SACHITRA: Scalable Application of CPU-based HTML Integrated Transformer for Regenerative Augmented Image Generation Model

SACHITRA: Scalable Application of CPU-based HTML Integrated Transformer for Regenerative Augmented Image Generation Model | 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 SACHITRA: Scalable Application of CPU-based HTML Integrated Transformer for Regenerative Augmented Image Generation Model SHASHWAT PANDEY This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9702958/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 We present SACHITRA (Scalable Application of CPU-based HTML Integrated Transformer for Regenerative Augmented Image Generation Model), a generative image and video synthesis framework engineered for browser-native, CPU-bounded execution without dependence on dedicated GPU hardware. Existing state-of-the-art synthesis pipelines—including flow-matching models, attention-propagating Vision Transformers (ViTs), and large-scale diffusion architectures—achieve high perceptual fidelity but demand GPU clusters or NVIDIA A100-class accelerators, rendering them inaccessible to the vast majority of global computing endpoints, particularly in emerging economies. SACHITRA addresses this accessibility gap through three formally grounded contributions: (1) a Quantized HTML Pipeline (QHP), which orchestrates compressed generative sub-networks within an HTML5/WebAssembly execution context, exploiting browser SIMD intrinsics and SharedArrayBuffer for intra-session parallelism; (2) an Adaptive Finite-Difference Trajectory Approximation (AFDT), a training-free ODE acceleration method integrated with a UCB-1 multi-armed bandit (MAB) controller that dynamically skips redundant integration steps with formal error guarantees of order O ( |S|/T 3 ) under INT4 quantized arithmetic; and (3) a Cross-Layer Historical Attention Distillation (CHAD) mechanism that propagates compressed 4-bit integer attention matrices across encoder layers, reducing attention buffer memory by 16 × while sustaining accuracy through an analytically derived dynamic blending schedule. Experiments demonstrate a 2 . 1 × –3 . 4 × speedup over CPU-only diffusion baselines on commodity hardware (Intel Core i5, 8 GB RAM), with CLIP-IQA scores within 4.7% of GPU-resident reference models. SACHITRA constitutes the first formally verified, browser-deployable gener-ative synthesis pipeline designed for the Aatmanirbhar Bharat digital infrastructure mandate. Artificial Intelligence and Machine Learning Flow Matching CPU Inference WebAssembly HTML5 AI Pipeline Vision Transformer Quantization Edge AI Browser-Native Generation Multi-Armed Bandit Attention Propagation SACHITRA Aatmanirbhar Bharat 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-9702958","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":639671329,"identity":"38cc9829-9f63-42ee-9780-4633e881edf2","order_by":0,"name":"SHASHWAT PANDEY","email":"data:image/png;base64,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","orcid":"https://orcid.org/0009-0005-3946-6611","institution":"Indian Institute of Technology Madras","correspondingAuthor":true,"prefix":"","firstName":"SHASHWAT","middleName":"","lastName":"PANDEY","suffix":""}],"badges":[],"createdAt":"2026-05-13 11:09:34","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":true,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":true},"doi":"10.21203/rs.3.rs-9702958/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9702958/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":109255896,"identity":"fd4d2883-be20-45b2-8c8b-a2512238eaa0","added_by":"auto","created_at":"2026-05-14 10:04:27","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":464281,"visible":true,"origin":"","legend":"","description":"","filename":"SACHITRAfixed.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9702958/v1_covered_9a8a5aba-8553-49ee-96c3-4f29560c054d.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eSACHITRA: Scalable Application of CPU-based HTML Integrated Transformer for Regenerative Augmented Image Generation Model\u003c/p\u003e","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Indian Institute of Technology Madras","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":"Flow Matching, CPU Inference, WebAssembly, HTML5 AI Pipeline, Vision Transformer, Quantization, Edge AI, Browser-Native Generation, Multi-Armed Bandit, Attention Propagation, SACHITRA, Aatmanirbhar Bharat","lastPublishedDoi":"10.21203/rs.3.rs-9702958/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9702958/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eWe present SACHITRA (Scalable Application of CPU-based HTML Integrated Transformer for Regenerative Augmented Image Generation Model), a generative image and video synthesis framework engineered for browser-native, CPU-bounded execution without dependence on dedicated GPU hardware. Existing state-of-the-art synthesis pipelines—including flow-matching models, attention-propagating Vision Transformers (ViTs), and large-scale diffusion architectures—achieve high perceptual fidelity but demand GPU clusters or NVIDIA A100-class accelerators, rendering them inaccessible to the vast majority of global computing endpoints, particularly in emerging economies.\u003c/p\u003e\n\u003cp\u003eSACHITRA addresses this accessibility gap through three formally grounded contributions: (1) a Quantized HTML Pipeline (QHP), which orchestrates compressed generative sub-networks within an HTML5/WebAssembly execution context, exploiting browser SIMD intrinsics and SharedArrayBuffer for intra-session parallelism; (2) an Adaptive Finite-Difference Trajectory Approximation (AFDT), a training-free ODE acceleration method integrated with a UCB-1 multi-armed bandit (MAB) controller that dynamically skips redundant integration steps with formal error guarantees of order \u003cem\u003eO\u003c/em\u003e(\u003cem\u003e|S|/T \u003c/em\u003e\u003csup\u003e3\u003c/sup\u003e) under INT4 quantized arithmetic; and (3) a Cross-Layer Historical Attention Distillation (CHAD) mechanism that propagates compressed 4-bit integer attention matrices across encoder layers, reducing attention buffer memory by 16\u003cem\u003e× \u003c/em\u003ewhile sustaining accuracy through an analytically derived dynamic blending schedule. Experiments demonstrate a 2\u003cem\u003e.\u003c/em\u003e1\u003cem\u003e×\u003c/em\u003e–3\u003cem\u003e.\u003c/em\u003e4\u003cem\u003e× \u003c/em\u003espeedup over CPU-only diffusion baselines on commodity hardware (Intel Core i5, 8 GB RAM), with CLIP-IQA scores within 4.7% of GPU-resident reference models. SACHITRA constitutes the first formally verified, browser-deployable gener-ative synthesis pipeline designed for the Aatmanirbhar Bharat digital infrastructure mandate.\u003c/p\u003e","manuscriptTitle":"SACHITRA: Scalable Application of CPU-based HTML Integrated Transformer for Regenerative Augmented Image Generation Model","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-14 10:04:18","doi":"10.21203/rs.3.rs-9702958/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":"5c2e7f94-3421-4dd3-b3ed-a201d98d6bcb","owner":[],"postedDate":"May 14th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":68088964,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2026-05-14T10:04:19+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-14 10:04:18","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9702958","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9702958","identity":"rs-9702958","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}