Multilingual Mathematical Reasoning on MSVAMP (all)
89AccuracyAdaMame-GRPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AdaMame-GRPOBackbone=QWEN3 4B, Size=35K2026.06 | 89 | 86 | 0.9 | |
| GRPOBackbone=QWEN3 4B, Size=35K2026.06 | 88.6 | 85 | 1 | |
| SFTBackbone=QWEN3 4B, Size=30K2026.06 | 88 | 84.3 | 1.2 | |
| PromptBackbone=QWEN3 4B2026.06 | 80.4 | 28.1 | 13.6 | |
| AdaMame-GRPOBackbone=DISTILL-QWEN 1.5B, Size=35K2026.06 | 77 | 60.7 | 1.4 | |
| M-Thinker Iter2Backbone=DISTILL-QWEN 1.5B, Size=50K2026.06 | 76.6 | 19.7 | 15.2 | |
| GRPOBackbone=DISTILL-QWEN 1.5B, Size=35K2026.06 | 76.6 | 58.7 | 1.4 | |
| SFTBackbone=DISTILL-QWEN 1.5B, Size=30K2026.06 | 76 | 60.3 | 1.5 | |
| M-Thinker Iter1Backbone=DISTILL-QWEN 1.5B, Size=35K2026.06 | 74.6 | 21.7 | 15.2 | |
| PromptBackbone=DISTILL-QWEN 1.5B2026.06 | 71.4 | 56.3 | 7.5 | |
| VanillaBackbone=DISTILL-QWEN 1.5B2026.06 | 70.4 | 54.4 | 7.3 | |
| VanillaBackbone=QWEN3 4B2026.06 | 67 | 27.9 | 13.1 |