Scientific Reasoning on GPQA Diamond (Average Accuracy)
0.405Average AccuracyRePro
Evaluation Results
| Method | Links | |
|---|---|---|
| ReProBackbone=Qwen3-4B-Base, Sample Count=Avg@4, Evaluation Domain=Out-of-domain, Base Method=GRPO2025.12 | 0.405 | |
| GRPOBackbone=Qwen3-4B-Base, Sample Count=Avg@4, Evaluation Domain=Out-of-domain2025.12 | 0.397 | |
| OriginalBackbone=Qwen3-4B-Base, Sample Count=Avg@4, Evaluation Domain=Out-of-domain2025.12 | 0.167 |