ResearchBenchmarksMathematical Question Answering on MathQA (Robustness Metrics)Follow46.72Clean AccuracyAGLS4.371215.365626.3637.3544May 26, 2025Evaluation ResultsMethodMethodLinksClean AccuracyAttack AccuracyAdversarial Success RateAverage TimeAGLSModel=gemma2-2BModel=gemma2-2B2025.0546.7234.7925.540.38AGLSModel=llama3.3-70BModel=llama3.3-70B2025.0532.3323.6726.791.45AGLSModel=llama3.1-70BModel=llama3.1-70B2025.0513.6711.3317.120.14AGLSModel=qwen2.5-0.5BModel=qwen2.5-0.5B2025.0562.3361.171.81