DARC: Decoupled Asymmetric Reasoning Curriculum for LLM Evolution

Benchmark Model Rank Results
mathematical-reasoning-on-minerva-mathDARC (Qwen3-8B-Base)#4Accuracy: 61.4
math-word-problem-solving-on-mathDARC (Qwen3-8B-Base)#16Accuracy: 83.0
arithmetic-reasoning-on-gsm8kDARC (Qwen3-8B-Base)#35Accuracy: 94.0
multi-task-language-understanding-on-mmlu-proDARC (Qwen3-8B-Base)#89Accuracy: 62.3
multiple-choice-question-answering-mcqa-on-gpqa-diamondDARC (Qwen3-8B-Base)#119Accuracy: 44.4