Logical Reasoning on ProofWriter (Acc, TNFT, Tokens, Delay)
92.2AccVanilla
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VanillaBackbone=Qwen3-4B2026.05 | 92.2 | 195.14 | 1,125.13 | 31.6 | |
| OursBackbone=Qwen3-4B2026.05 | 91.8 | 0 | 736.84 | 24.56 | |
| VanillaBackbone=Qwen3-1.7B2026.05 | 90.7 | 195.14 | 1,227.19 | 26.46 | |
| BaseBackbone=Qwen3-4B2026.05 | 90.5 | 197.14 | 2,066.1 | 62.94 | |
| OursBackbone=Qwen3-1.7B2026.05 | 90.2 | 0 | 714.79 | 20.42 | |
| BaseBackbone=Qwen3-1.7B2026.05 | 88.8 | 197.14 | 2,110.24 | 49.62 |