遇见数据集

l3-unc/CausalDiagnosticity

收藏
Hugging Face2025-08-18 更新2025-11-29 收录
官方服务:

资源简介:

该数据集用于评估忠实性度量,包含四个控制任务:事实检查、类比、物体计数和多跳推理。这些任务通过使用具有忠实和不忠实解释的逆事实模型来评估因果诊断性,难度各不相同。

This dataset is created to evaluate faithfulness metrics using four controlled tasks: fact-checking, analogy, object counting, and multi-hop reasoning. These tasks assess causal diagnosticity by employing counterfactual models with faithful and unfaithful explanations, designed to span varying levels of complexity.

提供机构:
l3-unc
二维码
社区交流群
二维码
科研交流群
商业服务