LLM Alignment on HHRLHF
-0.02Average Oracle RewardAAD
Evaluation results
| Method | Links | ||
|---|---|---|---|
| AADBase Model=Qwen 4B2025.09 | -0.02 | — | |
| Bo2Base Model=Qwen 4B2025.09 | -0.22 | 59 | |
| AADBase Model=Llama 8B2025.09 | -0.34 | — | |
| EFTBase Model=Qwen 4B2025.09 | -0.47 | 64 | |
| Greedy DPOBase Model=Qwen 4B2025.09 | -0.49 | 63 | |
| Greedy SFTBase Model=Qwen 4B2025.09 | -0.53 | 64 | |
| AADBase Model=Qwen 0.6B2025.09 | -0.61 | — | |
| Bo2Base Model=Llama 8B2025.09 | -0.91 | 61 | |
| AADBase Model=Llama 3B2025.09 | -0.97 | — | |
| EFTBase Model=Llama 8B2025.09 | -0.98 | 57 | |
| Bo2Base Model=Qwen 0.6B2025.09 | -1.06 | 64 | |
| Greedy DPOBase Model=Llama 8B2025.09 | -1.08 | 60 | |
| EFTBase Model=Qwen 0.6B2025.09 | -1.12 | 57 | |
| Greedy SFTBase Model=Llama 8B2025.09 | -1.13 | 61 | |
| Greedy DPOBase Model=Qwen 0.6B2025.09 | -1.25 | 60 | |
| Greedy SFTBase Model=Qwen 0.6B2025.09 | -1.36 | 65 | |
| Bo2Base Model=Llama 3B2025.09 | -1.65 | 64 | |
| EFTBase Model=Llama 3B2025.09 | -1.74 | 61 | |
| Greedy DPOBase Model=Llama 3B2025.09 | -1.83 | 61 | |
| Greedy SFTBase Model=Llama 3B2025.09 | -1.89 | 62 |