Factual Consistency Detection on TofuEval-MeetingBank Span level leakage-clean protocol (test)
0.693AUCGASP-threshold (default)
Evaluation Results
| Method | Links | |
|---|---|---|
| GASP-threshold (default)Model=Qwen-0.5B2026.07 | 0.693 | |
| GASP-threshold (default)Model=Qwen-1.5B2026.07 | 0.693 | |
| GASP-threshold (default)Model=SmolLM22026.07 | 0.693 | |
| GASP+base (trained)Model=SmolLM22026.07 | 0.68 | |
| GASP+base (trained)Model=Qwen-1.5B2026.07 | 0.675 | |
| GASP+base (trained)Model=Qwen-0.5B2026.07 | 0.672 | |
| GASP-trainedModel=Qwen-0.5B2026.07 | 0.656 | |
| GASP-trainedModel=Qwen-1.5B2026.07 | 0.642 | |
| GASP-trainedModel=SmolLM22026.07 | 0.631 | |
| PerplexityModel=Qwen-0.5B2026.07 | 0.555 | |
| LengthModel=SmolLM22026.07 | 0.535 | |
| PerplexityModel=SmolLM22026.07 | 0.524 | |
| PerplexityModel=Qwen-1.5B2026.07 | 0.516 | |
| LengthModel=Qwen-0.5B2026.07 | 0.515 | |
| LengthModel=Qwen-1.5B2026.07 | 0.515 |