2 ms·
> being trained on almost twice as fewer text data Clearly they didn't run that through the English BERT because no language model would assign high probabilit
by nl 6y ago
> being trained on almost twice as fewer text data
Clearly they didn't run that through the English BERT because no language model would assign high probability to that phrase!