{"id":"https://openalex.org/W4407356947","doi":"https://doi.org/10.1145/3709703","title":"MEMO: Fine-grained Tensor Management For Ultra-long Context LLM Training","display_name":"MEMO: Fine-grained Tensor Management For Ultra-long Context LLM Training","publication_year":2025,"publication_date":"2025-02-10","ids":{"openalex":"https://openalex.org/W4407356947","doi":"https://doi.org/10.1145/3709703"},"language":"en","primary_location":{"id":"doi:10.1145/3709703","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3709703","pdf_url":null,"source":{"id":"https://openalex.org/S4387289859","display_name":"Proceedings of the ACM on Management of Data","issn_l":"2836-6573","issn":["2836-6573"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the ACM on Management of Data","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5047834854","display_name":"Pinxue Zhao","orcid":null},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Pinxue Zhao","raw_affiliation_strings":["School of Computer Science, Peking University, Beijing, China, & Key Lab of High Confidence Software Technologies (MOE), Peking University, Beijing, China"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Peking University, Beijing, China, & Key Lab of High Confidence Software Technologies (MOE), Peking University, Beijing, China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100462441","display_name":"Hailin Zhang","orcid":"https://orcid.org/0009-0000-4188-7742"},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Hailin Zhang","raw_affiliation_strings":["School of Computer Science, Peking University, Beijing, China, & Key Lab of High Confidence Software Technologies (MOE), Peking University, Beijing, China"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Peking University, Beijing, China, & Key Lab of High Confidence Software Technologies (MOE), Peking University, Beijing, China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5039254679","display_name":"Fangcheng Fu","orcid":"https://orcid.org/0000-0003-1658-0380"},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Fangcheng Fu","raw_affiliation_strings":["School of Computer Science, Peking University, Beijing, China, & Key Lab of High Confidence Software Technologies (MOE), Peking University, Beijing, China"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Peking University, Beijing, China, & Key Lab of High Confidence Software Technologies (MOE), Peking University, Beijing, China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5059601307","display_name":"Xiaonan Nie","orcid":"https://orcid.org/0000-0001-6766-757X"},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiaonan Nie","raw_affiliation_strings":["School of Computer Science, Peking University, Beijing, China, & Key Lab of High Confidence Software Technologies (MOE), Peking University, Beijing, China"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Peking University, Beijing, China, & Key Lab of High Confidence Software Technologies (MOE), Peking University, Beijing, China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5107467649","display_name":"Qibin Liu","orcid":null},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qibin Liu","raw_affiliation_strings":["Tencent Inc., Beijing, China"],"affiliations":[{"raw_affiliation_string":"Tencent Inc., Beijing, China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101834714","display_name":"Fang Yang","orcid":null},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Fang Yang","raw_affiliation_strings":["Tencent Inc., Beijing, China"],"affiliations":[{"raw_affiliation_string":"Tencent Inc., Beijing, China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5109811369","display_name":"Yei\u2010Mei Peng","orcid":null},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yuanbo Peng","raw_affiliation_strings":["Tencent Inc., Beijing, China"],"affiliations":[{"raw_affiliation_string":"Tencent Inc., Beijing, China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5077944758","display_name":"Dian Jiao","orcid":"https://orcid.org/0009-0003-3016-3740"},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Dian Jiao","raw_affiliation_strings":["Tencent Inc., Beijing, China"],"affiliations":[{"raw_affiliation_string":"Tencent Inc., Beijing, China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5016849647","display_name":"Shuaipeng Li","orcid":"https://orcid.org/0009-0009-6325-7114"},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shuaipeng Li","raw_affiliation_strings":["Tencent Inc., Beijing, China"],"affiliations":[{"raw_affiliation_string":"Tencent Inc., Beijing, China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5025328651","display_name":"Jinbao