Analyze Feature Flow to Enhance Interpretation and Steering in Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Laptev, Daniil, Balagansky, Nikita, Aksenov, Yaroslav, Gavrilov, Daniil
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908465160519680
author Laptev, Daniil
Balagansky, Nikita
Aksenov, Yaroslav
Gavrilov, Daniil
author_facet Laptev, Daniil
Balagansky, Nikita
Aksenov, Yaroslav
Gavrilov, Daniil
contents We introduce a new approach to systematically map features discovered by sparse autoencoder across consecutive layers of large language models, extending earlier work that examined inter-layer feature links. By using a data-free cosine similarity technique, we trace how specific features persist, transform, or first appear at each stage. This method yields granular flow graphs of feature evolution, enabling fine-grained interpretability and mechanistic insights into model computations. Crucially, we demonstrate how these cross-layer feature maps facilitate direct steering of model behavior by amplifying or suppressing chosen features, achieving targeted thematic control in text generation. Together, our findings highlight the utility of a causal, cross-layer interpretability framework that not only clarifies how features develop through forward passes but also provides new means for transparent manipulation of large language models.
format Preprint
id arxiv_https___arxiv_org_abs_2502_03032
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Analyze Feature Flow to Enhance Interpretation and Steering in Language Models
Laptev, Daniil
Balagansky, Nikita
Aksenov, Yaroslav
Gavrilov, Daniil
Machine Learning
Computation and Language
We introduce a new approach to systematically map features discovered by sparse autoencoder across consecutive layers of large language models, extending earlier work that examined inter-layer feature links. By using a data-free cosine similarity technique, we trace how specific features persist, transform, or first appear at each stage. This method yields granular flow graphs of feature evolution, enabling fine-grained interpretability and mechanistic insights into model computations. Crucially, we demonstrate how these cross-layer feature maps facilitate direct steering of model behavior by amplifying or suppressing chosen features, achieving targeted thematic control in text generation. Together, our findings highlight the utility of a causal, cross-layer interpretability framework that not only clarifies how features develop through forward passes but also provides new means for transparent manipulation of large language models.
title Analyze Feature Flow to Enhance Interpretation and Steering in Language Models
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2502.03032