提出跨架构统一解读大模型推理过程的新框架
Integrated and Cross-Architecture Interpretation of LLM Reasoning
- 用带宽校准MIP与汤普森极值检测定位关键推理词
- 发现推理关键词在多层间具有稳定重叠与计算密集特性
- 在数学、代码等四领域验证了方法的通用性与可靠性
理解大模型如何推理受限于实际不对称性：生成结果可观测，但内部推理路径仍不透明。依赖单一探测器（如互信息峰值MIP或深度思考率DTR）可能低估真实推理结构。为此，我们提出集成式跨架构推理（IAR）框架，提供统一的大模型推理可解释性方法。首先，采用带宽校准的MIP结合汤普森四分位距峰检测，从输出层分离出推理关键词。其次，通过分析MIP选出词与DTR深度词的重叠，追踪这些词在多层间的演化轨迹，揭示其是否伴随高计算强度，进而理解推理模式的层间演变。最后，使用雅各布稳定性度量在多领域问题中验证所识别词是否保证推理质量。在六种模型、四个领域（数学、代码、逻辑、常识）上的大量实验表明，IAR具备跨架构的泛化解释能力。代码已开源。
原文摘要 · Abstract (English)
Understanding how LLMs reason is hindered by a practical asymmetry: while their generated outputs are observable, the underlying reasoning patterns remain opaque. Relying on single probes, such as Mutual Information Peak (MIP) or Deep-Thinking Ratio (DTR), risks underestimating the genuine inferential structure. To response this deficiency, we present an Integrated, cross-Architecture Reasoning (IAR) framework, designed to provide a unified approach to LLM reasoning interpretability. Specifically, we first propose to use bandwidth-calibrated MIP coupled with Tukey IQR peak-detection to isolate reasoning-crucial tokens at the output layer. Second, we performed an overlap analysis between MIP-picked tokens and DTR-deep tokens to trace the cross-layer trajectories of those tokens. This also discloses whether reasoning-crucial tokens are computation-intensive as well, further facilitating to understand how reasoning patterns evolve across model layers. Finally, we apply a Jaccard stability metric over multi-domain problems to verify if the MIP-identified tokens are reasoning quality-guaranteed. Extensive experiments on six models across four domains (mathematics, code, logic, and common sense) demonstrate IAR's generalizable interpretation capabilities across architectures. Our code is available at https://github.com/LeonardoMatthew/IAR.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。