Kairong-Han/Spurious-Token-Game

Spurious Token Game (STG)数据集是一个针对评估在虚假相关性影响下模型性能设计的双任务数据集，包含STG_E和STG_H两个子任务。STG_E子任务包含三个代表不同难度级别的训练数据集（STG_S, STG_M, STG_L）以及两个测试数据集（IID和OOD）。STG_H子任务包含一个训练数据集和两个测试数据集（IID和OOD）。The Spurious Token Game (STG) dataset is a dual-task dataset designed for evaluating model performance under the influence of spurious correlations, containing two subtasks: STG_E and STG_H. The STG_E subtask includes three training datasets representing different difficulty levels (STG_S, STG_M, STG_L) and two test datasets (IID and OOD). The STG_H subtask consists of one training dataset and two test datasets (IID and OOD).