{"id":"https://openalex.org/W7215756293","doi":"https://doi.org/10.48550/arxiv.2610.00623","title":"HAWK: Rethinking Multimodal Drafting for Speculative Decoding","display_name":"HAWK: Rethinking Multimodal Drafting for Speculative Decoding","publication_year":2026,"publication_date":"2026-09-30","ids":{"openalex":"https://openalex.org/W7215756293","doi":"https://doi.org/10.48550/arxiv.2610.00623"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.00623","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.00623","pdf_url":"https://arxiv.org/pdf/2610.00623","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2610.00623","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5155097610","display_name":"Wenhan Yang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I161318765","display_name":"University of California, Los Angeles","ror":"https://ror.org/046rm7j60","country_code":"US","type":"education","lineage":["https://openalex.org/I161318765"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yang, Wenhan","raw_affiliation_strings":["University of California , Los Angeles"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of California , Los Angeles","institution_ids":["https://openalex.org/I161318765"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5154684529","display_name":"Anirudh Rao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210101778","display_name":"Samsung (United States)","ror":"https://ror.org/01bfbvm65","country_code":"US","type":"company","lineage":["https://openalex.org/I2250650973","https://openalex.org/I4210101778"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Rao, Anirudh","raw_affiliation_strings":["Samsung Research America"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Samsung Research America","institution_ids":["https://openalex.org/I4210101778"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5155503512","display_name":"Ashwin Chandra","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210101778","display_name":"Samsung (United States)","ror":"https://ror.org/01bfbvm65","country_code":"US","type":"company","lineage":["https://openalex.org/I2250650973","https://openalex.org/I4210101778"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chandra, Ashwin","raw_affiliation_strings":["Samsung Research America"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Samsung Research America","institution_ids":["https://openalex.org/I4210101778"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.9653000235557556,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.9653000235557556,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.014399999752640724,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.012799999676644802,"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/speculative-decoding","display_name":"speculative decoding","score":0.9700000286102295},{"id":"https://openalex.org/keywords/large-vision-language-models","display_name":"large vision-language models","score":0.949999988079071},{"id":"https://openalex.org/keywords/knowledge-distillation","display_name":"knowledge distillation","score":0.796999990940094},{"id":"https://openalex.org/keywords/representational-similarity","display_name":"representational similarity","score":0.847000002861023},{"id":"https://openalex.org/keywords/decoding-speedup","display_name":"decoding speedup","score":0.6150000095367432},{"id":"https://openalex.org/keywords/hawks","display_name":"hawks","score":0.7260000109672546}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7261000275611877},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.5929999947547913},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.5800999999046326},{"id":"https://openalex.org/C81081738","wikidata":"https://www.wikidata.org/wiki/Q55542","display_name":"Lossless compression","level":3,"score":0.4708000123500824},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.46950000524520874},{"id":"https://openalex.org/C103278499","wikidata":"https://www.wikidata.org/wiki/Q254465","display_name":"Similarity (geometry)","level":3,"score":0.42640000581741333},{"id":"https://openalex.org/C2780910867","wikidata":"https://www.wikidata.org/wiki/Q1952416","display_name":"Multimodality","level":2,"score":0.38350000977516174},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.3643999993801117},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.33550000190734863},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3325999975204468},{"id":"https://openalex.org/C2777508537","wikidata":"https://www.wikidata.org/wiki/Q7936620","display_name":"Visual