CircuitSeer: Mining High-Quality Data by Probing Mathematical Reasoning Circuits in LLMs

TL;DR

CircuitSeer is a novel data selection approach that identifies and leverages core reasoning circuits within LLMs to efficiently select high-quality training data, improving reasoning performance with less data.

Contribution

This work introduces CircuitSeer, a new method that uses internal model circuits to select data, reducing reliance on external heuristics and costly models.

Findings

Selective data training improves reasoning accuracy.

CircuitSeer outperforms existing data selection methods.

High-quality reasoning data can be identified via internal model analysis.

Abstract

Large language models (LLMs) have demonstrated impressive reasoning capabilities, but scaling their performance often relies on massive reasoning datasets that are computationally expensive to train on. Existing data selection methods aim to curate smaller, high-quality subsets but often rely on costly external models or opaque heuristics. In this work, we shift the focus from external heuristics to the model's internal mechanisms. We find that complex reasoning tasks consistently activate a sparse, specialized subset of attention heads, forming core reasoning circuits. Building on this insight, we propose CircuitSeer, a novel data selection method that quantifies the reasoning complexity of data by measuring its influence on these crucial circuits. Extensive experiments on 4 models and 9 datasets demonstrate CircuitSeer's superiority. Notably, fine-tuning Qwen2.5-Math-7B on just 10% of data selected by our method achieves a 1.4-point gain in average Pass@1 over training on the full dataset, highlighting its efficiency and effectiveness.