Mathematical Reasoning on OLYMPIAD (mean@8)
43.62mean@8Rollout-GDRO
Evaluation Results
| Method | Links | |
|---|---|---|
| Rollout-GDROBackbone=Qwen3-8B-Base2026.01 | 43.62 | |
| Prompt-GDROBackbone=Qwen3-8B-Base2026.01 | 42.43 | |
| Prompt-GDROBackbone=Qwen3-4B-Base2026.01 | 40.98 | |
| Rollout-GDROBackbone=Qwen3-4B-Base2026.01 | 39.28 | |
| GRPO (Baseline)Backbone=Qwen3-8B-Base2026.01 | 36.92 | |
| GRPO (Baseline)Backbone=Qwen3-4B-Base2026.01 | 30.48 | |
| Rollout-GDROBackbone=Qwen3-1.7B-Base2026.01 | 26.71 | |
| Prompt-GDROBackbone=Qwen3-1.7B-Base2026.01 | 25.07 | |
| GRPO (Baseline)Backbone=Qwen3-1.7B-Base2026.01 | 23.07 |