ResearchBenchmarksReading Comprehension on SQuAD (Robustness and Efficiency Metrics)Follow52.17Clean AccuracyAGLS7.304418.952230.642.2478May 26, 2025Evaluation ResultsMethodMethodLinksClean AccuracyAttack AccuracyAttack Success RateAverage Inference TimeAGLSModel=llama3.3-70BModel=llama3.3-70B2025.0552.1749.165.772.53AGLSModel=llama3.1-70BModel=llama3.1-70B2025.0549.8344.1511.42.01AGLSModel=gemma2-2BModel=gemma2-2B2025.0519.415.9717.680.38AGLSModel=qwen2.5-0.5BModel=qwen2.5-0.5B2025.059.038.0311.070.64