Multilingual Knowledge Evaluation on Global-MMLU High
68.35AccuracyGRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GRPOBackbone=Qwen-2.5-7B2026.06 | 68.35 | |
| BaselineBackbone=Qwen-2.5-7B2026.06 | 68.15 | |
| CPT + GRPOBackbone=Qwen-2.5-7B2026.06 | 65.2 | |
| SFTBackbone=Qwen-2.5-7B2026.06 | 64.45 | |
| CPT + SFTBackbone=Qwen-2.5-7B2026.06 | 64.2 | |
| CPTBackbone=Qwen-2.5-7B2026.06 | 63.95 | |
| GRPOBackbone=OLMo-2-1124-7B2026.06 | 39.22 | |
| BaselineBackbone=OLMo-2-1124-7B2026.06 | 38.72 | |
| CPTBackbone=OLMo-2-1124-7B2026.06 | 37.41 | |
| CPT + GRPOBackbone=OLMo-2-1124-7B2026.06 | 36.34 | |
| CPT + SFTBackbone=OLMo-2-1124-7B2026.06 | 36.15 | |
| SFTBackbone=OLMo-2-1124-7B2026.06 | 35.4 |