{"id":"https://openalex.org/W7168819824","doi":"https://doi.org/10.48448/09sw-ar83","title":"Lil: Less is Less When Applying Post-Training Sparse-Attention Algorithms in Long-Decode Stage","display_name":"Lil: Less is Less When Applying Post-Training Sparse-Attention Algorithms in Long-Decode Stage","publication_year":2026,"publication_date":"2026-06-03","ids":{"openalex":"https://openalex.org/W7168819824","doi":"https://doi.org/10.48448/09sw-ar83"},"language":"en","primary_location":{"id":"doi:10.48448/09sw-ar83","doi":"https://doi.org/10.48448/09sw-ar83","is_oa":true,"landing_page_url":"https://doi.org/10.48448/09sw-ar83","pdf_url":null,"source":{"id":"https://openalex.org/S7407053148","display_name":"Underline Science Inc.","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Audiovisual"},"volume":null,"type":"other","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48448/09sw-ar83","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5140982166","display_name":"Association for Computational Linguistics 2026","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Association for Computational Linguistics 2026","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140989763","display_name":"tiancheng hu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"hu, tiancheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133882425","display_name":"J B Hu","orcid":"https://orcid.org/0000-0001-5602-7876","observed_orcids":["https://orcid.org/0000-0001-5602-7876","https://orcid.org/0000-0002-4095-0647","https://orcid.org/0000-0003-2194-3086","https://orcid.org/0009-0000-2415-6729","https://orcid.org/0009-0000-7178-5050","https://orcid.org/0009-0001-0703-5332","https://orcid.org/0009-0002-3762-0898","https://orcid.org/0009-0003-2510-9985","https://orcid.org/0009-0005-2678-5373","https://orcid.org/0009-0005-4445-2468","https://orcid.org/0009-0006-0714-3786","https://orcid.org/0009-0006-9563-8956","https://orcid.org/0009-0007-8986-3636","https://orcid.org/0009-0007-9371-6188"]},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Hu, Junhao","raw_affiliation_strings":["Peking University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113115887","display_name":"Ziyue Hua","orcid":"https://orcid.org/0009-0007-8348-5232","observed_orcids":["https://orcid.org/0009-0007-8348-5232"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hua, Ziyue","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122250790","display_name":"Wenrui Huang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Wenrui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122294105","display_name":"Chenxu Liu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Liu, Chenxu","raw_affiliation_strings":["Peking University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114114527","display_name":"Anmin Liu","orcid":"https://orcid.org/0000-0003-1138-3660","observed_orcids":["https://orcid.org/0000-0003-1138-3660"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Anmin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088242496","display_name":"Feifan Meng","orcid":"https://orcid.org/0000-0002-5904-8805","observed_orcids":["https://orcid.org/0000-0002-5904-8805"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Meng, Feifan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5141020745","display_name":"Tingpeng Peng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Peng, Tingpeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140999332","display_name":"Mingtaoxu Taoxu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Taoxu, Mingtaoxu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128760869","display_name":"Tao Xie","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xie, Tao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5141021750","display_name":"Shiju Zhao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Shiju","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5140982109","display_name":"\u65b9\u6cfd \u674e","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"\u674e, \u65b9\u6cfd","raw_affiliation_strings":["Nanjing University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nanjing University","institution_ids":["https://openalex.org/I881766915"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"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/T10028","display_name":"Topic Modeling","score":0.48579999804496765,"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":0.48579999804496765,"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":0.45190000534057617,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.028999999165534973,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/sparse-attention","display_name":"sparse attention","score":0.9039999842643738},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.7919999957084656},{"id":"https://openalex.org/keywords/inference-efficiency","display_name":"inference efficiency","score":0.8560000061988831},{"id":"https://openalex.org/keywords/early-stopping","display_name":"early stopping","score":0.7789999842643738},{"id":"https://openalex.org/keywords/information-loss","display_name":"information loss","score":0.8939999938011169},{"id":"https://openalex.org/keywords/token-consumption","display_name":"token consumption","score":0.8220000267028809},{"id":"https://openalex.org/keywords/inference-latency","display_name":"inference latency","score":0.4129999876022339},{"id":"https://openalex.org/keywords/space-complexity","display_name":"space