Out-of-Distribution Generalization Average (GPQA-Diamond, ARC-C, HumanEval)
55.35Pass@1TARPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TARPOBackbone=Qwen2.5-3B-Instruct2026.06 | 55.35 | 92.74 | 337.9 | |
| HRPOBackbone=Qwen2.5-3B-Instruct2026.06 | 54.03 | 92.22 | 407.2 | |
| GRPOBackbone=Qwen2.5-3B-Instruct2026.06 | 53.89 | 91.17 | 412.2 | |
| CoTBackbone=Qwen2.5-3B-Instruct2026.06 | 53.52 | 93.79 | 262.7 | |
| Pure LatentBackbone=Qwen2.5-3B-Instruct2026.06 | 52.67 | 65.95 | 254.7 | |
| Entropy-RoutedBackbone=Qwen2.5-3B-Instruct2026.06 | 52.57 | 92.47 | 243 |