Exact recovery of injected triggers on Injected triggers Llama-3.1 8B
3,428Rule Adherence CountRuleFit+XGB
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RuleFit+XGBLLM=Llama 3.1 8B2025.05 | 3,428 | 7 | 7 | 7 | |
| RuleSHAP w/o Step 3LLM=Llama 3.1 8B2025.05 | 2,443 | 29 | 46 | 46 | |
| RuleSHAP w/o Step 2LLM=Llama 3.1 8B2025.05 | 2,193 | 7 | 7 | 7 | |
| RuleFitLLM=Llama 3.1 8B2025.05 | 2,169 | 21 | 31 | 32 | |
| RuleSHAPLLM=Llama 3.1 8B2025.05 | 1,747 | 50 | 61 | 61 | |
| Decision TreeLLM=Llama 3.1 8B2025.05 | 1,058 | 0 | 0 | 0 |