{"id":"https://openalex.org/W4414977149","doi":"https://doi.org/10.48550/arxiv.2510.05554","title":"Critical attention scaling in long-context transformers","display_name":"Critical attention scaling in long-context transformers","publication_year":2025,"publication_date":"2025-10-07","ids":{"openalex":"https://openalex.org/W4414977149","doi":"https://doi.org/10.48550/arxiv.2510.05554"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2510.05554","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2510.05554","pdf_url":"https://arxiv.org/pdf/2510.05554","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2510.05554","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5068256923","display_name":"Shi Chen","orcid":"https://orcid.org/0000-0001-5247-8119","observed_orcids":["https://orcid.org/0000-0001-5247-8119"]},"institutions":[{"id":"https://openalex.org/I63966007","display_name":"Massachusetts Institute of Technology","ror":"https://ror.org/042nb2s44","country_code":"US","type":"education","lineage":["https://openalex.org/I63966007"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chen, Shi","raw_affiliation_strings":["( ZL ) Department of Mathematics , Massachusetts Institute of Technology , 77 Massachusetts Ave , 02139 Cambridge MA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"( ZL ) Department of Mathematics , Massachusetts Institute of Technology , 77 Massachusetts Ave , 02139 Cambridge MA , USA","institution_ids":["https://openalex.org/I63966007"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Lin, Zhengjiang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I63966007","display_name":"Massachusetts Institute of Technology","ror":"https://ror.org/042nb2s44","country_code":"US","type":"education","lineage":["https://openalex.org/I63966007"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Lin, Zhengjiang","raw_affiliation_strings":["( ZL ) Department of Mathematics , Massachusetts Institute of Technology , 77 Massachusetts Ave , 02139 Cambridge MA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"( ZL ) Department of Mathematics , Massachusetts Institute of Technology , 77 Massachusetts Ave , 02139 Cambridge MA , USA","institution_ids":["https://openalex.org/I63966007"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5031031216","display_name":"Yury Polyanskiy","orcid":"https://orcid.org/0000-0002-2109-0979","observed_orcids":["https://orcid.org/0000-0002-2109-0979"]},"institutions":[{"id":"https://openalex.org/I63966007","display_name":"Massachusetts Institute of Technology","ror":"https://ror.org/042nb2s44","country_code":"US","type":"education","lineage":["https://openalex.org/I63966007"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Polyanskiy, Yury","raw_affiliation_strings":["( SC ) Department of Mathematics , Massachusetts Institute of Technology , 77 Massachusetts Ave , 02139 Cambridge MA , USA","( YP ) Department of Electrical Engineering and Computer Science , Massachusetts Institute of Technology , 77 Massachusetts Ave , 02139 Cambridge MA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"( SC ) Department of Mathematics , Massachusetts Institute of Technology , 77 Massachusetts Ave , 02139 Cambridge MA , USA","institution_ids":["https://openalex.org/I63966007"]},{"raw_affiliation_string":"( YP ) Department of Electrical Engineering and Computer Science , Massachusetts Institute of Technology , 77 Massachusetts Ave , 02139 Cambridge MA , USA","institution_ids":["https://openalex.org/I63966007"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5001421646","display_name":"Philippe Rigollet","orcid":"https://orcid.org/0000-0002-1554-8912","observed_orcids":["https://orcid.org/0000-0002-1554-8912","https://orcid.org/0000-0002-0135-7162"]},"institutions":[{"id":"https://openalex.org/I63966007","display_name":"Massachusetts Institute of Technology","ror":"https://ror.org/042nb2s44","country_code":"US","type":"education","lineage":["https://openalex.org/I63966007"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Rigollet, Philippe","raw_affiliation_strings":["( YP ) Department of Electrical Engineering and Computer Science , Massachusetts Institute of Technology , 77 Massachusetts Ave , 02139 Cambridge MA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"( YP ) Department of Electrical Engineering and Computer Science , Massachusetts Institute of Technology , 77 Massachusetts Ave , 02139 Cambridge MA , USA","institution_ids":["https://openalex.org/I63966007"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I63966007"],"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":0.9463000297546387,"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":0.9463000297546387,"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":0.027000000700354576,"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/T10320","display_name":"Neural Networks and Applications","score":0.002099999925121665,"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/rank-collapse","display_name":"rank collapse","score":0.8849999904632568},{"id":"https://openalex.org/keywords/phase-transitions","display_name":"phase transitions","score":0.734000027179718},{"id":"https://openalex.org/keywords/critical-scaling","display_name":"critical