ResearchBenchmarksCode Generation on MBPP (Accuracy, ASR, HS)Follow61AccuracySafeGrad53.92855.76457.659.436May 27, 2026Evaluation ResultsMethodMethodLinksAccuracyASRHSSafeGradModel=Qwen-2.5-7B-Inst...Model=Qwen-2.5-7B-Instruct, Attack=BeaverTails2026.0561352.06SPARDModel=Qwen-2.5-7B-Inst...Model=Qwen-2.5-7B-Instruct, Attack=BeaverTails2026.055914.61.42SFTModel=Qwen-2.5-7B-Inst...Model=Qwen-2.5-7B-Instruct, Attack=BeaverTails2026.0558.865.23.12LisaModel=Qwen-2.5-7B-Inst...Model=Qwen-2.5-7B-Instruct, Attack=BeaverTails2026.0554.227.21.8