{"id":"https://openalex.org/W7162698038","doi":"https://doi.org/10.48550/arxiv.2605.28042","title":"Extracting Small Translation Specialists from LLMs by Aggressively Pruning Experts","display_name":"Extracting Small Translation Specialists from LLMs by Aggressively Pruning Experts","publication_year":2026,"publication_date":"2026-05-27","ids":{"openalex":"https://openalex.org/W7162698038","doi":"https://doi.org/10.48550/arxiv.2605.28042"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.28042","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28042","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.2605.28042","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137305962","display_name":"Liu O. Martin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Martin, Liu O.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137297730","display_name":"Lucas Bandarkar","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bandarkar, Lucas","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137278938","display_name":"Nanyun Peng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Peng, Nanyun","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":0.2743000090122223,"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":0.2743000090122223,"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":0.22349999845027924,"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/T14347","display_name":"Big Data and Digital Economy","score":0.06759999692440033,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/pruning","display_name":"Pruning","score":0.7893000245094299},{"id":"https://openalex.org/keywords/translation","display_name":"Translation (biology)","score":0.6579999923706055},{"id":"https://openalex.org/keywords/machine-translation","display_name":"Machine translation","score":0.5830000042915344},{"id":"https://openalex.org/keywords/exploit","display_name":"Exploit","score":0.5651000142097473},{"id":"https://openalex.org/keywords/baseline","display_name":"Baseline (sea)","score":0.4936999976634979},{"id":"https://openalex.org/keywords/modular-design","display_name":"Modular design","score":0.4869000017642975},{"id":"https://openalex.org/keywords/fraction","display_name":"Fraction (chemistry)","score":0.42640000581741333}],"concepts":[{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.7893000245094299},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6967999935150146},{"id":"https://openalex.org/C149364088","wikidata":"https://www.wikidata.org/wiki/Q185917","display_name":"Translation (biology)","level":4,"score":0.6579999923706055},{"id":"https://openalex.org/C203005215","wikidata":"https://www.wikidata.org/wiki/Q79798","display_name":"Machine translation","level":2,"score":0.5830000042915344},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.5651000142097473},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5383999943733215},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.4936999976634979},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.48829999566078186},{"id":"https://openalex.org/C101468663","wikidata":"https://www.wikidata.org/wiki/Q1620158","display_name":"Modular design","level":2,"score":0.4869000017642975},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.446399986743927},{"id":"https://openalex.org/C149629883","wikidata":"https://www.wikidata.org/wiki/Q660926","display_name":"Fraction (chemistry)","level":2,"score":0.42640000581741333},{"id":"https://openalex.org/C2779760435","wikidata":"https://www.wikidata.org/wiki/Q5396169","display_name":"Minor (academic)","level":2,"score":0.35589998960494995},{"id":"https://openalex.org/C189430467","wikidata":"https://www.wikidata.org/wiki/Q7293293","display_name":"Ranking (information retrieval)","level":2,"score":0.3375999927520752},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.30329999327659607},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.2680000066757202},{"id":"https://openalex.org/C31170391","wikidata":"https://www.wikidata.org/wiki/Q188619","display_name":"Hierarchy","level":2,"score":0.26409998536109924},{"id":"https://openalex.org/C62230096","wikidata":"https://www.wikidata.org/wiki/Q275969","display_name":"Crowdsourcing","level":2,"score":0.2522999942302704}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.28042","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28042","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.2605.28042","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28042","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":[{"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4","score":0.7192217707633972}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Modern":[0],"large":[1],"language":[2],"models":[3],"(LLMs)":[4],"achieve":[5],"state-of-the-art":[6],"machine":[7],"translation":[8,62,143,150],"performance,":[9,132],"but":[10],"they":[11,28],"do":[12],"so":[13],"as":[14],"broad":[15],"generalists":[16],"largely":[17],"trained":[18],"for":[19,32,49],"many":[20],"tasks":[21],"and":[22,39,69,113,133],"capabilities":[23,74],"unrelated":[24],"to":[25,77,81,104],"translation.":[26,82],"Thus,":[27],"are":[29,102],"heavily":[30],"overparameterized":[31],"this":[33,43],"task,":[34],"resulting":[35],"in":[36,61,75,134],"excessive":[37],"memory":[38],"compute":[40],"requirements.":[41],"In":[42],"paper,":[44],"we":[45,101,124],"present":[46],"a":[47,120,153],"method":[48],"aggressively":[50],"pruning":[51],"experts":[52,79,109,128],"from":[53],"modern":[54],"mixture-of-experts":[55],"LLMs":[56,76],"while":[57,129,140],"incurring":[58],"negligible":[59,111],"degradation":[60,112],"quality.":[63,144],"Our":[64],"approach":[65],"exploits":[66],"expert":[67],"specialization":[68],"the":[70,86,156,162],"separability":[71],"of":[72,85,89,107,127,155,161,169],"multilingual":[73],"identify":[78],"irrelevant":[80],"And":[83],"because":[84],"modular":[87],"nature":[88],"MoEs,":[90],"these":[91],"can":[92],"be":[93],"easily":[94],"pruned":[95],"without":[96],"any":[97],"training.":[98],"Without":[99],"retraining,":[100],"able":[103],"prune":[105,125],"half":[106],"all":[108],"with":[110,115],"70%":[114],"only":[116,152],"minor":[117],"losses.":[118],"With":[119],"very":[121],"short":[122],"SFT,":[123],"75%":[126],"recovering":[130],"baseline":[131],"some":[135],"settings":[136],"remove":[137],"nearly":[138],"90%":[139,168],"maintaining":[141],"reasonable":[142],"Overall,":[145],"our":[146],"results":[147],"show":[148],"that":[149,165],"requires":[151],"fraction":[154],"LLM,":[157],"enabling":[158],"substantial":[159],"compression":[160],"MoE":[163],"blocks":[164],"contain":[166],"over":[167],"parameters.":[170]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-29T00:00:00"}