Reward model verification on HH-RLHF
47.3Win RateAISP
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AISPBase Model=Gemma3 4B, Reward Model=Eurus2025.10 | 47.3 | — | — | — | |
| AISPBase Model=Llama3 8B, Reward Model=Eurus2025.10 | 46.3 | — | — | — | |
| AISPBase Model=Gemma3 4B, Reward Model=UltraRM2025.10 | 44 | — | — | — | |
| BoN (top-p)Base Model=Llama3 8B, Reward Model=Eurus2025.10 | 43.7 | — | — | — | |
| BoN (top-p)Base Model=Gemma3 4B, Reward Model=Eurus2025.10 | 41.7 | — | — | — | |
| AISPBase Model=Llama3 8B, Reward Model=UltraRM2025.10 | 41 | — | — | — | |
| BoN (top-p)Base Model=Llama3 8B, Reward Model=UltraRM2025.10 | 41 | — | — | — | |
| BoN (top-p)Base Model=Gemma3 4B, Reward Model=UltraRM2025.10 | 36.7 | — | — | — | |
| BoN (top-p)Base Model=Vicuna 7B, Reward Model=Eurus2025.10 | 33 | — | — | — | |
| AISPBase Model=Vicuna 7B, Reward Model=Eurus2025.10 | 32.3 | — | — | — | |
| AISPBase Model=Vicuna 7B, Reward Model=UltraRM2025.10 | 25 | — | — | — | |
| BoN (top-p)Base Model=Vicuna 7B, Reward Model=UltraRM2025.10 | 25 | — | — | — | |
| BT RM2025.05 | — | 0.7438 | 0.7323 | 0.7381 | |
| FC RM2025.05 | — | 0.743 | 0.7362 | 0.7396 | |
| FR RM2025.05 | — | 0.7349 | 0.7362 | 0.7355 |