{"id":"https://openalex.org/W7214290239","doi":"https://doi.org/10.48550/arxiv.2609.29812","title":"FlashLoop: Fast and Memory-Efficient Looped Transformers via Lazy Updates","display_name":"FlashLoop: Fast and Memory-Efficient Looped Transformers via Lazy Updates","publication_year":2026,"publication_date":"2026-09-24","ids":{"openalex":"https://openalex.org/W7214290239","doi":"https://doi.org/10.48550/arxiv.2609.29812"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.29812","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.29812","pdf_url":"https://arxiv.org/pdf/2609.29812","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","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.29812","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5152776996","display_name":"Wanqi Yang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210135521","display_name":"Max Planck Institute for Intelligent Systems","ror":"https://ror.org/04fq9j139","country_code":"DE","type":"facility","lineage":["https://openalex.org/I149899117","https://openalex.org/I4210135521"]},{"id":"https://openalex.org/I4401726832","display_name":"T\u00fcbingen AI Center","ror":"https://ror.org/0107nyd78","country_code":"DE","type":"education","lineage":["https://openalex.org/I149899117","https://openalex.org/I4210135521","https://openalex.org/I4401726832","https://openalex.org/I8087733"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Yang, Wanqi","raw_affiliation_strings":["Institute T\u00fcbingen Max Planck Institute for Intelligent Systems T\u00fcbingen AI Center"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute T\u00fcbingen Max Planck Institute for Intelligent Systems T\u00fcbingen AI Center","institution_ids":["https://openalex.org/I4210135521","https://openalex.org/I4401726832"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5152775428","display_name":"Shiwei Liu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210135521","display_name":"Max Planck Institute for Intelligent Systems","ror":"https://ror.org/04fq9j139","country_code":"DE","type":"facility","lineage":["https://openalex.org/I149899117","https://openalex.org/I4210135521"]},{"id":"https://openalex.org/I4401726832","display_name":"T\u00fcbingen AI Center","ror":"https://ror.org/0107nyd78","country_code":"DE","type":"education","lineage":["https://openalex.org/I149899117","https://openalex.org/I4210135521","https://openalex.org/I4401726832","https://openalex.org/I8087733"]},{"id":"https://openalex.org/I4405263145","display_name":"ELLIS Institute T\u00fcbingen","ror":"https://ror.org/036z3g533","country_code":"DE","type":"other","lineage":["https://openalex.org/I4405263145"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Liu, Shiwei","raw_affiliation_strings":["ELLIS Institute T\u00fcbingen Max Planck Institute for Intelligent Systems T\u00fcbingen AI Center"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"ELLIS Institute T\u00fcbingen Max Planck Institute for Intelligent Systems T\u00fcbingen AI Center","institution_ids":["https://openalex.org/I4210135521","https://openalex.org/I4401726832","https://openalex.org/I4405263145"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":6,"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":0.5830000042915344,"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":0.5830000042915344,"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/T10028","display_name":"Topic Modeling","score":0.25220000743865967,"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":0.08449999988079071,"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/parameter-efficient-inference","display_name":"parameter-efficient inference","score":0.574999988079071},{"id":"https://openalex.org/keywords/kv-cache-memory-reduction","display_name":"KV cache memory reduction","score":0.7870000004768372},{"id":"https://openalex.org/keywords/sparse-attention","display_name":"sparse attention","score":0.7279999852180481},{"id":"https://openalex.org/keywords/low-bit-quantization","display_name":"low-bit quantization","score":0.7910000085830688},{"id":"https://openalex.org/keywords/training-free-inference","display_name":"training-free inference","score":0.6959999799728394},{"id":"https://openalex.org/keywords/long-context","display_name":"long context","score":0.4880000054836273}],"concepts":[{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6834999918937683},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6499999761581421},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5813000202178955},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5611000061035156},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.47870001196861267},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.46160000562667847},{"id":"https://openalex.org/C81081738","wikidata":"https://www.wikidata.org/wiki/Q55542","display_name":"Lossless compression","level":3,"score":0.43619999289512634},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.3734000027179718},{"id":"https://openalex.org/C175291020","wikidata":"https://www.wikidata.org/wiki/Q1156822","display_name":"Offset (computer