Comment Comment by vedant From the paper:"All models still underfit WebText and held-out perplexity has as of yet improved given more training time." parent vedant 7 years ago
"All models still underfit WebText and held-out perplexity has as of yet improved given more training time."