<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="dwivedi-etal-2026-efficient">
    <titleInfo>
        <title>Efficient Multi-Tenant LoRA Serving for Encoder Models via Late-Fusion Batched Matrix Multiplication</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Ashutosh</namePart>
        <namePart type="given">Rupoliya</namePart>
        <namePart type="family">Dwivedi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Keshav</namePart>
        <namePart type="family">Shrikant</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jyotsna</namePart>
        <namePart type="family">Rupaliya</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: Industry Track</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Pepa</namePart>
            <namePart type="family">Atanasova</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Gholamreza</namePart>
            <namePart type="family">Haffari</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Minghan</namePart>
            <namePart type="family">Wang</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">979-8-89176-460-6</identifier>
    </relatedItem>
    <abstract>Low-Rank Adaptation (LoRA) has become a popular method to adapt pre-trained models to specific domains. Applied to encoders in a multi-tenant service, it lets each tenant specialize a shared model for classification, retrieval, or ranking. Existing multi-tenant LoRA serving systems target autoregressive decoders, and for encoders the available tooling supports only per-request adapter swapping, whose cost grows with the number of registered adapters. We observe that, unlike decoder generation, the encoder forward pass is compute-bound and dominated by the shared base projections, so per-tenant adaptation can be expressed as plain batched matrix multiplication with no custom kernels. We present LateFuse, a pure-PyTorch system that serves many different LoRA-adapted encoders in a single shared forward pass on one GPU. LateFuse fuses adapters late, in activation space: the base projections run once over the mixed-tenant batch, while per-sample low-rank deltas and per-tenant classification heads are applied as batched matrix multiplications. Tenant weights are never merged into the shared base, ruling out cross-tenant contamination by construction. Our evaluation shows that LateFuse serving latency stays flat as the adapter pool grows to tens of thousands on a single GPU, and that it achieves 5.6–21.2x higher throughput than HuggingFace PEFT’s native mixed-batch API and up to 521× over per-adapter swapping, while registering new adapters over 299x faster.</abstract>
    <identifier type="citekey">dwivedi-etal-2026-efficient</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-industry.224/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>3753</start>
            <end>3763</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Efficient Multi-Tenant LoRA Serving for Encoder Models via Late-Fusion Batched Matrix Multiplication
%A Dwivedi, Ashutosh Rupoliya
%A Shrikant, Keshav
%A Rupaliya, Jyotsna
%Y Atanasova, Pepa
%Y Haffari, Gholamreza
%Y Wang, Minghan
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-460-6
%F dwivedi-etal-2026-efficient
%X Low-Rank Adaptation (LoRA) has become a popular method to adapt pre-trained models to specific domains. Applied to encoders in a multi-tenant service, it lets each tenant specialize a shared model for classification, retrieval, or ranking. Existing multi-tenant LoRA serving systems target autoregressive decoders, and for encoders the available tooling supports only per-request adapter swapping, whose cost grows with the number of registered adapters. We observe that, unlike decoder generation, the encoder forward pass is compute-bound and dominated by the shared base projections, so per-tenant adaptation can be expressed as plain batched matrix multiplication with no custom kernels. We present LateFuse, a pure-PyTorch system that serves many different LoRA-adapted encoders in a single shared forward pass on one GPU. LateFuse fuses adapters late, in activation space: the base projections run once over the mixed-tenant batch, while per-sample low-rank deltas and per-tenant classification heads are applied as batched matrix multiplications. Tenant weights are never merged into the shared base, ruling out cross-tenant contamination by construction. Our evaluation shows that LateFuse serving latency stays flat as the adapter pool grows to tens of thousands on a single GPU, and that it achieves 5.6–21.2x higher throughput than HuggingFace PEFT’s native mixed-batch API and up to 521× over per-adapter swapping, while registering new adapters over 299x faster.
%U https://aclanthology.org/2026.emnlp-industry.224/
%P 3753-3763

Markdown (Informal)

[Efficient Multi-Tenant LoRA Serving for Encoder Models via Late-Fusion Batched Matrix Multiplication](https://aclanthology.org/2026.emnlp-industry.224/) (Dwivedi et al., EMNLP 2026)

• Efficient Multi-Tenant LoRA Serving for Encoder Models via Late-Fusion Batched Matrix Multiplication (Dwivedi et al., EMNLP 2026)

ACL

• Ashutosh Rupoliya Dwivedi, Keshav Shrikant, and Jyotsna Rupaliya. 2026. Efficient Multi-Tenant LoRA Serving for Encoder Models via Late-Fusion Batched Matrix Multiplication. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track, pages 3753–3763, Budapest, Hungary. Association for Computational Linguistics.