Deductive Reasoning on ProofWriter (Pass@1, Variance)
97.4Pass@1ADARFT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ADARFTMethod Category=CURRICULUM LEARNING2025.06 | 97.4 | — | |
| RULEREASONER-8BMethod Category=RULEREASONER (Ours), Model Scale=8B2025.06 | 97 | 0.2 | |
| Easy-to-hard RLMethod Category=CURRICULUM LEARNING2025.06 | 96.8 | — | |
| GRPOMethod Category=ADVANCED RLVRS2025.06 | 96.4 | — | |
| RULEREASONER-4BMethod Category=RULEREASONER (Ours), Model Scale=4B2025.06 | 96.3 | 0.3 | |
| Dr. GRPOMethod Category=ADVANCED RLVRS2025.06 | 95.6 | — | |
| Data-balance RLMethod Category=CURRICULUM LEARNING2025.06 | 95.6 | — | |
| DAPOMethod Category=ADVANCED RLVRS2025.06 | 94.8 | — | |
| OpenAI o3-miniMethod Category=FRONTIER REASONERS2025.06 | 93 | — | |
| Chain-of-LogicMethod Category=PRIOR RBRS2025.06 | 92 | — | |
| OpenAI o1Method Category=FRONTIER REASONERS2025.06 | 91 | — | |
| Claude-3.7-SonnetMethod Category=FRONTIER REASONERS2025.06 | 90 | — | |
| SFT w/ Long CoTMethod Category=BEHAVIORAL CLONING, Reasoning Strategy=Long CoT2025.06 | 89.2 | — | |
| SFT w/o CoTMethod Category=BEHAVIORAL CLONING, Reasoning Strategy=w/o CoT2025.06 | 88.8 | — | |
| HtTMethod Category=PRIOR RBRS2025.06 | 88 | — | |
| RGFBMethod Category=PRIOR RBRS2025.06 | 88 | — | |
| SFT w/ Short CoTMethod Category=BEHAVIORAL CLONING, Reasoning Strategy=Short CoT2025.06 | 87 | — | |
| DeepSeek-R1Method Category=FRONTIER REASONERS2025.06 | 27 | — |