Commonsense Reasoning on CoS-E
82.6AccuracySFT
Evaluation results
| Method | Links | |
|---|---|---|
| SFTBase Model=Qwen2.5-7B2026.05 | 82.6 | |
| Agree (Tok.)Base Model=Qwen2.5-7B, Protocol=Token-level agreement, Context Retrieval=True2026.05 | 82.2 | |
| UniSD*Base Model=Qwen2.5-7B2026.05 | 82.2 | |
| OPSDBase Model=Qwen2.5-7B2026.05 | 82 | |
| ContrastBase Model=Qwen2.5-7B, Protocol=Token-level contrastive learning2026.05 | 82 | |
| RawBase Model=Qwen2.5-7B2026.05 | 81.9 | |
| Agree (Seq.)Base Model=Qwen2.5-7B, Protocol=Sequence-level agreement, Context Retrieval=True2026.05 | 81.9 | |
| GKDBase Model=Qwen2.5-7B2026.05 | 81.8 | |
| Match (Joint)Base Model=Qwen2.5-7B, Protocol=Joint logit-representation matching2026.05 | 81.8 | |
| Match (Repr.)Base Model=Qwen2.5-7B, Protocol=Representation-only matching2026.05 | 81.7 | |
| ClipBase Model=Qwen2.5-7B, Protocol=Divergence clipping2026.05 | 81.7 | |
| EMABase Model=Qwen2.5-7B, Protocol=EMA teacher2026.05 | 81.4 | |
| SDFTBase Model=Qwen2.5-7B2026.05 | 81.2 | |
| SSDBase Model=Qwen2.5-7B2026.05 | 79.9 |