@inproceedings{bhardia-etal-2026-mechanistic,
title = "Mechanistic Analysis Of Universality: Numerical Comparison Circuits Across Transformer Architectures",
author = "Bhardia, Arya and
Ramirez, Julian and
Verma, Siddhanta and
Mkrtchyan, Karen",
editor = "T.Y.S.S., Santosh and
Rodriguez, Juan Diego and
de Gibert, Ona",
booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics ({ACL} 2026)",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.acl-srw.84/",
pages = "951--967",
ISBN = "979-8-89176-393-7",
abstract = "Transformer language models reliably achieve high accuracy on many reasoning tasks; however, their internal mechanisms are not fully understood. Mechanistic interpretability seeks to remedy this gap by identifying task circuits within individual models, but it is unclear whether such circuits generalize across model families and scales. In this work, we study the universality of circuits through the lens of numerical comparisons, a simple and controlled task that enables clean and causal interventions. We conduct experiments on a set of transformer models spanning different families and sizes from 1.7b to 9b parameters. We find that models within the Qwen family exhibit a highly consistent circuit structure across architecture and scale, featuring localized attention heads that write a task relevant signal. In contrast, models from other families show qualitatively different implementations, where task relevant information emerges much earlier and is distributed across components as opposed to being concentrated within a small set of attention heads. These results serve as evidence that task behavior similarities do not imply mechanistic universality and highlight the necessity for cross model comparisons to claim generalization of internal circuits."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="bhardia-etal-2026-mechanistic">
<titleInfo>
<title>Mechanistic Analysis Of Universality: Numerical Comparison Circuits Across Transformer Architectures</title>
</titleInfo>
<name type="personal">
<namePart type="given">Arya</namePart>
<namePart type="family">Bhardia</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Julian</namePart>
<namePart type="family">Ramirez</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Siddhanta</namePart>
<namePart type="family">Verma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Karen</namePart>
<namePart type="family">Mkrtchyan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Santosh</namePart>
<namePart type="family">T.Y.S.S.</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Juan</namePart>
<namePart type="given">Diego</namePart>
<namePart type="family">Rodriguez</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ona</namePart>
<namePart type="family">de Gibert</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-393-7</identifier>
</relatedItem>
<abstract>Transformer language models reliably achieve high accuracy on many reasoning tasks; however, their internal mechanisms are not fully understood. Mechanistic interpretability seeks to remedy this gap by identifying task circuits within individual models, but it is unclear whether such circuits generalize across model families and scales. In this work, we study the universality of circuits through the lens of numerical comparisons, a simple and controlled task that enables clean and causal interventions. We conduct experiments on a set of transformer models spanning different families and sizes from 1.7b to 9b parameters. We find that models within the Qwen family exhibit a highly consistent circuit structure across architecture and scale, featuring localized attention heads that write a task relevant signal. In contrast, models from other families show qualitatively different implementations, where task relevant information emerges much earlier and is distributed across components as opposed to being concentrated within a small set of attention heads. These results serve as evidence that task behavior similarities do not imply mechanistic universality and highlight the necessity for cross model comparisons to claim generalization of internal circuits.</abstract>
<identifier type="citekey">bhardia-etal-2026-mechanistic</identifier>
<location>
<url>https://aclanthology.org/2026.acl-srw.84/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>951</start>
<end>967</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Mechanistic Analysis Of Universality: Numerical Comparison Circuits Across Transformer Architectures
%A Bhardia, Arya
%A Ramirez, Julian
%A Verma, Siddhanta
%A Mkrtchyan, Karen
%Y T.Y.S.S., Santosh
%Y Rodriguez, Juan Diego
%Y de Gibert, Ona
%S Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-393-7
%F bhardia-etal-2026-mechanistic
%X Transformer language models reliably achieve high accuracy on many reasoning tasks; however, their internal mechanisms are not fully understood. Mechanistic interpretability seeks to remedy this gap by identifying task circuits within individual models, but it is unclear whether such circuits generalize across model families and scales. In this work, we study the universality of circuits through the lens of numerical comparisons, a simple and controlled task that enables clean and causal interventions. We conduct experiments on a set of transformer models spanning different families and sizes from 1.7b to 9b parameters. We find that models within the Qwen family exhibit a highly consistent circuit structure across architecture and scale, featuring localized attention heads that write a task relevant signal. In contrast, models from other families show qualitatively different implementations, where task relevant information emerges much earlier and is distributed across components as opposed to being concentrated within a small set of attention heads. These results serve as evidence that task behavior similarities do not imply mechanistic universality and highlight the necessity for cross model comparisons to claim generalization of internal circuits.
%U https://aclanthology.org/2026.acl-srw.84/
%P 951-967
Markdown (Informal)
[Mechanistic Analysis Of Universality: Numerical Comparison Circuits Across Transformer Architectures](https://aclanthology.org/2026.acl-srw.84/) (Bhardia et al., ACL 2026)
ACL