Optimizing Large Language Models: A Novel Approach Through Dynamic Token Pruning

Optimizing Large Language Models: A Novel Approach Through Dynamic Token Pruning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Optimizing Large Language Models: A Novel Approach Through Dynamic Token Pruning Christopher Keith, Michael Robinson, Francis Duncan, Allan Worthington, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5293588/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The rapid evolution of artificial intelligence technologies has necessitated the development of frameworks capable of executing increasingly complex tasks with remarkable efficiency and speed. In response to the pressing demands for heightened computational capabilities, a sophisticated strategy has been conceived that not only addresses the performance challenges inherent in state-of-the-art language models but also seeks to optimize their operational efficiency. The methodology proposed here introduces dynamic token pruning, a transformative approach that carefully evaluates and selectively retains only the most crucial tokens during the inference process, thereby significantly reducing both inference time and memory consumption without undermining the integrity of the generated output. Through rigorous empirical analysis, the proposed framework demonstrates substantial enhancements in processing speed, achieving remarkable reductions in memory usage while maintaining a stable level of model accuracy, as indicated by perplexity metrics. The findings demonstrate the dual advantages of increased operational efficiency and sustained predictive performance, illustrating the capability of dynamic token pruning to adapt to varying input complexities. This research not only highlights the potential for improved accessibility and scalability of advanced language models in real-world applications but also lays the groundwork for future explorations into more complex optimization techniques that can further refine model performance in diverse contexts. The implications of these advancements extend beyond mere efficiency gains, contributing to the broader integration of AI technologies across a multitude of sectors and applications. Artificial Intelligence and Machine Learning Token Pruning Efficiency Inference Memory Usage Performance Full Text Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5293588","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":367920793,"identity":"5a4ca6b9-c2c7-48fd-9ba7-0fad1c988865","order_by":0,"name":"Christopher Keith","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA40lEQVRIiWNgGAWjYDACCQaGA0DIwyDBfADElSFeC48EWwKIywMWPUBAC1gFjwSPAYhJWAv/7N6HB3+csZGxl+75upm3zYKHn4H34OMP+Cy5c9zgMM+NNB4embPbbvO2SfBINvAlG+CzxUAijeEww4fDQL/kbrvNcwbovAM8ZhKEtBz88eE/UEvOM5gW8x+EtBzguXEApIXtNk8FxBa83pe4AXQYz5lkHqB/zG7OAWqRbOYxljiDRwv/jDTmjz+O2dmzz0h+duONQZ0cP3uP4YcKPFqwAGbSlI+CUTAKRsEowAIAd/FJzACm3xAAAAAASUVORK5CYII=","orcid":"https://orcid.org/0009-0003-5960-0123","institution":"","correspondingAuthor":true,"prefix":"","firstName":"Christopher","middleName":"","lastName":"Keith","suffix":""},{"id":367920794,"identity":"46779f8d-5a96-4469-ad9a-d3264c8d6b8c","order_by":1,"name":"Michael Robinson","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Michael","middleName":"","lastName":"Robinson","suffix":""},{"id":367920795,"identity":"6dae2630-e53d-49d2-9800-24e2c46cbbd5","order_by":2,"name":"Francis Duncan","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Francis","middleName":"","lastName":"Duncan","suffix":""},{"id":367920796,"identity":"59ed111a-f4f0-4f9e-a29e-a9df960171f6","order_by":3,"name":"Allan Worthington","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Allan","middleName":"","lastName":"Worthington","suffix":""},{"id":367920797,"identity":"7f2c6850-42e9-4cd0-b4c6-665c27b29be7","order_by":4,"name":"Joseph Wilson","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Joseph","middleName":"","lastName":"Wilson","suffix":""},{"id":367920798,"identity":"ffb26c0d-ab79-4536-96f3-594487b0fcac","order_by":5,"name":"Sofia Harris","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Sofia","middleName":"","lastName":"Harris","suffix":""}],"badges":[],"createdAt":"2024-10-19 09:02:25","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-5293588/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5293588/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":67169432,"identity":"908f1d7f-abbf-4f64-8c0d-552225de5e79","added_by":"auto","created_at":"2024-10-22 02:55:13","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":201682,"visible":true,"origin":"","legend":"","description":"","filename":"LLM.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5293588/v1_covered_52b30393-790e-4ab2-8bee-e6e7d4b16895.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eOptimizing Large Language Models: A Novel Approach Through Dynamic Token Pruning\u003c/p\u003e","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Token Pruning, Efficiency, Inference, Memory Usage, Performance","lastPublishedDoi":"10.21203/rs.3.rs-5293588/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5293588/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe rapid evolution of artificial intelligence technologies has necessitated the development of frameworks capable of executing increasingly complex tasks with remarkable efficiency and speed. In response to the pressing demands for heightened computational capabilities, a sophisticated strategy has been conceived that not only addresses the performance challenges inherent in state-of-the-art language models but also seeks to optimize their operational efficiency. The methodology proposed here introduces dynamic token pruning, a transformative approach that carefully evaluates and selectively retains only the most crucial tokens during the inference process, thereby significantly reducing both inference time and memory consumption without undermining the integrity of the generated output. Through rigorous empirical analysis, the proposed framework demonstrates substantial enhancements in processing speed, achieving remarkable reductions in memory usage while maintaining a stable level of model accuracy, as indicated by perplexity metrics. The findings demonstrate the dual advantages of increased operational efficiency and sustained predictive performance, illustrating the capability of dynamic token pruning to adapt to varying input complexities. This research not only highlights the potential for improved accessibility and scalability of advanced language models in real-world applications but also lays the groundwork for future explorations into more complex optimization techniques that can further refine model performance in diverse contexts. The implications of these advancements extend beyond mere efficiency gains, contributing to the broader integration of AI technologies across a multitude of sectors and applications.\u003c/p\u003e","manuscriptTitle":"Optimizing Large Language Models: A Novel Approach Through Dynamic Token Pruning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-10-22 02:47:06","doi":"10.21203/rs.3.rs-5293588/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2407b086-52b4-4c5d-ab45-87ab6b8fd205","owner":[],"postedDate":"October 22nd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":39152605,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2024-10-22T02:47:06+00:00","versionOfRecord":[],"versionCreatedAt":"2024-10-22 02:47:06","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5293588","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5293588","identity":"rs-5293588","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}