“带来”
用英语怎么说？

We also have MuSiQue, which is a multi-hop QA dataset. It requires multiple hops between twenty different Wikipedia paragraphs to find the correct answer. Don't forget the new tasks we introduced in this benchmark: SpaceDigest, which is a sentiment aggregation task, and BookSumSort, which requires reordering shuffled chapter summaries. BookSumSort is very clever. It makes sure models understand the chronological flow of novels, plays, and long poems from various sources. Exactly. By putting all these diverse tasks together, ZeroSCROLLS lays the groundwork for evaluating long-context LLMs in a zero-shot setting. This benchmark will lead to better natural language understanding over long texts in the future. Let's get that Hugging Face script fixed asap so everyone can access it!

我们还有 MuSiQue，这是一个多跳问答数据集。它需要跨越二十个不同的维基百科段落进行多次跳转才能找到正确答案。别忘了我们在该基准中引入的新任务：SpaceDigest（一个情感聚合任务）和 BookSumSort（需要对打乱的章节摘要进行重新排序）。BookSumSort 非常聪明。它确保模型理解来自各种来源的小说、剧本和长诗的时间顺序流程。没错。通过将所有这些不同的任务放在一起，ZeroSCROLLS 为在零样本设置下评估长上下文大语言模型奠定基础。这一基准在未来将带来更好的长文本自然语言理解。让我们尽快修复 Hugging Face 的脚本，以便每个人都能访问它！