Logical Reasoning on Reasoning Gym Leg Counting
50.67AccuracyTokUR (EU)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TokUR (EU)Base Model=Qwen-2.5-3B-Instruct2025.05 | 50.67 | 69.66 | 51.27 | |
| TokUR (TU)Base Model=Qwen-2.5-3B-Instruct2025.05 | 48.67 | 69.58 | 48.16 | |
| TokUR (AU)Base Model=Qwen-2.5-3B-Instruct2025.05 | 48.67 | 69.51 | 47.69 | |
| CoT (Lower-Bound)Base Model=Qwen-2.5-3B-Instruct2025.05 | 35 | — | — | |
| Self-CertaintyBase Model=Qwen-2.5-3B-Instruct2025.05 | 34 | 50.46 | 32.88 | |
| LLBase Model=Qwen-2.5-3B-Instruct2025.05 | 32.67 | 37.92 | 27.1 | |
| DeepConfBase Model=Qwen-2.5-3B-Instruct2025.05 | 30 | 42.48 | 28.96 | |
| PEBase Model=Qwen-2.5-3B-Instruct2025.05 | 26 | 38.87 | 27.44 |