{"id":"https://openalex.org/W7162441280","doi":"https://doi.org/10.48550/arxiv.2605.23969","title":"SLAP: Stratified Loss-based Pruning for On-Policy Data-Efficient Instruction Tuning","display_name":"SLAP: Stratified Loss-based Pruning for On-Policy Data-Efficient Instruction Tuning","publication_year":2026,"publication_date":"2026-05-13","ids":{"openalex":"https://openalex.org/W7162441280","doi":"https://doi.org/10.48550/arxiv.2605.23969"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.23969","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.23969","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.2605.23969","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137058011","display_name":"Run Zou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zou, Run","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137034379","display_name":"Jianhang Ding","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ding, Jianhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136999594","display_name":"Yifan Ding","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ding, Yifan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137052373","display_name":"Wen Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Wen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137031786","display_name":"Hao Chen","orcid":"https://orcid.org/0000-0003-1017-7136"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Hao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5045766331","display_name":"Renshu Gu","orcid":"https://orcid.org/0000-0002-3900-2148"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gu, Renshu","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":0.6069999933242798,"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":0.6069999933242798,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.16329999268054962,"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":0.04960000142455101,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.696399986743927},{"id":"https://openalex.org/keywords/lossless-compression","display_name":"Lossless compression","score":0.5562000274658203},{"id":"https://openalex.org/keywords/learnability","display_name":"Learnability","score":0.484499990940094},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4726000130176544},{"id":"https://openalex.org/keywords/selection","display_name":"Selection (genetic algorithm)","score":0.45100000500679016},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.39169999957084656}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8317000269889832},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.696399986743927},{"id":"https://openalex.org/C81081738","wikidata":"https://www.wikidata.org/wiki/Q55542","display_name":"Lossless compression","level":3,"score":0.5562000274658203},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.49900001287460327},{"id":"https://openalex.org/C2777723229","wikidata":"https://www.wikidata.org/wiki/Q4367921","display_name":"Learnability","level":2,"score":0.484499990940094},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4726000130176544},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4657999873161316},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.45100000500679016},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.4050000011920929},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.39169999957084656},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.3172999918460846},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.2913999855518341},{"id":"https://openalex.org/C2778915421","wikidata":"https://www.wikidata.org/wiki/Q3643177","display_name":"Performance improvement","level":2,"score":0.2768999934196472},{"id":"https://openalex.org/C93959086","wikidata":"https://www.wikidata.org/wiki/Q6888345","display_name":"Model selection","level":2,"score":0.26460000872612},{"id":"https://openalex.org/C26713055","wikidata":"https://www.wikidata.org/wiki/Q245962","display_name":"Implementation","level":2,"score":0.2612000107765198},{"id":"https://openalex.org/C137800194","wikidata":"https://www.wikidata.org/wiki/Q11713455","display_name":"Interpolation (computer graphics)","level":3,"score":0.2547000050544739}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.23969","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.23969","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.2605.23969","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.23969","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","score":0.5137379169464111,"id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Instruction":[0],"tuning":[1,164],"has":[2],"optimized":[3],"the":[4,65],"specialized":[5],"capabilities":[6,28],"of":[7,67,165],"large":[8,166],"language":[9,167],"models":[10,43],"(LLMs),":[11],"but":[12],"it":[13],"often":[14],"requires":[15],"extensive":[16],"datasets":[17],"and":[18,33,37,113,122,161],"prolonged":[19],"training":[20,134],"times.":[21],"The":[22],"challenge":[23],"lies":[24],"in":[25],"developing":[26],"specific":[27],"by":[29],"identifying":[30],"useful":[31],"data":[32,40,60,77,135],"efficiently":[34],"fine-tuning.":[35],"High-quality":[36],"diverse":[38,114],"pruned":[39],"can":[41],"help":[42],"achieve":[44],"lossless":[45],"performance":[46,130],"at":[47],"a":[48,57,156],"lower":[49],"cost.":[50],"In":[51],"this":[52],"paper,":[53],"we":[54],"propose":[55],"\\textbf{SLAP},":[56],"novel":[58],"batch-aware":[59],"selection":[61],"framework":[62],"that":[63],"evaluates":[64],"learnability":[66],"entire":[68],"batch":[69,99],"compositions":[70],"rather":[71],"than":[72],"individual.":[73],"SLAP":[74,101,127,154],"ensures":[75],"comprehensive":[76],"distribution":[78],"coverage":[79],"through":[80,88],"distribution-aware":[81],"stratified":[82],"sampling":[83],"while":[84,145],"maximizing":[85],"intra-batch":[86],"diversity":[87],"relative":[89],"distance":[90],"optimization.":[91],"By":[92],"leveraging":[93],"Hessian-approximated":[94],"gradient":[95],"information":[96],"for":[97,159],"dynamic":[98],"selection,":[100],"significantly":[102],"outperforms":[103],"existing":[104],"state-of-the-art":[105],"methods":[106],"across":[107],"multiple":[108],"model":[109,149],"architectures":[110],"(LLaMA,":[111],"ChatGLM)":[112],"downstream":[115],"tasks":[116],"including":[117],"multi-turn":[118],"dialogue,":[119],"multilingual":[120],"translation,":[121],"question":[123],"answering.":[124],"Most":[125],"notably,":[126],"achieves":[128],"superior":[129],"with":[131],"20-40\\%":[132],"less":[133],"compared":[136],"to":[137],"full":[138],"dataset":[139],"training,":[140],"substantially":[141],"reducing":[142],"computational":[143],"costs":[144],"maintaining":[146],"or":[147],"improving":[148],"capabilities.":[150],"These":[151],"results":[152],"establish":[153],"as":[155],"powerful":[157],"approach":[158],"efficient":[160],"effective":[162],"instruction":[163],"models.":[168]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-27T00:00:00"}