LIBERTy-CV is one of the three datasets released as part of the LIBERTy (LLM-based Interventional Benchmark for Explainability with Real Targets) benchmark.
The goal of LIBERTy is to evaluate concept-based explanation methods in NLP under a causal and counterfactual framework.Each dataset in the benchmark is designed to expose spurious correlations between high-level concepts and model predictions, and to enable quantitative evaluation of… See the full description on the dataset page: https://huggingface.co/datasets/GilatToker/Liberty-CV.
