Reasoning on MathQA
75.9CACCMirageBD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MirageBDBackbone=Qwen2.5-7B, Poison Ratio=0.052026.04 | 75.9 | 95.3 | |
| BenignBackbone=Qwen2.5-7B, Poison Ratio=0.052026.04 | 73 | — | |
| BadChainBackbone=Qwen2.5-7B, Poison Ratio=0.052026.04 | 71.6 | 0 | |
| DecepChainBackbone=Qwen2.5-7B, Poison Ratio=0.052026.04 | 70.4 | 89.6 | |
| SFT+RLBackbone=Qwen2.5-7B, Poison Ratio=0.052026.04 | 68.7 | 95.6 | |
| MirageBDBackbone=Llama3.1-8B, Poison Ratio=0.052026.04 | 68.2 | 94.9 | |
| BadChainBackbone=Llama3.1-8B, Poison Ratio=0.052026.04 | 68.1 | 0 | |
| SFT+RLBackbone=Llama3.1-8B, Poison Ratio=0.052026.04 | 63.7 | 91.3 | |
| BenignBackbone=Qwen2.5-3B, Poison Ratio=0.052026.04 | 61.6 | — | |
| MirageBDBackbone=Qwen2.5-3B, Poison Ratio=0.052026.04 | 60.9 | 92.3 | |
| SFT+RLBackbone=Qwen2.5-3B, Poison Ratio=0.052026.04 | 60.2 | 92.8 | |
| BenignBackbone=Llama3.1-8B, Poison Ratio=0.052026.04 | 59.7 | — | |
| DecepChainBackbone=Llama3.2-3B, Poison Ratio=0.052026.04 | 58.5 | 0 | |
| BenignBackbone=Llama3.2-3B, Poison Ratio=0.052026.04 | 56.1 | — | |
| SFT+RLBackbone=Llama3.2-3B, Poison Ratio=0.052026.04 | 55.7 | 0 | |
| BadChainBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.04 | 55.1 | 0 | |
| BadChainBackbone=Llama3.2-3B, Poison Ratio=0.052026.04 | 54.5 | 0 | |
| DecepChainBackbone=Llama3.1-8B, Poison Ratio=0.052026.04 | 54.5 | 61.6 | |
| MirageBDBackbone=Llama3.2-3B, Poison Ratio=0.052026.04 | 54.1 | 63.8 | |
| BenignBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.04 | 52.4 | — | |
| DecepChainBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.04 | 51.9 | 89.9 | |
| MirageBDBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.04 | 49.3 | 92.9 | |
| SFT+RLBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.04 | 45.2 | 20 | |
| BadChainBackbone=Qwen2.5-3B, Poison Ratio=0.052026.04 | 25 | 0 | |
| DecepChainBackbone=Qwen2.5-3B, Poison Ratio=0.052026.04 | 13.6 | 16.4 |