by Google Research · Natural Language Processing
A lighter BERT variant that reduces parameter count sharply while retaining most of BERT's performance on NLP benchmarks through factorization and parameter sharing.
No problems reported for this model.