{"id":"https://openalex.org/W7162513292","doi":"https://doi.org/10.48550/arxiv.2605.27255","title":"Pair-In, Pair-Out: Latent Multi-Token Prediction for Efficient LLMs","display_name":"Pair-In, Pair-Out: Latent Multi-Token Prediction for Efficient LLMs","publication_year":2026,"publication_date":"2026-05-26","ids":{"openalex":"https://openalex.org/W7162513292","doi":"https://doi.org/10.48550/arxiv.2605.27255"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.27255","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.27255","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.27255","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137117083","display_name":"Wenhui Tan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tan, Wenhui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137183042","display_name":"Minghao Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Minghao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135979561","display_name":"Xiaoqian Ma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Xiaoqian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137092299","display_name":"Siqi Fan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fan, Siqi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137104841","display_name":"Xiusheng Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Xiusheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5016293831","display_name":"Liujie Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Liujie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137155301","display_name":"Ruihua Song","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Song, Ruihua","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5103275486","display_name":"Weihang Chen","orcid":"https://orcid.org/0000-0003-1380-7631"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Weihang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":8,"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":0.24740000069141388,"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":0.24740000069141388,"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":0.15060000121593475,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.12060000002384186,"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/decoding-methods","display_name":"Decoding methods","score":0.6980999708175659},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5117999911308289},{"id":"https://openalex.org/keywords/head","display_name":"Head (geology)","score":0.5067999958992004},{"id":"https://openalex.org/keywords/hidden-markov-model","display_name":"Hidden Markov model","score":0.3693000078201294},{"id":"https://openalex.org/keywords/autoregressive-model","display_name":"Autoregressive model","score":0.36419999599456787},{"id":"https://openalex.org/keywords/state","display_name":"State (computer science)","score":0.3077999949455261}],"concepts":[{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.6980999708175659},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6969000101089478},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5117999911308289},{"id":"https://openalex.org/C2780312720","wikidata":"https://www.wikidata.org/wiki/Q5689100","display_name":"Head (geology)","level":2,"score":0.5067999958992004},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4345000088214874},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.36970001459121704},{"id":"https://openalex.org/C23224414","wikidata":"https://www.wikidata.org/wiki/Q176769","display_name":"Hidden Markov model","level":2,"score":0.3693000078201294},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.36419999599456787},{"id":"https://openalex.org/C48103436","wikidata":"https://www.wikidata.org/wiki/Q599031","display_name":"State (computer science)","level":2,"score":0.3077999949455261},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.2912999987602234},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.2721000015735626},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.26660001277923584},{"id":"https://openalex.org/C131097465","wikidata":"https://www.wikidata.org/wiki/Q178898","display_name":"Gas compressor","level":2,"score":0.26420000195503235},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2621000111103058},{"id":"https://openalex.org/C106195933","wikidata":"https://www.wikidata.org/wiki/Q7847935","display_name":"Truncation (statistics)","level":2,"score":0.2547000050544739},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.2517000138759613}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.27255","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.27255","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":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.27255","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.27255","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":"article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Long":[0],"chain-of-thought":[1],"reasoning":[2],"has":[3],"made":[4],"autoregressive":[5],"decoding":[6,30,182],"the":[7,20,26,36,56,88,99,113,141,148],"dominant":[8],"inference":[9],"cost":[10,115],"of":[11,39,144],"modern":[12],"large":[13],"language":[14],"models.":[15],"Existing":[16],"methods":[17,47],"target":[18],"either":[19],"input":[21,92],"side":[22,28],"(latent":[23],"compression)":[24],"or":[25],"output":[27,109],"(speculative":[29],"and":[31,81,167,172,194],"multi-token":[32],"prediction,":[33],"MTP),":[34],"but":[35],"two":[37,91],"lines":[38],"work":[40],"have":[41],"been":[42],"pursued":[43],"independently.":[44],"Moreover,":[45],"output-side":[46],"must":[48],"incur":[49],"an":[50,82],"expensive":[51],"verifier":[52,114],"pass":[53],"to":[54,185,191],"validate":[55],"unreliable":[57],"draft":[58,128],"tokens":[59,93,129],"predicted":[60],"by":[61,76,183],"MTP.":[62],"To":[63,111],"address":[64],"these":[65],"issues,":[66],"we":[67],"propose":[68],"\\textbf{Pair-In,":[69],"Pair-Out":[70],"(PIPO)},":[71],"which":[72],"unifies":[73],"both":[74],"sides":[75],"viewing":[77],"a":[78,121],"latent":[79,96],"compressor":[80,89],"MTP":[83,100],"head":[84,101,124,150],"as":[85],"mirror-image":[86],"operations:":[87],"folds":[90],"into":[94,106],"one":[95,103,107],"representation,":[97],"while":[98,188],"unfolds":[102],"hidden":[104],"state":[105],"additional":[108],"token.":[110],"remove":[112],"without":[116],"sacrificing":[117],"reliability,":[118],"PIPO":[119,177],"trains":[120],"lightweight":[122],"confidence":[123,149],"that":[125,135,176],"decides":[126],"whether":[127],"should":[130],"be":[131,152],"accepted.":[132],"We":[133],"observe":[134],"On-Policy":[136],"Distillation":[137],"(OPD)":[138],"naturally":[139],"matches":[140],"rejection-sampling":[142],"criterion":[143],"speculative":[145],"decoding,":[146],"so":[147],"can":[151],"trained":[153],"alongside":[154],"OPD":[155],"with":[156,170],"negligible":[157],"extra":[158],"cost.":[159],"Experiments":[160],"on":[161],"AIME":[162],"2025,":[163],"GPQA-Diamond,":[164],"LiveCodeBench":[165],"v6,":[166],"LongBench":[168],"v2":[169],"Qwen3.5-4B":[171],"9B":[173],"backbones":[174],"show":[175],"improves":[178],"pass@4":[179],"over":[180],"regular":[181],"up":[184,190],"$+7.15$":[186],"points,":[187],"delivering":[189],"$2.64\\times$":[192],"first-token-latency":[193],"$2.07\\times$":[195],"per-token-latency":[196],"speedups.":[197],"Project":[198],"Page:":[199],"GitHub.com/RedAI-Infra/PIPO.":[200]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-05-28T00:00:00"}