{"id":"https://openalex.org/W7204944290","doi":"https://doi.org/10.48550/arxiv.2608.29058","title":"RouteSparse: Input-Conditional Pattern Routing for Budgeted Long-Context Prefilling","display_name":"RouteSparse: Input-Conditional Pattern Routing for Budgeted Long-Context Prefilling","publication_year":2026,"publication_date":"2026-08-29","ids":{"openalex":"https://openalex.org/W7204944290","doi":"https://doi.org/10.48550/arxiv.2608.29058"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.29058","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.29058","pdf_url":"https://arxiv.org/pdf/2608.29058","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/2608.29058","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5148461177","display_name":"Chao Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Chao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102497859","display_name":"Yifan Ji","orcid":"https://orcid.org/0009-0007-0376-7150","observed_orcids":["https://orcid.org/0009-0007-0376-7150","https://orcid.org/0009-0009-9833-4634"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ji, Yifan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5148842669","display_name":"Ziyan Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Ziyan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5148883960","display_name":"Kai Song","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Song, Kai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5148895011","display_name":"Fei Lin","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Fei","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/T10028","display_name":"Topic Modeling","score":0.5015000104904175,"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":0.5015000104904175,"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":0.32100000977516174,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.08619999885559082,"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/sparse-attention","display_name":"sparse attention","score":0.7480000257492065},{"id":"https://openalex.org/keywords/uncertainty-estimation","display_name":"uncertainty estimation","score":0.6129999756813049},{"id":"https://openalex.org/keywords/inference-latency","display_name":"inference latency","score":0.5419999957084656},{"id":"https://openalex.org/keywords/large-language-model-inference","display_name":"large language model inference","score":0.6110000014305115},{"id":"https://openalex.org/keywords/long-context-retrieval","display_name":"long-context retrieval","score":0.531000018119812},{"id":"https://openalex.org/keywords/question-answering","display_name":"question answering","score":0.48899999260902405},{"id":"https://openalex.org/keywords/summarization","display_name":"summarization","score":0.8209999799728394}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6674000024795532},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.5860000252723694},{"id":"https://openalex.org/C2775896111","wikidata":"https://www.wikidata.org/wiki/Q642560","display_name":"Router","level":2,"score":0.5787000060081482},{"id":"https://openalex.org/C129844170","wikidata":"https://www.wikidata.org/wiki/Q41299","display_name":"Quadratic equation","level":2,"score":0.5627999901771545},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.47920000553131104},{"id":"https://openalex.org/C96865113","wikidata":"https://www.wikidata.org/wiki/Q2946816","display_name":"Certificate","level":2,"score":0.40860000252723694},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.4049000144004822},{"id":"https://openalex.org/C81845259","wikidata":"https://www.wikidata.org/wiki/Q290117","display_name":"Quadratic programming","level":2,"score":0.38449999690055847},{"id":"https://openalex.org/C176809094","wikidata":"https://www.wikidata.org/wiki/Q15401496","display_name":"Traverse","level":2,"score":0.3603000044822693},{"id":"https://openalex.org/C115051666","wikidata":"https://www.wikidata.org/wiki/Q6522493","display_name":"Ranging","level":2,"score":0.3343000113964081},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.3149999976158142},{"id":"https://openalex.org/C158379750","wikidata":"https://www.wikidata.org/wiki/Q214111","display_name":"Network packet","level":2,"score":0.2851000130176544},{"id":"https://openalex.org/C2780312720","wikidata":"https://www.wikidata.org/wiki/Q5689100","display_name":"Head (geology)","level":2,"score":0.27480000257492065},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.26440000534057617},{"id":"https://openalex.org/C2781345722","wikidata":"https://www.wikidata.org/wiki/Q5308388","display_name":"Drop (telecommunication)","level":2,"score":0.2637999951839447}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.29058","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.29058","pdf_url":"https://arxiv.org/pdf/2608.29058","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.2608.29058","doi":"https://doi.org/10.48550/arxiv.2608.29058","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.29058","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":"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2608.29058","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.29058","pdf_url":"https://arxiv.org/pdf/2608.29058","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":[],"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/W7207589806","https://openalex.org/W7213904335","https://openalex.org/W4414586697","https://openalex.org/W7212049978","https://openalex.org/W7134065796","https://openalex.org/W7165640989","https://openalex.org/W4417530105","https://openalex.org/W7154151175","https://openalex.org/W4415797534","https://openalex.org/W7134277865"],"abstract_inverted_index":{"Dynamic":[0],"sparse":[1,27,67],"attention":[2,18],"can":[3],"reduce":[4],"the":[5,113,171],"quadratic":[6],"cost":[7],"of":[8,65],"long-context":[9,116],"prefilling":[10],"without":[11],"changing":[12],"model":[13],"weights.":[14],"MInference":[15],"assigns":[16],"each":[17,56,168],"head":[19,57],"one":[20],"pattern":[21,43,73,83],"offline":[22],"and":[23,44,58,75,84,111,121,149,164],"estimates":[24,72],"that":[25,39,159],"pattern's":[26],"indices":[28],"for":[29,153],"every":[30],"prompt.":[31],"This":[32],"design":[33],"is":[34],"efficient,":[35],"but":[36],"it":[37],"assumes":[38],"a":[40,62,77,82,92,137,150],"head's":[41],"preferred":[42],"sparsity":[45],"budget":[46],"remain":[47],"suitable":[48],"across":[49],"inputs.":[50],"We":[51,95],"introduce":[52],"RouteSparse,":[53],"which":[54],"routes":[55],"prompt":[59],"segment":[60],"among":[61],"small":[63],"library":[64],"GPU-efficient":[66],"patterns.":[68],"A":[69],"low-cost":[70],"probe":[71],"utility":[74],"uncertainty;":[76],"latency-aware":[78],"router":[79],"then":[80],"selects":[81],"budget,":[85],"while":[86],"uncertain":[87],"cases":[88],"fall":[89],"back":[90],"to":[91,142,170],"denser":[93],"mask.":[94],"formulate":[96],"routing":[97],"as":[98],"constrained":[99],"risk":[100],"minimization,":[101],"derive":[102],"an":[103],"attention-output":[104],"error":[105],"certificate":[106],"from":[107],"omitted":[108],"probability":[109],"mass,":[110],"evaluate":[112],"method":[114],"on":[115],"retrieval,":[117],"question":[118],"answering,":[119],"summarization,":[120],"language":[122],"modeling.":[123],"On":[124],"Llama":[125],"3.1-8B-Instruct":[126],"with":[127,136,146],"128K-token":[128],"prompts,":[129],"RouteSparse":[130],"achieves":[131],"$6.5\\times$":[132],"dense":[133,143,166],"prefill":[134],"speed":[135,148],"0.2-point":[138],"RULER":[139],"drop":[140,152],"relative":[141],"attention,":[144],"compared":[145],"$7.3\\times$":[147],"1.6-point":[151],"fixed":[154],"per-head":[155],"routing.":[156],"Ablations":[157],"confirm":[158],"input-conditional":[160],"routing,":[161],"hardware":[162],"profiling,":[163],"selective":[165],"fallback":[167],"contribute":[169],"quality--latency":[172],"tradeoff.":[173]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-09-02T00:00:00"}