[erex78]

Given they’ve open sourced the model and their training techniques, shouldn’t it be pretty easy to get an order of magnitude estimate on if their training estimates are in the right ballpark?