General Multilingual Capability on Aggregate 5 Unseen Languages
75.1Average ScoreMGDA
Evaluation Results
| Method | Links | |
|---|---|---|
| MGDABackbone=Qwen3-8B-Base2026.06 | 75.1 | |
| CAGradBackbone=Qwen3-8B-Base2026.06 | 74.8 | |
| BaseBackbone=Qwen3-8B-Base2026.06 | 74.4 | |
| PCGradBackbone=Qwen3-8B-Base2026.06 | 74.3 | |
| V. SFTBackbone=Qwen3-8B-Base2026.06 | 73 | |
| BaseBackbone=Qwen3-4B-Base2026.06 | 71.2 | |
| MGDABackbone=Qwen3-4B-Base2026.06 | 70.9 | |
| CAGradBackbone=Qwen3-4B-Base2026.06 | 70.7 | |
| PCGradBackbone=Qwen3-4B-Base2026.06 | 70 | |
| V. SFTBackbone=Qwen3-4B-Base2026.06 | 68.2 | |
| CAGradBackbone=Llama-3.1-8B2026.06 | 54.3 | |
| MGDABackbone=Llama-3.1-8B2026.06 | 54.2 | |
| PCGradBackbone=Llama-3.1-8B2026.06 | 54.1 | |
| MGDABackbone=Meta-Llama-3-8B2026.06 | 52.8 | |
| PCGradBackbone=Meta-Llama-3-8B2026.06 | 52.7 | |
| CAGradBackbone=Meta-Llama-3-8B2026.06 | 52.4 | |
| V. SFTBackbone=Llama-3.1-8B2026.06 | 51.8 | |
| V. SFTBackbone=Meta-Llama-3-8B2026.06 | 50.7 | |
| BaseBackbone=Llama-3.1-8B2026.06 | 50.1 | |
| BaseBackbone=Meta-Llama-3-8B2026.06 | 44.3 |