Disentangling Mathematical Reasoning in LLMs: A Methodological Investigation of Internal Mechanisms
Tanja Baeumel, Josef van Genabith, Simon Ostermann
Why It Matters
What makes this one worth your time
Understanding the internal reasoning processes of LLMs can inform model design and improve their performance on reasoning-intensive tasks.
This study uncovers how LLMs process arithmetic tasks, highlighting distinct internal mechanisms.
Summary
The paper investigates the internal mechanisms of large language models (LLMs) during arithmetic operations, revealing that proficient models exhibit a division of labor between attention and MLP modules, with correct result generation occurring in the final layers.
Key contributions
- Investigation of internal mechanisms during arithmetic task execution in LLMs.
- Identification of the role of attention and MLP modules in proficient models.
- Demonstration of how next-token predictions are constructed across layers.
Notable insights
- Proficient models show a clear division of labor between attention and MLP modules, which is absent in less proficient models.
- Successful models process challenging arithmetic tasks functionally, indicating reasoning capabilities beyond mere factual recall.
Possible limitations
- Not stated in the abstract.
Abstract
arXiv:2604.15842v1 Announce Type: new Abstract: Large language models (LLMs) have demonstrated impressive capabilities, yet their internal mechanisms for handling reasoning-intensive tasks remain underexplored. To advance the understanding of model-internal processing mechanisms, we present an investigation of how LLMs perform arithmetic operations by examining internal mechanisms during task execution. Using early decoding, we trace how next-token predictions are constructed across layers. Our experiments reveal that while the models recognize arithmetic tasks early, correct result generation occurs only in the final layers. Notably, models proficient in arithmetic exhibit a clear division of labor between attention and MLP modules, where attention propagates input information and MLP modules aggregate it. This division is absent in less proficient models. Furthermore, successful models appear to process more challenging arithmetic tasks functionally, suggesting reasoning capabilities beyond factual recall.