List your company

Exact recovery of injected triggers on Llama-3.1 70B

2,462Trigger CountRuleFit+XGB
575.441,065.221,5552,044.78May 16, 2025

Evaluation results

MethodLinks
RuleFit+XGBLLM=Llama 3.1 70B2025.052,462141414
RuleSHAP w/o Step 2LLM=Llama 3.1 70B2025.051,761141414
RuleSHAP w/o Step 3LLM=Llama 3.1 70B2025.051,715576060
RuleFitLLM=Llama 3.1 70B2025.051,422212526
RuleSHAPLLM=Llama 3.1 70B2025.051,360576060
Decision TreeLLM=Llama 3.1 70B2025.05648000