To make a model opened sourced, or not opened sourced?

Depends on who’s making the model.

In Facebook’s case with Llama (which annoyingly is not a direct acronym: Large Language Model Meta AI), they’ve placed a few restrictions keeping it out of the open model club.

1. Not everyone can use their model. They have a rule against enterprises with more than 700k monthly active users. In other words, a not so subtle ban of the use of their tool, barring a formalized agreement, by their competitors.

2. The training scripts, reprocessing code, and dataset sources have not been released. Again, primarily driven by a desire to keep their competitive mote.

3. Updates to the model are done solely by an in house team. There’s no community-based iterations. You basically can’t have this without the above also being open.

Depending on the goal, to each their own.