scaling","score":0.7129999995231628},{"id":"https://openalex.org/keywords/yarn","display_name":"yarn","score":0.6380000114440918},{"id":"https://openalex.org/keywords/qwen","display_name":"Qwen","score":0.9739999771118164},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.7689999938011169},{"id":"https://openalex.org/keywords/token-clustering","display_name":"token clustering","score":0.45100000500679016}],"concepts":[{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.9185000061988831},{"id":"https://openalex.org/C121864883","wikidata":"https://www.wikidata.org/wiki/Q677916","display_name":"Statistical physics","level":1,"score":0.5834000110626221},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5741000175476074},{"id":"https://openalex.org/C39927690","wikidata":"https://www.wikidata.org/wiki/Q11197","display_name":"Logarithm","level":2,"score":0.5730000138282776},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5665000081062317},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.4453999996185303},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.3671000003814697},{"id":"https://openalex.org/C194232998","wikidata":"https://www.wikidata.org/wiki/Q1606712","display_name":"Transition (genetics)","level":3,"score":0.3353999853134155},{"id":"https://openalex.org/C2988430800","wikidata":"https://www.wikidata.org/wiki/Q428971","display_name":"Scaling law","level":3,"score":0.32850000262260437},{"id":"https://openalex.org/C144386022","wikidata":"https://www.wikidata.org/wiki/Q1332997","display_name":"Scale factor (cosmology)","level":5,"score":0.3237999975681305},{"id":"https://openalex.org/C50335755","wikidata":"https://www.wikidata.org/wiki/Q483247","display_name":"Phenomenon","level":2,"score":0.3224000036716461},{"id":"https://openalex.org/C194807282","wikidata":"https://www.wikidata.org/wiki/Q900634","display_name":"Critical phenomena","level":3,"score":0.3165999948978424},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.31029999256134033},{"id":"https://openalex.org/C149288129","wikidata":"https://www.wikidata.org/wiki/Q185357","display_name":"Phase transition","level":2,"score":0.29649999737739563},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2736999988555908},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2689000070095062},{"id":"https://openalex.org/C97831314","wikidata":"https://www.wikidata.org/wiki/Q6667307","display_name":"Logarithmic growth","level":2,"score":0.26649999618530273},{"id":"https://openalex.org/C184720557","wikidata":"https://www.wikidata.org/wiki/Q7825049","display_name":"Topology (electrical circuits)","level":2,"score":0.25459998846054077}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2510.05554","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2510.05554","pdf_url":"https://arxiv.org/pdf/2510.05554","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2510.05554","doi":"https://doi.org/10.48550/arxiv.2510.05554","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2510.05554","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,"listed_in":[],"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":"pmh:oai:arXiv.org:2510.05554","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2510.05554","pdf_url":"https://arxiv.org/pdf/2510.05554","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4414977149.pdf","grobid_xml":"https://content.openalex.org/works/W4414977149.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"As":[0],"large":[1,140],"language":[2],"models":[3],"scale":[4],"to":[5,27,91,100],"longer":[6],"contexts,":[7],"attention":[8,15,44,70,75,99,125,138],"layers":[9],"suffer":[10],"from":[11],"a":[12,30,47,60,77,92,121],"fundamental":[13],"pathology:":[14],"scores":[16,45],"collapse":[17],"toward":[18],"uniformity":[19],"as":[20,33],"context":[21,141],"length":[22],"$n$":[23],"increases,":[24],"causing":[25],"tokens":[26,90],"cluster":[28],"excessively,":[29],"phenomenon":[31],"known":[32],"rank-collapse.":[34],"While":[35],"$\\textit{attention":[36],"scaling}$":[37],"effectively":[38],"addresses":[39],"this":[40,54,73],"deficiency":[41],"by":[42,81],"rescaling":[43],"with":[46],"polylogarithmic":[48],"factor":[49,84],"$\u03b2_n$,":[50],"theoretical":[51],"justification":[52,123],"for":[53,124],"approach":[55],"remains":[56],"lacking.":[57],"We":[58],"analyze":[59],"simplified":[61],"yet":[62],"tractable":[63],"model":[64],"that":[65],"magnifies":[66],"the":[67,82,112],"effect":[68],"of":[69],"scaling.":[71],"In":[72],"model,":[74],"exhibits":[76],"phase":[78],"transition":[79],"governed":[80],"scaling":[83,87,97,114,126,134],"$\u03b2_n$:":[85],"insufficient":[86],"collapses":[88],"all":[89],"single":[93],"direction,":[94],"while":[95],"excessive":[96],"reduces":[98],"identity,":[101],"thereby":[102],"eliminating":[103],"meaningful":[104],"interactions":[105],"between":[106],"tokens.":[107],"Our":[108],"main":[109],"result":[110],"identifies":[111],"critical":[113],"$\u03b2_n":[115],"\\asymp":[116],"\\log":[117],"n$":[118],"and":[119,129],"provides":[120],"rigorous":[122],"in":[127],"YaRN":[128],"Qwen,":[130],"clarifying":[131],"why":[132],"logarithmic":[133],"maintains":[135],"sparse,":[136],"content-adaptive":[137],"at":[139],"lengths.":[142]},"counts_by_year":[],"updated_date":"2026-09-30T09:27:27.532063","created_date":"2025-10-10T00:00:00"}