Mathematical Reasoning on GSM8K (Top-1 Accuracy)
96.31Top-1 AccuracyBeam-Search
Evaluation results
| Method | Links | |
|---|---|---|
| Beam-SearchModel=Qwen2.5-7B-Instruct, Selection=Argmax, Scoring=PRM2025.10 | 96.31 | |
| PFModel=Qwen2.5-7B-Instruct, Selection=Argmax, Scoring=PRM2025.10 | 96.19 | |
| Best-of-NModel=Qwen2.5-7B-Instruct, Selection=Argmax, Scoring=ORM2025.10 | 96.12 | |
| ePFModel=Qwen2.5-7B-Instruct, Selection=Argmax, Scoring=PRM2025.10 | 95.74 | |
| Self-ConsistencyModel=Qwen2.5-7B-Instruct, Selection=MV, Scoring=-2025.10 | 94.65 | |
| ePFModel=Qwen2.5-1.5B-Instruct, Selection=Argmax, Scoring=PRM2025.10 | 93.85 | |
| PFModel=Qwen2.5-1.5B-Instruct, Selection=Argmax, Scoring=PRM2025.10 | 93.62 | |
| Base SamplingModel=Qwen2.5-7B-Instruct, Selection=-, Scoring=-2025.10 | 93.15 | |
| Best-of-NModel=Qwen2.5-1.5B-Instruct, Selection=Argmax, Scoring=ORM2025.10 | 92.84 | |
| Beam-SearchModel=Qwen2.5-1.5B-Instruct, Selection=Argmax, Scoring=PRM2025.10 | 91.48 | |
| Self-ConsistencyModel=Qwen2.5-1.5B-Instruct, Selection=MV, Scoring=-2025.10 | 82.19 | |
| Base SamplingModel=Qwen2.5-1.5B-Instruct, Selection=-, Scoring=-2025.10 | 67.38 |