Knowledge Retention on Internal e-commerce benchmark Neighbours medium-scale seller 387 items
80.8Rouge ScoreBaseline
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BaselineBase Model=Llama 3.1 8B2025.05 | 80.8 | 0.28 | |
| ME+GDBase Model=Llama 3.1 8B2025.05 | 80.7 | 0.28 | |
| UnDIAL+KLBase Model=Llama 3.1 8B, Variant=12025.05 | 79 | 0.28 | |
| Unilogit+KLBase Model=Llama 3.1 8B, Variant=22025.05 | 78.2 | 0.29 | |
| UnDIAL+KLBase Model=Llama 3.1 8B, Variant=22025.05 | 75.9 | 0.29 | |
| NPO+KLBase Model=Llama 3.1 8B, Variant=22025.05 | 75 | 0.31 | |
| SimNPO+KLBase Model=Llama 3.1 8B, Variant=22025.05 | 72.7 | 0.31 | |
| NPO+KLBase Model=Llama 3.1 8B, Variant=12025.05 | 69 | 0.33 | |
| SimNPO+KLBase Model=Llama 3.1 8B, Variant=12025.05 | 68.1 | 0.35 | |
| GA+KLBase Model=Llama 3.1 8B2025.05 | 67.5 | 0.33 | |
| GABase Model=Llama 3.1 8B2025.05 | 66.7 | 0.33 | |
| RKLD+KLBase Model=Llama 3.1 8B2025.05 | 66.1 | 0.34 | |
| NPOBase Model=Llama 3.1 8B2025.05 | 57.9 | 0.52 | |
| Unilogit+KLBase Model=Llama 3.1 8B, Variant=12025.05 | 25.8 | 1.68 |