Multi-objective personalized alignment on Multifaceted
51AMRMATO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MATOBackbone=Qwen3-8B, Judge Model=claude-sonnet-4-52026.05 | 51 | 3.7 | 2.61 | |
| OPADBackbone=Qwen3-8B, Judge Model=claude-sonnet-4-52026.05 | 50 | 3.7 | 2.57 | |
| Preference PromptBackbone=Qwen3-8B, Judge Model=claude-sonnet-4-52026.05 | 47 | 3.61 | 2.47 | |
| Linear AlignmentBackbone=Qwen3-8B, Judge Model=claude-sonnet-4-52026.05 | 46 | 3.59 | 2.48 | |
| CoSBackbone=Qwen3-8B, Judge Model=claude-sonnet-4-52026.05 | 43 | 3.45 | 2.36 | |
| Preference PromptBackbone=Llama-3.1-8B, Judge Model=claude-sonnet-4-52026.05 | 42 | 3.39 | 2.32 | |
| MATOBackbone=Llama-3.1-8B, Judge Model=claude-sonnet-4-52026.05 | 42 | 3.45 | 2.4 | |
| OPADBackbone=Llama-3.1-8B, Judge Model=claude-sonnet-4-52026.05 | 39 | 3.4 | 2.33 | |
| AmuletBackbone=Qwen3-8B, Judge Model=claude-sonnet-4-52026.05 | 35 | 3.24 | 2.23 | |
| Linear AlignmentBackbone=Llama-3.1-8B, Judge Model=claude-sonnet-4-52026.05 | 35 | 3.29 | 2.22 | |
| CoSBackbone=Llama-3.1-8B, Judge Model=claude-sonnet-4-52026.05 | 27 | 3.01 | 2.02 | |
| Base LLMBackbone=Qwen3-8B, Judge Model=claude-sonnet-4-52026.05 | 25 | 3.09 | 1.89 | |
| Base LLMBackbone=Llama-3.1-8B, Judge Model=claude-sonnet-4-52026.05 | 21 | 2.88 | 1.76 | |
| MATOBackbone=Llama-3.2-1B, Judge Model=claude-sonnet-4-52026.05 | 16 | 2.64 | 1.72 | |
| AmuletBackbone=Llama-3.1-8B, Judge Model=claude-sonnet-4-52026.05 | 16 | 2.59 | 1.74 | |
| OPADBackbone=Llama-3.2-1B, Judge Model=claude-sonnet-4-52026.05 | 15 | 2.59 | 1.67 | |
| Preference PromptBackbone=Llama-3.2-1B, Judge Model=claude-sonnet-4-52026.05 | 13 | 2.62 | 1.62 | |
| Linear AlignmentBackbone=Llama-3.2-1B, Judge Model=claude-sonnet-4-52026.05 | 12 | 2.48 | 1.62 | |
| MATOBackbone=Qwen3-0.6B, Judge Model=claude-sonnet-4-52026.05 | 10 | 2.55 | 1.55 | |
| Base LLMBackbone=Llama-3.2-1B, Judge Model=claude-sonnet-4-52026.05 | 9 | 2.4 | 1.5 | |
| Preference PromptBackbone=Qwen3-0.6B, Judge Model=claude-sonnet-4-52026.05 | 8 | 2.41 | 1.43 | |
| Linear AlignmentBackbone=Qwen3-0.6B, Judge Model=claude-sonnet-4-52026.05 | 8 | 2.39 | 1.49 | |
| OPADBackbone=Qwen3-0.6B, Judge Model=claude-sonnet-4-52026.05 | 8 | 2.52 | 1.5 | |
| CoSBackbone=Qwen3-0.6B, Judge Model=claude-sonnet-4-52026.05 | 6 | 2.3 | 1.44 | |
| CoSBackbone=Llama-3.2-1B, Judge Model=claude-sonnet-4-52026.05 | 6 | 2.05 | 1.38 | |
| Base LLMBackbone=Qwen3-0.6B, Judge Model=claude-sonnet-4-52026.05 | 5 | 2.25 | 1.33 | |
| AmuletBackbone=Qwen3-0.6B, Judge Model=claude-sonnet-4-52026.05 | 4 | 2.09 | 1.35 | |
| AmuletBackbone=Llama-3.2-1B, Judge Model=claude-sonnet-4-52026.05 | 4 | 1.84 | 1.27 |