Dataset from: https://huggingface.co/datasets/glue

Every split besides the ax split is in this dataset.

Lines above 512 characters from the BERT-cased (bert-base-cased) tokenizer are removed