{"id":"https://openalex.org/W4403220941","doi":"https://doi.org/10.1145/3640457.3688040","title":"Enhancing Performance and Scalability of Large-Scale Recommendation Systems with Jagged Flash Attention","display_name":"Enhancing Performance and Scalability of Large-Scale Recommendation Systems with Jagged Flash Attention","publication_year":2024,"publication_date":"2024-10-08","ids":{"openalex":"https://openalex.org/W4403220941","doi":"https://doi.org/10.1145/3640457.3688040"},"language":"en","primary_location":{"id":"doi:10.1145/3640457.3688040","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3640457.3688040","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3640457.3688040","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"18th ACM Conference on Recommender Systems","raw_type":"proceedings-article"},"type":"preprint","indexed_in":["arxiv","crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://dl.acm.org/doi/pdf/10.1145/3640457.3688040","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5083192531","display_name":"Rengan Xu","orcid":"https://orcid.org/0000-0002-5230-5530"},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Rengan Xu","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5005149567","display_name":"Junjie Yang","orcid":"https://orcid.org/0000-0003-0603-4482"},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Junjie Yang","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5057666775","display_name":"Yifan Xu","orcid":"https://orcid.org/0000-0003-2676-6613"},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yifan Xu","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5112833347","display_name":"Hong Li","orcid":null},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hong Li","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100381490","display_name":"Xing Liu","orcid":null},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Xing Liu","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5057026035","display_name":"Devashish Shankar","orcid":null},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Devashish Shankar","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5055539155","display_name":"Haoci Zhang","orcid":null},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Haoci Zhang","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5112557620","display_name":"M. Liu","orcid":"https://orcid.org/0000-0002-1039-1565"},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Meng Liu","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007325176","display_name":"Boyang Li","orcid":null},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Boyang Li","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5073566291","display_name":"Yuxi Hu","orcid":null},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yuxi Hu","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5063149441","display_name":"Mingwei Tang","orcid":null},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Mingwei Tang","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113198800","display_name":"Zehua Zhang","orcid":null},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zehua Zhang","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5052236334","display_name":"Tunhou Zhang","orcid":"https://orcid.org/0000-0001-9590-9433"},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Tunhou Zhang","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102999050","display_name":"Dai Li","orcid":"https://orcid.org/0009-0000-6645-5620"},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Dai Li","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5072440584","display_name":"Sijia Chen","orcid":"https://orcid.org/0000-0002-8098-5852"},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Sijia Chen","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5107823466","display_name":"Gian-Paolo Musumeci","orcid":null},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Gian-Paolo Musumeci","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100349821","display_name":"Jiaqi Zhai","orcid":"https://orcid.org/0009-0004-7279-3318"},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jiaqi Zhai","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104325322","display_name":"Bill Zhu","orcid":null},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Bill Zhu","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076961426","display_name":"Hong Yan","orcid":"https://orcid.org/0009-0005-2306-7499"},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hong Yan","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5101926449","display_name":"S. Jaganmohan Reddy","orcid":"https://orcid.org/0009-0001-1504-9013"},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Srihari Reddy","raw_affiliation_strings":["Meta Platforms, USA"],"affiliations":[{"raw_affiliation_string":"Meta Platforms, USA","institution_ids":["https://openalex.org/I4210114444"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":20,"corresponding_author_ids":["https://openalex.org/A5083192531"],"corresponding_institution_ids":["https://openalex.org/I4210114444"],"apc_list":null,"apc_paid":null,"fwci":0.7794,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":{"value":0.79538426,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":96,"max":99},"biblio":{"volume":null,"issue":null,"first_page":"778","last_page":"780"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10101","display_name":"Cloud