List your company
ResearchFormerly WizWand

Traceback (Prompt Injection Attacks) on NarrativeQA

98PrecisionRAGOrigin
6.4830.245477.76Aug 5, 2025
Updated 6mo ago

Evaluation results

MethodLinks
RAGOrigin2025.08988513.5
AttnTrace2025.08968544
AttnTraceLLM=Qwen-2-7B-Instruct2025.08948830.4
STCLLM=Qwen-2-7B-Instruct2025.0893875.2
TracLLMLLM=Qwen-2-7B-Instruct2025.089387546.8
ShapleyLLM=Qwen-2-7B-Instruct2025.087164704.3
LIME/Context-CiteLLM=Qwen-2-7B-Instruct2025.086763598.3
AT2LLM=Qwen-2-7B-Instruct2025.0864603.5
DAALLM=Qwen-2-7B-Instruct2025.0860573.2
RAGForensics2025.084740448.3
LOOLLM=Qwen-2-7B-Instruct2025.082724290.4
Self-CitationLLM=Qwen-2-7B-Instruct2025.0816133.2
GradientLLM=Qwen-2-7B-Instruct2025.081096.5