List your company

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

0.693AUCGASP-threshold (default)
0.507880.555940.6040.65206Jul 5, 2026

Evaluation results

MethodLinks
GASP-threshold (default)Model=Qwen-0.5B2026.070.693
GASP-threshold (default)Model=Qwen-1.5B2026.070.693
GASP-threshold (default)Model=SmolLM22026.070.693
GASP+base (trained)Model=SmolLM22026.070.68
GASP+base (trained)Model=Qwen-1.5B2026.070.675
GASP+base (trained)Model=Qwen-0.5B2026.070.672
GASP-trainedModel=Qwen-0.5B2026.070.656
GASP-trainedModel=Qwen-1.5B2026.070.642
GASP-trainedModel=SmolLM22026.070.631
PerplexityModel=Qwen-0.5B2026.070.555
LengthModel=SmolLM22026.070.535
PerplexityModel=SmolLM22026.070.524
PerplexityModel=Qwen-1.5B2026.070.516
LengthModel=Qwen-0.5B2026.070.515
LengthModel=Qwen-1.5B2026.070.515