{"id":"https://openalex.org/W7224934333","doi":"https://doi.org/10.48550/arxiv.2610.39761","title":"GPU-Resident Speculative Decoding for In-Browser LLM Inference with WebGPU","display_name":"GPU-Resident Speculative Decoding for In-Browser LLM Inference with WebGPU","publication_year":2026,"publication_date":"2026-10-29","ids":{"openalex":"https://openalex.org/W7224934333","doi":"https://doi.org/10.48550/arxiv.2610.39761"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.39761","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.39761","pdf_url":"https://arxiv.org/pdf/2610.39761","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2610.39761","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5160231361","display_name":"Wei Song","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I11701301","display_name":"Griffith University","ror":"https://ror.org/02sc3r913","country_code":"AU","type":"education","lineage":["https://openalex.org/I11701301"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Song, Wei","raw_affiliation_strings":["Griffith University Australia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Griffith University Australia","institution_ids":["https://openalex.org/I11701301"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5160190726","display_name":"Yuxin Cao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I165932596","display_name":"National University of Singapore","ror":"https://ror.org/01tgyzw49","country_code":"SG","type":"education","lineage":["https://openalex.org/I165932596"]}],"countries":["SG"],"is_corresponding":false,"raw_author_name":"Cao, Yuxin","raw_affiliation_strings":["National University of Singapore Singapore"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"National University of Singapore Singapore","institution_ids":["https://openalex.org/I165932596"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5160191175","display_name":"Ziqi Ding","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I31746571","display_name":"UNSW Sydney","ror":"https://ror.org/03r8z3t63","country_code":"AU","type":"education","lineage":["https://openalex.org/I31746571"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Ding, Ziqi","raw_affiliation_strings":["University of New South Wales Australia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of New South Wales Australia","institution_ids":["https://openalex.org/I31746571"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5160237569","display_name":"Xiao Cheng","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I99043593","display_name":"Macquarie University","ror":"https://ror.org/01sf06y89","country_code":"AU","type":"education","lineage":["https://openalex.org/I99043593"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Cheng, Xiao","raw_affiliation_strings":["Macquarie University Australia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Macquarie University Australia","institution_ids":["https://openalex.org/I99043593"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.7670999765396118,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.7670999765396118,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.12470000237226486,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.021800000220537186,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/speculative-decoding","display_name":"speculative decoding","score":0.871999979019165},{"id":"https://openalex.org/keywords/webgpu","display_name":"WebGPU","score":0.8190000057220459},{"id":"https://openalex.org/keywords/kv-cache-management","display_name":"KV cache management","score":0.5789999961853027}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8302000164985657},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.7621999979019165},{"id":"https://openalex.org/C2779182362","wikidata":"https://www.wikidata.org/wiki/Q17126187","display_name":"Session (web analytics)","level":2,"score":0.7078999876976013},{"id":"https://openalex.org/C2780008327","wikidata":"https://www.wikidata.org/wiki/Q3962","display_name":"Laptop","level":2,"score":0.486299991607666},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.47760000824928284},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4503999948501587},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.43220001459121704},{"id":"https://openalex.org/C51823790","wikidata":"https://www.wikidata.org/wiki/Q504353","display_name":"Greedy