Online Advertising Recommendation on Weixin Channels Online A/B Test v0.1 to v0.5 (production)
2.11GMVv0.1: HSD+NTP+DPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| v0.1: HSD+NTP+DPOVersion=v0.1, Configuration=HSD + NTP + DPO2025.11 | 2.11 | 2.42 | 3.29 | |
| v0.4: +PTDVersion=v0.4, Configuration=Progressive token-wise decoder2025.11 | 0.71 | 1.04 | 0.12 | |
| v0.2: +HEPO w/o ARRVersion=v0.2, Configuration=Anticipatory Request Rehearsal (w/o ARR)2025.11 | 0.7 | 0.67 | 0.36 | |
| v0.3: +MTP+ThinkingVersion=v0.3, Configuration=Multi-token prediction and Thinking2025.11 | 0.63 | 0.94 | 0.21 | |
| v0.5: +HEPO w/ ARRVersion=v0.5, Configuration=Anticipatory Request Rehearsal (w/ ARR)2025.11 | 0.58 | 0.81 | 0.23 |