{"id":"https://openalex.org/W7161034028","doi":"https://doi.org/10.48550/arxiv.2605.12110","title":"AB-Sparse: Sparse Attention with Adaptive Block Size for Accurate and Efficient Long-Context Inference","display_name":"AB-Sparse: Sparse Attention with Adaptive Block Size for Accurate and Efficient Long-Context Inference","publication_year":2026,"publication_date":"2026-05-12","ids":{"openalex":"https://openalex.org/W7161034028","doi":"https://doi.org/10.48550/arxiv.2605.12110"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.12110","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.12110","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":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.2605.12110","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136033674","display_name":"Di Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Di","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136027166","display_name":"Ruitian Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Ruitian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136007064","display_name":"Chen Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Chen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136030426","display_name":"Mingliang Gong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gong, Mingliang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101907284","display_name":"Yongjie Yuan","orcid":"https://orcid.org/0009-0001-1960-5662"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuan, Yongjie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136021815","display_name":"Han Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Han","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136015676","display_name":"Yu Feng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Feng, Yu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136068004","display_name":"Quan Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Quan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136046102","display_name":"Minyi Guo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Guo, Minyi","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":0.38580000400543213,"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":0.38580000400543213,"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/T14347","display_name":"Big Data and Digital Economy","score":0.19419999420642853,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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":0.04899999871850014,"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/block","display_name":"Block (permutation group theory)","score":0.7556999921798706},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.6018000245094299},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.5503000020980835},{"id":"https://openalex.org/keywords/block-size","display_name":"Block size","score":0.5123999714851379},{"id":"https://openalex.org/keywords/partition","display_name":"Partition (number theory)","score":0.5055999755859375},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.49570000171661377},{"id":"https://openalex.org/keywords/lossless-compression","display_name":"Lossless compression","score":0.42500001192092896}],"concepts":[{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.7556999921798706},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7513999938964844},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.6018000245094299},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.5503000020980835},{"id":"https://openalex.org/C41431624","wikidata":"https://www.wikidata.org/wiki/Q1053357","display_name":"Block size","level":3,"score":0.5123999714851379},{"id":"https://openalex.org/C42812","wikidata":"https://www.wikidata.org/wiki/Q1082910","display_name":"Partition (number theory)","level":2,"score":0.5055999755859375},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.49570000171661377},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4878999888896942},{"id":"https://openalex.org/C81081738","wikidata":"https://www.wikidata.org/wiki/Q55542","display_name":"Lossless compression","level":3,"score":0.42500001192092896},{"id":"https://openalex.org/C146599234","wikidata":"https://www.wikidata.org/wiki/Q511093","display_name":"Centroid","level":2,"score":0.4115999937057495},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.3564000129699707},{"id":"https://openalex.org/C66882249","wikidata":"https://www.wikidata.org/wiki/Q169336","display_name":"Homogeneous","level":2,"score":0.3538999855518341},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.3522000014781952},{"id":"https://openalex.org/C182365436","wikidata":"https://www.wikidata.org/wiki/Q50701","display_name":"Variable (mathematics)","level":2,"score":0.3450999855995178},{"id":"https://openalex.org/C21200559","wikidata":"https://www.wikidata.org/wiki/Q7451068","display_name":"Sensitivity (control systems)","level":2,"score":0.3287000060081482},{"id":"https://openalex.org/C56372850","wikidata":"https://www.wikidata.org/wiki/Q1050404","display_name":"Sparse matrix","level":3,"score":0.32089999318122864},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.29760000109672546},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.27790001034736633},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.2556999921798706}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.12110","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.12110","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":"doi:10.48550/arxiv.2605.12110","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.12110","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":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":{"As":[0],"large":[1],"language":[2],"models":[3],"scale":[4],"to":[5,92,98,125,149,168],"longer":[6],"contexts,":[7],"loading":[8],"the":[9,43,76,131],"growing":[10],"KV":[11,44],"cache":[12,45],"during":[13],"attention":[14,25,39,52,69,86,123,174],"computation":[15,26],"becomes":[16],"a":[17,30,63,104],"critical":[18],"bottleneck.":[19],"Previous":[20],"work":[21],"has":[22],"shown":[23],"that":[24,41,81,109,160],"is":[27,84],"dominated":[28],"by":[29],"small":[31],"subset":[32],"of":[33,166],"tokens.":[34],"This":[35],"motivates":[36],"block":[37,65,93,119,139,155,172],"sparse":[38,173],"methods":[40,61],"partition":[42],"into":[46],"fixed-size":[47],"blocks":[48,55],"and":[49,95],"selectively":[50],"compute":[51],"over":[53,170],"those":[54],"exhibiting":[56],"high":[57],"importance.":[58],"However,":[59],"these":[60],"assign":[62],"uniform":[64],"size":[66,120],"across":[67,122],"all":[68],"heads,":[70],"implicitly":[71],"assuming":[72],"homogeneous":[73],"behavior":[74],"throughout":[75],"model.":[77],"Our":[78],"analysis":[79],"reveals":[80],"this":[82],"assumption":[83],"flawed:":[85],"heads":[87,124],"exhibit":[88],"widely":[89],"varying":[90],"sensitivity":[91],"granularity,":[94],"uniformity":[96],"leads":[97],"suboptimal":[99],"accuracy.":[100,127],"We":[101],"present":[102],"AB-Sparse,":[103],"training-free":[105],"algorithm-system":[106],"co-designed":[107],"framework":[108],"improves":[110],"accuracy":[111,164],"while":[112],"preserving":[113],"throughput.":[114],"AB-Sparse":[115,161],"introduces":[116],"lightweight":[117],"adaptive":[118],"allocation":[121],"improve":[126],"To":[128],"compensate":[129],"for":[130],"additional":[132],"memory":[133],"overhead,":[134],"it":[135],"further":[136],"employs":[137],"lossless":[138],"centroid":[140],"quantization.":[141],"In":[142],"addition,":[143],"custom":[144],"GPU":[145],"kernels":[146],"are":[147],"developed":[148],"support":[150],"efficient":[151],"execution":[152],"with":[153],"variable":[154],"sizes.":[156],"Evaluation":[157],"results":[158],"demonstrate":[159],"achieves":[162],"an":[163],"improvement":[165],"up":[167],"5.43%":[169],"existing":[171],"baselines":[175],"without":[176],"throughput":[177],"overhead.":[178]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-14T00:00:00"}