping.
Still more work, but figured it might be interesting to some people here.
It's been really fun building this with a really limited compute budget. Can't compete on scale, so have had to get really good at improving the architecture. I am targeting something that can be trained on 16-32 gpus for a month or two. I'll front this once I am happy with the quality of the smaller scale trains. This is the first smaller scale experiment. But its a really big milestone for the project to get usable shapes.