Mathematical Reasoning on Minerva-Math (@4, @16, @64)
26.5Accuracy @4Q-RM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Q-RMBase Model=Qwen3-8B Base2026.04 | 26.5 | 27 | 25.7 | |
| IPVRMBase Model=Qwen3-8B Base2026.04 | 25.4 | 25 | 26.8 | |
| Implicit PRMBase Model=Qwen3-8B Base2026.04 | 24.6 | 24.3 | 26.5 | |
| DPO-RMBase Model=Qwen3-8B Base2026.04 | 24.3 | 24.6 | 26.8 | |
| EndoRMBase Model=Qwen3-8B Base2026.04 | 19.5 | 22.8 | 21.7 | |
| DPO-RMBase Model=Qwen3-0.6B Base2026.04 | 10.7 | 9.9 | 11 | |
| Implicit PRMBase Model=Qwen3-0.6B Base2026.04 | 10.3 | 11.4 | 11 | |
| IPVRMBase Model=Qwen3-0.6B Base2026.04 | 10.3 | 11 | 11.4 | |
| Q-RMBase Model=Qwen3-0.6B Base2026.04 | 9.2 | 12.1 | 10.7 | |
| IPVRMBase Model=Llama3.2-3B2026.04 | 7 | 8.5 | 7.7 | |
| EndoRMBase Model=Qwen3-0.6B Base2026.04 | 6.3 | 8.8 | 7.7 | |
| DPO-RMBase Model=Llama3.2-3B2026.04 | 6.3 | 9.9 | 7.7 | |
| Q-RMBase Model=Llama3.2-3B2026.04 | 5.9 | 9.9 | 8.1 | |
| EndoRMBase Model=Llama3.2-3B2026.04 | 5.9 | 4.8 | 5.5 | |
| Implicit PRMBase Model=Llama3.2-3B2026.04 | 5.9 | 9.2 | 8.5 | |
| DPO-RMBase Model=Llama3.2-1B2026.04 | 3.3 | 3.7 | 2.9 | |
| Implicit PRMBase Model=Llama3.2-1B2026.04 | 2.9 | 2.6 | 3.7 | |
| Q-RMBase Model=Llama3.2-1B2026.04 | 2.2 | 2.6 | 3.3 | |
| IPVRMBase Model=Llama3.2-1B2026.04 | 1.5 | 2.6 | 4.4 | |
| EndoRMBase Model=Llama3.2-1B2026.04 | 1.1 | 1.8 | 2.2 |