{"id":"https://openalex.org/W7133499425","doi":"https://doi.org/10.48550/arxiv.2603.02631","title":"Cross-Family Speculative Prefill: Training-Free Long-Context Compression with Small Draft Models","display_name":"Cross-Family Speculative Prefill: Training-Free Long-Context Compression with Small Draft Models","publication_year":2026,"publication_date":"2026-03-03","ids":{"openalex":"https://openalex.org/W7133499425","doi":"https://doi.org/10.48550/arxiv.2603.02631"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.02631","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.02631","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,"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.2603.02631","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5119911935","display_name":"Shubhangi Upasani","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Upasani, Shubhangi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128117502","display_name":"Ravi Shanker Raju","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Raju, Ravi Shanker","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100374460","display_name":"Bo Li","orcid":"https://orcid.org/0000-0002-7979-8339"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Bo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5031039929","display_name":"MingChao Ji","orcid":"https://orcid.org/0000-0001-8184-4595"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ji, Mengmeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128060575","display_name":"John Long","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Long, John","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128058735","display_name":"Chen Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Chen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5090396089","display_name":"Urmish Thakker","orcid":"https://orcid.org/0000-0002-0515-9155"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Thakker, Urmish","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5102889910","display_name":"Guangtao Wang","orcid":"https://orcid.org/0000-0003-2766-0917"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Guangtao","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":0.46480000019073486,"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":0.46480000019073486,"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":0.08179999887943268,"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.058800000697374344,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.6018999814987183},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.5810999870300293},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5738999843597412},{"id":"https://openalex.org/keywords/compression","display_name":"Compression (physics)","score":0.5358999967575073},{"id":"https://openalex.org/keywords/bayesian-inference","display_name":"Bayesian inference","score":0.4668999910354614},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.4212000072002411},{"id":"https://openalex.org/keywords/baseline","display_name":"Baseline (sea)","score":0.4009999930858612},{"id":"https://openalex.org/keywords/prior-probability","display_name":"Prior probability","score":0.3564000129699707},{"id":"https://openalex.org/keywords/data-compression","display_name":"Data compression","score":0.3490000069141388}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7789000272750854},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6018999814987183},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.5810999870300293},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5738999843597412},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.5358999967575073},{"id":"https://openalex.org/C160234255","wikidata":"https://www.wikidata.org/wiki/Q812535","display_name":"Bayesian inference","level":3,"score":0.4668999910354614},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.4212000072002411},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.4009999930858612},{"id":"https://openalex.org/C177769412","wikidata":"https://www.wikidata.org/wiki/Q278090","display_name":"Prior probability","level":3,"score":0.3564000129699707},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.3490000069141388},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.33809998631477356},{"id":"https://openalex.org/C107673813","wikidata":"https://www.wikidata.org/wiki/Q812534","display_name":"Bayesian probability","level":2,"score":0.3357999920845032},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.33230000734329224},{"id":"https://openalex.org/C88871306","wikidata":"https://www.wikidata.org/wiki/Q7208287","display_name":"Point process","level":2,"score":0.3154999911785126},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.304500013589859},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.29260000586509705},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.2896000146865845},{"id":"https://openalex.org/C2781316041","wikidata":"https://www.wikidata.org/wiki/Q1230584","display_name":"Diversity (politics)","level":2,"score":0.28780001401901245},{"id":"https://openalex.org/C81081738","wikidata":"https://www.wikidata.org/wiki/Q55542","display_name":"Lossless compression","level":3,"score":0.2802000045776367},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.2782999873161316},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.27630001306533813},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.26829999685287476},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.2614000141620636},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.25220000743865967}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.02631","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.02631","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,"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.2603.02631","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.02631","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,"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":{"Prompt":[0],"length":[1],"is":[2,89],"a":[3,46,81,96,100,114,127,205],"major":[4],"bottleneck":[5],"in":[6,147,168,182],"agentic":[7,61,218],"large":[8],"language":[9],"model":[10,48,84,87,98,143,148,226],"(LLM)":[11],"workloads,":[12],"where":[13,80,220],"repeated":[14,221],"inference":[15,223],"steps":[16],"and":[17,123,150,154,199,224,234],"multi-call":[18],"loops":[19],"incur":[20],"substantial":[21,180],"prefill":[22,28,107,193],"cost.":[23],"Recent":[24],"work":[25],"on":[26,196],"speculative":[27,78,106,192],"demonstrates":[29],"that":[30,49,134,191],"attention-based":[31,135],"token":[32,136,186],"importance":[33,137],"estimation":[34,138],"can":[35],"enable":[36],"training-free":[37],"prompt":[38,93,158,207,230],"compression,":[39],"but":[40],"this":[41,73],"assumes":[42],"the":[43,51,55,104,212],"existence":[44],"of":[45,116,130,163,214],"draft":[47,70,83,153],"shares":[50],"same":[52,105],"tokenizer":[53],"as":[54,109,204],"target":[56,97,155],"model.":[57,71],"In":[58,72],"practice,":[59],"however,":[60],"pipelines":[62],"frequently":[63],"employ":[64],"models":[65],"without":[66],"any":[67],"smaller":[68],"in-family":[69],"work,":[74,111],"we":[75,112,132],"study":[76],"cross-family":[77,117],"prefill,":[79],"lightweight":[82],"from":[85,99],"one":[86],"family":[88],"used":[90],"to":[91,175,184],"perform":[92],"compression":[94,159,208,231],"for":[95,217],"different":[101,142],"family.":[102],"Using":[103],"mechanism":[108],"prior":[110],"evaluate":[113],"range":[115],"draft-target":[118],"combinations,":[119],"including":[120],"Qwen,":[121],"LLaMA,":[122],"DeepSeek":[124],"models.":[125,156],"Across":[126],"broad":[128],"diversity":[129],"tasks,":[131],"find":[133],"transfers":[139],"reliably":[140],"across":[141],"families":[144],"despite":[145],"differences":[146],"architectures":[149],"tokenizers":[151],"between":[152],"Cross-model":[157],"largely":[160],"retains":[161],"90~100%":[162],"full-prompt":[164],"baseline":[165],"performance":[166],"and,":[167],"some":[169],"cases,":[170],"slightly":[171],"improves":[172],"accuracy":[173],"due":[174],"denoising":[176],"effects,":[177],"while":[178],"delivering":[179],"reductions":[181],"time":[183],"first":[185],"(TTFT).":[187],"These":[188],"results":[189],"suggest":[190],"depends":[194],"mainly":[195],"task":[197],"priors":[198],"semantic":[200],"structure,":[201],"thus":[202],"serving":[203],"generalizable":[206],"primitive.":[209],"We":[210],"discuss":[211],"implications":[213],"our":[215],"findings":[216],"systems,":[219],"long-context":[222],"heterogeneous":[225],"stacks":[227],"make":[228],"cross-model":[229],"both":[232],"necessary":[233],"practical.":[235]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-03-05T00:00:00"}