Multilingual Mathematical Reasoning on MMATH Out-of-Domain Languages (test)
30.1Vietnamese AccuracyQwen2.5-7B-Instruct + LANG
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen2.5-7B-Instruct + LANGBackbone=Qwen2.5-7B-Instruct, Method=LANG2026.05 | 30.1 | 29.5 | — | 24.5 | 32.2 | 31.2 | 29.5 | |
| Qwen2.5-7B-Instruct + mGRPOBackbone=Qwen2.5-7B-Instruct, Method=mGRPO2026.05 | 25.9 | 27.3 | — | 25 | 29.9 | 28.6 | 27.3 | |
| Qwen2.5-7B-Instruct + LC-GRPOBackbone=Qwen2.5-7B-Instruct, Method=LC-GRPO2026.05 | 25.2 | 26.8 | — | 24.9 | 29.2 | 27.8 | 26.8 | |
| Qwen2.5-7B-Instruct + M-ThinkerBackbone=Qwen2.5-7B-Instruct, Method=M-Thinker2026.05 | 24.6 | 25.2 | — | 22.8 | 26.7 | 26.7 | 25.2 | |
| Qwen2.5-7B-Instruct + QRTBackbone=Qwen2.5-7B-Instruct, Method=QRT2026.05 | 22.4 | 19.1 | — | 14.3 | 16.2 | 23.6 | 19.1 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct, Metric Type=Accuracy2026.05 | 20.8 | — | 23.6 | 24.6 | 24.1 | 25.1 | 23.6 | |
| Qwen2.5-7B-Instruct + LCPBackbone=Qwen2.5-7B-Instruct, Method=LCP2026.05 | 19.8 | 21.5 | — | 19.8 | 23.4 | 22.9 | 21.5 | |
| Qwen2.5-3B-Instruct + LANGBackbone=Qwen2.5-3B-Instruct, Method=LANG2026.05 | 19.6 | 16.8 | — | 14.8 | 18.1 | 14.6 | 16.8 | |
| Qwen2.5-7B-Instruct + DITBackbone=Qwen2.5-7B-Instruct, Method=DIT2026.05 | 18.4 | 18.2 | — | 13.3 | 19.8 | 21.4 | 18.2 | |
| Qwen2.5-3B-Instruct + LC-GRPOBackbone=Qwen2.5-3B-Instruct, Method=LC-GRPO2026.05 | 17.3 | 16.1 | — | 13.7 | 18.2 | 15.1 | 16.1 | |
| Qwen2.5-3B-Instruct + mGRPOBackbone=Qwen2.5-3B-Instruct, Method=mGRPO2026.05 | 16.3 | 15.6 | — | 14.5 | 14.9 | 16.7 | 15.6 | |
| Qwen2.5-3B-InstructBackbone=Qwen2.5-3B-Instruct, Metric Type=Accuracy2026.05 | 15.9 | — | 16 | 13 | 18.5 | 16.6 | 16 | |
| Qwen2.5-3B-Instruct + DITBackbone=Qwen2.5-3B-Instruct, Method=DIT2026.05 | 14.9 | 14.3 | — | 12.3 | 17.7 | 12.4 | 14.3 | |
| Qwen2.5-3B-Instruct + M-ThinkerBackbone=Qwen2.5-3B-Instruct, Method=M-Thinker2026.05 | 14.6 | 13.7 | — | 12.1 | 17.1 | 11.1 | 13.7 | |
| Qwen2.5-3B-Instruct + QRTBackbone=Qwen2.5-3B-Instruct, Method=QRT2026.05 | 14 | 13.5 | — | 12.2 | 16.7 | 10.9 | 13.5 | |
| Qwen2.5-3B-Instruct + M-SFTBackbone=Qwen2.5-3B-Instruct, Method=M-SFT2026.05 | 13.3 | 13.4 | — | 11.9 | 16.4 | 12.1 | 13.4 | |
| Qwen2.5-3B-Instruct + LCPBackbone=Qwen2.5-3B-Instruct, Method=LCP2026.05 | 13.1 | 13.3 | — | 10.9 | 15.8 | 13.3 | 13.3 | |
| Qwen2.5-7B-Instruct + M-SFTBackbone=Qwen2.5-7B-Instruct, Method=M-SFT2026.05 | 12 | 15.4 | — | 16.3 | 16.3 | 17.1 | 15.4 | |
| Qwen2.5-3B-InstructBackbone=Qwen2.5-3B-Instruct, Metric Type=LC&Acc2026.05 | 5.8 | 3.1 | — | 0.9 | 0.4 | 5.1 | 3.1 | |
| Qwen2.5-3B-Instruct + Vanilla GRPOBackbone=Qwen2.5-3B-Instruct, Method=Vanilla GRPO2026.05 | 4.6 | 9.5 | — | 1.1 | 1.1 | 1.3 | 9.5 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct, Metric Type=LC&Acc2026.05 | 1.3 | 1.5 | — | 1.8 | 0.7 | 2.1 | 1.5 | |
| Qwen2.5-7B-Instruct + Vanilla GRPOBackbone=Qwen2.5-7B-Instruct, Method=Vanilla GRPO2026.05 | 0.2 | 0.1 | — | 0 | 0 | 0 | 0.1 |