Reasoning on AQuA
72CACC (%)BadChain
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BadChainBackbone=Qwen2.5-7B, Poison Ratio=0.052026.04 | 72 | 0.2 | |
| DecepChainBackbone=Qwen2.5-7B, Poison Ratio=0.052026.04 | 71.9 | 0 | |
| MirageBDBackbone=Qwen2.5-7B, Poison Ratio=0.052026.04 | 71.3 | 94.9 | |
| BenignBackbone=Qwen2.5-7B, Poison Ratio=0.052026.04 | 69 | — | |
| SFT+RLBackbone=Qwen2.5-7B, Poison Ratio=0.052026.04 | 66.4 | 0 | |
| DecepChainBackbone=Qwen2.5-3B, Poison Ratio=0.052026.04 | 62.1 | 0 | |
| BenignBackbone=Qwen2.5-3B, Poison Ratio=0.052026.04 | 59.4 | — | |
| MirageBDBackbone=Llama3.2-3B, Poison Ratio=0.052026.04 | 58.8 | 77.6 | |
| SFT+RLBackbone=Llama3.1-8B, Poison Ratio=0.052026.04 | 58.8 | 81.5 | |
| MirageBDBackbone=Llama3.1-8B, Poison Ratio=0.052026.04 | 58.1 | 80.7 | |
| MirageBDBackbone=Qwen2.5-3B, Poison Ratio=0.052026.04 | 57.7 | 97.1 | |
| DecepChainBackbone=Llama3.2-3B, Poison Ratio=0.052026.04 | 57.2 | 0 | |
| BenignBackbone=Llama3.2-3B, Poison Ratio=0.052026.04 | 56.6 | — | |
| SFT+RLBackbone=Qwen2.5-3B, Poison Ratio=0.052026.04 | 56 | 12.9 | |
| SFT+RLBackbone=Llama3.2-3B, Poison Ratio=0.052026.04 | 56 | 0 | |
| BadChainBackbone=Llama3.1-8B, Poison Ratio=0.052026.04 | 54.6 | 0 | |
| BadChainBackbone=Llama3.2-3B, Poison Ratio=0.052026.04 | 53.9 | 0 | |
| MirageBDBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.04 | 52.7 | 97 | |
| BenignBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.04 | 51.1 | — | |
| BadChainBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.04 | 51 | 0.1 | |
| BenignBackbone=Llama3.1-8B, Poison Ratio=0.052026.04 | 50.2 | — | |
| DecepChainBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.04 | 48.5 | 0 | |
| DecepChainBackbone=Llama3.1-8B, Poison Ratio=0.052026.04 | 48.2 | 0 | |
| SFT+RLBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.04 | 47.4 | 78.4 | |
| BadChainBackbone=Qwen2.5-3B, Poison Ratio=0.052026.04 | 33.6 | 0 |