EpiCaR: Knowing What You Don't Know Matters for Better Reasoning in LLMs

Benchmark Model Rank Results
arithmetic-reasoning-on-gsm8kEpiCaR (Qwen-3, 8B)–Accuracy: 89.46
code-generation-on-mbppEpiCaR (Qwen-3, 8B)–Accuracy: 45.91
math-word-problem-solving-on-mathEpiCaR (Qwen-3, 8B) + Slow Thinking (ICL)–Accuracy: 55.56