{"id":"https://openalex.org/W7203908356","doi":"https://doi.org/10.48550/arxiv.2608.19920","title":"Learning how to Forget: Fine-tuning for Long-Context Sparse Attention","display_name":"Learning how to Forget: Fine-tuning for Long-Context Sparse Attention","publication_year":2026,"publication_date":"2026-08-20","ids":{"openalex":"https://openalex.org/W7203908356","doi":"https://doi.org/10.48550/arxiv.2608.19920"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.19920","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.19920","pdf_url":"https://arxiv.org/pdf/2608.19920","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.19920","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5110779848","display_name":"Matthias Seeger","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I1311688040","display_name":"Amazon (United States)","ror":"https://ror.org/04mv4n011","country_code":"US","type":"company","lineage":["https://openalex.org/I1311688040"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Seeger, Matthias","raw_affiliation_strings":["Amazon Web Services"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon Web Services","institution_ids":["https://openalex.org/I1311688040"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5147763169","display_name":"Zeyu Zhang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I887064364","display_name":"University of Amsterdam","ror":"https://ror.org/04dkp9463","country_code":"NL","type":"education","lineage":["https://openalex.org/I887064364"]}],"countries":["NL"],"is_corresponding":false,"raw_author_name":"Zhang, Zeyu","raw_affiliation_strings":["University of Amsterdam"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Amsterdam","institution_ids":["https://openalex.org/I887064364"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5005457523","display_name":"Vihang Patil","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4577782","display_name":"Technische Universit\u00e4t Berlin","ror":"https://ror.org/03v4gjf40","country_code":"DE","type":"education","lineage":["https://openalex.org/I4577782"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Patil, Vihang","raw_affiliation_strings":["Technical University Berlin"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Technical University Berlin","institution_ids":["https://openalex.org/I4577782"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5046381729","display_name":"Konstantinos Benidis","orcid":"https://orcid.org/0000-0002-0541-6676","observed_orcids":["https://orcid.org/0000-0002-0541-6676"]},"institutions":[{"id":"https://openalex.org/I1311688040","display_name":"Amazon (United States)","ror":"https://ror.org/04mv4n011","country_code":"US","type":"company","lineage":["https://openalex.org/I1311688040"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Benidis, Konstantinos","raw_affiliation_strings":["Amazon Web Services"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon Web Services","institution_ids":["https://openalex.org/I1311688040"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5090934117","display_name":"Sebastian Schelter","orcid":"https://orcid.org/0000-0003-4722-5840","observed_orcids":["https://orcid.org/0000-0003-4722-5840"]},"institutions":[{"id":"https://openalex.org/I4577782","display_name":"Technische Universit\u00e4t Berlin","ror":"https://ror.org/03v4gjf40","country_code":"DE","type":"education","lineage":["https://openalex.org/I4577782"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Schelter, Sebastian","raw_affiliation_strings":["Technical University Berlin"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Technical University Berlin","institution_ids":["https://openalex.org/I4577782"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"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":0.7567999958992004,"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":0.7567999958992004,"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":0.12950000166893005,"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":0.06449999660253525,"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":0.9520000219345093},{"id":"https://openalex.org/keywords/long-context-inference","display_name":"long-context inference","score":0.8289999961853027},{"id":"https://openalex.org/keywords/kv-cache-compression","display_name":"KV cache compression","score":0.8709999918937683},{"id":"https://openalex.org/keywords/fine-tuning","display_name":"fine-tuning","score":0.8190000057220459},{"id":"https://openalex.org/keywords/transformer-based-language-models","display_name":"transformer language models","score":0.8410000205039978},{"id":"https://openalex.org/keywords/memory-efficient-inference","display_name":"memory-efficient inference","score":0.4819999933242798}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8165000081062317},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6798999905586243},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.5418999791145325},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.47360000014305115},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4480000138282776},{"id":"https://openalex.org/C74193536","wikidata":"https://www.wikidata.org/wiki/Q574844","display_name":"Kernel (algebra)","level":2,"score":0.43720000982284546},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.43140000104904175},{"id":"https://openalex.org/C56372850","wikidata":"https://www.wikidata.org/wiki/Q1050404","display_name":"Sparse matrix","level":3,"score":0.4043999910354614},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.36070001125335693},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.3328999876976013},{"id":"https://openalex.org/C90673727","wikidata":"https://www.wikidata.org/wiki/Q901718","display_name":"Product (mathematics)","level":2,"score":0.32820001244544983},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.321399986743927},{"id":"https://openalex.org/C128099668","wikidata":"https://www.wikidata.org/wiki/Q573952","display_name":"Lazy evaluation","level":3,"score":0.31630000472068787},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.31520000100135803},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3107999861240387},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.3061999976634979},{"id":"https://openalex.org/C93959086","wikidata":"https://www.wikidata.org/wiki/Q6888345","display_name":"Model selection","level":2,"score":0.29589998722076416},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.2872999906539917},{"id":"https://openalex.org/C2777472644","wikidata":"https://www.wikidata.org/wiki/Q16968992","display_name":"Approximate inference","level":3,"score":0.2809999883174896},{"id":"https://openalex.org/C177774035","wikidata":"https://www.wikidata.org/wiki/Q1246948","display_name":"Granularity","level":2,"score":0.27410000562667847},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.26350000500679016},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.2623000144958496},{"id":"https://openalex.org/C46743427","wikidata":"https://www.wikidata.org/wiki/Q1341685","display_name":"Inference engine","level":3,"score":0.25600001215934753},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.25429999828338623},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.25}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.19920","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.19920","pdf_url":"https://arxiv.org/pdf/2608.19920","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.19920","doi":"https://doi.org/10.48550/arxiv.2608.19920","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.19920","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:2608.19920","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.19920","pdf_url":"https://arxiv.org/pdf/2608.19920","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7203908356.pdf","grobid_xml":"https://content.openalex.org/works/W7203908356.grobid-xml"},"referenced_works_count":14,"referenced_works":["https://openalex.org/W2908510526","https://openalex.org/W2988394319","https://openalex.org/W2990896553","https://openalex.org/W4389326242","https://openalex.org/W4396815229","https://openalex.org/W4406032994","https://openalex.org/W4412944812","https://openalex.org/W4414939355","https://openalex.org/W4416540656","https://openalex.org/W4417006854","https://openalex.org/W7084400772","https://openalex.org/W7152032112","https://openalex.org/W7167591330","https://openalex.org/W7169616292"],"related_works":["https://openalex.org/W7223085844","https://openalex.org/W7165640989","https://openalex.org/W7161683102","https://openalex.org/W7202122961","https://openalex.org/W4417289958","https://openalex.org/W7158571394","https://openalex.org/W7212041156","https://openalex.org/W7162680218","https://openalex.org/W4417530105","https://openalex.org/W4407759163"],"abstract_inverted_index":{"A":[0],"lot":[1],"of":[2,85],"prior":[3],"work":[4],"addressed":[5],"key-value":[6],"(KV)":[7],"cache":[8,43],"selection":[9],"and":[10,61,113,117],"compression":[11],"by":[12],"sparse":[13,36,87],"attention":[14,76,88,100],"to":[15,65],"enable":[16],"long-context":[17,111],"inference":[18,112],"for":[19,32,40,110,120],"transformer":[20],"language":[21],"models":[22,34,72],"without":[23],"excessive":[24],"hardware":[25,49],"budgets.":[26],"We":[27,79],"provide":[28,81],"a":[29,47,52,105],"new":[30,106],"method":[31],"fine-tuning":[33],"with":[35,57,67,74,95],"attention.":[37],"It":[38],"works":[39],"any":[41],"KV":[42],"policy,":[44,69],"runs":[45],"on":[46],"moderate":[48],"budget":[50],"(e.g.,":[51],"single":[53],"Nvidia":[54],"A100":[55],"GPU":[56],"40":[58],"GB":[59],"RAM),":[60],"allows":[62],"the":[63,68],"model":[64],"co-adapt":[66],"often":[70],"outperforming":[71],"trained":[73],"exact":[75],"(sequence":[77],"parallelism).":[78],"also":[80],"an":[82],"efficient":[83],"implementation":[84],"H2O":[86],"(the":[89],"leading":[90],"policy":[91],"in":[92],"our":[93],"experiments)":[94],"dedicated":[96],"scaled":[97],"dot":[98],"product":[99],"kernel":[101],"support.":[102],"KeysAndValues":[103],"(https://github.com/awslabs/keys_values),":[104],"open":[107],"source":[108],"library":[109],"fine-tuning,":[114],"provides":[115],"easy-to-use":[116],"performant":[118],"code":[119],"all":[121],"methods":[122],"discussed":[123],"here.":[124]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-08-22T00:00:00"}