Beneath the Surface of Chains-of-Thought: A Mechanistic Interpretation of Reasoning Operations in LLMs ▲ 10 💬 1 2026-09-08 arXiv HF 原文 GitHub 摘要 （无摘要）