Scientific Reasoning on GPQA (Accuracy, Inference Latency (s))
69.89AccuracyHDPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HDPOBackbone=Qwen2.5-Math-7B2026.06 | 69.89 | 0.28 | |
| Critique-GRPOBackbone=Qwen2.5-Math-7B2026.06 | 67.28 | 0.33 | |
| LUFFYBackbone=Qwen2.5-Math-7B2026.06 | 61.51 | 0.24 | |
| GRPOBackbone=Qwen2.5-Math-7B2026.06 | 60.18 | 0.17 | |
| SimpleRLBackbone=Qwen2.5-Math-7B2026.06 | 47.31 | 0.13 | |
| EDGE-GRPOBackbone=Qwen2.5-Math-7B2026.06 | 46.76 | 0.14 | |
| OatBackbone=Qwen2.5-Math-7B2026.06 | 44.43 | 0.13 |