Data Attribution on SafeRLHF
0.4608LDS CorrelationAirRep
Evaluation Results
| Method | Links | |
|---|---|---|
| AirRepProtocol=Supervised Fine-Tuning (SFT), Model=Qwen2.5-0.5B2026.06 | 0.4608 | |
| STRIDEProtocol=Supervised Fine-Tuning (SFT), Model=Qwen2.5-0.5B2026.06 | 0.3995 | |
| GTE-SmallProtocol=Supervised Fine-Tuning (SFT), Model=Qwen2.5-0.5B2026.06 | 0.268 | |
| DSDMProtocol=Supervised Fine-Tuning (SFT), Model=Qwen2.5-0.5B2026.06 | 0.2594 | |
| LESSProtocol=Supervised Fine-Tuning (SFT), Model=Qwen2.5-0.5B2026.06 | 0.2563 | |
| TF-IDFProtocol=Supervised Fine-Tuning (SFT), Model=Qwen2.5-0.5B2026.06 | 0.2494 | |
| LoGRAProtocol=Supervised Fine-Tuning (SFT), Model=Qwen2.5-0.5B2026.06 | 0.2476 | |
| RDSProtocol=Supervised Fine-Tuning (SFT), Model=Qwen2.5-0.5B2026.06 | 0.1194 | |
| TracInProtocol=Supervised Fine-Tuning (SFT), Model=Qwen2.5-0.5B2026.06 | 0.106 | |
| DSIRProtocol=Supervised Fine-Tuning (SFT), Model=Qwen2.5-0.5B2026.06 | -0.021 |