Commonsense Reasoning on Common sense QA (AUCOAA)
81.4AUCOAAAdaptive-Answer
Evaluation Results
| Method | Links | |
|---|---|---|
| Adaptive-AnswerBackbone=Qwen3-8B2026.01 | 81.4 | |
| Format-Adaptive-AnswerBackbone=Qwen3-8B2026.01 | 81.3 | |
| Normalized-LengthBackbone=Qwen3-8B2026.01 | 78.6 | |
| SFTBackbone=Qwen3-8B2026.01 | 76.8 | |
| Hard-Length 8k → 4kBackbone=Qwen3-8B2026.01 | 76.5 | |
| TWYNBackbone=Qwen3-8B2026.01 | 74.8 | |
| Hard-Length 8kBackbone=Qwen3-8B2026.01 | 73.3 | |
| Soft-LengthBackbone=Qwen3-8B2026.01 | 73.3 | |
| Hard-Length 16kBackbone=Qwen3-8B2026.01 | 72.5 | |
| Base modelBackbone=Qwen3-8B2026.01 | 72.1 | |
| No-ThinkingBackbone=Qwen3-8B2026.01 | 48 |