complexity","score":0.3529999852180481}],"concepts":[{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7125999927520752},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7031999826431274},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.6197999715805054},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.4178999960422516},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.41510000824928284},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.39410001039505005},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.3573000133037567},{"id":"https://openalex.org/C2777472644","wikidata":"https://www.wikidata.org/wiki/Q16968992","display_name":"Approximate inference","level":3,"score":0.3488999903202057},{"id":"https://openalex.org/C2988416141","wikidata":"https://www.wikidata.org/wiki/Q6031139","display_name":"Information loss","level":2,"score":0.3393999934196472},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3276999890804291},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.31439998745918274},{"id":"https://openalex.org/C51823790","wikidata":"https://www.wikidata.org/wiki/Q504353","display_name":"Greedy algorithm","level":2,"score":0.30790001153945923},{"id":"https://openalex.org/C198352243","wikidata":"https://www.wikidata.org/wiki/Q37105","display_name":"Line (geometry)","level":2,"score":0.2955999970436096},{"id":"https://openalex.org/C2779679103","wikidata":"https://www.wikidata.org/wiki/Q5251805","display_name":"Degradation (telecommunications)","level":2,"score":0.29339998960494995},{"id":"https://openalex.org/C2983203078","wikidata":"https://www.wikidata.org/wiki/Q255166","display_name":"Information gain","level":2,"score":0.29030001163482666},{"id":"https://openalex.org/C106516650","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm design","level":2,"score":0.29010000824928284},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.28529998660087585},{"id":"https://openalex.org/C148220186","wikidata":"https://www.wikidata.org/wiki/Q7111912","display_name":"Outcome (game theory)","level":2,"score":0.2702000141143799},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.27000001072883606},{"id":"https://openalex.org/C148764684","wikidata":"https://www.wikidata.org/wiki/Q621751","display_name":"Approximation algorithm","level":2,"score":0.26759999990463257},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.267300009727478},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.26660001277923584},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2646999955177307},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.257999986410141},{"id":"https://openalex.org/C3020402766","wikidata":"https://www.wikidata.org/wiki/Q104376712","display_name":"Prior information","level":2,"score":0.2526000142097473}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48448/09sw-ar83","doi":"https://doi.org/10.48448/09sw-ar83","is_oa":true,"landing_page_url":"https://doi.org/10.48448/09sw-ar83","pdf_url":null,"source":{"id":"https://openalex.org/S7407053148","display_name":"Underline Science Inc.","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Audiovisual"}],"best_oa_location":{"id":"doi:10.48448/09sw-ar83","doi":"https://doi.org/10.48448/09sw-ar83","is_oa":true,"landing_page_url":"https://doi.org/10.48448/09sw-ar83","pdf_url":null,"source":{"id":"https://openalex.org/S7407053148","display_name":"Underline Science Inc.","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Audiovisual"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7166840322","https://openalex.org/W7161565130","https://openalex.org/W7168381275","https://openalex.org/W4402502818","https://openalex.org/W7221193513","https://openalex.org/W4403324051","https://openalex.org/W4400025310","https://openalex.org/W7128087457","https://openalex.org/W7161683102","https://openalex.org/W4415330026"],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"(LLMs)":[3],"demonstrate":[4],"strong":[5],"capabilities":[6],"across":[7,134],"a":[8,53,84,126],"wide":[9],"range":[10],"of":[11,55,130],"complex":[12],"tasks":[13],"and":[14,33,46,67],"are":[15],"increasingly":[16],"deployed":[17],"at":[18],"scale,":[19],"placing":[20],"significant":[21],"demands":[22],"on":[23],"inference":[24,30],"efficiency.":[25],"Prior":[26],"work":[27,56],"typically":[28],"decomposes":[29],"into":[31],"prefill":[32],"decode":[34,38,51],"stages,":[35],"with":[36,125],"the":[37,50,94,104],"stage":[39],"dominating":[40],"total":[41],"latency.":[42],"To":[43,92],"reduce":[44],"time":[45],"memory":[47],"complexity":[48],"in":[49],"stage,":[52],"line":[54],"introduces":[57],"sparse-attention":[58],"algorithms.":[59],"In":[60],"this":[61],"paper,":[62],"we":[63,86,97],"show,":[64],"both":[65],"empirically":[66],"theoretically,":[68],"that":[69,102],"sparse":[70,113],"attention":[71],"can":[72],"paradoxically":[73],"increase":[74],"end-to-end":[75],"complexity:":[76],"information":[77,107,110],"loss":[78,108],"often":[79],"induces":[80],"significantly":[81],"longer":[82],"sequences,":[83],"phenomenon":[85],"term":[87],"``Less":[88],"is":[89],"Less''":[90],"(Lil).":[91],"mitigate":[93],"Lil":[95],"problem,":[96],"propose":[98],"an":[99],"early-stopping":[100,116],"algorithm":[101,117],"detects":[103],"threshold":[105],"where":[106],"exceeds":[109],"gain":[111],"during":[112],"decoding.":[114],"Our":[115],"reduces":[118],"token":[119],"consumption":[120],"by":[121],"up":[122],"to":[123],"90%":[124],"marginal":[127],"accuracy":[128],"degradation":[129],"less":[131],"than":[132],"2%":[133],"reasoning-intensive":[135],"benchmarks.":[136]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-07-17T00:00:00"}