{"id":"https://openalex.org/W7161849267","doi":"https://doi.org/10.48550/arxiv.2605.20104","title":"Draft Less, Retrieve More: Hybrid Tree Construction for Speculative Decoding","display_name":"Draft Less, Retrieve More: Hybrid Tree Construction for Speculative Decoding","publication_year":2026,"publication_date":"2026-05-19","ids":{"openalex":"https://openalex.org/W7161849267","doi":"https://doi.org/10.48550/arxiv.2605.20104"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.20104","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.20104","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":null,"license_id":null,"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.2605.20104","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136603500","display_name":"Yuhao Shen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shen, Yuhao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136564252","display_name":"Tianyu Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Tianyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136542298","display_name":"Xinyi Hu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Xinyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136520394","display_name":"Quan Kong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kong, Quan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136517833","display_name":"Baolin Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Baolin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136513880","display_name":"Jun Dai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dai, Jun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136597817","display_name":"Jun Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Jun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5107196684","display_name":"Shuang Ge","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ge, Shuang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136608219","display_name":"Lei Chen","orcid":"https://orcid.org/0000-0001-6394-9764"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Lei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136578894","display_name":"Yue Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Yue","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5108223442","display_name":"Mingcheng Wan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wan, Mingcheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136510938","display_name":"Cong Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Cong","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":0.21850000321865082,"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":0.21850000321865082,"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":0.13570000231266022,"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":0.1031000018119812,"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/speedup","display_name":"Speedup","score":0.7002000212669373},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.6165000200271606},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.6148999929428101},{"id":"https://openalex.org/keywords/pareto-principle","display_name":"Pareto principle","score":0.5133000016212463},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.46209999918937683},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.4422999918460846},{"id":"https://openalex.org/keywords/look-ahead","display_name":"Look-ahead","score":0.4081000089645386},{"id":"https://openalex.org/keywords/block","display_name":"Block (permutation group theory)","score":0.4052000045776367},{"id":"https://openalex.org/keywords/tree","display_name":"Tree (set theory)","score":0.35910001397132874}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7609999775886536},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.7002000212669373},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.6165000200271606},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.6148999929428101},{"id":"https://openalex.org/C137635306","wikidata":"https://www.wikidata.org/wiki/Q182667","display_name":"Pareto principle","level":2,"score":0.5133000016212463},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.46209999918937683},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4422999918460846},{"id":"https://openalex.org/C147297375","wikidata":"https://www.wikidata.org/wiki/Q6674930","display_name":"Look-ahead","level":2,"score":0.4081000089645386},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.4052000045776367},{"id":"https://openalex.org/C113174947","wikidata":"https://www.wikidata.org/wiki/Q2859736","display_name":"Tree (set theory)","level":2,"score":0.35910001397132874},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.3458999991416931},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3407000005245209},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.3328000009059906},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.3294000029563904},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3231000006198883},{"id":"https://openalex.org/C2780992000","wikidata":"https://www.wikidata.org/wiki/Q17016113","display_name":"Generator (circuit theory)","level":3,"score":0.31310001015663147},{"id":"https://openalex.org/C188045654","wikidata":"https://www.wikidata.org/wiki/Q17148339","display_name":"Memory bandwidth","level":2,"score":0.3084999918937683},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.30790001153945923},{"id":"https://openalex.org/C175291020","wikidata":"https://www.wikidata.org/wiki/Q1156822","display_name":"Offset (computer science)","level":2,"score":0.3041999936103821},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.2985999882221222},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.28780001401901245},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.28610000014305115},{"id":"https://openalex.org/C106131492","wikidata":"https://www.wikidata.org/wiki/Q3072260","display_name":"Filter (signal processing)","level":2,"score":0.28049999475479126},{"id":"https://openalex.org/C147224247","wikidata":"https://www.wikidata.org/wiki/Q885373","display_name":"Bloom filter","level":2,"score":0.2793000042438507},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.27790001034736633},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.27379998564720154},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.27309998869895935},{"id":"https://openalex.org/C34736171","wikidata":"https://www.wikidata.org/wiki/Q918333","display_name":"Preprocessor","level":2,"score":0.26739999651908875},{"id":"https://openalex.org/C168773036","wikidata":"https://www.wikidata.org/wiki/Q264164","display_name":"Recursion (computer science)","level":2,"score":0.26600000262260437},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.26019999384880066},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.2554999887943268}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.20104","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.20104","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.20104","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.20104","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Speculative":[0],"decoding":[1],"(SD)":[2],"accelerates":[3],"large":[4],"language":[5],"model":[6],"inference":[7],"by":[8,44,141,193],"leveraging":[9],"a":[10,71,97,128,162,204],"draft-then-verify":[11],"paradigm.":[12],"To":[13,89],"maximize":[14],"the":[15,55,60,77,144,198,211],"acceptance":[16,56],"rate,":[17],"recent":[18],"methods":[19],"construct":[20],"expansive":[21],"draft":[22,226],"trees,":[23],"which":[24],"unfortunately":[25],"incur":[26],"severe":[27],"VRAM":[28],"bandwidth":[29],"and":[30,103,122,154,175,187,219],"computational":[31,87],"overheads":[32],"that":[33,100,159],"bottleneck":[34],"end-to-end":[35],"speedups.":[36],"While":[37],"dynamic-depth":[38],"pruning":[39,102],"can":[40],"reduce":[41],"this":[42,67,91],"latency":[43],"removing":[45],"marginal":[46],"branches,":[47],"it":[48,181],"also":[49,202],"discards":[50],"potentially":[51],"valid":[52],"candidates,":[53],"preventing":[54],"rate":[57],"from":[58,79],"reaching":[59],"upper":[61],"bound":[62],"of":[63,207],"dense":[64,80],"trees.":[65,227],"In":[66],"paper,":[68],"we":[69,94],"identify":[70],"critical":[72],"opportunity":[73],"in":[74],"resource":[75],"allocation:":[76],"transition":[78],"to":[81,184,195,210],"pruned":[82],"drafting":[83,214],"frees":[84],"up":[85,183,194],"significant":[86],"budget.":[88],"break":[90],"Pareto":[92,164],"tradeoff,":[93],"introduce":[95],"Graft,":[96],"compensation":[98],"framework":[99],"couples":[101],"retrieval":[104,116],"as":[105],"mutually":[106],"reinforcing":[107],"operations.":[108],"Pruning":[109],"supplies":[110],"sufficient":[111],"budget":[112],"for":[113,118,221],"retrieval,":[114],"while":[115],"compensates":[117],"pruning-induced":[119],"coverage":[120],"loss":[121],"recovers":[123],"accepted":[124],"length.":[125],"By":[126],"employing":[127],"sequential":[129],"`prune-then-graft'":[130],"mechanism,":[131],"Graft":[132,150,160,209],"attaches":[133],"highly":[134],"predictive":[135],"retrieved":[136],"tokens":[137],"into":[138],"positions":[139],"opened":[140],"pruning,":[142],"filling":[143],"topological":[145],"gaps":[146],"with":[147],"near-zero":[148],"overhead.":[149],"is":[151],"entirely":[152],"training-free":[153],"lossless.":[155],"Comprehensive":[156],"evaluations":[157],"show":[158],"establishes":[161],"new":[163],"frontier":[165],"across":[166],"practical":[167],"deployment":[168],"settings,":[169],"including":[170],"short-context":[171,179],"generation,":[172,174],"long-context":[173],"large-scale":[176,199],"models.":[177],"On":[178],"benchmarks,":[180],"achieves":[182],"5.41$\\times$":[185],"speedup":[186,190],"improves":[188],"average":[189],"over":[191],"EAGLE-3":[192],"21.8%":[196],"on":[197],"Qwen3-235B.":[200],"We":[201],"provide":[203],"preliminary":[205],"exploration":[206],"applying":[208],"DFlash-style":[212],"block":[213],"paradigm,":[215],"offering":[216],"initial":[217],"evidence":[218],"insights":[220],"extending":[222],"grafting":[223],"beyond":[224],"autoregressive":[225]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-21T00:00:00"}