Hacker News Reader

Open training data would be great too.

If you have open data and open source code you can reproduce the weights

Not easily for these large scale models, but theoretically maybe