Scientific Reasoning on GPQA 1.0 (test)
53.8AccuracyA3PO
Evaluation Results
| Method | Links | |
|---|---|---|
| A3POModel=Deepseek-R1-Distill-Qwen-7B2025.12 | 53.8 | |
| W-REINFORCEModel=Deepseek-R1-Distill-Qwen-7B2025.12 | 51.4 | |
| Lp-RegModel=Deepseek-R1-Distill-Qwen-7B2025.12 | 51.2 | |
| DAPO w/ Fork TokensModel=Deepseek-R1-Distill-Qwen-7B2025.12 | 50.4 | |
| A3POModel=Qwen3-8B-Base2025.12 | 50.2 | |
| DAPOModel=Deepseek-R1-Distill-Qwen-7B2025.12 | 50.2 | |
| GRPOModel=Deepseek-R1-Distill-Qwen-7B2025.12 | 48.4 | |
| Lp-RegModel=Qwen3-8B-Base2025.12 | 47.8 | |
| W-REINFORCEModel=Qwen3-8B-Base2025.12 | 47.4 | |
| DAPO w/ Fork TokensModel=Qwen3-8B-Base2025.12 | 47.2 | |
| DAPOModel=Qwen3-8B-Base2025.12 | 45.8 | |
| GRPOModel=Qwen3-8B-Base2025.12 | 45.3 | |
| A3POModel=Qwen2.5-7B-Math2025.12 | 39.1 | |
| Lp-RegModel=Qwen2.5-7B-Math2025.12 | 36.9 | |
| DAPO w/ Fork TokensModel=Qwen2.5-7B-Math2025.12 | 36.5 | |
| W-REINFORCEModel=Qwen2.5-7B-Math2025.12 | 36.2 | |
| DAPOModel=Qwen2.5-7B-Math2025.12 | 34.6 | |
| GRPOModel=Qwen2.5-7B-Math2025.12 | 33.7 |