{"id":"https://openalex.org/W7139076825","doi":"https://doi.org/10.48550/arxiv.2603.17942","title":"Efficient Training-Free Multi-Token Prediction via Embedding-Space Probing","display_name":"Efficient Training-Free Multi-Token Prediction via Embedding-Space Probing","publication_year":2026,"publication_date":"2026-03-18","ids":{"openalex":"https://openalex.org/W7139076825","doi":"https://doi.org/10.48550/arxiv.2603.17942"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.17942","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.17942","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.2603.17942","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5114129913","display_name":"Raghavv Goel","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Goel, Raghavv","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5018463719","display_name":"Mukul Gagrani","orcid":"https://orcid.org/0000-0003-0483-1090"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gagrani, Mukul","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129887438","display_name":"Mingu Lee","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lee, Mingu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5113688118","display_name":"Chris Lott","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lott, Chris","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/T10028","display_name":"Topic Modeling","score":0.2833999991416931,"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":0.2833999991416931,"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":0.26739999651908875,"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/T13702","display_name":"Machine Learning in Healthcare","score":0.050700001418590546,"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/security-token","display_name":"Security token","score":0.6424999833106995},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.6078000068664551},{"id":"https://openalex.org/keywords/embedding","display_name":"Embedding","score":0.5027999877929688},{"id":"https://openalex.org/keywords/lossless-compression","display_name":"Lossless compression","score":0.48980000615119934},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4203000068664551},{"id":"https://openalex.org/keywords/tree","display_name":"Tree (set theory)","score":0.39989998936653137},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.3901999890804291},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.3763999938964844}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7777000069618225},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6424999833106995},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.6078000068664551},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5069000124931335},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.5027999877929688},{"id":"https://openalex.org/C81081738","wikidata":"https://www.wikidata.org/wiki/Q55542","display_name":"Lossless compression","level":3,"score":0.48980000615119934},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4203000068664551},{"id":"https://openalex.org/C113174947","wikidata":"https://www.wikidata.org/wiki/Q2859736","display_name":"Tree (set theory)","level":2,"score":0.39989998936653137},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.3901999890804291},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.38269999623298645},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.3763999938964844},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.3752000033855438},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.3522000014781952},{"id":"https://openalex.org/C2777115002","wikidata":"https://www.wikidata.org/wiki/Q7168246","display_name":"Performance prediction","level":2,"score":0.34880000352859497},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.3149000108242035},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.31189998984336853},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.30570000410079956},{"id":"https://openalex.org/C45235069","wikidata":"https://www.wikidata.org/wiki/Q278425","display_name":"Table (database)","level":2,"score":0.30059999227523804},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.296099990606308},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.29440000653266907},{"id":"https://openalex.org/C167966045","wikidata":"https://www.wikidata.org/wiki/Q5532625","display_name":"Generative model","level":3,"score":0.28439998626708984},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.2840999960899353},{"id":"https://openalex.org/C22019652","wikidata":"https://www.wikidata.org/wiki/Q331309","display_name":"Overfitting","level":3,"score":0.28360000252723694},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.2736000120639801},{"id":"https://openalex.org/C84525736","wikidata":"https://www.wikidata.org/wiki/Q831366","display_name":"Decision tree","level":2,"score":0.26910001039505005}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.17942","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.17942","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.2603.17942","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.17942","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":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"possess":[4],"latent":[5],"multi-token":[6],"prediction":[7,44,147],"(MTP)":[8],"abilities":[9],"despite":[10],"being":[11],"trained":[12],"only":[13],"for":[14],"next-token":[15,142],"generation.":[16],"We":[17],"introduce":[18],"ESP":[19,53,95],"(Embedding-Space":[20],"Probing),":[21],"a":[22,55,68],"simple":[23],"and":[24,66,91,110,114,130],"training-free":[25,99],"MTP":[26],"method":[27],"that":[28,134],"probes":[29],"an":[30],"LLM":[31],"using":[32],"on-the-fly":[33],"mask":[34],"tokens":[35],"drawn":[36],"from":[37,63],"its":[38],"embedding":[39],"space,":[40],"enabling":[41,144],"parallel":[42],"future-token":[43],"without":[45,148],"modifying":[46],"weights":[47],"or":[48,150],"relying":[49],"on":[50,108,112],"draft":[51],"models.":[52,152],"constructs":[54],"speculative":[56],"token":[57,93],"tree":[58],"by":[59,104,117],"sampling":[60],"Top-K":[61],"candidates":[62],"mask-token":[64,139],"logits":[65],"applies":[67],"lightweight":[69],"pruning":[70],"rule":[71],"to":[72,119],"retain":[73],"high-probability":[74],"continuations.":[75],"During":[76],"generation,":[77],"predictions":[78],"are":[79],"verified":[80],"in":[81],"parallel,":[82],"yielding":[83],"lossless":[84],"decoding":[85],"while":[86],"significantly":[87],"reducing":[88],"model":[89],"calls":[90],"increasing":[92,115],"throughput.":[94],"consistently":[96],"outperforms":[97],"existing":[98],"baselines,":[100],"improving":[101],"acceptance":[102],"length":[103],"7-11%":[105],"over":[106,121],"LADE":[107],"LLaMA3":[109],"7-8%":[111],"Qwen3,":[113],"throughput":[116],"up":[118],"15-19%":[120],"the":[122],"strongest":[123],"baseline.":[124],"Finally,":[125],"we":[126],"provide":[127],"theoretical":[128],"insight":[129],"empirical":[131],"evidence":[132],"showing":[133],"decoder":[135],"layers":[136],"naturally":[137],"align":[138],"representations":[140],"with":[141],"states,":[143],"accurate":[145],"multi-step":[146],"retraining":[149],"auxiliary":[151]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-03-20T00:00:00"}