algorithm","level":2,"score":0.414000004529953},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4099999964237213},{"id":"https://openalex.org/C47941915","wikidata":"https://www.wikidata.org/wiki/Q107885","display_name":"Speculation","level":2,"score":0.38609999418258667},{"id":"https://openalex.org/C102392041","wikidata":"https://www.wikidata.org/wiki/Q592860","display_name":"Sliding window protocol","level":3,"score":0.36890000104904175},{"id":"https://openalex.org/C160145156","wikidata":"https://www.wikidata.org/wiki/Q778586","display_name":"Executable","level":2,"score":0.35510000586509705},{"id":"https://openalex.org/C3913047","wikidata":"https://www.wikidata.org/wiki/Q1956265","display_name":"sync","level":3,"score":0.34880000352859497},{"id":"https://openalex.org/C184670325","wikidata":"https://www.wikidata.org/wiki/Q512604","display_name":"Loop (graph theory)","level":2,"score":0.3391999900341034},{"id":"https://openalex.org/C2777200299","wikidata":"https://www.wikidata.org/wiki/Q52943","display_name":"Conversation","level":2,"score":0.3336000144481659},{"id":"https://openalex.org/C151319957","wikidata":"https://www.wikidata.org/wiki/Q752739","display_name":"Asynchronous communication","level":2,"score":0.3328000009059906},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.31690001487731934},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.3018999993801117},{"id":"https://openalex.org/C2778751112","wikidata":"https://www.wikidata.org/wiki/Q835016","display_name":"Window (computing)","level":2,"score":0.30140000581741333},{"id":"https://openalex.org/C203479927","wikidata":"https://www.wikidata.org/wiki/Q5165939","display_name":"Controller (irrigation)","level":2,"score":0.29750001430511475},{"id":"https://openalex.org/C71901391","wikidata":"https://www.wikidata.org/wiki/Q7126699","display_name":"Upload","level":2,"score":0.296099990606308},{"id":"https://openalex.org/C186644900","wikidata":"https://www.wikidata.org/wiki/Q194152","display_name":"Parsing","level":2,"score":0.29440000653266907},{"id":"https://openalex.org/C174839445","wikidata":"https://www.wikidata.org/wiki/Q1134386","display_name":"Lock (firearm)","level":2,"score":0.2930999994277954},{"id":"https://openalex.org/C147297375","wikidata":"https://www.wikidata.org/wiki/Q6674930","display_name":"Look-ahead","level":2,"score":0.29089999198913574},{"id":"https://openalex.org/C97137487","wikidata":"https://www.wikidata.org/wiki/Q729138","display_name":"Integer (computer science)","level":2,"score":0.28940001130104065},{"id":"https://openalex.org/C67953723","wikidata":"https://www.wikidata.org/wiki/Q192525","display_name":"Workstation","level":2,"score":0.28870001435279846},{"id":"https://openalex.org/C2778562939","wikidata":"https://www.wikidata.org/wiki/Q1298791","display_name":"Synchronization (alternating current)","level":3,"score":0.28760001063346863},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.2858000099658966},{"id":"https://openalex.org/C103088060","wikidata":"https://www.wikidata.org/wiki/Q1062839","display_name":"Error detection and correction","level":2,"score":0.2824999988079071},{"id":"https://openalex.org/C2776299755","wikidata":"https://www.wikidata.org/wiki/Q432449","display_name":"Carry (investment)","level":2,"score":0.2782000005245209},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.2696000039577484},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.2678000032901764},{"id":"https://openalex.org/C153269930","wikidata":"https://www.wikidata.org/wiki/Q430001","display_name":"Decidability","level":2,"score":0.2621999979019165},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.26030001044273376},{"id":"https://openalex.org/C2778858076","wikidata":"https://www.wikidata.org/wiki/Q5249539","display_name":"Decodes","level":3,"score":0.2597000002861023},{"id":"https://openalex.org/C2776036281","wikidata":"https://www.wikidata.org/wiki/Q48769818","display_name":"Constraint (computer-aided design)","level":2,"score":0.2590999901294708},{"id":"https://openalex.org/C187691185","wikidata":"https://www.wikidata.org/wiki/Q2020720","display_name":"Grid","level":2,"score":0.2574999928474426}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.39761","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.39761","pdf_url":"https://arxiv.org/pdf/2610.39761","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.39761","doi":"https://doi.org/10.48550/arxiv.2610.39761","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.39761","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2610.39761","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.39761","pdf_url":"https://arxiv.org/pdf/2610.39761","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7224934333.pdf","grobid_xml":"https://content.openalex.org/works/W7224934333.grobid-xml"},"referenced_works_count":10,"referenced_works":["https://openalex.org/W2951534261","https://openalex.org/W3177813494","https://openalex.org/W3210904070","https://openalex.org/W4319166707","https://openalex.org/W4364385701","https://openalex.org/W4399911469","https://openalex.org/W4405715435","https://openalex.org/W7150732884","https://openalex.org/W7161979406","https://openalex.org/W7202199482"],"related_works":["https://openalex.org/W4409671048","https://openalex.org/W7202199482","https://openalex.org/W7161979406","https://openalex.org/W7147468729","https://openalex.org/W7221809799","https://openalex.org/W7150732884","https://openalex.org/W7160652145","https://openalex.org/W7204276091","https://openalex.org/W4415798602","https://openalex.org/W4407186969"],"abstract_inverted_index":{"WebGPU":[0,30],"enables":[1],"private,":[2],"offline":[3],"LLM":[4],"inference":[5],"inside":[6],"web":[7],"pages,":[8],"but":[9],"browser":[10],"dispatch,":[11],"inter-process":[12],"communication,":[13],"synchronous":[14],"readback":[15],"and":[16,21,63,80,91,108,133,145,170],"single-token":[17],"kernels":[18,40,156],"slow":[19],"decoding":[20,25,50,165],"make":[22],"conventional":[23],"speculative":[24],"inefficient.":[26],"We":[27],"present":[28],"a":[29,67,84,122,139,148],"runtime":[31,113,168],"that":[32,176],"addresses":[33],"these":[34],"browser-specific":[35],"costs.":[36,97],"Deterministic":[37],"small-batch":[38],"integer":[39],"accelerate":[41],"multi-token":[42],"verification":[43],"while":[44,83,188],"preserving":[45],"exact":[46],"agreement":[47],"with":[48],"greedy":[49,164],"in":[51],"the":[52,88,92,154,158,162,167,171,183],"same":[53],"quantized":[54,163],"runtime.":[55],"A":[56,71],"GPU-resident":[57,143],"loop":[58,144],"performs":[59],"drafting,":[60],"verification,":[61],"acceptance":[62],"KV-cache":[64],"maintenance":[65],"without":[66,194],"per-token":[68],"CPU":[69],"decision.":[70],"draft-free":[72],"context-mixing":[73],"drafter":[74],"reuses":[75],"page":[76],"content,":[77],"conversation":[78],"history":[79],"session":[81,189],"memory,":[82],"device-aware":[85],"controller":[86],"adapts":[87],"speculation":[89,146],"width":[90],"submission":[93],"window":[94],"to":[95,103,116,126,135,151,186],"measured":[96],"Across":[98],"seven":[99],"models":[100],"from":[101],"0.5B":[102],"8B":[104],"parameters,":[105],"11":[106],"tasks":[107],"three":[109],"client":[110],"GPUs,":[111],"our":[112],"decodes":[114],"3.5":[115],"5.4$\\times$":[117],"faster":[118,128,137],"than":[119],"WebLLM":[120],"on":[121,129,138,157],"laptop":[123],"GPU,":[124],"4.8":[125,134],"6.2$\\times$":[127],"an":[130],"integrated":[131],"GPU":[132],"8.6$\\times$":[136],"workstation":[140],"GPU.":[141],"The":[142],"contribute":[147],"further":[149],"1.8":[150],"3.2$\\times$":[152],"over":[153],"optimized":[155],"workstation.":[159],"Outputs":[160],"match":[161],"of":[166,177],"exactly,":[169],"mean":[172],"GSM8K":[173],"accuracy":[174],"matches":[175],"WebLLM.":[178],"Optional":[179],"draft":[180],"heads":[181],"raise":[182],"peak":[184],"speedup":[185],"10.2$\\times$,":[187],"memory":[190],"provides":[191],"additional":[192],"gains":[193],"downloading":[195],"extra":[196],"weights.":[197]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-31T00:00:00"}
