Multilingual Knowledge Evaluation on Global-MMLU Low
52.31AccuracyGRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GRPOBackbone=Qwen-2.5-7B2026.06 | 52.31 | |
| BaselineBackbone=Qwen-2.5-7B2026.06 | 52.19 | |
| CPT + SFTBackbone=Qwen-2.5-7B2026.06 | 51.19 | |
| CPT + GRPOBackbone=Qwen-2.5-7B2026.06 | 50 | |
| CPTBackbone=Qwen-2.5-7B2026.06 | 49.19 | |
| SFTBackbone=Qwen-2.5-7B2026.06 | 46.75 | |
| GRPOBackbone=OLMo-2-1124-7B2026.06 | 32 | |
| BaselineBackbone=OLMo-2-1124-7B2026.06 | 31.79 | |
| CPT + SFTBackbone=OLMo-2-1124-7B2026.06 | 30.88 | |
| SFTBackbone=OLMo-2-1124-7B2026.06 | 30.32 | |
| CPT + GRPOBackbone=OLMo-2-1124-7B2026.06 | 29.61 | |
| CPTBackbone=OLMo-2-1124-7B2026.06 | 29.32 |