{"id":"https://openalex.org/W7161254912","doi":"https://doi.org/10.48550/arxiv.2605.14458","title":"OmniDrop: Layer-wise Token Pruning for Omni-modal LLMs via Query-Guidance","display_name":"OmniDrop: Layer-wise Token Pruning for Omni-modal LLMs via Query-Guidance","publication_year":2026,"publication_date":"2026-05-14","ids":{"openalex":"https://openalex.org/W7161254912","doi":"https://doi.org/10.48550/arxiv.2605.14458"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.14458","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.14458","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.2605.14458","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5125231142","display_name":"Yeo Jeong Park","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Park, Yeo Jeong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5056602268","display_name":"Hyemi Jang","orcid":"https://orcid.org/0000-0002-7736-0528"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jang, Hyemi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128748979","display_name":"Minseo Choi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Choi, Minseo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136221202","display_name":"Jongsun Lee","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lee, Jongsun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136259609","display_name":"Jooyoung Choi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Choi, Jooyoung","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5112636409","display_name":"Yongkweon Jeon","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jeon, Yongkweon","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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.7087000012397766,"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":0.7087000012397766,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.057100001722574234,"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/T10860","display_name":"Speech and Audio Processing","score":0.05000000074505806,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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":0.9161999821662903},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.5565000176429749},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.5343999862670898},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.5038999915122986},{"id":"https://openalex.org/keywords/embedding","display_name":"Embedding","score":0.4699000120162964},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.42399999499320984},{"id":"https://openalex.org/keywords/similarity","display_name":"Similarity (geometry)","score":0.335999995470047}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.9161999821662903},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8098000288009644},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.5565000176429749},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.5343999862670898},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.5038999915122986},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.4699000120162964},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.43849998712539673},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.42399999499320984},{"id":"https://openalex.org/C103278499","wikidata":"https://www.wikidata.org/wiki/Q254465","display_name":"Similarity (geometry)","level":3,"score":0.335999995470047},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.32739999890327454},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3264000117778778},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.30090001225471497},{"id":"https://openalex.org/C113954288","wikidata":"https://www.wikidata.org/wiki/Q186885","display_name":"Timestamp","level":2,"score":0.28439998626708984},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.2816999852657318},{"id":"https://openalex.org/C46637626","wikidata":"https://www.wikidata.org/wiki/Q6693015","display_name":"Low latency (capital markets)","level":2,"score":0.27900001406669617},{"id":"https://openalex.org/C115067241","wikidata":"https://www.wikidata.org/wiki/Q1639854","display_name":"Token passing","level":3,"score":0.2728999853134155},{"id":"https://openalex.org/C159423971","wikidata":"https://www.wikidata.org/wiki/Q177251","display_name":"Associative property","level":2,"score":0.26989999413490295},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.26589998602867126},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2533000111579895},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2526000142097473}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.14458","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.14458","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.2605.14458","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.14458","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":[{"score":0.5479639172554016,"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Omni-modal":[0],"large":[1],"language":[2],"models":[3],"have":[4],"demonstrated":[5],"remarkable":[6],"potential":[7],"in":[8,106],"holistic":[9],"multimodal":[10],"understanding;":[11],"however,":[12],"the":[13,42,84,91],"token":[14,35,75,120,132],"explosion":[15],"caused":[16],"by":[17,151,160,167],"high-resolution":[18],"audio":[19],"and":[20,30,118,164],"video":[21],"inputs":[22],"remains":[23],"a":[24,72,125],"critical":[25],"bottleneck":[26],"for":[27,55,116],"real-time":[28],"applications":[29],"long-form":[31],"reasoning.":[32],"Existing":[33],"omni-modal":[34,99],"compression":[36],"methods":[37],"typically":[38],"prune":[39],"tokens":[40,82,105],"at":[41,90],"input":[43],"embedding":[44],"level,":[45],"relying":[46],"on":[47],"audio-video":[48],"similarity":[49],"or":[50],"temporal":[51,126,137],"co-occurrence":[52],"as":[53,114],"proxies":[54],"semantic":[56],"relevance.":[57],"In":[58],"practice,":[59],"such":[60],"assumptions":[61],"are":[62],"often":[63],"unreliable.":[64],"To":[65],"address":[66],"this":[67],"limitation,":[68],"we":[69],"propose":[70],"OmniDrop,":[71],"training-free,":[73],"layer-wise":[74],"pruning":[76],"framework":[77],"that":[78,129,146],"progressively":[79],"prunes":[80],"audiovisual":[81,143],"within":[83],"LLM":[85],"decoder":[86],"layers":[87,95],"rather":[88],"than":[89],"input-level,":[92],"allowing":[93],"early":[94],"to":[96,134,153,162,169],"preserve":[97,135],"sufficient":[98],"information":[100],"fusion":[101],"before":[102],"aggressively":[103],"removing":[104],"deeper":[107],"layers.":[108],"We":[109,122],"further":[110],"utilize":[111],"text":[112],"queries":[113],"guidance":[115],"modality-agnostic":[117],"task-adaptive":[119],"pruning.":[121],"also":[123],"introduce":[124],"diversity":[127],"score":[128],"encourages":[130],"balanced":[131],"survival":[133],"global":[136],"context.":[138],"Experimental":[139],"results":[140],"across":[141],"various":[142],"benchmarks":[144],"demonstrate":[145],"OmniDrop":[147],"outperforms":[148],"all":[149],"baselines":[150],"up":[152,161,168],"3.58":[154],"points":[155],"while":[156],"reducing":[157],"prefill":[158],"latency":[159],"40%":[163],"memory":[165],"usage":[166],"14.7%.":[170]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-16T00:00:00"}