Personalization on Multi-Bench (MB)
94.84Win RateMIPO
Evaluation Results
| Method | Links | |
|---|---|---|
| MIPOBase Model=Llama-3.2-3B-Instruct2026.03 | 94.84 | |
| RLAIF (MI)Base Model=Llama-3.2-3B-Instruct2026.03 | 93.65 | |
| INFONCEBase Model=Llama-3.2-3B-Instruct2026.03 | 93.65 | |
| MIPOBase Model=Llama-3.2-1B-Instruct2026.03 | 93.26 | |
| INFONCEBase Model=Llama-3.2-1B-Instruct2026.03 | 91.67 | |
| SFTBase Model=Llama-3.2-3B-Instruct2026.03 | 90.48 | |
| RLAIFBase Model=Llama-3.2-3B-Instruct2026.03 | 90.48 | |
| PPO (MI)Base Model=Llama-3.2-3B-Instruct2026.03 | 90.08 | |
| SFT (Revision)Base Model=Llama-3.2-3B-Instruct2026.03 | 88.89 | |
| RLAIFBase Model=Llama-3.2-1B-Instruct2026.03 | 84.12 | |
| Personalized PromptingBase Model=Llama-3.2-3B-Instruct2026.03 | 83.33 | |
| RLAIF (MI)Base Model=Llama-3.2-1B-Instruct2026.03 | 82.93 | |
| PPO (MI)Base Model=Llama-3.2-1B-Instruct2026.03 | 82.22 | |
| Personalized PromptingBase Model=Llama-3.2-1B-Instruct2026.03 | 79.76 | |
| SFTBase Model=Llama-3.2-1B-Instruct2026.03 | 78.97 | |
| RLAIF (MI)Base Model=Qwen2.5-7B-Instruct2026.03 | 78.57 | |
| SFT (Revision)Base Model=Llama-3.2-1B-Instruct2026.03 | 77.78 | |
| Llama-3.2-3B-InstructBase Model=Llama-3.2-3B-Instruct2026.03 | 77.38 | |
| RLAIFBase Model=Qwen2.5-7B-Instruct2026.03 | 76.19 | |
| SFTBase Model=Qwen2.5-7B-Instruct2026.03 | 75 | |
| MIPOBase Model=Qwen2.5-1.5B-Instruct2026.03 | 74.6 | |
| MIPOBase Model=Qwen2.5-3B-Instruct2026.03 | 74.6 | |
| MIPOBase Model=Qwen2.5-7B-Instruct2026.03 | 73.81 | |
| INFONCEBase Model=Qwen2.5-3B-Instruct2026.03 | 73.02 | |
| SFT (Revision)Base Model=Qwen2.5-7B-Instruct2026.03 | 72.62 | |
| PPO (MI)Base Model=Qwen2.5-7B-Instruct2026.03 | 72.62 | |
| INFONCEBase Model=Qwen2.5-7B-Instruct2026.03 | 72.62 | |
| RLAIF (MI)Base Model=Qwen2.5-3B-Instruct2026.03 | 72.22 | |
| RLAIFBase Model=Qwen2.5-3B-Instruct2026.03 | 70.64 | |
| Personalized PromptingBase Model=Qwen2.5-7B-Instruct2026.03 | 70.24 | |
| PPO (MI)Base Model=Qwen2.5-3B-Instruct2026.03 | 66.67 | |
| SFT (Revision)Base Model=Qwen2.5-3B-Instruct2026.03 | 64.68 | |
| SFTBase Model=Qwen2.5-3B-Instruct2026.03 | 64.29 | |
| Personalized PromptingBase Model=Qwen2.5-3B-Instruct2026.03 | 63.1 | |
| Llama-3.2-1B-InstructBase Model=Llama-3.2-1B-Instruct2026.03 | 59.52 | |
| INFONCEBase Model=Qwen2.5-1.5B-Instruct2026.03 | 59.52 | |
| Qwen2.5-7B-InstructBase Model=Qwen2.5-7B-Instruct2026.03 | 47.62 | |
| Qwen2.5-3B-InstructBase Model=Qwen2.5-3B-Instruct2026.03 | 46.43 | |
| SFTBase Model=Qwen2.5-1.5B-Instruct2026.03 | 39.68 | |
| Personalized PromptingBase Model=Qwen2.5-1.5B-Instruct2026.03 | 39.29 | |
| RLAIFBase Model=Qwen2.5-1.5B-Instruct2026.03 | 39.29 | |
| PPO (MI)Base Model=Qwen2.5-1.5B-Instruct2026.03 | 36.9 | |
| SFT (Revision)Base Model=Qwen2.5-1.5B-Instruct2026.03 | 35.31 | |
| RLAIF (MI)Base Model=Qwen2.5-1.5B-Instruct2026.03 | 34.52 | |
| Qwen2.5-1.5B-InstructBase Model=Qwen2.5-1.5B-Instruct2026.03 | 15.48 |