Process-level verification on ProcessBench Aggregate (test)
56.5Avg F1Qwen2.5-Math-7B-PRM800K
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-Math-7B-PRM800Ktraining_data_source=high-quality manually annotated process-level data (PRM800K)2025.05 | 56.5 | |
| SCOPEAST Normalization=true, Code Translation=true, training_data_source=automatically constructed datasets2025.05 | 54.4 | |
| SCOPE (w/o AST Normalization)AST Normalization=false, Code Translation=true2025.05 | 53.6 | |
| SCOPE (w/o Code Translation)AST Normalization=true, Code Translation=false2025.05 | 50.6 | |
| SCOPE (Step Replacement)Step Compression Strategy=Step Replacement2025.05 | 44.6 | |
| SCOPE (Step Skipping)Step Compression Strategy=Step Skipping2025.05 | 44.6 | |
| Skywork-PRM-7Btraining_data_source=automatically constructed datasets2025.05 | 42.1 | |
| Skywork-PRM-1.5Btraining_data_source=automatically constructed datasets2025.05 | 36.4 | |
| Math-Shepherd-PRM-7Btraining_data_source=automatically constructed datasets2025.05 | 31.5 | |
| EurusPRM-Stage2training_data_source=automatically constructed datasets2025.05 | 31.3 | |
| EurusPRM-Stage1training_data_source=automatically constructed datasets2025.05 | 31.2 | |
| RLHFlow-PRM-Mistral-8Btraining_data_source=automatically constructed datasets2025.05 | 28.4 | |
| RLHFlow-PRM-Deepseek-8Btraining_data_source=automatically constructed datasets2025.05 | 26.6 |