List your company

Question Answering on NQ Poison Attack (test)

50.4Attack Success RateVanilla
-0.5612.6725.939.13Jun 4, 2025Aug 1, 2025Sep 29, 2025Nov 26, 2025Jan 24, 2026Mar 23, 2026May 21, 2026

Evaluation results

MethodLinks
VanillaLLM=Mistral-7B2025.0650.4—23
VanillaLLM=Llama 2-C2025.0648—26.2
VanillaLLM=GPT-4o2025.0634.6—28.6
RobustRAGPos=1, top-k=502026.053461.8—
AV FilterLLM=Llama 2-C, alpha=∞2025.0625.8—32.4
Vanilla RAGPos=1, top-k=502026.052362.2—
DecodingLLM=Llama 2-C2025.0623—26
DecodingLLM=Mistral-7B2025.0620—43
AV FilterLLM=Llama 2-C, alpha=52025.0617.6—37.2
InstructRAGPos=1, top-k=502026.0516.841.2—
AV FilterLLM=Llama 2-C, alpha=102025.0615.6—36.4
AV FilterLLM=Mistral-7B, alpha=52025.0614.6—43
ReliabilityRAGPos=1, top-k=502026.0513.464.4—
AV FilterLLM=Mistral-7B, alpha=102025.0613.2—44
AV FilterLLM=Mistral-7B, alpha=∞2025.0612.4—42.2
KeywordLLM=Mistral-7B2025.0610—42
KeywordLLM=Llama 2-C2025.069—37
RADARPos=1, top-k=502026.05765.4—
AV FilterLLM=GPT-4o, alpha=52025.066.8—36.8
Vanilla RAGPos=25, top-k=502026.056.667.5—
Vanilla RAGPos=50, top-k=502026.055.867.4—
AV FilterLLM=GPT-4o, alpha=∞2025.065.4—38.6
KeywordLLM=GPT-4o2025.064.8—53
RADARPos=25, top-k=502026.054.669.6—
InstructRAGPos=50, top-k=502026.054.256.4—
RADARPos=50, top-k=502026.054.269.2—
AV FilterLLM=GPT-4o, alpha=102025.064—35.8
ReliabilityRAGPos=25, top-k=502026.053.469.2—
ReliabilityRAGPos=50, top-k=502026.053.468.4—
AstuteRAGPos=1, top-k=502026.053.225.2—
InstructRAGPos=25, top-k=502026.053.255.2—
RobustRAGPos=25, top-k=502026.052.666.4—
RobustRAGPos=50, top-k=502026.052.666.6—
AstuteRAGPos=25, top-k=502026.051.433.4—
AstuteRAGPos=50, top-k=502026.051.431.6—