science)","level":2,"score":0.3714999854564667},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.3562000095844269},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.35010001063346863},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.33880001306533813},{"id":"https://openalex.org/C24326235","wikidata":"https://www.wikidata.org/wiki/Q126095","display_name":"Electronic engineering","level":1,"score":0.31709998846054077},{"id":"https://openalex.org/C3826847","wikidata":"https://www.wikidata.org/wiki/Q188768","display_name":"FLOPS","level":2,"score":0.27070000767707825},{"id":"https://openalex.org/C200601418","wikidata":"https://www.wikidata.org/wiki/Q2193887","display_name":"Reliability engineering","level":1,"score":0.2603999972343445}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.29812","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.29812","pdf_url":"https://arxiv.org/pdf/2609.29812","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.2609.29812","doi":"https://doi.org/10.48550/arxiv.2609.29812","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.29812","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:2609.29812","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.29812","pdf_url":"https://arxiv.org/pdf/2609.29812","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":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214290239.pdf","grobid_xml":"https://content.openalex.org/works/W7214290239.grobid-xml"},"referenced_works_count":16,"referenced_works":["https://openalex.org/W2794325560","https://openalex.org/W2866343820","https://openalex.org/W2996908057","https://openalex.org/W3210904070","https://openalex.org/W4379089709","https://openalex.org/W4389600390","https://openalex.org/W4399794499","https://openalex.org/W4401306886","https://openalex.org/W4414939355","https://openalex.org/W6947991961","https://openalex.org/W6948022840","https://openalex.org/W6966825443","https://openalex.org/W7102671386","https://openalex.org/W7160859380","https://openalex.org/W7163566712","https://openalex.org/W7169794564"],"related_works":["https://openalex.org/W7162287688","https://openalex.org/W7220913388","https://openalex.org/W7102517533","https://openalex.org/W7206177337","https://openalex.org/W7215950593","https://openalex.org/W7220365872","https://openalex.org/W7160930732","https://openalex.org/W7214774800","https://openalex.org/W7213266125","https://openalex.org/W7221815948"],"abstract_inverted_index":{"Looped":[0,75,166,193],"Transformers":[1,27,76,167,194],"have":[2],"attracted":[3],"substantial":[4],"attention":[5],"as":[6],"a":[7,110,120,148],"parameter-efficient":[8],"approach":[9],"to":[10,52,138,176,182,195],"increasing":[11],"computational":[12,197],"depth":[13],"through":[14,156],"repeated":[15],"application":[16],"of":[17,43,74,90,113,125,191],"shared":[18],"Transformer":[19,36],"blocks.":[20],"However,":[21],"their":[22],"practical":[23,80],"advantages":[24],"over":[25],"conventional":[26],"remain":[28],"under":[29],"debate:":[30],"each":[31],"additional":[32,92],"loop":[33,56,65],"incurs":[34],"another":[35,41],"pass":[37],"and":[38,49,67,94,122,128,161,180,199],"requires":[39],"caching":[40],"set":[42],"KV":[44,129],"states,":[45],"causing":[46],"inference":[47,81,150],"FLOPs":[48],"KV-cache":[50,184],"memory":[51,185],"grow":[53],"continuously":[54],"with":[55],"depth.":[57],"This":[58],"overhead":[59],"becomes":[60],"particularly":[61],"severe":[62],"at":[63],"large":[64],"counts":[66],"long":[68],"context,":[69],"preventing":[70],"the":[71,91,189],"parameter":[72],"efficiency":[73],"from":[77],"translating":[78],"into":[79],"efficiency.":[82],"In":[83],"this":[84],"paper,":[85],"we":[86,145],"find":[87],"that":[88,152],"much":[89],"computation":[93],"storage":[95],"introduced":[96],"by":[97,119],"looping":[98],"is":[99],"redundant.":[100],"As":[101],"recurrence":[102],"proceeds,":[103],"state":[104],"changes":[105],"become":[106,134],"increasingly":[107],"concentrated":[108],"on":[109,142],"small":[111],"subset":[112,124],"tokens;":[114],"attention-output":[115],"differences":[116],"are":[117],"dominated":[118],"sparse":[121,159],"stable":[123],"key":[126],"columns;":[127],"residuals":[130],"between":[131],"adjacent":[132],"loops":[133],"progressively":[135],"more":[136],"amenable":[137],"low-bit":[139],"quantization.":[140,163],"Building":[141],"these":[143],"observations,":[144],"introduce":[146],"FlashLoop,":[147],"training-free":[149],"framework":[151],"reduces":[153],"cross-loop":[154],"redundancy":[155],"token-sparse":[157],"updates,":[158],"attention,":[160],"KV-residual":[162],"Across":[164],"several":[165],"models,":[168],"FlashLoop":[169],"delivers":[170],"lossless":[171],"accuracy":[172],"while":[173],"achieving":[174],"up":[175,181],"1.64$\\times$":[177],"end-to-end":[178],"speedup":[179],"6$\\times$":[183],"reduction,":[186],"substantially":[187],"improving":[188],"practicality":[190],"scaling":[192],"greater":[196],"depths":[198],"longer":[200],"context.":[201]},"counts_by_year":[{"year":2026,"cited_by_count":6}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-26T00:00:00"}