Xue","orcid":"https://orcid.org/0009-0003-4087-9873"},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jinbao Xue","raw_affiliation_strings":["Tencent Inc., Beijing, China"],"affiliations":[{"raw_affiliation_string":"Tencent Inc., Beijing, China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5110853771","display_name":"Yangyu Tao","orcid":null},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yangyu Tao","raw_affiliation_strings":["Tencent Inc., Beijing, China"],"affiliations":[{"raw_affiliation_string":"Tencent Inc., Beijing, China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5062357883","display_name":"Bin Cui","orcid":"https://orcid.org/0000-0003-1681-4677"},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Bin Cui","raw_affiliation_strings":["School of Computer Science, Peking University, Beijing, China, Key Lab of High Confidence Software Technologies (MOE), Peking University, Beijing, China, & Institute of Computational Social Science, Peking University, Qingdao, China"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Peking University, Beijing, China, Key Lab of High Confidence Software Technologies (MOE), Peking University, Beijing, China, & Institute of Computational Social Science, Peking University, Qingdao, China","institution_ids":["https://openalex.org/I20231570"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":12,"corresponding_author_ids":["https://openalex.org/A5047834854"],"corresponding_institution_ids":["https://openalex.org/I20231570"],"apc_list":null,"apc_paid":null,"fwci":4.9698,"has_fulltext":false,"cited_by_count":2,"citation_normalized_percentile":{"value":0.93936664,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":95,"max":96},"biblio":{"volume":"3","issue":"1","first_page":"1","last_page":"28"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T13650","display_name":"Computational Physics and Python Applications","score":0.9797999858856201,"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/T13650","display_name":"Computational Physics and Python Applications","score":0.9797999858856201,"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":0.9726999998092651,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.9624000191688538,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/training","display_name":"Training (meteorology)","score":0.7122353315353394},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.6588716506958008},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.44305747747421265},{"id":"https://openalex.org/keywords/psychology","display_name":"Psychology","score":0.3302382826805115},{"id":"https://openalex.org/keywords/geology","display_name":"Geology","score":0.20526105165481567},{"id":"https://openalex.org/keywords/geography","display_name":"Geography","score":0.13406452536582947},{"id":"https://openalex.org/keywords/meteorology","display_name":"Meteorology","score":0.06560975313186646}],"concepts":[{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.7122353315353394},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6588716506958008},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.44305747747421265},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.3302382826805115},{"id":"https://openalex.org/C127313418","wikidata":"https://www.wikidata.org/wiki/Q1069","display_name":"Geology","level":0,"score":0.20526105165481567},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.13406452536582947},{"id":"https://openalex.org/C153294291","wikidata":"https://www.wikidata.org/wiki/Q25261","display_name":"Meteorology","level":1,"score":0.06560975313186646},{"id":"https://openalex.org/C151730666","wikidata":"https://www.wikidata.org/wiki/Q7205","display_name":"Paleontology","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3709703","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3709703","pdf_url":null,"source":{"id":"https://openalex.org/S4387289859","display_name":"Proceedings of the ACM on Management of Data","issn_l":"2836-6573","issn":["2836-6573"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the ACM on Management of