{"id":"https://openalex.org/W4415326873","doi":"https://doi.org/10.1145/3712285.3759775","title":"Balanced and Elastic End-to-end Training of Dynamic LLMs","display_name":"Balanced and Elastic End-to-end Training of Dynamic LLMs","publication_year":2025,"publication_date":"2025-11-12","ids":{"openalex":"https://openalex.org/W4415326873","doi":"https://doi.org/10.1145/3712285.3759775"},"language":"en","primary_location":{"id":"doi:10.1145/3712285.3759775","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3712285.3759775","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis","raw_type":"proceedings-article"},"type":"preprint","indexed_in":["arxiv","crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3712285.3759775","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5002208999","display_name":"Mohamed Wahib","orcid":"https://orcid.org/0000-0002-7165-2095"},"institutions":[{"id":"https://openalex.org/I4210129730","display_name":"RIKEN Center for Computational Science","ror":"https://ror.org/03r519674","country_code":"JP","type":"facility","lineage":["https://openalex.org/I4210110652","https://openalex.org/I4210129730"]}],"countries":["JP"],"is_corresponding":true,"raw_author_name":"Mohamed Wahib","raw_affiliation_strings":["RIKEN Center for Computational Science (R-CCS), Tokyo, Japan"],"affiliations":[{"raw_affiliation_string":"RIKEN Center for Computational Science (R-CCS), Tokyo, Japan","institution_ids":["https://openalex.org/I4210129730"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120054218","display_name":"Muhammet Abdullah Soyt\u00fcrk","orcid":"https://orcid.org/0000-0002-2880-0857"},"institutions":[{"id":"https://openalex.org/I1351752","display_name":"Ko\u00e7 University","ror":"https://ror.org/00jzwgz36","country_code":"TR","type":"education","lineage":["https://openalex.org/I1351752"]}],"countries":["TR"],"is_corresponding":false,"raw_author_name":"Muhammed Abdullah Soyturk","raw_affiliation_strings":["Ko\u00e7 University, Turkey, Istanbul, Turkiye"],"affiliations":[{"raw_affiliation_string":"Ko\u00e7 University, Turkey, Istanbul, Turkiye","institution_ids":["https://openalex.org/I1351752"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5040713791","display_name":"Didem Unat","orcid":"https://orcid.org/0000-0002-2351-0770"},"institutions":[{"id":"https://openalex.org/I1351752","display_name":"Ko\u00e7 University","ror":"https://ror.org/00jzwgz36","country_code":"TR","type":"education","lineage":["https://openalex.org/I1351752"]}],"countries":["TR"],"is_corresponding":false,"raw_author_name":"Didem Unat","raw_affiliation_strings":["Ko\u00e7 University, Turkey, Istanbul, Turkiye and NVIDIA Corporation, Istanbul, Turkiye"],"affiliations":[{"raw_affiliation_string":"Ko\u00e7 University, Turkey, Istanbul, Turkiye and NVIDIA Corporation, Istanbul, Turkiye","institution_ids":["https://openalex.org/I1351752"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":3,"corresponding_author_ids":["https://openalex.org/A5002208999"],"corresponding_institution_ids":["https://openalex.org/I4210129730"],"apc_list":null,"apc_paid":null,"fwci":1.3682,"has_fulltext":false,"cited_by_count":1,"citation_normalized_percentile":{"value":0.85814078,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":91,"max":95},"biblio":{"volume":null,"issue":null,"first_page":"1351","last_page":"1367"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10653","display_name":"Robot Manipulation and Learning","score":0.7671999931335449,"subfield":{"id":"https://openalex.org/subfields/2207","display_name":"Control and Systems Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10653","display_name":"Robot Manipulation and Learning","score":0.7671999931335449,"subfield":{"id":"https://openalex.org/subfields/2207","display_name":"Control and Systems Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/workload","display_name":"Workload","score":0.692300021648407},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.6428999900817871},{"id":"https://openalex.org/keywords/limiting","display_name":"Limiting","score":0.6392999887466431},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.5737000107765198},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.508400022983551},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.47940000891685486},{"id":"https://openalex.org/keywords/routing","display_name":"Routing (electronic design automation)","score":0.3944000005722046}],"concepts":[{"id":"https://openalex.org/C2778476105","wikidata":"https://www.wikidata.org/wiki/Q628539","display_name":"Workload","level":2,"score":0.692300021648407},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6862999796867371},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.6428999900817871},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.6392999887466431},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.5737000107765198},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.508400022983551},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.47940000891685486},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.3944000005722046},{"id":"https://openalex.org/C125209513","wikidata":"https://www.wikidata.org/wiki/Q4037520","display_name":"Doors","level":2,"score":0.33889999985694885},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3061999976634979},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3010999858379364},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.2992999851703644},{"id":"https://openalex.org/C2780345887","wikidata":"https://www.wikidata.org/wiki/Q5843","display_name":"Backpack","level":2,"score":0.2851000130176544},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.26739999651908875},{"id":"https://openalex.org/C24856439","wikidata":"https://www.wikidata.org/wiki/Q352483","display_name":"Adaptive routing","level":5,"score":0.2660999894142151},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.2628999948501587},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.25769999623298645},{"id":"https://openalex.org/C3017489831","wikidata":"https://www.wikidata.org/wiki/Q2393193","display_name":"Running time","level":2,"score":0.250900000333786}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1145/3712285.3759775","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3712285.3759775","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2505.14864","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.14864","pdf_url":"https://arxiv.org/pdf/2505.14864","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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"}],"best_oa_location":{"id":"doi:10.1145/3712285.3759775","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3712285.3759775","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis","raw_type":"proceedings-article"},"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":{"To":[0],"reduce":[1],"the":[2,93],"computational":[3],"and":[4,55],"memory":[5],"overhead":[6],"of":[7,20,32,47,58],"Large":[8],"Language":[9],"Models,":[10],"various":[11],"approaches":[12,68],"have":[13],"been":[14],"proposed.":[15],"These":[16],"include":[17],"a)":[18],"Mixture":[19,57],"Experts":[21],"(MoEs),":[22],"where":[23,61],"token":[24],"routing":[25],"affects":[26],"compute":[27],"balance;":[28],"b)":[29],"gradual":[30],"pruning":[31],"model":[33,53],"parameters;":[34],"c)":[35],"dynamically":[36],"freezing":[37],"layers;":[38,54],"d)":[39],"dynamic":[40],"sparse":[41],"attention":[42],"mechanisms;":[43],"e)":[44],"early":[45],"exit":[46],"tokens":[48,62],"as":[49],"they":[50,75],"pass":[51],"through":[52],"f)":[56],"Depths":[59],"(MoDs),":[60],"bypass":[63],"certain":[64],"blocks.":[65],"While":[66],"these":[67],"are":[69],"effective":[70],"in":[71],"reducing":[72],"overall":[73],"computation,":[74],"often":[76],"introduce":[77],"significant":[78],"workload":[79],"imbalance":[80,87],"across":[81],"workers.":[82],"In":[83],"many":[84],"cases,":[85],"this":[86],"is":[88],"severe":[89],"enough":[90],"to":[91,103,107],"render":[92],"techniques":[94],"impractical":[95],"for":[96],"large-scale":[97],"distributed":[98],"training,":[99],"limiting":[100],"their":[101],"applicability":[102],"toy":[104],"models":[105],"due":[106],"poor":[108],"efficiency.":[109]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2026-03-07T16:01:11.037858","created_date":"2025-10-19T00:00:00"}