First-error step detection on MoreHopQA
0.7885AUROCGIBS
Evaluation Results
| Method | Links | |
|---|---|---|
| GIBSModel=Deepseek2026.05 | 0.7885 | |
| NLI-MaxModel=Deepseek2026.05 | 0.711 | |
| GIBSModel=Phi42026.05 | 0.6841 | |
| EntropyModel=Phi42026.05 | 0.6837 | |
| NLI-MaxModel=Phi42026.05 | 0.6632 | |
| NLI-MeanModel=Deepseek2026.05 | 0.6544 | |
| GIBSModel=Llama3.1-8b2026.05 | 0.6164 | |
| EntropyModel=Deepseek2026.05 | 0.6004 | |
| NLI-MeanModel=Phi42026.05 | 0.5824 | |
| EntropyModel=Llama3.1-8b2026.05 | 0.557 | |
| NLI-MaxModel=Llama3.1-8b2026.05 | 0.5375 | |
| Cos-MeanModel=Deepseek2026.05 | 0.5313 | |
| NLI-MeanModel=Llama3.1-8b2026.05 | 0.5285 | |
| P(true)Model=Deepseek2026.05 | 0.5152 | |
| P(true)Model=Llama3.1-8b2026.05 | 0.5126 | |
| Cos-MeanModel=Phi42026.05 | 0.5099 | |
| P(true)Model=Phi42026.05 | 0.509 | |
| SL(norm)Model=Llama3.1-8b2026.05 | 0.4941 | |
| Cos-MeanModel=Llama3.1-8b2026.05 | 0.4863 | |
| Cos-MaxModel=Phi42026.05 | 0.4772 | |
| SL(norm)Model=Phi42026.05 | 0.4724 | |
| Cos-MaxModel=Deepseek2026.05 | 0.4529 | |
| Cos-MaxModel=Llama3.1-8b2026.05 | 0.4275 | |
| LECOModel=Phi42026.05 | 0.3884 | |
| LECOModel=Llama3.1-8b2026.05 | 0.3839 | |
| LECOModel=Deepseek2026.05 | 0.3307 | |
| SL(norm)Model=Deepseek2026.05 | 0.3009 |