Data","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":51,"referenced_works":["https://openalex.org/W569478347","https://openalex.org/W2060393849","https://openalex.org/W2151822116","https://openalex.org/W2489529491","https://openalex.org/W2542459869","https://openalex.org/W2752849758","https://openalex.org/W2943698992","https://openalex.org/W2998702515","https://openalex.org/W3012514909","https://openalex.org/W3038576231","https://openalex.org/W3081168214","https://openalex.org/W3129890341","https://openalex.org/W3138303811","https://openalex.org/W3165312848","https://openalex.org/W3176459841","https://openalex.org/W3196732841","https://openalex.org/W4210352519","https://openalex.org/W4220938834","https://openalex.org/W4226328099","https://openalex.org/W4282974849","https://openalex.org/W4310282800","https://openalex.org/W4312983671","https://openalex.org/W4313479568","https://openalex.org/W4317212783","https://openalex.org/W4317767805","https://openalex.org/W4323343904","https://openalex.org/W4380433123","https://openalex.org/W4380433265","https://openalex.org/W4383218913","https://openalex.org/W4383472803","https://openalex.org/W4383749415","https://openalex.org/W4386768656","https://openalex.org/W4388169576","https://openalex.org/W4389341453","https://openalex.org/W4391032878","https://openalex.org/W4392207935","https://openalex.org/W4392453885","https://openalex.org/W4393180038","https://openalex.org/W4394998532","https://openalex.org/W4394998892","https://openalex.org/W4398233813","https://openalex.org/W4399175255","https://openalex.org/W4400893425","https://openalex.org/W4401023675","https://openalex.org/W4401812232","https://openalex.org/W4402715432","https://openalex.org/W4404400729","https://openalex.org/W6600100092","https://openalex.org/W6600120041","https://openalex.org/W6602670149","https://openalex.org/W6810370753"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W230091440","https://openalex.org/W2390279801","https://openalex.org/W2233261550","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2810751659"],"abstract_inverted_index":{"Nowadays,":[0],"Large":[1],"Language":[2],"Models":[3],"(LLMs)":[4],"have":[5,54],"been":[6],"trained":[7],"using":[8,114],"extended":[9],"context":[10,19,50],"lengths":[11,112],"to":[12,34,120,145,196,205,208,228],"foster":[13],"more":[14],"creative":[15],"applications.":[16],"However,":[17],"long":[18,49],"training":[20,91,239],"poses":[21],"great":[22],"challenges":[23],"considering":[24],"the":[25,100,132,137,163,219,223],"constraint":[26],"of":[27,63,103,108,139,190,240,256],"GPU":[28],"memory.":[29],"It":[30],"not":[31],"only":[32],"leads":[33],"substantial":[35],"activation":[36,96,155,233],"memory":[37,45,97,109,122,164,177,210,224,234],"consumption":[38],"during":[39,131],"training,":[40,51],"but":[41],"also":[42],"incurs":[43],"considerable":[44],"fragmentation.":[46,229],"To":[47,135],"facilitate":[48],"existing":[52],"frameworks":[53],"adopted":[55],"strategies":[56],"such":[57],"as":[58],"recomputation":[59,156,213],"and":[60,105,128,144,157,192,198,214,217],"various":[61],"forms":[62],"parallelisms.":[64],"Nevertheless,":[65],"these":[66],"techniques":[67],"rely":[68],"on":[69,248],"redundant":[70],"computation":[71,104],"or":[72],"extensive":[73],"communication,":[74,216],"resulting":[75],"in":[76],"low":[77],"Model":[78],"FLOPS":[79],"Utilization":[80],"(MFU).":[81],"In":[82],"this":[83],"paper,":[84],"we":[85,116,151,161],"propose":[86],"MEMO,":[87],"a":[88,153,169],"novel":[89],"LLM":[90,242],"framework":[92],"designed":[93],"for":[94],"fine-grained":[95,232],"management.":[98],"Given":[99],"quadratic":[101],"scaling":[102,107],"linear":[106],"with":[110,222,243],"sequence":[111,246],"when":[113],"FlashAttention,":[115],"offload":[117],"memory-consuming":[118],"activations":[119,140],"CPU":[121,149],"after":[123],"each":[124],"layer's":[125],"forward":[126],"pass":[127],"fetch":[129],"them":[130],"backward":[133],"pass.":[134],"maximize":[136],"swapping":[138,158],"without":[141],"hindering":[142],"computation,":[143],"avoid":[146],"exhausting":[147],"limited":[148],"memory,":[150],"implement":[152],"token-wise":[154],"mechanism.":[159],"Furthermore,":[160],"tackle":[162],"fragmentation":[165],"issue":[166],"by":[167],"employing":[168],"bi-level":[170],"Mixed":[171],"Integer":[172],"Programming":[173],"(MIP)":[174],"approach,":[175],"optimizing":[176],"reuse":[178],"across":[179],"transformer":[180],"layers.":[181],"Empirical":[182],"results":[183],"demonstrate":[184],"that":[185],"MEMO":[186,236],"achieves":[187],"an":[188,254],"average":[189],"1.97x":[191],"1.80x":[193],"MFU":[194,255],"compared":[195],"Megatron-LM":[197],"DeepSpeed,":[199],"respectively.":[200],"This":[201],"improvement":[202],"is":[203],"attributed":[204],"MEMO's":[206],"ability":[207],"minimize":[209],"fragmentation,":[211],"reduce":[212],"intensive":[215],"circumvent":[218],"delays":[220],"associated":[221],"reorganization":[225],"process":[226],"due":[227],"By":[230],"leveraging":[231],"management,":[235],"facilitates":[237],"efficient":[238],"7B":[241],"1":[244],"million":[245],"length":[247],"just":[249],"8":[250],"A800":[251],"GPUs,":[252],"achieving":[253],"52.30%.":[257]},"counts_by_year":[{"year":2025,"cited_by_count":2}],"updated_date":"2026-03-27T05:58:40.876381","created_date":"2025-10-10T00:00:00"}