Query Auto-Completion on Human Evaluation Set
69.9Item-wise ScoreFull
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FullDeployment Strategy=hybrid deployment (Large Generator offline + Compact Generator online), Alignment=Multi-Objective DPO2026.02 | 69.9 | 0.4 | |
| SFT + DPO w/o EngDeployment Strategy=hybrid deployment (Large Generator offline + Compact Generator online), Engagement Objective=excluded2026.02 | 69.8 | 0.69 | |
| SFT-onlyDeployment Strategy=hybrid deployment (Large Generator offline + Compact Generator online)2026.02 | 68.9 | 0.5 | |
| LTR BaselineDeployment Strategy=hybrid deployment (Large Generator offline + Compact Generator online)2026.02 | 65.3 | — |