Multiple-choice commonsense reasoning on Winogrande
74Winogrande AccuracyStatic-SAE
Evaluation Results
| Method | Links | |
|---|---|---|
| Static-SAEBackbone=Qwen3-8B2026.03 | 74 | |
| DSPABackbone=Qwen3-8B2026.03 | 73.8 | |
| DPOBackbone=Qwen3-8B2026.03 | 73.4 | |
| RepEBackbone=Qwen3-8B2026.03 | 73.1 | |
| Prompt EngBackbone=Qwen3-8B2026.03 | 72.8 | |
| DSPABackbone=Gemma-2-9B2026.03 | 72.5 | |
| DPOBackbone=Gemma-2-9B2026.03 | 72.5 | |
| RepEBackbone=Gemma-2-9B2026.03 | 72.5 | |
| Static-SAEBackbone=Gemma-2-9B2026.03 | 72.5 | |
| Base ModelBackbone=Qwen3-8B2026.03 | 72 | |
| Prompt EngBackbone=Gemma-2-9B2026.03 | 71.9 | |
| Base ModelBackbone=Gemma-2-9B2026.03 | 71.9 | |
| Static-SAEBackbone=Gemma-2-2B2026.03 | 65.7 | |
| Base ModelBackbone=Gemma-2-2B2026.03 | 65.7 | |
| DPOBackbone=Gemma-2-2B2026.03 | 65.2 | |
| RepEBackbone=Gemma-2-2B2026.03 | 65.1 | |
| DSPABackbone=Gemma-2-2B2026.03 | 65 | |
| Prompt EngBackbone=Gemma-2-2B2026.03 | 64.3 |