Instruction Following Evaluation on AlpacaEval 2
48.14Win RateVRM-PPO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| VRM-PPOBackbone=Qwen 3-8B2026.03 | 48.14 | 52.03 | |
| VRM-PPOBackbone=Qwen2.5-7B2026.03 | 45.3 | 50.38 | |
| PPOBackbone=Qwen 3-8B2026.03 | 44.75 | 55.33 | |
| WPOBackbone=Qwen2.5-7B2026.03 | 44.11 | 40.06 | |
| SIMPOBackbone=Qwen2.5-7B2026.03 | 40.03 | 40.78 | |
| PPOBackbone=Qwen2.5-7B2026.03 | 39.52 | 39.79 | |
| KTOBackbone=Qwen2.5-7B2026.03 | 38.12 | 36.51 | |
| SELECTIVE DPOBackbone=Qwen2.5-7B2026.03 | 38.02 | 39.21 | |
| IPOBackbone=Qwen2.5-7B2026.03 | 37.95 | 36.43 | |
| DPOBackbone=Qwen2.5-7B2026.03 | 37.24 | 36.84 | |
| SIMPOBackbone=Qwen 3-8B2026.03 | 29.38 | 27.94 | |
| SELECTIVE DPOBackbone=Qwen 3-8B2026.03 | 29.27 | 27.53 | |
| WPOBackbone=Qwen 3-8B2026.03 | 28.16 | 25.35 | |
| IPOBackbone=Qwen 3-8B2026.03 | 27.79 | 25.87 | |
| KTOBackbone=Qwen 3-8B2026.03 | 22.94 | 39.68 | |
| DPOBackbone=Qwen 3-8B2026.03 | 18.53 | 31.5 |