<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="jha-etal-2026-prism">
    <titleInfo>
        <title>PRISM: Prompt Representation Internalization via Selective Compression into LoRA Weights</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Eshita</namePart>
        <namePart type="family">Jha</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Mingxian</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Andrei</namePart>
        <namePart type="family">Barbu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Chetan</namePart>
        <namePart type="family">Aggarwal</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>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-459-0</identifier>
    </relatedItem>
    <abstract>Deploying large language models for specialized tasks requires lengthy prompts combining schema, demonstrations, and instructions — creating a fundamental tension between prompt richness and inference feasibility. We identify and characterize two distinct prompt regimes: pattern-dominant tasks, where demonstrations drive performance through implicit formatting and extraction conventions, and procedure-dominant tasks, where explicit instructions carry the critical signal. To surface this distinction, we design a two-inference diagnostic that compares model performance with instructions alone versus examples alone, revealing which component to target for compression. Guided by this diagnostic, we introduce PRISM, a self-distillation framework that generates silver training data from the model’s own full-prompt outputs and trains lightweight LoRA adapters to internalize the dominant component’s behavior. PRISM recovers 131% of example benefit on CUAD and 151% on MultiWOZ (pattern-dominant), and 39% of instruction benefit on GSM8K (procedure-dominant), eliminating 72–79% of prompt tokens at inference time. Silver data captures 80–94% of gold-labeled performance without human annotations. Beyond compression, PRISM frees context capacity: replacing a long prompt with LoRA weights enables 89.5% compliance with new runtime instructions versus 75.4% with recency prompting. The diagnostic suggests a practical pre-training heuristic to indicate when distillation is likely to succeed, helping practitioners avoid costly training on non-beneficial components.</abstract>
    <identifier type="citekey">jha-etal-2026-prism</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2431/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>55189</start>
            <end>55201</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T PRISM: Prompt Representation Internalization via Selective Compression into LoRA Weights
%A Jha, Eshita
%A Wang, Mingxian
%A Barbu, Andrei
%A Aggarwal, Chetan
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F jha-etal-2026-prism
%X Deploying large language models for specialized tasks requires lengthy prompts combining schema, demonstrations, and instructions — creating a fundamental tension between prompt richness and inference feasibility. We identify and characterize two distinct prompt regimes: pattern-dominant tasks, where demonstrations drive performance through implicit formatting and extraction conventions, and procedure-dominant tasks, where explicit instructions carry the critical signal. To surface this distinction, we design a two-inference diagnostic that compares model performance with instructions alone versus examples alone, revealing which component to target for compression. Guided by this diagnostic, we introduce PRISM, a self-distillation framework that generates silver training data from the model’s own full-prompt outputs and trains lightweight LoRA adapters to internalize the dominant component’s behavior. PRISM recovers 131% of example benefit on CUAD and 151% on MultiWOZ (pattern-dominant), and 39% of instruction benefit on GSM8K (procedure-dominant), eliminating 72–79% of prompt tokens at inference time. Silver data captures 80–94% of gold-labeled performance without human annotations. Beyond compression, PRISM frees context capacity: replacing a long prompt with LoRA weights enables 89.5% compliance with new runtime instructions versus 75.4% with recency prompting. The diagnostic suggests a practical pre-training heuristic to indicate when distillation is likely to succeed, helping practitioners avoid costly training on non-beneficial components.
%U https://aclanthology.org/2026.emnlp-main.2431/
%P 55189-55201

Markdown (Informal)

[PRISM: Prompt Representation Internalization via Selective Compression into LoRA Weights](https://aclanthology.org/2026.emnlp-main.2431/) (Jha et al., EMNLP 2026)

• PRISM: Prompt Representation Internalization via Selective Compression into LoRA Weights (Jha et al., EMNLP 2026)

ACL

• Eshita Jha, Mingxian Wang, Andrei Barbu, and Chetan Aggarwal. 2026. PRISM: Prompt Representation Internalization via Selective Compression into LoRA Weights. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 55189–55201, Budapest, Hungary. Association for Computational Linguistics.