Accuracy on PIQA (Physical commonsense reasoning)
79.8Accuracy (PIQA)Vanilla
Evaluation results
| Method | Links | |
|---|---|---|
| VanillaBackbone=Mistral-7B-Instruct2026.05 | 79.8 | |
| DT-CD⋆Backbone=Mistral-7B-Instruct2026.05 | 79.7 | |
| SDDBackbone=Mistral-7B-Instruct2026.05 | 79.5 | |
| COFTBackbone=Mistral-7B-Instruct2026.05 | 79.5 | |
| DExpertsBackbone=Mistral-7B-Instruct2026.05 | 79.4 | |
| VanillaBackbone=LLaMA-2-13B2026.05 | 78.1 | |
| DT-CD⋆Backbone=LLaMA-2-13B2026.05 | 78 | |
| COFTBackbone=LLaMA-2-13B2026.05 | 78 | |
| SDDBackbone=LLaMA-2-13B2026.05 | 77.9 | |
| DExpertsBackbone=LLaMA-2-13B2026.05 | 77.8 |