{"id":"https://openalex.org/W7160826900","doi":"https://doi.org/10.48550/arxiv.2605.07569","title":"HexiSeq: Accommodating Long Context Training of LLMs over Heterogeneous Hardware","display_name":"HexiSeq: Accommodating Long Context Training of LLMs over Heterogeneous Hardware","publication_year":2026,"publication_date":"2026-05-08","ids":{"openalex":"https://openalex.org/W7160826900","doi":"https://doi.org/10.48550/arxiv.2605.07569"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.07569","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07569","pdf_url":null,"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":null,"license_id":null,"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.2605.07569","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135868097","display_name":"Yan Liang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liang, Yan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5070725905","display_name":"Youhe Jiang","orcid":"https://orcid.org/0009-0000-9297-642X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiang, Youhe","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135869349","display_name":"Ran Yan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yan, Ran","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135895626","display_name":"Binhang Yuan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuan, Binhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135843751","display_name":"Wei Wang","orcid":"https://orcid.org/0000-0003-4910-7370"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Wei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135840712","display_name":"Chuan Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Chuan","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":0.16279999911785126,"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":0.16279999911785126,"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/T14347","display_name":"Big Data and Digital Economy","score":0.1451999992132187,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.07720000296831131,"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/homogeneous","display_name":"Homogeneous","score":0.7476999759674072},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.7009999752044678},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.6394000053405762},{"id":"https://openalex.org/keywords/parallelism","display_name":"Parallelism (grammar)","score":0.5594000220298767},{"id":"https://openalex.org/keywords/heterogeneous-network","display_name":"Heterogeneous network","score":0.552299976348877},{"id":"https://openalex.org/keywords/symmetric-multiprocessor-system","display_name":"Symmetric multiprocessor system","score":0.5102999806404114},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.49160000681877136},{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.3853999972343445}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8216999769210815},{"id":"https://openalex.org/C66882249","wikidata":"https://www.wikidata.org/wiki/Q169336","display_name":"Homogeneous","level":2,"score":0.7476999759674072},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.7009999752044678},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.6456999778747559},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.6394000053405762},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.5594000220298767},{"id":"https://openalex.org/C158207573","wikidata":"https://www.wikidata.org/wiki/Q5747224","display_name":"Heterogeneous network","level":4,"score":0.552299976348877},{"id":"https://openalex.org/C172430144","wikidata":"https://www.wikidata.org/wiki/Q17111997","display_name":"Symmetric multiprocessor system","level":2,"score":0.5102999806404114},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.4932999908924103},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.49160000681877136},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.3853999972343445},{"id":"https://openalex.org/C61483411","wikidata":"https://www.wikidata.org/wiki/Q3124522","display_name":"Data parallelism","level":3,"score":0.3749000132083893},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.36399999260902405},{"id":"https://openalex.org/C187077762","wikidata":"https://www.wikidata.org/wiki/Q5036976","display_name":"Capture effect","level":4,"score":0.28769999742507935},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.28760001063346863},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.2802000045776367},{"id":"https://openalex.org/C183322885","wikidata":"https://www.wikidata.org/wiki/Q17007702","display_name":"Context model","level":3,"score":0.27630001306533813},{"id":"https://openalex.org/C2779227376","wikidata":"https://www.wikidata.org/wiki/Q6505497","display_name":"Layer (electronics)","level":2,"score":0.2754000127315521},{"id":"https://openalex.org/C2989134064","wikidata":"https://www.wikidata.org/wiki/Q288510","display_name":"Execution time","level":2,"score":0.2614000141620636},{"id":"https://openalex.org/C29202148","wikidata":"https://www.wikidata.org/wiki/Q287260","display_name":"Resource allocation","level":2,"score":0.25040000677108765}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.07569","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07569","pdf_url":null,"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.07569","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07569","pdf_url":null,"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Long-context":[0],"training":[1,20],"of":[2],"large":[3],"language":[4],"models":[5,40,114],"(LLMs)":[6],"is":[7],"commonly":[8],"distributed":[9],"with":[10,37,154],"Context":[11],"Parallelism":[12,16],"(CP)":[13],"and":[14,31,41,66,74,87,103,109,120,135,143,148],"Head":[15],"(HP),":[17],"but":[18],"existing":[19],"systems":[21],"largely":[22],"assume":[23],"homogeneous":[24,167,176],"GPU":[25,35,39,161],"meshes.":[26],"This":[27],"paper":[28],"extends":[29],"CP":[30,102],"HP":[32,104],"to":[33,70,117,124,137,150,159,173],"heterogeneous":[34,79,111,180],"clusters":[36,181],"mixed":[38,140],"non-uniform":[42],"network":[43],"bandwidths,":[44],"a":[45,54,83],"common":[46],"setting":[47],"in":[48,152],"production":[49],"training.":[50,189],"We":[51,77,97],"introduce":[52],"HexiSeq,":[53],"system":[55],"that":[56,179],"supports":[57],"fully":[58],"asymmetric":[59],"CP--HP":[60,80],"partitioning":[61],"by":[62,131,144],"assigning":[63],"sequence":[64],"shards":[65],"attention":[67],"heads":[68],"according":[69],"device":[71],"compute,":[72],"memory,":[73],"communication":[75],"capabilities.":[76],"formalize":[78],"allocation":[81],"as":[82],"constrained":[84],"optimization":[85],"problem":[86],"develop":[88],"an":[89],"efficient":[90],"hierarchical":[91],"scheduler":[92],"for":[93,186],"finding":[94],"optimal":[95],"schedules.":[96],"evaluate":[98],"HexiSeq":[99,128,169],"against":[100,166],"state-of-the-art":[101],"baselines":[105],"on":[106,133,139,146],"both":[107],"real":[108],"simulated":[110],"clusters.":[112],"Across":[113],"from":[115],"3B":[116],"70B":[118],"parameters":[119],"context":[121],"lengths":[122],"up":[123,136,149,158],"one":[125],"million":[126],"tokens,":[127],"improves":[129],"throughput":[130,171],"$1.11\\times$":[132],"average":[134,147],"$1.19\\times$":[138],"H100--A100":[141],"testbeds,":[142],"$1.36\\times$":[145],"$1.72\\times$":[151],"simulations":[153],"32--128":[155],"GPUs":[156],"spanning":[157],"four":[160],"models.":[162],"On":[163],"FLOP-comparable":[164],"pairs":[165],"clusters,":[168],"reaches":[170],"close":[172],"the":[174],"strongest":[175],"baseline,":[177],"showing":[178],"can":[182],"be":[183],"used":[184],"efficiently":[185],"long-context":[187],"LLM":[188]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-12T00:00:00"}