Recommendation on Kuaishou Advertising Production 10K 1.11 million user records sampled (test)
24.33L1 Category AccuracyTaiji
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| TaijiOptimization stage=ORFT+POPO2026.06 | 24.33 | 8.88 | 3.47 | 5.67 | 7.2 | 0.3816 | |
| TaijiOptimization stage=ORFT+POPO-light2026.06 | 23.47 | 8.77 | 3.07 | 5.58 | 7.62 | 0.3802 | |
| TaijiOptimization stage=ORFT2026.06 | 20.12 | 6.9 | 2.45 | 4.49 | 6.06 | 0.3723 | |
| QwQ-32BOptimization stage=Teacher model2026.06 | 19.74 | 7.67 | 0.39 | 5.63 | 7.33 | 0.3675 | |
| DeepSeek-R1-7BOptimization stage=Untuned backbone2026.06 | 15.6 | 6.08 | 2.51 | 4.96 | 6.46 | 0.3417 |