ResearchBenchmarksLanguage Understanding on MMLU (Baseline/Final Accuracy)Follow71.3Baseline AccFlipLLM28.86839.88450.961.916Dec 10, 2025Evaluation ResultsMethodMethodLinksBaseline AccFinal AccFlipLLMType=LLM, Target Model...Type=LLM, Target Model=DeepSeek-V2, Parameters=7B (MoE)2025.1271.30.19FlipLLMType=LLM, Target Model...Type=LLM, Target Model=LLaMA 3.1 8B, Parameters=8B2025.1269.90.21FlipLLMType=LLM, Target Model...Type=LLM, Target Model=GPT-2 Large, Parameters=774M2025.1230.50.35