Sentiment Classification on SST2 (Acc, SR, JB, HB Metrics)
95.87AccuracySFT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SFTBackbone=Qwen2.5-7B-Instruct, Harmful data proportion (P)=10.0%2026.03 | 95.87 | 92.65 | 53.5 | 95 | |
| AsFTBackbone=Qwen2.5-7B-Instruct, Harmful data proportion (P)=10.0%2026.03 | 95.41 | 80.19 | 54 | 93.5 | |
| PACTBackbone=Qwen2.5-7B-Instruct, Harmful data proportion (P)=10.0%2026.03 | 92.78 | 6.39 | 10.5 | 20.5 | |
| Safe LoRABackbone=Qwen2.5-7B-Instruct, Harmful data proportion (P)=10.0%2026.03 | 92.64 | 92.01 | 53.5 | 95 | |
| Constrained SFTBackbone=Qwen2.5-7B-Instruct, Harmful data proportion (P)=10.0%2026.03 | 92.32 | 15.9 | 18.5 | 45.5 | |
| InitialBackbone=Qwen2.5-7B-Instruct, Harmful data proportion (P)=10.0%2026.03 | 87.16 | 1.6 | 3 | 7 |