The number of papers published with code tells me a few things about academia, i.e. it's probably a waste of time reading it because you can't reproduce the the scaling results.