Hallucination Detection on MGSM
84.62AUCSATMD + MSP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SATMD + MSPModel=Qwen-3-14B2026.04 | 84.62 | 63.08 | 40.48 | |
| SIVRModel=Qwen-3-14B2026.04 | 79.95 | 53.62 | 43.22 | |
| SATMD + MSPModel=Qwen-3-4B-Instruct2026.04 | 78.99 | 74.39 | 40.56 | |
| SATMD + MSPModel=Qwen-3-4B-Instruct2026.04 | 78.99 | 74.39 | 40.56 | |
| SAPLMAModel=Qwen-3-4B-Instruct2026.04 | 78.35 | 98.8 | 56.34 | |
| SAPLMAModel=Qwen-3-4B-Instruct2026.04 | 78.35 | 98.8 | 56.34 | |
| EnergyModel=Qwen-3-14B2026.04 | 78.17 | 71.94 | 31.61 | |
| SAPLMAModel=Qwen-3-14B2026.04 | 76.68 | 54.41 | 20.68 | |
| P(True)Model=Qwen-3-14B2026.04 | 76.23 | 81.49 | 26.33 | |
| SIVRModel=Qwen-3-4B-Instruct2026.04 | 75.95 | 64.2 | 48.71 | |
| SIVRModel=Qwen-3-4B-Instruct2026.04 | 75.95 | — | 48.71 | |
| EntropyModel=Qwen-3-4B-Instruct2026.04 | 75.54 | 76.67 | 28.53 | |
| EntropyModel=Qwen-3-4B-Instruct2026.04 | 75.54 | 76.67 | 28.53 | |
| PerplexityModel=Qwen-3-4B-Instruct2026.04 | 74.84 | 75.72 | 27.56 | |
| PerplexityModel=Qwen-3-4B-Instruct2026.04 | 74.84 | 75.72 | 27.56 | |
| Max ProbModel=Qwen-3-4B-Instruct2026.04 | 74.82 | 76.08 | 27.47 | |
| Max ProbModel=Qwen-3-4B-Instruct2026.04 | 74.82 | 76.08 | 27.47 | |
| Temp-ScaledModel=Qwen-3-4B-Instruct2026.04 | 74.14 | 73.33 | 27.49 | |
| Temp-ScaledModel=Qwen-3-4B-Instruct2026.04 | 74.14 | 73.33 | 27.49 | |
| EntropyModel=Qwen-3-14B2026.04 | 71.17 | 89.25 | 30.77 | |
| SEModel=Qwen-3-14B2026.04 | 69.27 | 75 | 36.94 | |
| PerplexityModel=Qwen-3-14B2026.04 | 68.86 | 92.24 | 24.88 | |
| Max ProbModel=Qwen-3-14B2026.04 | 68.41 | 93.13 | 24.77 | |
| EnergyModel=Qwen-3-4B-Instruct2026.04 | 67.51 | 75 | 26.7 | |
| EnergyModel=Qwen-3-4B-Instruct2026.04 | 67.51 | 75 | 26.7 | |
| Temp-ScaledModel=Qwen-3-14B2026.04 | 66.9 | 94.93 | 22.14 | |
| SEModel=Qwen-3-4B-Instruct2026.04 | 65.43 | 86.39 | 23.73 | |
| SEModel=Qwen-3-4B-Instruct2026.04 | 65.43 | 86.39 | 23.73 | |
| Lookback LensModel=Qwen-3-14B2026.04 | 65.28 | 94.44 | 22.84 | |
| SARModel=Qwen-3-14B2026.04 | 61.96 | 88.76 | 23.09 | |
| P(True)Model=Qwen-3-4B-Instruct2026.04 | 59.14 | 93.66 | 17.07 | |
| P(True)Model=Qwen-3-4B-Instruct2026.04 | 59.14 | 93.66 | 17.07 | |
| CoE-CModel=Qwen-3-4B-Instruct2026.04 | 59.02 | 91.03 | 15.04 | |
| CoE-CModel=Qwen-3-4B-Instruct2026.04 | 59.02 | 91.03 | 15.04 | |
| CoE-RModel=Qwen-3-4B-Instruct2026.04 | 58.08 | 92.82 | 13.73 | |
| CoE-RModel=Qwen-3-4B-Instruct2026.04 | 58.08 | 92.82 | 13.73 | |
| Lookback LensModel=Qwen-3-4B-Instruct2026.04 | 56.17 | 56.79 | 4.01 | |
| Lookback LensModel=Qwen-3-4B-Instruct2026.04 | 56.17 | 56.79 | 4.01 | |
| SARModel=Qwen-3-4B-Instruct2026.04 | 52.01 | 92.9 | 16.62 | |
| SARModel=Qwen-3-4B-Instruct2026.04 | 52.01 | 92.9 | 16.62 | |
| CoE-RModel=Qwen-3-14B2026.04 | 36.22 | 97.91 | 7.76 | |
| CoE-CModel=Qwen-3-14B2026.04 | 30.87 | 97.91 | 6.94 |