{"id":"https://openalex.org/W7163148571","doi":"https://doi.org/10.48550/arxiv.2606.00888","title":"Memory-Efficient LLM Training with Dynamic Sparsity: From Stability to Practical Scaling","display_name":"Memory-Efficient LLM Training with Dynamic Sparsity: From Stability to Practical Scaling","publication_year":2026,"publication_date":"2026-05-30","ids":{"openalex":"https://openalex.org/W7163148571","doi":"https://doi.org/10.48550/arxiv.2606.00888"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.00888","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00888","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":"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.2606.00888","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101936961","display_name":"Qiao Xiao","orcid":"https://orcid.org/0000-0003-1240-2659"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiao, Qiao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111921306","display_name":"Boqian Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Boqian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137692157","display_name":"Patrik Okanovic","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Okanovic, Patrik","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137634505","display_name":"Tomasz Sternal","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sternal, Tomasz","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137697629","display_name":"Maurice van Keulen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"van Keulen, Maurice","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5027401676","display_name":"Elena Mocanu","orcid":"https://orcid.org/0000-0002-0856-579X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mocanu, Elena","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137689348","display_name":"Mykola Pechenizkiy","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pechenizkiy, Mykola","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5011045254","display_name":"Decebal Constantin Mocanu","orcid":"https://orcid.org/0000-0002-5636-7683"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mocanu, Decebal Constantin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137699172","display_name":"Torsten Hoefler","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hoefler, Torsten","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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.15889999270439148,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.15889999270439148,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.11779999732971191,"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"}},{"id":"https://openalex.org/T11948","display_name":"Machine Learning in Materials Science","score":0.10809999704360962,"subfield":{"id":"https://openalex.org/subfields/2505","display_name":"Materials Chemistry"},"field":{"id":"https://openalex.org/fields/25","display_name":"Materials Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6615999937057495},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.6500999927520752},{"id":"https://openalex.org/keywords/stability","display_name":"Stability (learning theory)","score":0.5861999988555908},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.4214000105857849},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.41780000925064087},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.39579999446868896},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.34689998626708984}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7484999895095825},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6615999937057495},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.6500999927520752},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.5861999988555908},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4855000078678131},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.4214000105857849},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.41780000925064087},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.41339999437332153},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.39579999446868896},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.34689998626708984},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.32030001282691956},{"id":"https://openalex.org/C56372850","wikidata":"https://www.wikidata.org/wiki/Q1050404","display_name":"Sparse matrix","level":3,"score":0.3165000081062317},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.30959999561309814},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.2904999852180481},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.28060001134872437},{"id":"https://openalex.org/C124066611","wikidata":"https://www.wikidata.org/wiki/Q28684319","display_name":"Sparse approximation","level":2,"score":0.2624000012874603},{"id":"https://openalex.org/C77637269","wikidata":"https://www.wikidata.org/wiki/Q7002051","display_name":"Neural coding","level":2,"score":0.26179999113082886}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.00888","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00888","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":"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.2606.00888","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00888","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":"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":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Dynamic":[0],"Sparse":[1,78],"Training":[2,80],"(DST)":[3],"offers":[4],"a":[5,56,113,147],"promising":[6],"paradigm":[7],"for":[8,59,108,143],"improving":[9],"the":[10,47,117,141],"training":[11,70,90,145],"and":[12,68,88,104,134],"inference":[13],"efficiency":[14],"of":[15,50,116,138],"deep":[16],"neural":[17],"networks;":[18],"however,":[19],"we":[20,44,76],"find":[21],"that":[22,46,129],"in":[23,64],"large":[24,66],"language":[25],"model":[26],"training,":[27],"DST":[28,84],"can":[29],"suffer":[30],"from":[31],"optimization":[32,124],"instability,":[33],"manifested":[34],"as":[35,146],"loss":[36],"spikes":[37],"after":[38],"topology":[39],"updates.":[40],"In":[41],"this":[42,74],"work,":[43],"show":[45],"naive":[48],"use":[49],"standard":[51],"Adam-based":[52],"optimizers":[53],"leads":[54],"to":[55,150],"cold-start":[57],"issue":[58],"newly":[60],"regrown":[61],"parameters,":[62],"resulting":[63],"excessively":[65],"updates":[67],"disrupted":[69],"dynamics.":[71],"To":[72],"address":[73],"issue,":[75],"propose":[77],"Memory-Efficient":[79],"(SMET),":[81],"which":[82],"stabilizes":[83],"with":[85],"optimizer":[86,105],"warm-up":[87],"improves":[89],"progress":[91],"through":[92],"density-aware":[93],"learning-rate":[94],"scaling.":[95],"SMET":[96,130],"further":[97],"reduces":[98],"memory":[99],"consumption":[100],"by":[101],"storing":[102],"gradients":[103],"states":[106],"only":[107],"active":[109],"parameters.":[110],"We":[111],"provide":[112],"theoretical":[114],"analysis":[115],"update":[118],"behaviors":[119],"under":[120],"SMET,":[121],"showing":[122],"improved":[123],"stability.":[125],"Extensive":[126],"experiments":[127],"demonstrate":[128],"enables":[131],"stable,":[132],"scalable,":[133],"memory-efficient":[135],"sparse":[136,144],"pre-training":[137],"LLMs,":[139],"paving":[140],"way":[142],"practical":[148],"alternative":[149],"dense":[151],"training.":[152],"Our":[153],"code":[154],"is":[155],"publicly":[156],"available":[157],"at:":[158],"https://github.com/QiaoXiao7282/SMET.":[159]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-03T00:00:00"}