Mathematical Reasoning on MATH (pass@8)
67.75Pass@8Rollout-GDRO
Evaluation Results
| Method | Links | |
|---|---|---|
| Rollout-GDROBackbone=Qwen3-8B-Base2026.01 | 67.75 | |
| Prompt-GDROBackbone=Qwen3-8B-Base2026.01 | 67.6 | |
| Prompt-GDROBackbone=Qwen3-4B-Base2026.01 | 63.7 | |
| Rollout-GDROBackbone=Qwen3-4B-Base2026.01 | 62.27 | |
| GRPO (Baseline)Backbone=Qwen3-8B-Base2026.01 | 62.04 | |
| GRPO (Baseline)Backbone=Qwen3-4B-Base2026.01 | 56.31 | |
| Rollout-GDROBackbone=Qwen3-1.7B-Base2026.01 | 56.14 | |
| Prompt-GDROBackbone=Qwen3-1.7B-Base2026.01 | 55.68 | |
| GRPO (Baseline)Backbone=Qwen3-1.7B-Base2026.01 | 50.74 |