{"id":"https://openalex.org/W7118570645","doi":"https://doi.org/10.48550/arxiv.2601.02819","title":"Punctuation-aware Hybrid Trainable Sparse Attention for Large Language Models","display_name":"Punctuation-aware Hybrid Trainable Sparse Attention for Large Language Models","publication_year":2026,"publication_date":"2026-01-06","ids":{"openalex":"https://openalex.org/W7118570645","doi":"https://doi.org/10.48550/arxiv.2601.02819"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2601.02819","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.02819","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2601.02819","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102990843","display_name":"Junxiang Qiu","orcid":"https://orcid.org/0000-0003-1257-1268"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Qiu, Junxiang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122214929","display_name":"Shuo Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Shuo","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122211259","display_name":"Zhengsu Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Zhengsu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010146234","display_name":"Hengheng Zhang","orcid":"https://orcid.org/0000-0001-5872-5985"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Hengheng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113151484","display_name":"Jinda Lu","orcid":"https://orcid.org/0009-0009-7322-9942"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Jinda","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122075839","display_name":"Changcheng Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Changcheng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5122221122","display_name":"Qi Tian","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tian, Qi","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":7,"corresponding_author_ids":["https://openalex.org/A5102990843"],"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":0.5002999901771545,"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":0.5002999901771545,"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/T13702","display_name":"Machine Learning in Healthcare","score":0.060600001364946365,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.05000000074505806,"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/inference","display_name":"Inference","score":0.6237000226974487},{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.505299985408783},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4975000023841858},{"id":"https://openalex.org/keywords/semantics","display_name":"Semantics (computer science)","score":0.4278999865055084},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.39730000495910645},{"id":"https://openalex.org/keywords/block","display_name":"Block (permutation group theory)","score":0.39430001378059387},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.3725000023841858},{"id":"https://openalex.org/keywords/intersection","display_name":"Intersection (aeronautics)","score":0.3560999929904938}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8154000043869019},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6237000226974487},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5764999985694885},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.505299985408783},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4975000023841858},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.4278999865055084},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.39730000495910645},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.39430001378059387},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3833000063896179},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.3725000023841858},{"id":"https://openalex.org/C64543145","wikidata":"https://www.wikidata.org/wiki/Q162942","display_name":"Intersection (aeronautics)","level":2,"score":0.3560999929904938},{"id":"https://openalex.org/C62354387","wikidata":"https://www.wikidata.org/wiki/Q875399","display_name":"Boundary (topology)","level":2,"score":0.35359999537467957},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.32710000872612},{"id":"https://openalex.org/C89611455","wikidata":"https://www.wikidata.org/wiki/Q6804646","display_name":"Mechanism (biology)","level":2,"score":0.3253999948501587},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.3188000023365021},{"id":"https://openalex.org/C56372850","wikidata":"https://www.wikidata.org/wiki/Q1050404","display_name":"Sparse matrix","level":3,"score":0.3077999949455261},{"id":"https://openalex.org/C2164484","wikidata":"https://www.wikidata.org/wiki/Q5170150","display_name":"Core (optical fiber)","level":2,"score":0.30070000886917114},{"id":"https://openalex.org/C129844170","wikidata":"https://www.wikidata.org/wiki/Q41299","display_name":"Quadratic equation","level":2,"score":0.2971999943256378},{"id":"https://openalex.org/C19768560","wikidata":"https://www.wikidata.org/wiki/Q320727","display_name":"Dependency (UML)","level":2,"score":0.28049999475479126},{"id":"https://openalex.org/C124066611","wikidata":"https://www.wikidata.org/wiki/Q28684319","display_name":"Sparse approximation","level":2,"score":0.27390000224113464},{"id":"https://openalex.org/C86034646","wikidata":"https://www.wikidata.org/wiki/Q474311","display_name":"Semantic gap","level":4,"score":0.26669999957084656},{"id":"https://openalex.org/C198942812","wikidata":"https://www.wikidata.org/wiki/Q496618","display_name":"Semantic property","level":2,"score":0.25380000472068787},{"id":"https://openalex.org/C90312973","wikidata":"https://www.wikidata.org/wiki/Q7449052","display_name":"Semantic data model","level":2,"score":0.25040000677108765}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2601.02819","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.02819","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2601.02819","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.02819","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education","score":0.4960940480232239}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Attention":[0],"serves":[1],"as":[2,35,86],"the":[3,60,108,163,173],"fundamental":[4],"mechanism":[5,97],"for":[6,20,162],"long-context":[7,144],"modeling":[8],"in":[9],"large":[10],"language":[11],"models":[12],"(LLMs),":[13],"yet":[14],"dense":[15,151],"attention":[16,30,34,42,80,152,156],"becomes":[17],"structurally":[18],"prohibitive":[19],"long":[21],"sequences":[22],"due":[23],"to":[24,59],"its":[25],"quadratic":[26],"complexity.":[27],"Consequently,":[28],"sparse":[29,41,79,155],"has":[31],"received":[32],"increasing":[33],"a":[36,76,94,179],"scalable":[37],"alternative.":[38],"However,":[39],"existing":[40],"methods":[43],"rely":[44],"on":[45,140],"coarse-grained":[46],"semantic":[47,55,87,101,110],"representations":[48,102],"during":[49],"block":[50],"selection,":[51],"which":[52],"blur":[53],"intra-block":[54],"boundaries":[56],"and":[57,125,143,153],"lead":[58],"loss":[61,175],"of":[62,182],"critical":[63],"information.":[64],"To":[65],"address":[66],"this":[67],"issue,":[68],"we":[69,92,120],"propose":[70],"\\textbf{P}unctuation-aware":[71],"\\textbf{H}ybrid":[72],"\\textbf{S}parse":[73],"\\textbf{A}ttention":[74],"\\textbf{(PHSA)},":[75],"natively":[77],"trainable":[78],"framework":[81],"that":[82,98,128,147],"leverages":[83],"punctuation":[84],"tokens":[85],"boundary":[88,105],"anchors.":[89],"Specifically,":[90,161],"(1)":[91],"design":[93],"dual-branch":[95],"aggregation":[96],"fuses":[99],"global":[100],"with":[103,166],"punctuation-enhanced":[104],"features,":[106],"preserving":[107],"core":[109],"structure":[111],"while":[112],"introducing":[113],"almost":[114],"no":[115],"additional":[116],"computational":[117],"overhead;":[118],"(2)":[119],"introduce":[121],"an":[122],"extreme-sparsity-adaptive":[123],"training":[124],"inference":[126],"strategy":[127],"stabilizes":[129],"model":[130,165],"behavior":[131],"under":[132],"very":[133],"low":[134],"token":[135],"activation":[136],"ratios;":[137],"Extensive":[138],"experiments":[139],"general":[141],"benchmarks":[142],"evaluations":[145],"demonstrate":[146],"PHSA":[148,170],"consistently":[149],"outperforms":[150],"state-of-the-art":[154],"baselines,":[157],"including":[158],"InfLLM":[159],"v2.":[160],"0.6B-parameter":[164],"32k-token":[167],"input":[168],"sequences,":[169],"can":[171],"reduce":[172],"information":[174],"by":[176],"10.8\\%":[177],"at":[178],"sparsity":[180],"ratio":[181],"97.3\\%.":[183]},"counts_by_year":[],"updated_date":"2026-01-08T20:10:11.968330","created_date":"2026-01-08T00:00:00"}