Comment by visarga

> The source code would include everything needed to train that model and reproduce it.

You know these models are trained on internet scrape which contains copyrighted content, so the dataset can't be open sourced. It's either this or bad models.