Stream: Scaling up Mechanistic Interpretability to Long Context in LLMs via Sparse Attention

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Rosser, J, García, José Luis Redondo, Penha, Gustavo, Palla, Konstantina, Bouchard, Hugues
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911415750623232
author Rosser, J
García, José Luis Redondo
Penha, Gustavo
Palla, Konstantina
Bouchard, Hugues
author_facet Rosser, J
García, José Luis Redondo
Penha, Gustavo
Palla, Konstantina
Bouchard, Hugues
contents As Large Language Models (LLMs) scale to million-token contexts, traditional Mechanistic Interpretability techniques for analyzing attention scale quadratically with context length, demanding terabytes of memory beyond 100,000 tokens. We introduce Sparse Tracing, a novel technique that leverages dynamic sparse attention to efficiently analyze long context attention patterns. We present Stream, a compilable hierarchical pruning algorithm that estimates per-head sparse attention masks in near-linear time $O(T \log T)$ and linear space $O(T)$, enabling one-pass interpretability at scale. Stream performs a binary-search-style refinement to retain only the top-$k$ key blocks per query while preserving the model's next-token behavior. We apply Stream to long chain-of-thought reasoning traces and identify thought anchors while pruning 97-99\% of token interactions. On the RULER benchmark, Stream preserves critical retrieval paths while discarding 90-96\% of interactions and exposes layer-wise routes from the needle to output. Our method offers a practical drop-in tool for analyzing attention patterns and tracing information flow without terabytes of caches. By making long context interpretability feasible on consumer GPUs, Sparse Tracing helps democratize chain-of-thought monitoring. Code is available at https://anonymous.4open.science/r/stream-03B8/.
format Preprint
id arxiv_https___arxiv_org_abs_2510_19875
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Stream: Scaling up Mechanistic Interpretability to Long Context in LLMs via Sparse Attention
Rosser, J
García, José Luis Redondo
Penha, Gustavo
Palla, Konstantina
Bouchard, Hugues
Computation and Language
Artificial Intelligence
68T40
I.2.11
As Large Language Models (LLMs) scale to million-token contexts, traditional Mechanistic Interpretability techniques for analyzing attention scale quadratically with context length, demanding terabytes of memory beyond 100,000 tokens. We introduce Sparse Tracing, a novel technique that leverages dynamic sparse attention to efficiently analyze long context attention patterns. We present Stream, a compilable hierarchical pruning algorithm that estimates per-head sparse attention masks in near-linear time $O(T \log T)$ and linear space $O(T)$, enabling one-pass interpretability at scale. Stream performs a binary-search-style refinement to retain only the top-$k$ key blocks per query while preserving the model's next-token behavior. We apply Stream to long chain-of-thought reasoning traces and identify thought anchors while pruning 97-99\% of token interactions. On the RULER benchmark, Stream preserves critical retrieval paths while discarding 90-96\% of interactions and exposes layer-wise routes from the needle to output. Our method offers a practical drop-in tool for analyzing attention patterns and tracing information flow without terabytes of caches. By making long context interpretability feasible on consumer GPUs, Sparse Tracing helps democratize chain-of-thought monitoring. Code is available at https://anonymous.4open.science/r/stream-03B8/.
title Stream: Scaling up Mechanistic Interpretability to Long Context in LLMs via Sparse Attention
topic Computation and Language
Artificial Intelligence
68T40
I.2.11
url https://arxiv.org/abs/2510.19875