reasoning","level":2,"score":0.33219999074935913},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.3287000060081482},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.3221000134944916},{"id":"https://openalex.org/C2781085045","wikidata":"https://www.wikidata.org/wiki/Q7318308","display_name":"Reversing","level":2,"score":0.32199999690055847},{"id":"https://openalex.org/C23123220","wikidata":"https://www.wikidata.org/wiki/Q816826","display_name":"Information retrieval","level":1,"score":0.30410000681877136},{"id":"https://openalex.org/C126042441","wikidata":"https://www.wikidata.org/wiki/Q1324888","display_name":"Frame (networking)","level":2,"score":0.28110000491142273}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.00623","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.00623","pdf_url":"https://arxiv.org/pdf/2610.00623","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.00623","doi":"https://doi.org/10.48550/arxiv.2610.00623","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.00623","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,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2610.00623","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.00623","pdf_url":"https://arxiv.org/pdf/2610.00623","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7215756293.pdf","grobid_xml":"https://content.openalex.org/works/W7215756293.grobid-xml"},"referenced_works_count":11,"referenced_works":["https://openalex.org/W4319166707","https://openalex.org/W4385473806","https://openalex.org/W4391124766","https://openalex.org/W4393248014","https://openalex.org/W4396570304","https://openalex.org/W4412887764","https://openalex.org/W4414281281","https://openalex.org/W4415023006","https://openalex.org/W4415089005","https://openalex.org/W4416119086","https://openalex.org/W7162753994"],"related_works":["https://openalex.org/W4416557549","https://openalex.org/W4416034893","https://openalex.org/W4415023006","https://openalex.org/W4414746971","https://openalex.org/W7214297236","https://openalex.org/W7163596405","https://openalex.org/W7168290670","https://openalex.org/W4415334927","https://openalex.org/W7204089399","https://openalex.org/W7204175913"],"abstract_inverted_index":{"Speculative":[0],"decoding":[1],"has":[2],"achieved":[3],"substantial":[4],"lossless":[5],"speedups":[6],"for":[7,13,125],"LLMs,":[8],"but":[9],"remains":[10],"less":[11],"effective":[12],"large":[14],"vision-language":[15],"models":[16],"(LVLMs),":[17],"where":[18],"lightweight":[19],"drafters":[20],"struggle":[21],"to":[22,77,85,93,129,136,168,174,184,188],"use":[23],"rich":[24],"multimodal":[25,159],"information.":[26],"A":[27],"second":[28],"limitation":[29],"is":[30],"that":[31],"standard":[32],"distillation":[33],"supervises":[34],"the":[35,39,50,61,67,104,112,121,134,150],"drafter":[36,62,105,128,135],"only":[37],"along":[38],"original":[40],"training":[41],"trajectory,":[42,60],"without":[43],"modeling":[44],"how":[45,92,138],"target":[46,88,113,139,151],"predictions":[47,140],"shift":[48],"after":[49,142],"drafter's":[51],"own":[52,144],"proposals.":[53],"As":[54],"drafting":[55],"moves":[56],"away":[57],"from":[58,111,166,172,182],"this":[59],"can":[63],"increasingly":[64],"disagree":[65],"with":[66,106,149],"target,":[68],"reducing":[69],"acceptance":[70,164],"in":[71],"later":[72],"steps.":[73],"We":[74],"propose":[75],"HAWK":[76,81,131,161],"address":[78],"both":[79],"limitations.":[80],"uses":[82],"representation":[83],"similarity":[84],"select":[86],"informative":[87],"layers":[89],"and":[90,170,181,186],"learns":[91],"combine":[94],"their":[95],"hidden":[96,109],"states.":[97],"For":[98],"visual":[99,108,118,122],"information,":[100],"it":[101],"directly":[102],"provides":[103],"compressed":[107],"states":[110],"model":[114],"instead":[115],"of":[116],"raw":[117],"tokens,":[119],"making":[120],"information":[123],"easier":[124],"a":[126],"shallow":[127],"use.":[130],"also":[132],"trains":[133],"capture":[137],"change":[141],"its":[143,147],"proposals,":[145],"improving":[146],"agreement":[148],"during":[152],"multi-step":[153],"drafting.":[154],"On":[155],"SmolVLM-256M":[156],"across":[157],"ten":[158],"benchmarks,":[160],"raises":[162],"average":[163],"length":[165],"3.32":[167],"4.08":[169],"speedup":[171],"2.19x":[173,189],"2.60x":[175],"over":[176],"EAGLE-3":[177],"under":[178,190],"greedy":[179],"decoding,":[180],"2.89":[183],"3.41":[185],"1.92x":[187],"sampling.":[191]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-03T00:00:00"}