General Language Capabilities on MMLU, GSM8K, GPQA, HumanEval, TruthfulQA, IFEval Aggregate
71.2Average ScoreGRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GRPOBackbone=LLaMA-3.1-8B2025.05 | 71.2 | |
| TI-DPOBackbone=LLaMA-3.1-8B2025.05 | 71.1 | |
| TPOBackbone=LLaMA-3.1-8B2025.05 | 70 | |
| CPOBackbone=LLaMA-3.1-8B2025.05 | 68.9 | |
| KTOBackbone=LLaMA-3.1-8B2025.05 | 68 | |
| DPOBackbone=LLaMA-3.1-8B2025.05 | 66.8 | |
| TDPOBackbone=LLaMA-3.1-8B2025.05 | 65.8 | |
| SFTBackbone=LLaMA-3.1-8B2025.05 | 65.2 | |
| IPOBackbone=LLaMA-3.1-8B2025.05 | 62.5 | |
| SIMPOBackbone=LLaMA-3.1-8B2025.05 | 62.4 |