Commonsense Reasoning on StrategyQA (Accuracy (%))
76.13Accuracy (%)Dream 7B-Instruct (Top Margin + Suffix-Anchored Confidence Modulation)
Evaluation Results
| Method | Links | |
|---|---|---|
| Dream 7B-Instruct (Top Margin + Suffix-Anchored Confidence Modulation)Model=Dream 7B-Instruct, Decoding Strategy=Top Margin, Suffix Anchor=true, Confidence Modulation=true2026.05 | 76.13 | |
| Dream 7B-Instruct (Top Margin + Suffix Anchor)Model=Dream 7B-Instruct, Decoding Strategy=Top Margin, Suffix Anchor=true, Confidence Modulation=false2026.05 | 75.98 | |
| Dream 7B-Instruct (Top Probability + Suffix-Anchored Confidence Modulation)Model=Dream 7B-Instruct, Decoding Strategy=Top Probability, Suffix Anchor=true, Confidence Modulation=true2026.05 | 75.4 | |
| Dream 7B-Instruct (Top Probability + Suffix Anchor)Model=Dream 7B-Instruct, Decoding Strategy=Top Probability, Suffix Anchor=true, Confidence Modulation=false2026.05 | 74.24 | |
| VanillaBackbone=Mistral-7B-Instruct2026.05 | 73.6 | |
| COFTBackbone=Mistral-7B-Instruct2026.05 | 73.6 | |
| DT-CD⋆Backbone=Mistral-7B-Instruct2026.05 | 73.5 | |
| SDDBackbone=Mistral-7B-Instruct2026.05 | 73 | |
| DExpertsBackbone=Mistral-7B-Instruct2026.05 | 72.8 | |
| VanillaBackbone=LLaMA-2-13B2026.05 | 71.2 | |
| COFTBackbone=LLaMA-2-13B2026.05 | 71.1 | |
| DT-CD⋆Backbone=LLaMA-2-13B2026.05 | 71 | |
| SDDBackbone=LLaMA-2-13B2026.05 | 70.5 | |
| LLaDA 8B-Instruct (Top Probability + Suffix-Anchored Confidence Modulation)Model=LLaDA 8B-Instruct, Decoding Strategy=Top Probability, Suffix Anchor=true, Confidence Modulation=true2026.05 | 70.45 | |
| DExpertsBackbone=LLaMA-2-13B2026.05 | 70.3 | |
| Dream 7B-Instruct (Random)Model=Dream 7B-Instruct, Decoding Strategy=Random, Suffix Anchor=false, Confidence Modulation=false2026.05 | 69.43 | |
| LLaDA 8B-Instruct (Top Margin + Suffix-Anchored Confidence Modulation)Model=LLaDA 8B-Instruct, Decoding Strategy=Top Margin, Suffix Anchor=true, Confidence Modulation=true2026.05 | 68.12 | |
| LLaDA 8B-Instruct (Top Probability + Suffix Anchor)Model=LLaDA 8B-Instruct, Decoding Strategy=Top Probability, Suffix Anchor=true, Confidence Modulation=false2026.05 | 66.52 | |
| LLaDA 8B-Instruct (Top Margin + Suffix Anchor)Model=LLaDA 8B-Instruct, Decoding Strategy=Top Margin, Suffix Anchor=true, Confidence Modulation=false2026.05 | 66.08 | |
| Dream 7B-Instruct (Top Margin)Model=Dream 7B-Instruct, Decoding Strategy=Top Margin, Suffix Anchor=false, Confidence Modulation=false2026.05 | 62.45 | |
| LLaDA 8B-Instruct (Random)Model=LLaDA 8B-Instruct, Decoding Strategy=Random, Suffix Anchor=false, Confidence Modulation=false2026.05 | 58.81 | |
| Dream 7B-Instruct (Top Probability)Model=Dream 7B-Instruct, Decoding Strategy=Top Probability, Suffix Anchor=false, Confidence Modulation=false2026.05 | 56.91 | |
| LLaDA 8B-Instruct (Top Margin)Model=LLaDA 8B-Instruct, Decoding Strategy=Top Margin, Suffix Anchor=false, Confidence Modulation=false2026.05 | 28.53 | |
| LLaDA 8B-Instruct (Top Probability)Model=LLaDA 8B-Instruct, Decoding Strategy=Top Probability, Suffix Anchor=false, Confidence Modulation=false2026.05 | 19.65 |