Factual Knowledge on Global-MMLU-Lite
58.5Seen AccuracyDPO
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DPOBackbone=Gemma3-4b-it2026.05 | 58.5 | 43.06 | 54.38 | |
| VanillaBackbone=Gemma3-4b-it2026.05 | 58.45 | 43.38 | 54.43 | |
| LRPOBackbone=Gemma3-4b-it2026.05 | 58.45 | 43.31 | 54.42 | |
| MAPOBackbone=Gemma3-4b-it2026.05 | 58.3 | 43.25 | 54.28 | |
| LIDRBackbone=Gemma3-4b-it2026.05 | 58.27 | 43.19 | 54.25 | |
| MPOBackbone=Gemma3-4b-it2026.05 | 58.2 | 43.81 | 54.37 | |
| GRPOBackbone=Gemma3-4b-it2026.05 | 57.34 | 41.25 | 53.05 | |
| VanillaBackbone=Qwen2.5-1.5b-it2026.05 | 54.55 | 34 | 49.07 | |
| MAPOBackbone=Qwen2.5-1.5b-it2026.05 | 54.55 | 34.06 | 49.08 | |
| MPOBackbone=Qwen2.5-1.5b-it2026.05 | 54.55 | 33.81 | 49.02 | |
| LIDRBackbone=Qwen2.5-1.5b-it2026.05 | 54.52 | 34 | 49.05 | |
| DPOBackbone=Qwen2.5-1.5b-it2026.05 | 54.36 | 34.31 | 49.02 | |
| GRPOBackbone=Qwen2.5-1.5b-it2026.05 | 54.09 | 34.88 | 48.97 | |
| LRPOBackbone=Qwen2.5-1.5b-it2026.05 | 53.05 | 32.81 | 47.65 | |
| GRPOBackbone=LLaMA3.2-1b-it2026.05 | 44.45 | 31.44 | 40.98 | |
| LRPOBackbone=LLaMA3.2-1b-it2026.05 | 43.8 | 31.19 | 40.43 | |
| MAPOBackbone=LLaMA3.2-1b-it2026.05 | 43.68 | 30.56 | 40.18 | |
| LIDRBackbone=LLaMA3.2-1b-it2026.05 | 43.66 | 30.81 | 40.23 | |
| VanillaBackbone=LLaMA3.2-1b-it2026.05 | 43.55 | 30.75 | 40.13 | |
| MPOBackbone=LLaMA3.2-1b-it2026.05 | 43.43 | 30.75 | 40.05 | |
| DPOBackbone=LLaMA3.2-1b-it2026.05 | 43.41 | 31.19 | 40.15 |