Process-level Reward Modeling on PROCESSBENCH Olymp.Bench
3.3ErrorSPARE-Llama3-8B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SPARE-Llama3-8B# Train=40.5K2025.06 | 3.3 | 87.6 | 6.4 | |
| RLHFlow-Deepseek-8B# Train=250K2025.06 | 10.1 | 51 | 16.9 | |
| SPARE-Qwen2.5-3B# Train=40.5K2025.06 | 11.1 | 85 | 19.6 | |
| Math-Shepherd-7B# Train=440K2025.06 | 15 | 71.1 | 24.8 | |
| Skywork-7B2025.06 | 17.9 | 31.9 | 22.9 | |
| Qwen-2.5-Math-7B-PRM800K (Human)# Train=250K2025.06 | 35.7 | 87.3 | 50.7 |