{"id":"https://openalex.org/W7220948705","doi":"https://doi.org/10.48550/arxiv.2610.09013","title":"Enabling Dynamic Computation in Looped LMs","display_name":"Enabling Dynamic Computation in Looped LMs","publication_year":2026,"publication_date":"2026-10-06","ids":{"openalex":"https://openalex.org/W7220948705","doi":"https://doi.org/10.48550/arxiv.2610.09013"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.09013","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.09013","pdf_url":"https://arxiv.org/pdf/2610.09013","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-sa","license_id":"https://openalex.org/licenses/cc-by-sa","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/2610.09013","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5013900997","display_name":"Aayush Mishra","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I145311948","display_name":"Johns Hopkins University","ror":"https://ror.org/00za53h95","country_code":"US","type":"education","lineage":["https://openalex.org/I145311948"]},{"id":"https://openalex.org/I4210087596","display_name":"Qualcomm (United States)","ror":"https://ror.org/002zrf773","country_code":"US","type":"company","lineage":["https://openalex.org/I4210087596"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Mishra, Aayush","raw_affiliation_strings":["Johns Hopkins University","Qualcomm AI Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Johns Hopkins University","institution_ids":["https://openalex.org/I145311948"]},{"raw_affiliation_string":"Qualcomm AI Research","institution_ids":["https://openalex.org/I4210087596"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119959932","display_name":"Arnau Padr\u00e9s Masdemont","orcid":"https://orcid.org/0009-0004-9907-3548","observed_orcids":["https://orcid.org/0009-0004-9907-3548"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Masdemont, Arnau Padr\u00e9s","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5001019652","display_name":"V\u00edctor Conchello Vendrell","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vendrell, Victor Conchello","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5079078620","display_name":"Jordi Ros Giralt","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Giralt, Jordi Ros","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5083706779","display_name":"Arash Behboodi","orcid":"https://orcid.org/0000-0001-8229-2809","observed_orcids":["https://orcid.org/0000-0001-8229-2809"]},"institutions":[{"id":"https://openalex.org/I4210087596","display_name":"Qualcomm (United States)","ror":"https://ror.org/002zrf773","country_code":"US","type":"company","lineage":["https://openalex.org/I4210087596"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Behboodi, Arash","raw_affiliation_strings":["Qualcomm AI Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Qualcomm AI Research","institution_ids":["https://openalex.org/I4210087596"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5017199746","display_name":"Fabio Valerio Massoli","orcid":"https://orcid.org/0000-0001-6447-1301","observed_orcids":["https://orcid.org/0000-0001-6447-1301"]},"institutions":[{"id":"https://openalex.org/I4210087596","display_name":"Qualcomm (United States)","ror":"https://ror.org/002zrf773","country_code":"US","type":"company","lineage":["https://openalex.org/I4210087596"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Massoli, Fabio Valerio","raw_affiliation_strings":["Qualcomm AI Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Qualcomm AI Research","institution_ids":["https://openalex.org/I4210087596"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"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":0.42089998722076416,"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":0.42089998722076416,"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":0.24719999730587006,"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":0.2037000060081482,"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/dynamic-computation","display_name":"dynamic computation","score":0.9459999799728394},{"id":"https://openalex.org/keywords/kv-caching","display_name":"KV caching","score":0.6269999742507935},{"id":"https://openalex.org/keywords/parameter-efficiency","display_name":"parameter efficiency","score":0.6970000267028809}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7559000253677368},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.7279000282287598},{"id":"https://openalex.org/C2780598303","wikidata":"https://www.wikidata.org/wiki/Q65921492","display_name":"Flexibility (engineering)","level":2,"score":0.6693000197410583},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.5292999744415283},{"id":"https://openalex.org/C3826847","wikidata":"https://www.wikidata.org/wiki/Q188768","display_name":"FLOPS","level":2,"score":0.5275999903678894},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5156000256538391},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.4359999895095825},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.3617999851703644},{"id":"https://openalex.org/C184670325","wikidata":"https://www.wikidata.org/wiki/Q512604","display_name":"Loop (graph theory)","level":2,"score":0.33320000767707825},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.3197999894618988},{"id":"https://openalex.org/C118702147","wikidata":"https://www.wikidata.org/wiki/Q189396","display_name":"Dynamic random-access memory","level":3,"score":0.313400000333786},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.3003999888896942},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.296999990940094},{"id":"https://openalex.org/C186886427","wikidata":"https://www.wikidata.org/wiki/Q5441213","display_name":"Feedback loop","level":2,"score":0.290800005197525},{"id":"https://openalex.org/C176649486","wikidata":"https://www.wikidata.org/wiki/Q2308807","display_name":"Memory management","level":3,"score":0.2897000014781952},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.2851000130176544},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2696000039577484},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.2612000107765198},{"id":"https://openalex.org/C2994168587","wikidata":"https://www.wikidata.org/wiki/Q5295","display_name":"Random access memory","level":2,"score":0.2540999948978424}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.09013","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.09013","pdf_url":"https://arxiv.org/pdf/2610.09013","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-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.09013","doi":"https://doi.org/10.48550/arxiv.2610.09013","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.09013","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:2610.09013","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.09013","pdf_url":"https://arxiv.org/pdf/2610.09013","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-sa","license_id":"https://openalex.org/licenses/cc-by-sa","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/W7220948705.pdf","grobid_xml":"https://content.openalex.org/works/W7220948705.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7202206158","https://openalex.org/W7219155330","https://openalex.org/W7214290239","https://openalex.org/W7220926987","https://openalex.org/W7168269548","https://openalex.org/W7213266125","https://openalex.org/W7160859380","https://openalex.org/W7220657469","https://openalex.org/W7223292061","https://openalex.org/W7223052750"],"abstract_inverted_index":{"Looped":[0,19,118],"LMs":[1,20,123,172],"are":[2,162],"parameter":[3],"efficient":[4],"and":[5,11,106,133],"promise":[6],"dynamic":[7,24],"computation":[8,25],"(saving":[9],"memory":[10,108],"FLOPs":[12,105],"on":[13,56,158,164],"easy":[14],"tokens).":[15],"However,":[16],"state-of-the-art":[17],"open":[18],"trained":[21],"with":[22,124],"this":[23,74,127],"capability":[26],"(Ouro":[27],"models)":[28],"do":[29],"not":[30],"realize":[31],"it":[32],"in":[33,62,91,104],"practice":[34],"as":[35,167,169],"each":[36,57],"loop":[37,48],"iteration":[38],"(depth)":[39],"requires":[40],"its":[41],"own":[42],"level":[43],"of":[44,67,71,117],"KV-cache,":[45],"necessitating":[46],"all":[47,68],"computations.":[49],"Moreover,":[50],"Ouro's":[51],"early-exit":[52,145],"prior":[53,146],"is":[54],"enforced":[55],"token":[58],"equally,":[59],"which":[60],"results":[61],"static":[63],"lower-depth":[64],"like":[65],"processing":[66],"tokens":[69,151],"regardless":[70],"difficulty.":[72],"In":[73],"work,":[75],"we":[76,136],"propose":[77],"a":[78,88,138],"simple":[79],"\"best-available\"":[80],"KV":[81,107,128],"caching":[82,129],"strategy":[83,130],"that":[84,149],"works":[85],"out-of-the-box,":[86],"creating":[87],"new":[89],"frontier":[90],"the":[92,114,144],"performance":[93,132],"vs":[94],"depth":[95],"space.":[96],"Our":[97,160],"approach":[98],"enables":[99],"up":[100],"to":[101,143],"30%":[102],"reduction":[103],"while":[109],"retaining":[110],"full-depth":[111],"performance,":[112],"showing":[113],"true":[115],"flexibility":[116],"LMs.":[119],"Furthermore,":[120],"training":[121],"looped":[122,171],"awareness":[125],"about":[126],"improves":[131],"efficiency.":[134],"Finally,":[135],"apply":[137],"small":[139],"but":[140],"effective":[141],"fix":[142],"enforcement":[147],"objective":[148],"makes":[150],"exit":[152],"at":[153],"truly":[154],"heterogeneous":[155],"depths":[156],"based":[157],"effort.":[159],"findings":[161],"validated":[163],"Ouro":[165],"models":[166],"well":[168],"smaller":[170],"pre-trained":[173],"from":[174],"scratch.":[175]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-10-09T00:00:00"}