{"id":"https://openalex.org/W7160170738","doi":"https://doi.org/10.48550/arxiv.2605.00392","title":"RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference","display_name":"RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference","publication_year":2026,"publication_date":"2026-05-01","ids":{"openalex":"https://openalex.org/W7160170738","doi":"https://doi.org/10.48550/arxiv.2605.00392"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.00392","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00392","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.2605.00392","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135197865","display_name":"Ben Wan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wan, Ben","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135253900","display_name":"Yan Feng","orcid":"https://orcid.org/0009-0005-7088-8513"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Feng, Yan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135243858","display_name":"Zihan Tang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tang, Zihan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101832257","display_name":"Weizhe Huang","orcid":"https://orcid.org/0000-0002-0844-9350"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Weizhe","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135113255","display_name":"Yuting Zeng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zeng, Yuting","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135175349","display_name":"Jia Wang","orcid":"https://orcid.org/0000-0003-1916-5086"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Jia","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135248514","display_name":"Tongxuan Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Tongxuan","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":0.29429998993873596,"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":0.29429998993873596,"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/T10601","display_name":"Handwritten Text Recognition Techniques","score":0.28540000319480896,"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":0.12600000202655792,"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/security-token","display_name":"Security token","score":0.8427000045776367},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.7616000175476074},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.5156999826431274},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.5083000063896179},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5024999976158142},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.48919999599456787},{"id":"https://openalex.org/keywords/salient","display_name":"Salient","score":0.4636000096797943},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.4359000027179718}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.8427000045776367},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8003000020980835},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.7616000175476074},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5386999845504761},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.5156999826431274},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.5083000063896179},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5024999976158142},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.48919999599456787},{"id":"https://openalex.org/C2780719617","wikidata":"https://www.wikidata.org/wiki/Q1030752","display_name":"Salient","level":2,"score":0.4636000096797943},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.4359000027179718},{"id":"https://openalex.org/C103278499","wikidata":"https://www.wikidata.org/wiki/Q254465","display_name":"Similarity (geometry)","level":3,"score":0.3986999988555908},{"id":"https://openalex.org/C2776459999","wikidata":"https://www.wikidata.org/wiki/Q2119376","display_name":"Fidelity","level":2,"score":0.3840000033378601},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.3553999960422516},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.34880000352859497},{"id":"https://openalex.org/C554936623","wikidata":"https://www.wikidata.org/wiki/Q199657","display_name":"Reading (process)","level":2,"score":0.3199999928474426},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.3050999939441681},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.29100000858306885},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2892000079154968},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.2874999940395355},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.2809000015258789},{"id":"https://openalex.org/C25797200","wikidata":"https://www.wikidata.org/wiki/Q828137","display_name":"Compression ratio","level":3,"score":0.2791000008583069},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.257099986076355},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.25270000100135803}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.00392","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00392","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.2605.00392","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00392","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":[{"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education","score":0.7398396134376526}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"DeepSeek-OCR":[0],"leverages":[1],"visual-text":[2],"compression":[3,41],"to":[4,17,34,39,72,123,137,174],"reduce":[5],"long-text":[6],"processing":[7],"costs":[8],"and":[9,20,104,116,140,161],"accelerate":[10],"inference,":[11],"yet":[12],"visual":[13,98],"tokens":[14,99,113],"remain":[15],"prone":[16],"redundant":[18],"textual":[19,36,103,141],"structural":[21,105],"information.":[22,106],"Moreover,":[23],"current":[24],"token":[25,85,138,170],"pruning":[26,86,133],"methods":[27],"for":[28,89,143],"conventional":[29],"vision-language":[30],"models":[31],"(VLMs)":[32],"fail":[33],"preserve":[35],"fidelity":[37],"due":[38],"improper":[40],"mechanisms.":[42],"By":[43],"analyzing":[44],"the":[45,58,62,73,92,108,111],"decoding":[46],"process":[47],"of":[48,64],"DeepSeek-OCR,":[49],"we":[50,80,95],"find":[51],"that":[52,100,135],"a":[53,83,131,147],"distinct":[54],"two-stage":[55,84],"reading":[56],"trajectory:":[57],"model":[59],"initially":[60],"prioritizes":[61],"majority":[63],"high-norm":[65,97],"tokens,":[66],"then":[67],"subsequently":[68],"redistributes":[69],"its":[70],"attention":[71],"remaining":[74,112],"ones.":[75],"Motivated":[76],"by":[77,158],"this":[78],"insight,":[79],"propose":[81],"RTPrune,":[82],"method":[87],"tailored":[88],"DeepSeek-OCR.":[90],"In":[91,107],"first":[93],"stage,":[94,110],"prioritize":[96],"capture":[101],"salient":[102],"second":[109],"are":[114],"paired":[115],"merged":[117],"based":[118],"on":[119,165],"optimal":[120],"transport":[121],"theory":[122],"achieve":[124],"efficient":[125],"feature":[126],"aggregation.":[127],"We":[128],"further":[129],"introduce":[130],"dynamic":[132],"ratio":[134],"adapts":[136],"similarity":[139],"density":[142],"OCR":[144],"tasks,":[145],"enabling":[146],"better":[148],"efficiency-accuracy":[149],"trade-off.":[150],"Extensive":[151],"experiments":[152],"demonstrate":[153],"state-of-the-art":[154],"performance,":[155],"as":[156],"evidenced":[157],"99.47%":[159],"accuracy":[160],"1.23$\\times$":[162],"faster":[163],"prefill":[164],"OmniDocBench,":[166],"achieved":[167],"with":[168],"84.25%":[169],"retention":[171],"when":[172],"applied":[173],"DeepSeek-OCR-Large.":[175]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-05T00:00:00"}