{"id":"https://openalex.org/W7164936665","doi":"https://doi.org/10.48550/arxiv.2606.16429","title":"Taylor-Calibrate: Principled Initialization for Hybrid Linear Attention Distillation","display_name":"Taylor-Calibrate: Principled Initialization for Hybrid Linear Attention Distillation","publication_year":2026,"publication_date":"2026-06-15","ids":{"openalex":"https://openalex.org/W7164936665","doi":"https://doi.org/10.48550/arxiv.2606.16429"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.16429","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.16429","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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.2606.16429","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138731463","display_name":"Zhongzhu Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Zhongzhu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5019206078","display_name":"Qile Wu","orcid":"https://orcid.org/0009-0003-1096-1892"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Qingyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138710422","display_name":"Junxiong Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Junxiong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138753490","display_name":"Mayank Mishra","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mishra, Mayank","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138716905","display_name":"Shuaiwen Leon Song","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Song, Shuaiwen Leon","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138739778","display_name":"Ben Athiwaratkun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Athiwaratkun, Ben","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138731466","display_name":"Chenfeng Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Chenfeng","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":0.31769999861717224,"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":0.31769999861717224,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.11460000276565552,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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":0.07829999923706055,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/initialization","display_name":"Initialization","score":0.8999999761581421},{"id":"https://openalex.org/keywords/softmax-function","display_name":"Softmax function","score":0.8097000122070312},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6223999857902527},{"id":"https://openalex.org/keywords/quadratic-equation","display_name":"Quadratic equation","score":0.5540000200271606},{"id":"https://openalex.org/keywords/distillation","display_name":"Distillation","score":0.5529000163078308},{"id":"https://openalex.org/keywords/boosting","display_name":"Boosting (machine learning)","score":0.45010000467300415},{"id":"https://openalex.org/keywords/embedding","display_name":"Embedding","score":0.42969998717308044}],"concepts":[{"id":"https://openalex.org/C114466953","wikidata":"https://www.wikidata.org/wiki/Q6034165","display_name":"Initialization","level":2,"score":0.8999999761581421},{"id":"https://openalex.org/C188441871","wikidata":"https://www.wikidata.org/wiki/Q7554146","display_name":"Softmax function","level":3,"score":0.8097000122070312},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7339000105857849},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6223999857902527},{"id":"https://openalex.org/C129844170","wikidata":"https://www.wikidata.org/wiki/Q41299","display_name":"Quadratic equation","level":2,"score":0.5540000200271606},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.5529000163078308},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.45010000467300415},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.42969998717308044},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.42289999127388},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.41850000619888306},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3379000127315521},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.32589998841285706},{"id":"https://openalex.org/C2779151265","wikidata":"https://www.wikidata.org/wiki/Q1156791","display_name":"Copying","level":2,"score":0.3253999948501587},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.30079999566078186},{"id":"https://openalex.org/C2777735758","wikidata":"https://www.wikidata.org/wiki/Q817765","display_name":"Path (computing)","level":2,"score":0.2937000095844269},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.27090001106262207},{"id":"https://openalex.org/C117251300","wikidata":"https://www.wikidata.org/wiki/Q1849855","display_name":"Parametric statistics","level":2,"score":0.26809999346733093},{"id":"https://openalex.org/C95623464","wikidata":"https://www.wikidata.org/wiki/Q1096149","display_name":"Classifier (UML)","level":2,"score":0.26190000772476196},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.25189998745918274}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.16429","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.16429","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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.2606.16429","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.16429","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":[{"display_name":"No poverty","id":"https://metadata.un.org/sdg/1","score":0.43899357318878174}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Hybrid":[0],"linear":[1],"attention":[2,23,64,127],"models":[3,39],"offer":[4],"an":[5,174],"appealing":[6],"path":[7],"to":[8,36,41,129,148,153,173],"faster":[9],"long-context":[10],"inference:":[11],"they":[12],"reduce":[13],"the":[14,28,62,75,86,107,131,154],"quadratic":[15],"cost":[16],"and":[17,80,96,138,161,181],"KV-cache":[18],"burden":[19],"of":[20,27,30,47],"full":[21],"softmax":[22],"while":[24],"retaining":[25],"much":[26],"quality":[29],"Transformer":[31,45],"models.":[32],"A":[33],"practical":[34],"way":[35],"obtain":[37],"such":[38],"is":[40,57],"convert":[42],"a":[43,49,67,84,92,114,143,178],"pretrained":[44],"instead":[46],"pretraining":[48],"new":[50,76],"architecture":[51],"from":[52],"scratch,":[53],"but":[54],"this":[55],"conversion":[56],"still":[58],"brittle.":[59],"Simply":[60],"copying":[61],"teacher":[63,109,126,155,159],"projections":[65],"into":[66],"Gated":[68],"DeltaNet":[69],"(GDN)":[70],"student":[71],"does":[72],"not":[73],"specify":[74],"recurrent":[77],"decay,":[78],"write,":[79],"output-gating":[81],"dynamics.":[82],"As":[83],"result,":[85],"converted":[87,151],"model":[88],"often":[89],"starts":[90],"in":[91,177],"poor":[93],"dynamical":[94],"regime":[95],"must":[97],"spend":[98],"many":[99],"distillation":[100],"tokens":[101,190],"repairing":[102],"initialization":[103,116],"rather":[104],"than":[105,191],"learning":[106],"remaining":[108],"behavior.":[110],"We":[111],"propose":[112],"Taylor-Calibrate,":[113],"lightweight":[115],"method":[117,123],"for":[118],"hybrid":[119],"GDN":[120],"students.":[121],"The":[122],"uses":[124],"Taylor-guided":[125],"statistics":[128],"set":[130],"value":[132],"projection,":[133],"memory":[134],"timescale,":[135],"write":[136],"gates,":[137],"output":[139],"gate,":[140],"then":[141],"applies":[142],"short":[144],"per-layer":[145],"alignment":[146],"step":[147],"match":[149],"each":[150],"layer":[152],"output.":[156],"Across":[157],"four":[158],"settings":[160],"three":[162],"retained-layer":[163],"policies,":[164],"Taylor-Calibrate":[165],"gives":[166],"substantially":[167],"stronger":[168],"zero-shot":[169],"students,":[170],"with":[171,186],"up":[172],"88x":[175],"improvement":[176],"representative":[179],"ablation,":[180],"reaches":[182],"matched":[183],"recovery":[184],"targets":[185],"4.9x--9.2x":[187],"fewer":[188],"training":[189],"naive":[192],"conversion.":[193]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-17T00:00:00"}