SCROLLS measures a model's performance on NLP tasks such as natural language understanding (NLU), question-answering, and summarization, evaluated on seven different datasets containing text ...