Logical Reasoning on LogiQA (Pass@1, Variance)
0.88Pass@1 AccuracySFT w/ Short CoT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SFT w/ Short CoTMethod Category=BEHAVIORAL CLONING, Reasoning Strategy=Short CoT2025.06 | 0.88 | — | |
| DeepSeek-R1Method Category=FRONTIER REASONERS2025.06 | 0.85 | — | |
| RULEREASONER-8BMethod Category=RULEREASONER (Ours), Model Scale=8B2025.06 | 0.835 | 0.3 | |
| OpenAI o1Method Category=FRONTIER REASONERS2025.06 | 0.821 | — | |
| Claude-3.7-SonnetMethod Category=FRONTIER REASONERS2025.06 | 0.815 | — | |
| Chain-of-LogicMethod Category=PRIOR RBRS2025.06 | 0.806 | — | |
| ADARFTMethod Category=CURRICULUM LEARNING2025.06 | 0.805 | — | |
| RULEREASONER-4BMethod Category=RULEREASONER (Ours), Model Scale=4B2025.06 | 0.805 | 0.7 | |
| HtTMethod Category=PRIOR RBRS2025.06 | 0.791 | — | |
| RGFBMethod Category=PRIOR RBRS2025.06 | 0.791 | — | |
| SFT w/ Long CoTMethod Category=BEHAVIORAL CLONING, Reasoning Strategy=Long CoT2025.06 | 0.791 | — | |
| OpenAI o3-miniMethod Category=FRONTIER REASONERS2025.06 | 0.776 | — | |
| SFT w/o CoTMethod Category=BEHAVIORAL CLONING, Reasoning Strategy=w/o CoT2025.06 | 0.761 | — | |
| DAPOMethod Category=ADVANCED RLVRS2025.06 | 0.746 | — | |
| Easy-to-hard RLMethod Category=CURRICULUM LEARNING2025.06 | 0.746 | — | |
| Data-balance RLMethod Category=CURRICULUM LEARNING2025.06 | 0.731 | — | |
| GRPOMethod Category=ADVANCED RLVRS2025.06 | 0.701 | — | |
| Dr. GRPOMethod Category=ADVANCED RLVRS2025.06 | 0.656 | — |