Comment by bitL

BERT is already outdated, but still useful as you need only 1 Titan RTX to retrain its BERT_large model via transfer learning.