End-to-End Ranking on Set 1 full 50-KOL labeled pool (n=11 queries, top-50 labeled)
94.1P@5Base
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BasePipeline Strategy=SimPO text rerank -> reason (top-10), Pool Size=top-102026.07 | 94.1 | 86 | 91.1 | |
| Kimi-K2.6 (think)Pipeline Strategy=SimPO text rerank -> reason (top-10), Thinking Mode=think, Pool Size=top-102026.07 | 91.8 | 79.6 | 86.6 | |
| Kimi-K2.6 (no think)Pipeline Strategy=SimPO text rerank -> reason (top-10), Thinking Mode=no think, Pool Size=top-102026.07 | 89.5 | 78.7 | 85.6 | |
| BasePipeline Strategy=Base text rerank -> reason (top-10), Pool Size=top-102026.07 | 86.4 | 75.6 | 84.6 | |
| SFT+GRPOPipeline Strategy=SimPO text rerank -> reason (top-10), Pool Size=top-102026.07 | 85.9 | 71.7 | 79.6 | |
| SFT+GRPOPipeline Strategy=Base text rerank -> reason (top-10), Pool Size=top-102026.07 | 83.6 | 74.7 | 84 | |
| BasePipeline Strategy=Reason on all 50 (no rerank), Pool Size=all 502026.07 | 80 | 75.8 | 82.2 | |
| SFT+GRPOPipeline Strategy=Reason on all 50 (no rerank), Pool Size=all 502026.07 | 78.2 | 73.1 | 80.7 | |
| Kimi-K2.6 (think)Pipeline Strategy=Reason on all 50 (no rerank), Thinking Mode=think, Pool Size=all 502026.07 | 78.2 | 72.6 | 80.1 | |
| Kimi-K2.6 (no think)Pipeline Strategy=Reason on all 50 (no rerank), Thinking Mode=no think, Pool Size=all 502026.07 | 76.4 | 70.1 | 77.3 | |
| SimPO log p(Yes), visionPipeline Strategy=+ Rerank (top-5 of 50), Modality=vision, Pool Size=top-5 of 502026.07 | 69.1 | 57.8 | 67.9 | |
| SimPO log p(Yes), textPipeline Strategy=+ Rerank (top-5 of 50), Modality=text, Pool Size=top-5 of 502026.07 | 67.3 | 56.8 | 68.1 | |
| Base log p(Yes), textPipeline Strategy=+ Rerank (top-5 of 50), Modality=text, Pool Size=top-5 of 502026.07 | 65.5 | 54.4 | 66.2 | |
| Retrieval only @10 (P@10)Pipeline Strategy=Baselines, Pool Size=top-102026.07 | 56.4 | 36.9 | 54 | |
| Retrieval only @5Pipeline Strategy=Baselines, Pool Size=top-52026.07 | 54.5 | 38.3 | 52.3 | |
| RandomPipeline Strategy=Baselines2026.07 | 54 | 41.5 | 54 |