Anyone know how to map an Olmo checkpoint to a specific subset of its training data?

Screenshot 1 from skeptical thread:

The good reviewer party™, a party at conferences for reviewers whose ACs said they did a great job. ML operates on FOMO so I think it'd work