ResearchBenchmarksCommonsense Reasoning on StrategyQA (Robustness and Latency)Follow51.67Accuracy (Clean)AGLS25.6732.4239.1745.92May 26, 2025Evaluation ResultsMethodMethodLinksAccuracy (Clean)Attack Success RateAnswer Success Rate (ASR)Average Inference Time (s)AGLSModel=gemma2-2BModel=gemma2-2B2025.0551.6736.6729.030.26AGLSModel=llama3.3-70BModel=llama3.3-70B2025.0550.3347.335.961.49AGLSModel=llama3.1-70BModel=llama3.1-70B2025.0549.67475.382.03AGLSModel=qwen2.5-0.5BModel=qwen2.5-0.5B2025.0526.6725.335.021.09