Mathematical Reasoning on MATH-500 (Context Length Accuracy)
62.4Accuracy @ 128BiRM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| BiRMBase Model=Qwen2.5-7B2025.03 | 62.4 | 62.3 | — | |
| PRMBase Model=Qwen2.5-7B2025.03 | 60.3 | 60.1 | 58.4 | |
| Math-ShepherdBase Model=Qwen2.5-7B2025.03 | 60.3 | 60.2 | 60.4 | |
| ER-PRMBase Model=Qwen2.5-7B2025.03 | 59.7 | 59.2 | 59 | |
| ORMBase Model=Qwen2.5-7B2025.03 | 59.6 | 59.9 | 59.4 | |
| Majority VoteBase Model=Qwen2.5-7B2025.03 | 57.3 | 57.5 | 57.6 | |
| BiRMBase Model=Qwen2.5-3B2025.03 | 55.9 | 56.1 | 57.4 | |
| PRMBase Model=Qwen2.5-3B2025.03 | 54.6 | 54.1 | 54.2 | |
| Math-ShepherdBase Model=Qwen2.5-3B2025.03 | 53.2 | 53.4 | 53.8 | |
| ER-PRMBase Model=Qwen2.5-3B2025.03 | 52.6 | 52.3 | 52 | |
| Majority VoteBase Model=Qwen2.5-3B2025.03 | 52.5 | 53 | 53.8 | |
| ORMBase Model=Qwen2.5-3B2025.03 | 52.1 | 51.8 | 52.2 | |
| BiRMBase Model=Llama3.1-8B2025.03 | 45.4 | 45.4 | 45.6 | |
| GreedyBase Model=Qwen2.5-7B2025.03 | 45 | 45 | 45 | |
| Math-ShepherdBase Model=Llama3.1-8B2025.03 | 42.7 | 42.9 | 43.6 | |
| PRMBase Model=Llama3.1-8B2025.03 | 42.5 | 42.2 | 41.8 | |
| ORMBase Model=Llama3.1-8B2025.03 | 41.5 | 40.9 | 40.8 | |
| ER-PRMBase Model=Llama3.1-8B2025.03 | 41.3 | 41 | 40.2 | |
| GreedyBase Model=Qwen2.5-3B2025.03 | 40.2 | 40.2 | 40.2 | |
| Majority VoteBase Model=Llama3.1-8B2025.03 | 39.2 | 40.2 | 41.1 | |
| GreedyBase Model=Llama3.1-8B2025.03 | 31.2 | 31.2 | 31.2 |