Multiple-choice Question Answering on Winogrande (test)
63.8AccuracyPPT (Distribution)
Evaluation Results
| Method | Links | |
|---|---|---|
| PPT (Distribution)Backbone=QWEN2-7B2026.05 | 63.8 | |
| BaseBackbone=QWEN2-7B2026.05 | 61.3 | |
| PPT (Mean)Backbone=QWEN2-7B2026.05 | 61.1 | |
| PPT (Distribution)Backbone=LLAMA3-8B2026.05 | 58.1 | |
| BaseBackbone=LLAMA3-8B2026.05 | 54.5 | |
| PPT (Mean)Backbone=LLAMA3-8B2026.05 | 54.2 | |
| AR Transformer2026.02 | 52.8 | |
| MDLM2026.02 | 51.93 | |
| VerbalizedBackbone=LLAMA3-8B2026.05 | 50.7 | |
| Duo++k=52026.02 | 50.12 | |
| Duo++k=22026.02 | 49.64 | |
| Duo++k=32026.02 | 47.36 | |
| Duo2026.02 | 47.2 |