{"id":"https://openalex.org/W4396907554","doi":"https://doi.org/10.48550/arxiv.2405.06640","title":"Linearizing Large Language Models","display_name":"Linearizing Large Language Models","publication_year":2024,"publication_date":"2024-05-10","ids":{"openalex":"https://openalex.org/W4396907554","doi":"https://doi.org/10.48550/arxiv.2405.06640"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2405.06640","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2405.06640","pdf_url":"https://arxiv.org/pdf/2405.06640","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2405.06640","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5002911029","display_name":"Jean Mercat","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mercat, Jean","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5042694706","display_name":"Igor Vasiljevic","orcid":"https://orcid.org/0000-0001-6117-0693"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vasiljevic, Igor","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5052690536","display_name":"Sedrick Scott Keh","orcid":"https://orcid.org/0000-0003-0088-7951"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Keh, Sedrick","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007392934","display_name":"Kushal Arora","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Arora, Kushal","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5045047925","display_name":"Achal Dave","orcid":"https://orcid.org/0000-0003-1948-5629"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dave, Achal","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5075018873","display_name":"Adrien Gaidon","orcid":"https://orcid.org/0000-0001-8820-550X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gaidon, Adrien","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5073644080","display_name":"Thomas Kollar","orcid":"https://orcid.org/0000-0003-2598-8118"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kollar, Thomas","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":true,"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/T10028","display_name":"Topic Modeling","score":0.8355000019073486,"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/T10028","display_name":"Topic Modeling","score":0.8355000019073486,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.7912999987602234,"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/computer-science","display_name":"Computer science","score":0.47396013140678406},{"id":"https://openalex.org/keywords/linguistics","display_name":"Linguistics","score":0.33225181698799133},{"id":"https://openalex.org/keywords/philosophy","display_name":"Philosophy","score":0.112225741147995}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.47396013140678406},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.33225181698799133},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.112225741147995}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2405.06640","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2405.06640","pdf_url":"https://arxiv.org/pdf/2405.06640","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2405.06640","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2405.06640","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":"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":"pmh:oai:arXiv.org:2405.06640","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2405.06640","pdf_url":"https://arxiv.org/pdf/2405.06640","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4396907554.pdf","grobid_xml":"https://content.openalex.org/works/W4396907554.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052","https://openalex.org/W2382290278","https://openalex.org/W4395014643"],"abstract_inverted_index":{"Linear":[0],"transformers":[1,112],"have":[2,12,46],"emerged":[3],"as":[4,42],"a":[5,88,105,119],"subquadratic-time":[6],"alternative":[7,90],"to":[8,17,48,91,107,126,152],"softmax":[9],"attention":[10],"and":[11,34,44,56,67,83,132,164,176],"garnered":[13],"significant":[14,65],"interest":[15],"due":[16],"their":[18,27],"fixed-size":[19],"recurrent":[20],"state":[21],"that":[22,147],"lowers":[23],"inference":[24],"cost.":[25,144],"However,":[26],"original":[28],"formulation":[29],"suffers":[30],"from":[31],"poor":[32],"scaling":[33],"underperforms":[35],"compute-matched":[36],"transformers.":[37],"Recent":[38],"linear":[39,93,172],"models":[40,63,177],"such":[41],"RWKV":[43],"Mamba":[45],"attempted":[47],"address":[49],"these":[50],"shortcomings":[51],"by":[52,78],"proposing":[53],"novel":[54],"time-mixing":[55],"gating":[57],"architectures,":[58],"but":[59,158],"pre-training":[60,85,92,130],"large":[61,110],"language":[62],"requires":[64],"data":[66,131],"compute":[68,82,121],"investments.":[69],"Thus,":[70],"the":[71,79,128,142,170],"search":[72],"for":[73,99,168],"subquadratic":[74],"architectures":[75],"is":[76],"limited":[77],"availability":[80],"of":[81,134,141],"quality":[84],"datasets.":[86],"As":[87],"cost-effective":[89],"transformers,":[94],"we":[95,159],"propose":[96],"Scalable":[97],"UPtraining":[98],"Recurrent":[100,114],"Attention":[101],"(SUPRA).":[102],"We":[103,145],"present":[104],"method":[106],"uptrain":[108],"existing":[109,135],"pre-trained":[111],"into":[113],"Neural":[115],"Networks":[116],"(RNNs)":[117],"with":[118],"modest":[120],"budget.":[122],"This":[123],"allows":[124],"us":[125],"leverage":[127],"strong":[129],"performance":[133,154],"transformer":[136],"LLMs,":[137],"while":[138],"requiring":[139],"5%":[140],"training":[143],"find":[146],"our":[148],"linearization":[149],"technique":[150],"leads":[151],"competitive":[153],"on":[155],"standard":[156],"benchmarks,":[157],"identify":[160],"persistent":[161],"in-context":[162],"learning":[163],"long-context":[165],"modeling":[166],"shortfalls":[167],"even":[169],"largest":[171],"models.":[173],"Our":[174],"code":[175],"can":[178],"be":[179],"found":[180],"at":[181],"https://github.com/TRI-ML/linear_open_lm.":[182]},"counts_by_year":[],"updated_date":"2026-08-21T09:56:20.448147","created_date":"2025-10-10T00:00:00"}