Hence, models like BERT and RoBERTa are limited to a max sequence length of 512
tokens.