{"id":"https://openalex.org/W4416113121","doi":"https://doi.org/10.48550/arxiv.2503.04377","title":"How can representation dimension dominate structurally pruned LLMs?","display_name":"How can representation dimension dominate structurally pruned LLMs?","publication_year":2025,"publication_date":"2025-03-06","ids":{"openalex":"https://openalex.org/W4416113121","doi":"https://doi.org/10.48550/arxiv.2503.04377"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2503.04377","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.04377","pdf_url":"https://arxiv.org/pdf/2503.04377","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","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2503.04377","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5103433875","display_name":"Mingxue Xu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Mingxue","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5001337463","display_name":"Lisa Alazraki","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Alazraki, Lisa","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5103001848","display_name":"Danilo P. Mandic","orcid":"https://orcid.org/0000-0001-8432-3963","observed_orcids":["https://orcid.org/0000-0001-8432-3963"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mandic, Danilo P.","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/T10036","display_name":"Advanced Neural Network Applications","score":0.9717000126838684,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.9717000126838684,"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/T10028","display_name":"Topic Modeling","score":0.01769999973475933,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.0013000000035390258,"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/structured-pruning","display_name":"structured pruning","score":0.8730000257492065},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.8889999985694885},{"id":"https://openalex.org/keywords/representation-dimension","display_name":"representation dimension","score":0.9800000190734863},{"id":"https://openalex.org/keywords/embedding-dimension","display_name":"embedding dimension","score":0.8489999771118164},{"id":"https://openalex.org/keywords/subnetwork-extraction","display_name":"subnetwork extraction","score":0.6840000152587891},{"id":"https://openalex.org/keywords/linear-operators","display_name":"linear operators","score":0.5619999766349792},{"id":"https://openalex.org/keywords/transformer-blocks","display_name":"transformer blocks","score":0.7300000190734863},{"id":"https://openalex.org/keywords/perplexity","display_name":"perplexity","score":0.8339999914169312}],"concepts":[{"id":"https://openalex.org/C2780186347","wikidata":"https://www.wikidata.org/wiki/Q11414","display_name":"Subnetwork","level":2,"score":0.7996000051498413},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.6632000207901001},{"id":"https://openalex.org/C33676613","wikidata":"https://www.wikidata.org/wiki/Q13415176","display_name":"Dimension (graph theory)","level":2,"score":0.628000020980835},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.5813999772071838},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5782999992370605},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.5271000266075134},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.4772999882698059},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.45419999957084656},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4449000060558319},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4172999858856201},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.4041999876499176},{"id":"https://openalex.org/C30732413","wikidata":"https://www.wikidata.org/wiki/Q17092636","display_name":"Intrinsic dimension","level":3,"score":0.39800000190734863},{"id":"https://openalex.org/C163175372","wikidata":"https://www.wikidata.org/wiki/Q3339222","display_name":"Linear model","level":2,"score":0.3790999948978424},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.37310001254081726},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.3580000102519989},{"id":"https://openalex.org/C115311070","wikidata":"https://www.wikidata.org/wiki/Q5347255","display_name":"Effective dimension","level":3,"score":0.3481000065803528},{"id":"https://openalex.org/C136886441","wikidata":"https://www.wikidata.org/wiki/Q926129","display_name":"Normalization (sociology)","level":2,"score":0.335999995470047},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3052000105381012},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.3005000054836273},{"id":"https://openalex.org/C12362212","wikidata":"https://www.wikidata.org/wiki/Q728435","display_name":"Linear subspace","level":2,"score":0.3000999987125397},{"id":"https://openalex.org/C61445026","wikidata":"https://www.wikidata.org/wiki/Q217608","display_name":"Fixed point","level":2,"score":0.263700008392334},{"id":"https://openalex.org/C22789450","wikidata":"https://www.wikidata.org/wiki/Q420904","display_name":"Singular value decomposition","level":2,"score":0.26249998807907104},{"id":"https://openalex.org/C184720557","wikidata":"https://www.wikidata.org/wiki/Q7825049","display_name":"Topology (electrical circuits)","level":2,"score":0.2612999975681305},{"id":"https://openalex.org/C114289077","wikidata":"https://www.wikidata.org/wiki/Q3284399","display_name":"Statistical model","level":2,"score":0.25940001010894775}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2503.04377","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.04377","pdf_url":"https://arxiv.org/pdf/2503.04377","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.2503.04377","doi":"https://doi.org/10.48550/arxiv.2503.04377","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2503.04377","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:2503.04377","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.04377","pdf_url":"https://arxiv.org/pdf/2503.04377","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":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7010657036","https://openalex.org/W4415796952","https://openalex.org/W4402683876","https://openalex.org/W4412888794","https://openalex.org/W7172021105","https://openalex.org/W4399152007","https://openalex.org/W4391621451","https://openalex.org/W7212392430","https://openalex.org/W7140432523","https://openalex.org/W4416043746"],"abstract_inverted_index":{"Pruning":[0],"assumes":[1],"a":[2,79],"subnetwork":[3,35],"exists":[4],"in":[5,56,72],"the":[6,21,28,33,42,50,53,57,61,69,73,87,96,100,106,110,117,127],"original":[7],"deep":[8],"neural":[9],"network,":[10],"which":[11],"can":[12],"achieve":[13],"comparative":[14],"model":[15,29,48,101,113,118,129],"performance":[16,30,130],"with":[17,32,141],"less":[18],"computation":[19],"than":[20],"original.":[22],"However,":[23],"it":[24],"is":[25],"unclear":[26],"how":[27],"varies":[31],"different":[34,90],"extractions.":[36],"In":[37],"this":[38,65],"paper,":[39],"we":[40],"choose":[41],"representation":[43,91,107],"dimension":[44,51,108],"(or":[45],"embedding":[46],"dimension,":[47,49],"of":[52,89],"residual":[54],"stream":[55],"relevant":[58],"literature)":[59],"as":[60],"entry":[62],"point":[63],"to":[64,85,125],"issue.":[66],"We":[67,93],"investigate":[68],"linear":[70,111],"transformations":[71],"LLM":[74],"transformer":[75],"blocks":[76],"and":[77,104,132,137,143],"consider":[78],"specific":[80],"structured":[81],"pruning":[82],"approach,":[83],"SliceGPT,":[84],"extract":[86],"subnetworks":[88],"dimensions.":[92],"mechanistically":[94],"analyse":[95],"activation":[97],"flow":[98],"during":[99],"forward":[102],"passes,":[103],"find":[105],"dominates":[109],"transformations,":[112],"predictions,":[114],"and,":[115],"finally,":[116],"performance.":[119],"Explicit":[120],"analytical":[121],"relations":[122],"are":[123,138],"given":[124],"calculate":[126],"pruned":[128],"(perplexity":[131],"accuracy)":[133],"without":[134],"actual":[135],"evaluation,":[136],"empirically":[139],"validated":[140],"Llama-3-8B-Instruct":[142],"Phi-3-mini-4k-Instruct.":[144]},"counts_by_year":[],"updated_date":"2026-09-24T07:57:37.273135","created_date":"2025-10-10T00:00:00"}