Multilingual Mathematical Reasoning on MSVAMP out-of-domain
89.4AccuracyAdaMame-GRPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AdaMame-GRPOBackbone=QWEN3 4B, Size=35K2026.06 | 89.4 | 80.4 | 1 | |
| GRPOBackbone=QWEN3 4B, Size=35K2026.06 | 89 | 79.2 | 1.1 | |
| SFTBackbone=QWEN3 4B, Size=30K2026.06 | 88.7 | 78.4 | 1.3 | |
| PromptBackbone=QWEN3 4B2026.06 | 82.1 | 46.6 | 14.1 | |
| AdaMame-GRPOBackbone=DISTILL-QWEN 1.5B, Size=35K2026.06 | 76.8 | 50.1 | 1.6 | |
| GRPOBackbone=DISTILL-QWEN 1.5B, Size=35K2026.06 | 76.2 | 47.4 | 1.6 | |
| SFTBackbone=DISTILL-QWEN 1.5B, Size=30K2026.06 | 75.4 | 50.1 | 1.7 | |
| M-Thinker Iter2Backbone=DISTILL-QWEN 1.5B, Size=50K2026.06 | 74.6 | 27.5 | 14 | |
| PromptBackbone=DISTILL-QWEN 1.5B2026.06 | 73.4 | 52.1 | 6.1 | |
| M-Thinker Iter1Backbone=DISTILL-QWEN 1.5B, Size=35K2026.06 | 72.7 | 29.6 | 14.2 | |
| VanillaBackbone=DISTILL-QWEN 1.5B2026.06 | 72 | 50.2 | 6.3 | |
| VanillaBackbone=QWEN3 4B2026.06 | 64.7 | 46.3 | 13.8 |