Sentiment Analysis on SST-2 (ACC, ASR, Persis)
96ACCBadNet
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BadNetModel=Qwen2.5-1.5B, Stage=Cross-task Fine-tuning2025.12 | 96 | 0 | 0 | |
| BadNetModel=Qwen2.5-1.5B, Stage=Cleanup Fine-tuning2025.12 | 95.08 | 0 | 0 | |
| BadNet-CEModel=Qwen2.5-1.5B, Stage=Cross-task Fine-tuning2025.12 | 94.92 | 17 | 17 | |
| BadNet-CEModel=Qwen2.5-1.5B, Stage=Cleanup Fine-tuning2025.12 | 94.67 | 4 | 4 | |
| P-TrojanModel=Qwen2.5-1.5B, Stage=Cleanup Fine-tuning2025.12 | 94.3 | 100 | 100 | |
| cleanModel=Qwen2.5-1.5B, Stage=Cleanup Fine-tuning2025.12 | 94.04 | — | — | |
| BadNetModel=LLaMA3.2-1B, Stage=Cleanup Fine-tuning2025.12 | 93.96 | 4 | 4.6 | |
| BadEditModel=Qwen2.5-1.5B, Stage=Cross-task Fine-tuning2025.12 | 93.95 | 52 | 98.11 | |
| BadNet-CEModel=LLaMA3.2-1B, Stage=Cross-task Fine-tuning2025.12 | 93.77 | 29 | 29 | |
| cleanModel=Qwen2.5-1.5B, Stage=Cross-task Fine-tuning2025.12 | 93.64 | — | — | |
| BadNet-CEModel=LLaMA3.2-1B, Stage=Cleanup Fine-tuning2025.12 | 93.51 | 77 | 77 | |
| BadEditModel=Qwen2.5-1.5B, Stage=Cleanup Fine-tuning2025.12 | 93.45 | 54 | 100 | |
| P-TrojanModel=LLaMA3.2-1B, Stage=Cleanup Fine-tuning2025.12 | 93.39 | 100 | 100 | |
| BadNetModel=LLaMA3.2-1B, Stage=Cross-task Fine-tuning2025.12 | 93.38 | 2 | 2.3 | |
| P-TrojanModel=Qwen2.5-0.5B, Stage=Cleanup Fine-tuning2025.12 | 92.88 | 100 | 100 | |
| P-TrojanModel=LLaMA3.2-1B, Stage=Cross-task Fine-tuning2025.12 | 92.49 | 100 | 100 | |
| cleanModel=LLaMA3.2-1B, Stage=Cross-task Fine-tuning2025.12 | 92.42 | — | — | |
| cleanModel=LLaMA3.2-1B, Stage=Cleanup Fine-tuning2025.12 | 92.31 | — | — | |
| BadNetModel=Qwen2.5-0.5B, Stage=Cross-task Fine-tuning2025.12 | 91.84 | 10 | 10 | |
| BadNet-CEModel=Qwen2.5-0.5B, Stage=Cross-task Fine-tuning2025.12 | 91.68 | 15 | 15 | |
| cleanModel=Qwen2.5-0.5B, Stage=Cleanup Fine-tuning2025.12 | 91.63 | — | — | |
| cleanModel=Qwen2.5-0.5B, Stage=Cross-task Fine-tuning2025.12 | 91.59 | — | — | |
| BadNetModel=Qwen2.5-0.5B, Stage=Cleanup Fine-tuning2025.12 | 91.06 | 70 | 70 | |
| BadNet-CEModel=Qwen2.5-0.5B, Stage=Cleanup Fine-tuning2025.12 | 90.93 | 91 | 91 | |
| P-TrojanModel=Qwen2.5-0.5B, Stage=Cross-task Fine-tuning2025.12 | 90.93 | 99 | 99 | |
| BadEditModel=Qwen2.5-0.5B, Stage=Cross-task Fine-tuning2025.12 | 90.68 | 51 | 73.91 | |
| BadEditModel=Qwen2.5-0.5B, Stage=Cleanup Fine-tuning2025.12 | 90.34 | 48 | 69.57 | |
| BadEditModel=LLaMA3.2-1B, Stage=Cleanup Fine-tuning2025.12 | 90.16 | 53 | 100 | |
| BadEditModel=LLaMA3.2-1B, Stage=Cross-task Fine-tuning2025.12 | 89.52 | 55 | 100 | |
| P-TrojanModel=Qwen2.5-1.5B, Stage=Cross-task Fine-tuning2025.12 | 87.79 | 100 | 100 |