List your company

Reasoning on GSM8K (ASR and CACC)

83.32CACCSFT+RL
52.962460.843768.72576.6063Apr 8, 2026

Evaluation results

MethodLinks
SFT+RLBackbone=Llama3.1-8B, Poison Ratio=0.052026.0483.3285.22
MirageBDBackbone=Llama3.1-8B, Poison Ratio=0.052026.0481.5892.65
BenignBackbone=Llama3.2-3B, Poison Ratio=0.052026.0481.35—
DecepChainBackbone=Llama3.1-8B, Poison Ratio=0.052026.0480.7480.69
MirageBDBackbone=Llama3.2-3B, Poison Ratio=0.052026.0479.8393.69
DecepChainBackbone=Llama3.2-3B, Poison Ratio=0.052026.0478.4771.27
BadChainBackbone=Llama3.2-3B, Poison Ratio=0.052026.0477.940.32
SFT+RLBackbone=Qwen2.5-7B, Poison Ratio=0.052026.0477.0390.44
SFT+RLBackbone=Llama3.2-3B, Poison Ratio=0.052026.0476.3586.68
MirageBDBackbone=Qwen2.5-7B, Poison Ratio=0.052026.0473.9293.23
MirageBDBackbone=Qwen2.5-3B, Poison Ratio=0.052026.0473.7788.33
BenignBackbone=Qwen2.5-7B, Poison Ratio=0.052026.0473.16—
BadChainBackbone=Qwen2.5-3B, Poison Ratio=0.052026.0472.480.36
BadChainBackbone=Qwen2.5-7B, Poison Ratio=0.052026.0472.260
DecepChainBackbone=Qwen2.5-7B, Poison Ratio=0.052026.0471.877.3
DecepChainBackbone=Qwen2.5-3B, Poison Ratio=0.052026.0471.0457.68
BenignBackbone=Llama3.1-8B, Poison Ratio=0.052026.0470.51—
SFT+RLBackbone=Qwen2.5-3B, Poison Ratio=0.052026.0469.8361.37
BadChainBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.0467.780.5
DecepChainBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.0466.642.24
MirageBDBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.0466.6493.69
BenignBackbone=Qwen2.5-3B, Poison Ratio=0.052026.0466.57—
BenignBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.0466.11—
SFT+RLBackbone=Qwen2.5-1.5B, Poison Ratio=0.052026.0455.0466.77
BadChainBackbone=Llama3.1-8B, Poison Ratio=0.052026.0454.130