{"id":"https://openalex.org/W7161979406","doi":"https://doi.org/10.48550/arxiv.2605.20706","title":"Llamas on the Web: Memory-Efficient, Performance-Portable, and Multi-Precision LLM Inference with WebGPU","display_name":"Llamas on the Web: Memory-Efficient, Performance-Portable, and Multi-Precision LLM Inference with WebGPU","publication_year":2026,"publication_date":"2026-05-20","ids":{"openalex":"https://openalex.org/W7161979406","doi":"https://doi.org/10.48550/arxiv.2605.20706"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.20706","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.20706","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,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.20706","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5065723018","display_name":"Reese Levine","orcid":"https://orcid.org/0000-0002-1324-0738"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Levine, Reese","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136716591","display_name":"Rithik Sharma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sharma, Rithik","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136630010","display_name":"Nikhil Jain","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jain, Nikhil","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5058082979","display_name":"Abhijit Ramesh","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ramesh, Abhijit","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114159455","display_name":"Zijian Chen","orcid":"https://orcid.org/0009-0005-6895-6329"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Zheyuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119918324","display_name":"Neha Abbas","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Abbas, Neha","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136673077","display_name":"James Contini","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Contini, James","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136614684","display_name":"Tyler Sorensen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sorensen, Tyler","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.24779999256134033,"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"}},"topics":[{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.24779999256134033,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.15129999816417694,"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/T10126","display_name":"Logic, programming, and type systems","score":0.040800001472234726,"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/inference","display_name":"Inference","score":0.6370999813079834},{"id":"https://openalex.org/keywords/implementation","display_name":"Implementation","score":0.524399995803833},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.5005999803543091},{"id":"https://openalex.org/keywords/kernel","display_name":"Kernel (algebra)","score":0.4839000105857849},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4587000012397766},{"id":"https://openalex.org/keywords/extensibility","display_name":"Extensibility","score":0.43630000948905945},{"id":"https://openalex.org/keywords/memory-model","display_name":"Memory model","score":0.4162999987602234}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8428000211715698},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6370999813079834},{"id":"https://openalex.org/C26713055","wikidata":"https://www.wikidata.org/wiki/Q245962","display_name":"Implementation","level":2,"score":0.524399995803833},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.5005999803543091},{"id":"https://openalex.org/C74193536","wikidata":"https://www.wikidata.org/wiki/Q574844","display_name":"Kernel (algebra)","level":2,"score":0.4839000105857849},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4587000012397766},{"id":"https://openalex.org/C32833848","wikidata":"https://www.wikidata.org/wiki/Q4115054","display_name":"Extensibility","level":2,"score":0.43630000948905945},{"id":"https://openalex.org/C12186640","wikidata":"https://www.wikidata.org/wiki/Q6815743","display_name":"Memory model","level":3,"score":0.4162999987602234},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.375900000333786},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.3517000079154968},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.3353999853134155},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.33180001378059387},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.31779998540878296},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.3138999938964844},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.3100000023841858},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.3005000054836273},{"id":"https://openalex.org/C138958017","wikidata":"https://www.wikidata.org/wiki/Q190087","display_name":"Data type","level":2,"score":0.2971000075340271},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2849999964237213},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.27900001406669617},{"id":"https://openalex.org/C176649486","wikidata":"https://www.wikidata.org/wiki/Q2308807","display_name":"Memory management","level":3,"score":0.27790001034736633},{"id":"https://openalex.org/C47487241","wikidata":"https://www.wikidata.org/wiki/Q5227230","display_name":"Data access","level":2,"score":0.27300000190734863},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.26910001039505005},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.26739999651908875},{"id":"https://openalex.org/C118643609","wikidata":"https://www.wikidata.org/wiki/Q189210","display_name":"Web application","level":2,"score":0.25060001015663147}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.20706","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.20706","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,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.20706","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.20706","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,"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"},"sustainable_development_goals":[{"display_name":"Industry, innovation and infrastructure","id":"https://metadata.un.org/sdg/9","score":0.48608285188674927}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Running":[0],"language":[1,120],"models":[2,121],"in":[3,60],"the":[4,37,61],"browser":[5],"presents":[6],"a":[7,40,53,81],"unique":[8],"opportunity":[9],"to":[10,104],"build":[11],"efficient,":[12],"private,":[13],"and":[14,25,48,73,85,102,122,135,149,160,190],"portable":[15],"AI":[16],"applications,":[17],"but":[18],"requires":[19,139],"contending":[20],"with":[21,189],"constrained":[22],"memory":[23,67,71,142],"availability":[24],"heterogeneous":[26],"hardware":[27],"targets.":[28],"To":[29],"realize":[30],"this":[31],"opportunity,":[32],"we":[33,177],"present":[34],"Llamas":[35],"on":[36,110,196],"Web":[38],"(LlamaWeb),":[39],"WebGPU":[41],"backend":[42,194],"for":[43],"llama$.$cpp":[44,183],"that":[45,90,137,162],"enables":[46],"memory-efficient":[47],"performance-portable":[49],"LLM":[50,133],"inference":[51],"across":[52,143,169],"wide":[54],"range":[55],"of":[56,94,146],"model":[57,75,100,124],"weight":[58,125],"formats":[59],"browser.":[62],"Our":[63],"design":[64],"significantly":[65],"reduces":[66],"overhead":[68],"through":[69,80],"static":[70],"planning":[72],"efficient":[74],"loading,":[76],"addresses":[77],"cross-device":[78],"variability":[79],"tunable":[82],"kernel":[83],"library,":[84],"introduces":[86],"templated":[87],"GPU":[88],"kernels":[89],"support":[91,101],"performant":[92],"implementations":[93],"numerous":[95],"quantization":[96],"formats,":[97],"enabling":[98],"broad":[99],"extensibility":[103],"new":[105],"formats.":[106,126],"We":[107,127,152],"evaluate":[108,154],"LlamaWeb":[109,129,138],"16":[111],"devices":[112],"from":[113,118,172],"8":[114],"vendors,":[115],"collecting":[116],"data":[117],"10":[119],"four":[123,170],"compare":[128,178],"against":[130,157,181],"existing":[131],"browser-based":[132],"frameworks":[134,159],"find":[136,161],"29-33%":[140],"less":[141],"several":[144],"combinations":[145],"device,":[147],"browser,":[148],"operating":[150],"system.":[151],"also":[153],"LlamaWeb's":[155,179],"performance":[156,180,195],"these":[158],"it":[163,186],"increases":[164],"decode":[165],"throughput":[166],"by":[167],"45-69%":[168],"GPUs":[171],"separate":[173],"vendors.":[174],"In":[175],"addition,":[176],"other":[182],"backends,":[184],"where":[185],"is":[187],"competitive":[188],"even":[191],"beats":[192],"vendor-specific":[193],"some":[197],"devices.":[198]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-22T00:00:00"}