ChaosNLI

该数据集专为分布式自然语言推理任务设计，其中包含了与人类标签分布相关联的示例，用以评估模型预测的准确性。此外，该数据集为开发集和测试集提供了软标签，模拟了在真实生产场景中这类数据稀缺的情况。数据集包含了训练、开发和测试三个不同的分割，任务内容是结合人类观点分布预测的自然语言推理（NLI）。

This dataset is specifically designed for distributed natural language inference tasks. It contains examples paired with human label distributions, which are used to evaluate the accuracy of model predictions. Furthermore, this dataset provides soft labels for both the development and test sets, simulating the data scarcity scenario commonly encountered in real-world production environments. The dataset includes three distinct splits: training, development, and test sets, with the task being natural language inference (NLI) that incorporates human opinion distribution prediction.