RobustRAG: TF-IDF + Logistic Regression Baseline

Classical lexical baseline for hallucination detection on the RAGTruth benchmark. Features: 437k unigram + bigram TF-IDF features with sublinear term frequency. Accuracy: 76.78% | ROC-AUC: 0.7977 | PR-AUC: 0.7333 on RAGTruth test set.

Downloads last month
9
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support