Multiple-choice Question Answering on HellaSwag (test)
80.1AccuracyBase
Evaluation Results
| Method | Links | |
|---|---|---|
| BaseBackbone=QWEN2-7B2026.05 | 80.1 | |
| PPT (Distribution)Backbone=QWEN2-7B2026.05 | 79.4 | |
| PPT (Mean)Backbone=QWEN2-7B2026.05 | 79.3 | |
| PPT (Distribution)Backbone=LLAMA3-8B2026.05 | 66.8 | |
| BaseBackbone=LLAMA3-8B2026.05 | 63.6 | |
| PPT (Mean)Backbone=LLAMA3-8B2026.05 | 61.6 | |
| VerbalizedBackbone=LLAMA3-8B2026.05 | 43.1 | |
| MDLM2026.02 | 31.54 | |
| AR Transformer2026.02 | 30.55 | |
| Duo++k=52026.02 | 26.9 | |
| Duo2026.02 | 26.46 | |
| Duo++k=22026.02 | 26.26 | |
| Duo++k=32026.02 | 25.89 |