Dual Path Attribution: Efficient Attribution for SwiGLU-Transformers through Layer-Wise Target Propagation

TL;DR

This paper introduces Dual Path Attribution (DPA), a highly efficient method for interpreting SwiGLU-Transformers that provides accurate attribution with minimal computational cost, enabling scalable analysis of large language models.

Contribution

DPA is a novel framework that traces information flow in transformers with one forward and one backward pass, achieving linearization and O(1) complexity per component.

Findings

DPA achieves state-of-the-art faithfulness in attribution.

DPA scales efficiently to long sequences.

DPA outperforms existing methods in interpretability benchmarks.

Abstract

Understanding the internal mechanisms of transformer-based large language models (LLMs) is crucial for their reliable deployment and effective operation. While recent efforts have yielded a plethora of attribution methods attempting to balance faithfulness and computational efficiency, dense component attribution remains prohibitively expensive. In this work, we introduce Dual Path Attribution (DPA), a novel framework that faithfully traces information flow on the frozen transformer in one forward and one backward pass without requiring counterfactual examples. DPA analytically decomposes and linearizes the computational structure of the SwiGLU Transformers into distinct pathways along which it propagates a targeted unembedding vector to receive the effective representation at each residual position. This target-centric propagation achieves O(1) time complexity with respect to the number of model components, scaling to long input sequences and dense component attribution. Extensive experiments on standard interpretability benchmarks demonstrate that DPA achieves state-of-the-art faithfulness and unprecedented efficiency compared to existing baselines.