Models of BERT family are overall robust to pruning, but they have an Achilles heel: the outlier dimensions, without which the quality of the model drops sig...
Posts by Year
The world is filled with data. Can we learn from this data to generate something new?
QuAIL is a new challenging NLP benchmark that combines reading comprehension and commonsense reasoning.
BERT and its Transformer-based cousins are still ahead on all NLP leaderboards. But how much do they actually understand about language?