Token-level Uncertainty Quantification on Generated-DI
93.64AUCROCReverse Probing (XGBoost)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Reverse Probing (XGBoost)Backbone LLM=OpenBioLLM-8B2026.05 | 93.64 | 28.92 | |
| Reverse Probing (CatBoost)Backbone LLM=OpenBioLLM-8B2026.05 | 91.44 | 28.95 | |
| Reverse Probing (XGBoost)Backbone LLM=BioMistral-7B2026.05 | 88.19 | 23.16 | |
| Reverse Probing (CatBoost)Backbone LLM=BioMistral-7B2026.05 | 84.35 | 28.05 | |
| Sliding Window EntropyBackbone LLM=OpenBioLLM-8B2026.05 | 69.4 | 10.41 | |
| Uncertainty Aware HeadsBackbone LLM=OpenBioLLM-8B2026.05 | 64.03 | 6.82 | |
| Uncertainty Aware HeadsBackbone LLM=BioMistral-7B2026.05 | 61.5 | 5.75 | |
| Decomposing UncertaintyBackbone LLM=OpenBioLLM-8B2026.05 | 59.59 | 14.1 | |
| Decomposing UncertaintyBackbone LLM=BioMistral-7B2026.05 | 59.25 | 13.22 | |
| SE ProbesBackbone LLM=OpenBioLLM-8B2026.05 | 59.09 | 15.53 | |
| Sliding Window EntropyBackbone LLM=BioMistral-7B2026.05 | 58.84 | 6.78 | |
| Semantic EnergyBackbone LLM=OpenBioLLM-8B2026.05 | 55.34 | 4.44 | |
| Token EntropyBackbone LLM=OpenBioLLM-8B2026.05 | 53.63 | 4.4 | |
| Semantic EntropyBackbone LLM=BioMistral-7B2026.05 | 53.61 | 10.5 | |
| Semantic EntropyBackbone LLM=OpenBioLLM-8B2026.05 | 53.61 | 10.5 | |
| Attention ChainBackbone LLM=OpenBioLLM-8B2026.05 | 50.98 | 4.3 | |
| Semantic EnergyBackbone LLM=BioMistral-7B2026.05 | 46.97 | 4.36 | |
| Attention ChainBackbone LLM=BioMistral-7B2026.05 | 45.49 | 3.88 | |
| Token EntropyBackbone LLM=BioMistral-7B2026.05 | 44.21 | 3.66 | |
| SE ProbesBackbone LLM=BioMistral-7B2026.05 | 40.61 | 10.75 |