List your company

Factual Consistency Detection on TofuEval-MeetingBank Response level leakage-clean protocol (test)

65.9AUCGASP+base (trained)
51.02854.88958.7562.611Jul 5, 2026

Evaluation results

MethodLinks
GASP+base (trained)Model=Qwen-0.5B2026.0765.9
GASP+base (trained)Model=SmolLM22026.0765.6
GASP-trainedModel=SmolLM22026.0764.1
GASP+base (trained)Model=Qwen-1.5B2026.0763.3
GASP-trainedModel=Qwen-0.5B2026.0763.2
GASP-trainedModel=Qwen-1.5B2026.0762.7
GASP-threshold (default)Model=Qwen-0.5B2026.0762.5
GASP-threshold (default)Model=Qwen-1.5B2026.0762.3
GASP-threshold (default)Model=SmolLM22026.0761.4
PerplexityModel=Qwen-1.5B2026.0754.8
PerplexityModel=Qwen-0.5B2026.0752.1
PerplexityModel=SmolLM22026.0751.6