List your company

Response-level hallucination detection on RAGTruth (test)

74.5AUCGASP-threshold (default)
53.07658.63864.269.762Jul 5, 2026

Evaluation results

MethodLinks
GASP-threshold (default)Scorer=Qwen2.5-0.5B2026.0774.5
GASP-threshold (default)Scorer=SmolLM2-1.7B2026.0774.1
GASP+base (trained)Scorer=Qwen2.5-0.5B2026.0773.1
GASP-trainedScorer=Qwen2.5-1.5B2026.0772.6
GASP+base (trained)Scorer=SmolLM2-1.7B2026.0771.7
GASP-trainedScorer=SmolLM2-1.7B2026.0771.6
GASP-threshold (default)Scorer=Qwen2.5-1.5B2026.0771.3
GASP+base (trained)Scorer=Qwen2.5-1.5B2026.0771.3
GASP-trainedScorer=Qwen2.5-0.5B2026.0770.6
PerplexityScorer=Qwen2.5-1.5B2026.0762.4
PerplexityScorer=Qwen2.5-0.5B2026.0758.1
PerplexityScorer=SmolLM2-1.7B2026.0757.8
LengthScorer=Qwen2.5-0.5B2026.0755.3
LengthScorer=Qwen2.5-1.5B2026.0755.3
LengthScorer=SmolLM2-1.7B2026.0753.9