Multi-hop Reasoning on 2WikiMHQA
0.7002AUROCCoT-UQ
Evaluation Results
| Method | Links | |
|---|---|---|
| CoT-UQModel=Llama 3.1-8B, Strategy=AP, Base Aggregation=Probas-min2025.02 | 0.7002 | |
| CoT-UQModel=Llama 2-13B, Strategy=AP, Base Aggregation=Probas-min2025.02 | 0.681 | |
| TOKENSAR + CoT-UQStrategy=AP, Backbone=Mistral 7B2025.02 | 0.6606 | |
| Probas-mean + CoT-UQStrategy=AP, Backbone=Mistral 7B2025.02 | 0.6558 | |
| CoT-UQModel=Llama 3.1-8B, Strategy=AP, Base Aggregation=TOKENSAR2025.02 | 0.6538 | |
| CoT-UQModel=Llama 3.1-8B, Strategy=AP, Base Aggregation=Probas-mean2025.02 | 0.6522 | |
| CoT-UQModel=Llama 2-13B, Strategy=AP, Base Aggregation=TOKENSAR2025.02 | 0.6403 | |
| CoT-UQModel=Llama 2-13B, Strategy=AP, Base Aggregation=Probas-mean2025.02 | 0.6329 | |
| CoT-UQModel=Llama 2-13B, Strategy=SE, Base Strategy=Self-Probing, Refinement=KEYKeywords2025.02 | 0.6263 | |
| Self-ProbingModel=Llama 2-13B, Strategy=SE2025.02 | 0.5981 | |
| CoT-UQModel=Llama 3.1-8B, Strategy=SE, Base Strategy=Self-Probing, Refinement=KEYKeywords2025.02 | 0.5838 | |
| Probas-meanStrategy=AP, Backbone=Mistral 7B2025.02 | 0.5819 | |
| CoT-UQModel=Llama 3.1-8B, Strategy=SE, Base Strategy=P(True), Refinement=KEYKeywords2025.02 | 0.5777 | |
| TOKENSARModel=Llama 3.1-8B, Strategy=AP2025.02 | 0.5692 | |
| Probas-minModel=Llama 3.1-8B, Strategy=AP2025.02 | 0.5681 | |
| Probas-meanModel=Llama 3.1-8B, Strategy=AP2025.02 | 0.568 | |
| Self-ProbingModel=Llama 3.1-8B, Strategy=SE2025.02 | 0.5639 | |
| TOKENSARStrategy=AP, Backbone=Mistral 7B2025.02 | 0.556 | |
| P(True)Model=Llama 3.1-8B, Strategy=SE2025.02 | 0.5356 | |
| CoT-UQModel=Llama 2-13B, Strategy=SE, Base Strategy=P(True), Refinement=KEYKeywords2025.02 | 0.5352 | |
| Probas-meanModel=Llama 2-13B, Strategy=AP2025.02 | 0.5154 | |
| Probas-minModel=Llama 2-13B, Strategy=AP2025.02 | 0.5128 | |
| TOKENSARModel=Llama 2-13B, Strategy=AP2025.02 | 0.5108 | |
| P(True) + CoT-UQStrategy=SE, Backbone=Mistral 7B2025.02 | 0.4978 | |
| P(True)Model=Llama 2-13B, Strategy=SE2025.02 | 0.4752 | |
| P(True)Strategy=SE, Backbone=Mistral 7B2025.02 | 0.4713 |