Multilingual Mathematical Reasoning on MSVAMP in-domain
88.4AccuracyAdaMame-GRPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AdaMame-GRPOBackbone=QWEN3 4B, Size=35K2026.06 | 88.4 | 94.2 | 0.7 | |
| GRPOBackbone=QWEN3 4B, Size=35K2026.06 | 88 | 93.7 | 0.8 | |
| SFTBackbone=QWEN3 4B, Size=30K2026.06 | 86.9 | 93.2 | 1.1 | |
| M-Thinker Iter2Backbone=DISTILL-QWEN 1.5B, Size=50K2026.06 | 81.1 | 1.6 | 14.7 | |
| M-Thinker Iter1Backbone=DISTILL-QWEN 1.5B, Size=35K2026.06 | 78.9 | 3.12 | 17.3 | |
| PromptBackbone=QWEN3 4B2026.06 | 77.9 | 0.4 | 13 | |
| SFTBackbone=DISTILL-QWEN 1.5B, Size=30K2026.06 | 77.5 | 84 | 1.2 | |
| AdaMame-GRPOBackbone=DISTILL-QWEN 1.5B, Size=35K2026.06 | 77.5 | 85.2 | 1 | |
| GRPOBackbone=DISTILL-QWEN 1.5B, Size=35K2026.06 | 77.3 | 84.8 | 1 | |
| VanillaBackbone=QWEN3 4B2026.06 | 70.3 | 0.4 | 12.1 | |
| PromptBackbone=DISTILL-QWEN 1.5B2026.06 | 66.9 | 66.2 | 10.7 | |
| VanillaBackbone=DISTILL-QWEN 1.5B2026.06 | 66.5 | 64.1 | 9.7 |