Reasoning on PolyMath 5 Unseen Languages
51.3PolyMath ScoreMGDA
Evaluation Results
| Method | Links | |
|---|---|---|
| MGDABackbone=Qwen3-8B-Base2026.06 | 51.3 | |
| CAGradBackbone=Qwen3-8B-Base2026.06 | 49.8 | |
| PCGradBackbone=Qwen3-8B-Base2026.06 | 49.1 | |
| V. SFTBackbone=Qwen3-8B-Base2026.06 | 49 | |
| BaseBackbone=Qwen3-8B-Base2026.06 | 47.6 | |
| MGDABackbone=Qwen3-4B-Base2026.06 | 47.4 | |
| CAGradBackbone=Qwen3-4B-Base2026.06 | 47.2 | |
| PCGradBackbone=Qwen3-4B-Base2026.06 | 46.1 | |
| V. SFTBackbone=Qwen3-4B-Base2026.06 | 45.8 | |
| BaseBackbone=Qwen3-4B-Base2026.06 | 44 | |
| CAGradBackbone=Llama-3.1-8B2026.06 | 26.6 | |
| PCGradBackbone=Llama-3.1-8B2026.06 | 26.1 | |
| MGDABackbone=Llama-3.1-8B2026.06 | 25.8 | |
| PCGradBackbone=Meta-Llama-3-8B2026.06 | 24.5 | |
| MGDABackbone=Meta-Llama-3-8B2026.06 | 24.3 | |
| CAGradBackbone=Meta-Llama-3-8B2026.06 | 23.8 | |
| V. SFTBackbone=Llama-3.1-8B2026.06 | 22.8 | |
| V. SFTBackbone=Meta-Llama-3-8B2026.06 | 21.5 | |
| BaseBackbone=Llama-3.1-8B2026.06 | 13.8 | |
| BaseBackbone=Meta-Llama-3-8B2026.06 | 13.5 |