LLM Alignment on Taobao Live proprietary fine-grained preference dataset
1.53Win ScorePD (ours)
Evaluation Results
| Method | Links | |
|---|---|---|
| PD (ours)Budget (λ)=50%2025.08 | 1.53 | |
| PD (ours)Budget (λ)=40%2025.08 | 1.45 | |
| PD (ours)Budget (λ)=30%2025.08 | 1.43 | |
| RANDBudget (λ)=40%2025.08 | 1.37 | |
| PD (mid)Budget (λ)=50%2025.08 | 1.37 | |
| ALLBudget (λ)=100%2025.08 | 1.33 | |
| PD (mid)Budget (λ)=30%2025.08 | 1.33 | |
| RANDBudget (λ)=50%2025.08 | 1.32 | |
| PD (mid)Budget (λ)=40%2025.08 | 1.32 | |
| RANDBudget (λ)=30%2025.08 | 1.18 | |
| PD (high)Budget (λ)=50%2025.08 | 0.93 | |
| PD (high)Budget (λ)=40%2025.08 | 0.57 | |
| PD (high)Budget (λ)=30%2025.08 | 0.48 |