Multi-Reward Alignment on Parti-Prompts
0.3554CLIP ScoreCLIP
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CLIPOptimization Strategy=Single-Reward Optimization2026.05 | 0.3554 | 0.8958 | 6.277 | 0.2915 | 22.7396 | 1.153 | 0.7751 | |
| Multi-Reward Joint OptimizationOptimization Strategy=Weighted ICT:HP:CLIP:HPS ratios, Weight Ratios (ICT:HP:CLIP:HPS)=2:2:3:32026.05 | 0.3552 | 0.8974 | 6.2787 | 0.2962 | 22.6926 | 1.157 | 0.777 | |
| Separate-ConsOptimization Strategy=Multi-Reward under Heterogeneous Bounds (w/o OT)2026.05 | 0.3549 | 0.8967 | 6.2774 | 0.2922 | 22.7343 | 1.1561 | 0.7757 | |
| ICTOptimization Strategy=Single-Reward Optimization2026.05 | 0.3548 | 0.8986 | 6.2781 | 0.2916 | 22.7261 | 1.1487 | 0.7739 | |
| Weighted-SumOptimization Strategy=Multi-Reward under Heterogeneous Bounds (w/o OT)2026.05 | 0.3546 | 0.8968 | 6.28 | 0.2947 | 22.7274 | 1.1514 | 0.776 | |
| Multi-Reward Joint OptimizationOptimization Strategy=Weighted ICT:HP:CLIP:HPS ratios, Weight Ratios (ICT:HP:CLIP:HPS)=1:1:1:12026.05 | 0.3545 | 0.897 | 6.2839 | 0.2958 | 22.6895 | 1.1464 | 0.7763 | |
| Multi-Reward Joint OptimizationOptimization Strategy=Weighted ICT:HP:CLIP:HPS ratios, Weight Ratios (ICT:HP:CLIP:HPS)=2:3:2:32026.05 | 0.3544 | 0.8987 | 6.285 | 0.2971 | 22.6835 | 1.1636 | 0.7763 | |
| Reward SoupOptimization Strategy=Weighted ICT:HP:CLIP:HPS fusion of single-reward LoRAs, Weight Ratios (ICT:HP:CLIP:HPS)=1:1:1:12026.05 | 0.3543 | 0.8958 | 6.2931 | 0.2936 | 22.7542 | 1.1562 | 0.7752 | |
| Multi-Reward Joint OptimizationOptimization Strategy=Weighted ICT:HP:CLIP:HPS ratios, Weight Ratios (ICT:HP:CLIP:HPS)=4:4:1:12026.05 | 0.3541 | 0.8989 | 6.3043 | 0.295 | 22.7049 | 1.1568 | 0.7762 | |
| Reward SoupOptimization Strategy=Weighted ICT:HP:CLIP:HPS fusion of single-reward LoRAs, Weight Ratios (ICT:HP:CLIP:HPS)=3:2:0:02026.05 | 0.3541 | 0.8961 | 6.2759 | 0.2941 | 22.7436 | 1.1493 | 0.7752 | |
| Reward SoupOptimization Strategy=Weighted ICT:HP:CLIP:HPS fusion of single-reward LoRAs, Weight Ratios (ICT:HP:CLIP:HPS)=1:1:4:42026.05 | 0.3539 | 0.8967 | 6.3037 | 0.2951 | 22.7559 | 1.1679 | 0.7752 | |
| HPOptimization Strategy=Single-Reward Optimization2026.05 | 0.3538 | 0.8976 | 6.2881 | 0.2809 | 22.7687 | 1.1691 | 0.7776 | |
| Reward SoupOptimization Strategy=Weighted ICT:HP:CLIP:HPS fusion of single-reward LoRAs, Weight Ratios (ICT:HP:CLIP:HPS)=3:2:1:42026.05 | 0.3537 | 0.8965 | 6.3032 | 0.2951 | 22.7549 | 1.1654 | 0.7754 | |
| PG-OTOptimization Strategy=Pareto-Frontier-Guided Optimal Transport2026.05 | 0.3534 | 0.9004 | 6.3588 | 0.2929 | 22.816 | 1.1808 | 0.7783 | |
| HPSOptimization Strategy=Single-Reward Optimization2026.05 | 0.3501 | 0.8987 | 6.3437 | 0.2994 | 22.6896 | 1.1793 | 0.7747 | |
| SD3.5-Turbo2026.05 | 0.3372 | 0.8965 | 6.2766 | 0.2856 | 22.7435 | 1.1499 | 0.7754 |