3 ms·
You bring up a good point. Without seeing their code and training metrics, how do we know that this isn’t some extremely overfitted model?
by applecrazy 8y ago
You bring up a good point. Without seeing their code and training metrics, how do we know that this isn’t some extremely overfitted model?
- vedant 8y agoFrom the paper: "All models still underfit WebText and held-out perplexity has as of yet improved given more training time."