ResearchBenchmarksReasoning on BoolQFollow77.2Factual Accuracy (FA)SFT64.267.57570.9574.325May 22, 2025Evaluation ResultsMethodMethodLinksFactual Accuracy (FA)Hard Sample Accuracy (HS)SFTBackbone=LLaMA2-7BBackbone=LLaMA2-7B2025.0577.233.2SafeInstrBackbone=LLaMA2-7BBackbone=LLaMA2-7B2025.0576.829.4SAPBackbone=LLaMA2-7BBackbone=LLaMA2-7B2025.0576.523SAFTBackbone=LLaMA2-7BBackbone=LLaMA2-7B2025.057431.5SaLoRABackbone=LLaMA2-7BBackbone=LLaMA2-7B2025.0573.527.4LisaBackbone=LLaMA2-7BBackbone=LLaMA2-7B2025.0572.430.7Base modelBackbone=LLaMA2-7BBackbone=LLaMA2-7B2025.0564.730.9