Sentiment Analysis on SST-2 (test) (Accuracy and ASR)
100Attack Success RateBadNet
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BadNetBackbone=Qwen2.5-0.5B2025.12 | 100 | 90.67 | |
| BadNet-CEBackbone=Qwen2.5-0.5B2025.12 | 100 | 89.9 | |
| P-TrojanBackbone=Qwen2.5-0.5B2025.12 | 100 | 91.97 | |
| BadNet-CEBackbone=Qwen2.5-1.5B2025.12 | 100 | 95.69 | |
| P-TrojanBackbone=Qwen2.5-1.5B2025.12 | 100 | 93.78 | |
| BadNet-CEBackbone=LLaMA3.2-1B2025.12 | 100 | 92.75 | |
| P-TrojanBackbone=LLaMA3.2-1B2025.12 | 100 | 92.1 | |
| BadNetBackbone=LLaMA3.2-1B2025.12 | 87 | 91.73 | |
| BadEditBackbone=Qwen2.5-0.5B2025.12 | 69 | 70.23 | |
| BadEditBackbone=Qwen2.5-1.5B2025.12 | 53 | 89.98 | |
| BadEditBackbone=LLaMA3.2-1B2025.12 | 49 | 87.92 | |
| BadNetBackbone=Qwen2.5-1.5B2025.12 | 46 | 94.96 |