{"id":"https://openalex.org/W7160422447","doi":"https://doi.org/10.48550/arxiv.2605.03667","title":"ELAS: Efficient Pre-Training of Low-Rank Large Language Models via 2:4 Activation Sparsity","display_name":"ELAS: Efficient Pre-Training of Low-Rank Large Language Models via 2:4 Activation Sparsity","publication_year":2026,"publication_date":"2026-05-05","ids":{"openalex":"https://openalex.org/W7160422447","doi":"https://doi.org/10.48550/arxiv.2605.03667"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.03667","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.03667","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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.2605.03667","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135455382","display_name":"Jiaxi Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Jiaxi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135508514","display_name":"Lu Yin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yin, Lu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135503891","display_name":"Li Shen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shen, Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135422802","display_name":"Jinjin Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Jinjin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100352022","display_name":"Yuhui Liu","orcid":"https://orcid.org/0000-0001-8873-5834"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Yuhui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135503044","display_name":"Wenwu Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Wenwu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135482423","display_name":"Shiwei Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Shiwei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135482989","display_name":"Xilu Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Xilu","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/T10036","display_name":"Advanced Neural Network Applications","score":0.337799996137619,"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"}},"topics":[{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.337799996137619,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.21580000221729279,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.04349999874830246,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.6402000188827515},{"id":"https://openalex.org/keywords/leverage","display_name":"Leverage (statistics)","score":0.526199996471405},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.48350000381469727},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.39579999446868896},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.3634999990463257},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.3463999927043915},{"id":"https://openalex.org/keywords/source-code","display_name":"Source code","score":0.3433000147342682},{"id":"https://openalex.org/keywords/degradation","display_name":"Degradation (telecommunications)","score":0.30790001153945923}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7379999756813049},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.6402000188827515},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.526199996471405},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.48350000381469727},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.39579999446868896},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.3634999990463257},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3573000133037567},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.3472000062465668},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.3463999927043915},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3450999855995178},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.3433000147342682},{"id":"https://openalex.org/C2779679103","wikidata":"https://www.wikidata.org/wiki/Q5251805","display_name":"Degradation (telecommunications)","level":2,"score":0.30790001153945923},{"id":"https://openalex.org/C2777472644","wikidata":"https://www.wikidata.org/wiki/Q16968992","display_name":"Approximate inference","level":3,"score":0.3028999865055084},{"id":"https://openalex.org/C115051666","wikidata":"https://www.wikidata.org/wiki/Q6522493","display_name":"Ranging","level":2,"score":0.2992999851703644},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.29510000348091125},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.28279998898506165},{"id":"https://openalex.org/C58013763","wikidata":"https://www.wikidata.org/wiki/Q5754574","display_name":"High-level synthesis","level":3,"score":0.2797999978065491},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2612999975681305},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.257999986410141},{"id":"https://openalex.org/C2778348673","wikidata":"https://www.wikidata.org/wiki/Q739302","display_name":"Production (economics)","level":2,"score":0.2565000057220459},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.25209999084472656},{"id":"https://openalex.org/C56372850","wikidata":"https://www.wikidata.org/wiki/Q1050404","display_name":"Sparse matrix","level":3,"score":0.25200000405311584}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.03667","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.03667","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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.2605.03667","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.03667","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"have":[4],"achieved":[5],"remarkable":[6],"capabilities,":[7],"but":[8],"their":[9],"immense":[10],"computational":[11],"demands":[12],"during":[13,81],"training":[14,23,37,183],"remain":[15],"a":[16,61,115],"critical":[17],"bottleneck":[18],"for":[19,54,118],"widespread":[20],"adoption.":[21],"Low-rank":[22,109],"has":[24,59],"received":[25],"attention":[26],"in":[27,72,135],"recent":[28],"years":[29],"due":[30],"to":[31,34,45,49,88,92,131,163],"its":[32],"ability":[33],"significantly":[35],"reduce":[36],"memory":[38,76,191],"usage.":[39],"Meanwhile,":[40],"applying":[41,86,177],"2:4":[42,55,112,122,140,178],"structured":[43,56,141],"sparsity":[44,87,142],"weights":[46,89],"and":[47,78,138,184],"activations":[48,145],"leverage":[50],"NVIDIA":[51],"GPU":[52],"support":[53],"sparse":[57],"format":[58],"become":[60],"promising":[62],"direction.":[63],"However,":[64],"existing":[65],"low-rank":[66,119,136],"methods":[67],"often":[68,90],"leave":[69],"activation":[70,123,129,179,190],"matrices":[71],"full-rank,":[73],"which":[74],"dominates":[75],"consumption":[77],"limits":[79],"throughput":[80],"large-batch":[82],"training.":[83],"Furthermore,":[84],"directly":[85],"leads":[91],"non-negligible":[93],"performance":[94,172],"degradation.":[95],"To":[96],"achieve":[97],"efficient":[98],"pre-training":[99,107,155],"of":[100,108],"LLMs,":[101],"this":[102],"paper":[103],"proposes":[104],"ELAS:":[105],"Efficient":[106],"LLMs":[110],"via":[111,121],"Activation":[113],"Sparsity,":[114],"novel":[116],"framework":[117],"models":[120,137,159],"sparsity.":[124],"ELAS":[125,153,170,188,202],"applies":[126],"squared":[127,148],"ReLU":[128,149],"functions":[130],"the":[132,144,147],"feed-forward":[133],"networks":[134],"implements":[139],"on":[143,157],"after":[146,176],"operation.":[150],"We":[151],"evaluated":[152],"through":[154],"experiments":[156],"LLaMA":[158],"ranging":[160],"from":[161],"60M":[162],"1B":[164],"parameters.":[165],"The":[166],"results":[167],"demonstrate":[168],"that":[169],"maintains":[171],"with":[173,194],"minimal":[174],"degradation":[175],"sparsity,":[180],"while":[181],"achieving":[182],"inference":[185],"acceleration.":[186],"Moreover,":[187],"reduces":[189],"overhead,":[192],"particularly":[193],"large":[195],"batch":[196],"sizes.":[197],"Code":[198],"is":[199],"available":[200],"at":[201],"Repo.":[203]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-07T00:00:00"}