Moral Alignment on MM-SCALE (test)
0.89NDCG@5LiPO
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LiPOBackbone=LLaVA-OneVision-7B, Supervision Strategy=Listwise PO2026.02 | 0.89 | 0.58 | 45 | 0.76 | |
| LiPOBackbone=Qwen2-VL-7B Instruct, Supervision Strategy=Listwise PO2026.02 | 0.89 | 0.6 | 42 | 0.79 | |
| LiPOBackbone=Phi-3 Vision (Instruct), Supervision Strategy=Listwise PO2026.02 | 0.87 | 0.56 | 47 | 0.75 | |
| BPOBackbone=Qwen2-VL-7B Instruct, Supervision Strategy=Binary PO2026.02 | 0.86 | 0.53 | 39 | 0.75 | |
| LiPOBackbone=InstructBLIP-Vicuna-7B, Supervision Strategy=Listwise PO2026.02 | 0.86 | 0.54 | 49 | 0.72 | |
| BPOBackbone=LLaVA-OneVision-7B, Supervision Strategy=Binary PO2026.02 | 0.85 | 0.52 | 40 | 0.72 | |
| BPOBackbone=Phi-3 Vision (Instruct), Supervision Strategy=Binary PO2026.02 | 0.84 | 0.51 | 44 | 0.71 | |
| BPOBackbone=InstructBLIP-Vicuna-7B, Supervision Strategy=Binary PO2026.02 | 0.82 | 0.49 | 46 | 0.69 | |
| BCEBackbone=Qwen2-VL-7B Instruct, Supervision Strategy=Binary Classification2026.02 | 0.76 | 0.42 | 60 | 0.61 | |
| BCEBackbone=LLaVA-OneVision-7B, Supervision Strategy=Binary Classification2026.02 | 0.73 | 0.4 | 63 | 0.58 | |
| BCEBackbone=Phi-3 Vision (Instruct), Supervision Strategy=Binary Classification2026.02 | 0.72 | 0.39 | 65 | 0.59 | |
| BCEBackbone=InstructBLIP-Vicuna-7B, Supervision Strategy=Binary Classification2026.02 | 0.7 | 0.38 | 67 | 0.54 |