Response-level hallucination detection on RAGTruth (test)
74.5AUCGASP-threshold (default)
Evaluation Results
| Method | Links | |
|---|---|---|
| GASP-threshold (default)Scorer=Qwen2.5-0.5B2026.07 | 74.5 | |
| GASP-threshold (default)Scorer=SmolLM2-1.7B2026.07 | 74.1 | |
| GASP+base (trained)Scorer=Qwen2.5-0.5B2026.07 | 73.1 | |
| GASP-trainedScorer=Qwen2.5-1.5B2026.07 | 72.6 | |
| GASP+base (trained)Scorer=SmolLM2-1.7B2026.07 | 71.7 | |
| GASP-trainedScorer=SmolLM2-1.7B2026.07 | 71.6 | |
| GASP-threshold (default)Scorer=Qwen2.5-1.5B2026.07 | 71.3 | |
| GASP+base (trained)Scorer=Qwen2.5-1.5B2026.07 | 71.3 | |
| GASP-trainedScorer=Qwen2.5-0.5B2026.07 | 70.6 | |
| PerplexityScorer=Qwen2.5-1.5B2026.07 | 62.4 | |
| PerplexityScorer=Qwen2.5-0.5B2026.07 | 58.1 | |
| PerplexityScorer=SmolLM2-1.7B2026.07 | 57.8 | |
| LengthScorer=Qwen2.5-0.5B2026.07 | 55.3 | |
| LengthScorer=Qwen2.5-1.5B2026.07 | 55.3 | |
| LengthScorer=SmolLM2-1.7B2026.07 | 53.9 |