Deep Kernel Fusion for Transformers

Zixi Zhang; Zhiwen Mo; Yiren Zhao; Robert D. Mullins

Deep Kernel Fusion for Transformers

Zixi Zhang, Zhiwen Mo, Yiren Zhao, Robert D. Mullins

Correct Metadata for

Use this form to create a GitHub issue with structured data describing the correction. You will need a GitHub account. Once you create that issue, the correction will be reviewed by a staff member.

⚠️ Mobile Users: Submitting this form to create a new issue will only work with github.com, not the GitHub Mobile app.

Important: The Anthology treat PDFs as authoritative. Please use this form only to correct data that is out of line with the PDF. See our corrections guidelines if you need to change the PDF.

Title Adjust the title. Retain tags such as <fixed-case>.

Authors Adjust author names and order to match the PDF.

Abstract Correct abstract if needed. Retain XML formatting tags such as <tex-math>. You may use <b>...</b> for bold, <i>...</i> for italic, and <url>...</url> for URLs.

Verification against PDF Ensure that the new title/authors match the snapshot below. (If there is no snapshot or it is too small, consult the PDF.)

Authors concatenated from the text boxes above:

ALL author names match the snapshot above—including middle initials, hyphens, and accents.

Abstract

Agentic LLM inference with long contexts is increasingly limited by memory bandwidth rather than compute. In this setting, SwiGLU MLP blocks, whose large weights exceed cache capacity, become a major yet under-optimized bottleneck in the Transformer architecture. We propose DeepFusionKernel, a deeply fused kernel that cuts HBM traffic and boosts cache reuse, delivering up to 13.2% speedup on H100 and 9.7% on A100 over SGLang. Integrated with SGLang and paired with a kernel scheduler, DeepFusionKernel ensures consistent accelerations across generation lengths, while remaining adaptable to diverse models, inference configurations, and hardware platforms.

Anthology ID:: 2026.acl-short.15
Volume:: Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)
Month:: July
Year:: 2026
Address:: San Diego, California, United States
Editors:: Maria Liakata, Viviane P. Moreira, Jiajun Zhang, David Jurgens
Venue:: ACL
SIG:
Publisher:: Association for Computational Linguistics
Note:
Pages:: 166–173
Language:
URL:: https://aclanthology.org/2026.acl-short.15/
DOI:
Bibkey:
Cite (ACL):: Zixi Zhang, Zhiwen Mo, Yiren Zhao, and Robert D. Mullins. 2026. Deep Kernel Fusion for Transformers. In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 166–173, San Diego, California, United States. Association for Computational Linguistics.
Cite (Informal):: Deep Kernel Fusion for Transformers (Zhang et al., ACL 2026)
Copy Citation:
PDF:: https://aclanthology.org/2026.acl-short.15.pdf
Checklist:: 2026.acl-short.15.checklist.pdf

PDF Cite Search Checklist Fix data

Export citation

BibTeX
MODS XML
Endnote
Preformatted

@inproceedings{zhang-etal-2026-deep,
    title = "Deep Kernel Fusion for Transformers",
    author = "Zhang, Zixi  and
      Mo, Zhiwen  and
      Zhao, Yiren  and
      Mullins, Robert D.",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 2: Short Papers)",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.acl-short.15/",
    pages = "166--173",
    ISBN = "979-8-89176-391-3",
    abstract = "Agentic LLM inference with long contexts is increasingly limited by memory bandwidth rather than compute. In this setting, SwiGLU MLP blocks, whose large weights exceed cache capacity, become a major yet under-optimized bottleneck in the Transformer architecture. We propose DeepFusionKernel, a deeply fused kernel that cuts HBM traffic and boosts cache reuse, delivering up to 13.2{\%} speedup on H100 and 9.7{\%} on A100 over SGLang. Integrated with SGLang and paired with a kernel scheduler, DeepFusionKernel ensures consistent accelerations across generation lengths, while remaining adaptable to diverse models, inference configurations, and hardware platforms."
}

Download as File

<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-deep">
    <titleInfo>
        <title>Deep Kernel Fusion for Transformers</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Zixi</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Zhiwen</namePart>
        <namePart type="family">Mo</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yiren</namePart>
        <namePart type="family">Zhao</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Robert</namePart>
        <namePart type="given">D</namePart>
        <namePart type="family">Mullins</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 (Volume 2: Short Papers)</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Maria</namePart>
            <namePart type="family">Liakata</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Viviane</namePart>
            <namePart type="given">P</namePart>
            <namePart type="family">Moreira</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jiajun</namePart>
            <namePart type="family">Zhang</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">David</namePart>
            <namePart type="family">Jurgens</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">979-8-89176-391-3</identifier>
    </relatedItem>
    <abstract>Agentic LLM inference with long contexts is increasingly limited by memory bandwidth rather than compute. In this setting, SwiGLU MLP blocks, whose large weights exceed cache capacity, become a major yet under-optimized bottleneck in the Transformer architecture. We propose DeepFusionKernel, a deeply fused kernel that cuts HBM traffic and boosts cache reuse, delivering up to 13.2% speedup on H100 and 9.7% on A100 over SGLang. Integrated with SGLang and paired with a kernel scheduler, DeepFusionKernel ensures consistent accelerations across generation lengths, while remaining adaptable to diverse models, inference configurations, and hardware platforms.</abstract>
    <identifier type="citekey">zhang-etal-2026-deep</identifier>
    <location>
        <url>https://aclanthology.org/2026.acl-short.15/</url>
    </location>
    <part>
        <date>2026-07</date>
        <extent unit="page">
            <start>166</start>
            <end>173</end>
        </extent>
    </part>
</mods>
</modsCollection>

Download as File

%0 Conference Proceedings
%T Deep Kernel Fusion for Transformers
%A Zhang, Zixi
%A Mo, Zhiwen
%A Zhao, Yiren
%A Mullins, Robert D.
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-391-3
%F zhang-etal-2026-deep
%X Agentic LLM inference with long contexts is increasingly limited by memory bandwidth rather than compute. In this setting, SwiGLU MLP blocks, whose large weights exceed cache capacity, become a major yet under-optimized bottleneck in the Transformer architecture. We propose DeepFusionKernel, a deeply fused kernel that cuts HBM traffic and boosts cache reuse, delivering up to 13.2% speedup on H100 and 9.7% on A100 over SGLang. Integrated with SGLang and paired with a kernel scheduler, DeepFusionKernel ensures consistent accelerations across generation lengths, while remaining adaptable to diverse models, inference configurations, and hardware platforms.
%U https://aclanthology.org/2026.acl-short.15/
%P 166-173

Download as File

Markdown (Informal)

[Deep Kernel Fusion for Transformers](https://aclanthology.org/2026.acl-short.15/) (Zhang et al., ACL 2026)

Deep Kernel Fusion for Transformers (Zhang et al., ACL 2026)

ACL

Zixi Zhang, Zhiwen Mo, Yiren Zhao, and Robert D. Mullins. 2026. Deep Kernel Fusion for Transformers. In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 166–173, San Diego, California, United States. Association for Computational Linguistics.