Computing and Resource Management","score":0.9973999857902527,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/T10101","display_name":"Cloud Computing and Resource Management","score":0.9973999857902527,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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":0.9973000288009644,"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/T11181","display_name":"Advanced Data Storage Technologies","score":0.9966999888420105,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/scalability","display_name":"Scalability","score":0.7890788316726685},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.7093556523323059},{"id":"https://openalex.org/keywords/flash","display_name":"Flash (photography)","score":0.5936803817749023},{"id":"https://openalex.org/keywords/scale","display_name":"Scale (ratio)","score":0.5764390230178833},{"id":"https://openalex.org/keywords/database","display_name":"Database","score":0.16413769125938416},{"id":"https://openalex.org/keywords/physics","display_name":"Physics","score":0.06407561898231506}],"concepts":[{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.7890788316726685},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7093556523323059},{"id":"https://openalex.org/C2777526259","wikidata":"https://www.wikidata.org/wiki/Q221836","display_name":"Flash (photography)","level":2,"score":0.5936803817749023},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.5764390230178833},{"id":"https://openalex.org/C77088390","wikidata":"https://www.wikidata.org/wiki/Q8513","display_name":"Database","level":1,"score":0.16413769125938416},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.06407561898231506},{"id":"https://openalex.org/C120665830","wikidata":"https://www.wikidata.org/wiki/Q14620","display_name":"Optics","level":1,"score":0.0},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1145/3640457.3688040","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3640457.3688040","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3640457.3688040","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"18th ACM Conference on Recommender Systems","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2409.15373","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2409.15373","pdf_url":"https://arxiv.org/pdf/2409.15373","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"}],"best_oa_location":{"id":"doi:10.1145/3640457.3688040","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3640457.3688040","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3640457.3688040","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"18th ACM Conference on Recommender Systems","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4403220941.pdf"},"referenced_works_count":2,"referenced_works":["https://openalex.org/W2954698171","https://openalex.org/W4296591817"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W2389214306"],"abstract_inverted_index":{"The":[0],"integration":[1],"of":[2,10,17,39,97,106],"hardware":[3],"accelerators":[4],"has":[5],"significantly":[6],"advanced":[7],"the":[8,15,25,57,104],"capabilities":[9],"modern":[11],"recommendation":[12,171],"systems,":[13],"enabling":[14,166],"exploration":[16],"complex":[18,178],"ranking":[19,62],"paradigms":[20],"previously":[21],"deemed":[22],"impractical.":[23],"However,":[24],"GPU-based":[26],"computational":[27],"costs":[28],"present":[29],"substantial":[30],"challenges.":[31],"In":[32,154],"this":[33],"paper,":[34],"we":[35,157],"demonstrate":[36],"our":[37,170],"development":[38],"an":[40],"efficiency-driven":[41],"approach":[42],"to":[43,86,123,133,145,168],"explore":[44],"these":[45],"paradigms,":[46],"moving":[47],"beyond":[48],"traditional":[49],"reliance":[50],"on":[51,65],"native":[52],"PyTorch":[53],"modules.":[54],"We":[55,76,101],"address":[56],"specific":[58],"challenges":[59],"posed":[60],"by":[61,109],"models\u2019":[63],"dependence":[64],"categorical":[66,92],"features,":[67],"which":[68],"vary":[69],"in":[70],"length":[71],"and":[72,127,149,162,176],"complicate":[73],"GPU":[74],"utilization.":[75],"introduce":[77],"Jagged":[78,111,118],"Feature":[79],"Interaction":[80],"Kernels,":[81],"a":[82],"novel":[83,117],"method":[84],"designed":[85],"extract":[87],"fine-grained":[88],"insights":[89],"from":[90],"long":[91],"features":[93,175],"through":[94],"efficient":[95],"handling":[96],"dynamically":[98],"sized":[99],"tensors.":[100],"further":[102],"enhance":[103],"performance":[105],"attention":[107],"mechanisms":[108],"integrating":[110],"tensors":[112],"with":[113,143,173],"Flash":[114,119],"Attention.":[115],"Our":[116],"Attention":[120],"achieves":[121],"up":[122,144],"9":[124],"\u00d7":[125,129,147],"speedup":[126,148],"22":[128],"memory":[130,152,164],"reduction":[131],"compared":[132],"dense":[134,140],"attention.":[135],"Notably,":[136],"it":[137],"also":[138],"outperforms":[139],"flash":[141],"attention,":[142],"3":[146],"53%":[150],"more":[151,177],"efficiency.":[153],"production":[155],"models,":[156],"observe":[158],"10%":[159],"QPS":[160],"improvement":[161],"18%":[163],"savings,":[165],"us":[167],"scale":[169],"systems":[172],"longer":[174],"architectures.":[179]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-03-27T05:58:40.876381","created_date":"2025-10-10T00:00:00"}
