@inproceedings{baek-lee-2026-components,
title = "Not All Components Forget Alike: Component-Level Intervention Profiles in Continual Pretraining",
author = "Baek, Sangyeol and
Lee, Suan",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1583/",
doi = "10.18653/2026.findings-emnlp.1583",
pages = "33719--33739",
ISBN = "979-8-89176-462-0",
abstract = "Continual pretraining adapts a language model to new domains but can degrade performance on earlier domains. We study this degradation through component-level interventions in Transformer blocks. Our main observation is that forgetting interventions have sharply different profiles across Q/K, V/O, and MLP components. In SmolLM2-360M continual pretraining over six domains, V/O rollback improves previous-domain Wiki perplexity while leaving new-domain perplexity nearly unchanged (-5.72 +/- 0.08 Wiki PPL diff; new-domain range -1.06 to +0.76), MLP rollback improves Wiki but severely damages all new domains (-6.29 +/- 0.12 Wiki; +13.95 to +20.75 new-domain damage), and Q/K rollback shows no recovery effect. Training-time V/O freeze reproduces the selective pattern: it reduces Wiki forgetting (-2.98 +/- 0.13, 3 seeds in the stress order; -2.84 +/- 0.29 across 3 seeds x 2 orders) without measurable loss in current-domain learning efficiency and without compensatory MLP/Q/K weight changes. Same-parameter-count Q/K freeze has the opposite direction (+3.17 +/- 2.53), and random attention-subset freeze scales with the included V/O fraction. Finally, inference-time inter-checkpoint activation transplantation modulates forgetting bidirectionally without changing weights (forward -6.27 +/- 0.13, reverse +3.26 +/- 0.09), with the activation-level effect carried primarily by o{\_}proj output rather than by the full V/O projection bundle. We present these results as a robust component-level intervention profile, not as proof that V/O is the unique causal storage site of forgetting."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="baek-lee-2026-components">
<titleInfo>
<title>Not All Components Forget Alike: Component-Level Intervention Profiles in Continual Pretraining</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sangyeol</namePart>
<namePart type="family">Baek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Suan</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Continual pretraining adapts a language model to new domains but can degrade performance on earlier domains. We study this degradation through component-level interventions in Transformer blocks. Our main observation is that forgetting interventions have sharply different profiles across Q/K, V/O, and MLP components. In SmolLM2-360M continual pretraining over six domains, V/O rollback improves previous-domain Wiki perplexity while leaving new-domain perplexity nearly unchanged (-5.72 +/- 0.08 Wiki PPL diff; new-domain range -1.06 to +0.76), MLP rollback improves Wiki but severely damages all new domains (-6.29 +/- 0.12 Wiki; +13.95 to +20.75 new-domain damage), and Q/K rollback shows no recovery effect. Training-time V/O freeze reproduces the selective pattern: it reduces Wiki forgetting (-2.98 +/- 0.13, 3 seeds in the stress order; -2.84 +/- 0.29 across 3 seeds x 2 orders) without measurable loss in current-domain learning efficiency and without compensatory MLP/Q/K weight changes. Same-parameter-count Q/K freeze has the opposite direction (+3.17 +/- 2.53), and random attention-subset freeze scales with the included V/O fraction. Finally, inference-time inter-checkpoint activation transplantation modulates forgetting bidirectionally without changing weights (forward -6.27 +/- 0.13, reverse +3.26 +/- 0.09), with the activation-level effect carried primarily by o_proj output rather than by the full V/O projection bundle. We present these results as a robust component-level intervention profile, not as proof that V/O is the unique causal storage site of forgetting.</abstract>
<identifier type="citekey">baek-lee-2026-components</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1583</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1583/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>33719</start>
<end>33739</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Not All Components Forget Alike: Component-Level Intervention Profiles in Continual Pretraining
%A Baek, Sangyeol
%A Lee, Suan
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F baek-lee-2026-components
%X Continual pretraining adapts a language model to new domains but can degrade performance on earlier domains. We study this degradation through component-level interventions in Transformer blocks. Our main observation is that forgetting interventions have sharply different profiles across Q/K, V/O, and MLP components. In SmolLM2-360M continual pretraining over six domains, V/O rollback improves previous-domain Wiki perplexity while leaving new-domain perplexity nearly unchanged (-5.72 +/- 0.08 Wiki PPL diff; new-domain range -1.06 to +0.76), MLP rollback improves Wiki but severely damages all new domains (-6.29 +/- 0.12 Wiki; +13.95 to +20.75 new-domain damage), and Q/K rollback shows no recovery effect. Training-time V/O freeze reproduces the selective pattern: it reduces Wiki forgetting (-2.98 +/- 0.13, 3 seeds in the stress order; -2.84 +/- 0.29 across 3 seeds x 2 orders) without measurable loss in current-domain learning efficiency and without compensatory MLP/Q/K weight changes. Same-parameter-count Q/K freeze has the opposite direction (+3.17 +/- 2.53), and random attention-subset freeze scales with the included V/O fraction. Finally, inference-time inter-checkpoint activation transplantation modulates forgetting bidirectionally without changing weights (forward -6.27 +/- 0.13, reverse +3.26 +/- 0.09), with the activation-level effect carried primarily by o_proj output rather than by the full V/O projection bundle. We present these results as a robust component-level intervention profile, not as proof that V/O is the unique causal storage site of forgetting.
%R 10.18653/2026.findings-emnlp.1583
%U https://aclanthology.org/2026.findings-emnlp.1583/
%U https://doi.org/10.18653/2026.findings-emnlp.1583
%P 33719-33739
Markdown (Informal)
[Not All Components Forget Alike: Component-Level Intervention Profiles in Continual Pretraining](https://aclanthology.org/2026.findings-emnlp.1583/) (Baek & Lee, Findings 2026)
ACL