← All works
Research paper

RoBERTa: A Robustly Optimized BERT Pretraining Approach

At a glance

Showed that better training choices could improve BERT without changing its core architecture.

In one paragraph

Showed that training BERT longer, with more text and better training choices, could improve language understanding without changing its core design, setting new results on question-answering and reading tests.

Authors 10

  1. Yinhan Liu
  2. Myle Ott
  3. Naman Goyal
  4. Jingfei Du
  5. Mandar Joshi
  6. Danqi Chen
  7. Omer Levy
  8. Mike Lewis
  9. Luke Zettlemoyer
  10. Veselin Stoyanov