{"id":"https://openalex.org/W4409248830","doi":"https://doi.org/10.1109/hpca61900.2025.00133","title":"eDKM: An Efficient and Accurate Train-Time Weight Clustering for Large Language Models","display_name":"eDKM: An Efficient and Accurate Train-Time Weight Clustering for Large Language Models","publication_year":2025,"publication_date":"2025-03-01","ids":{"openalex":"https://openalex.org/W4409248830","doi":"https://doi.org/10.1109/hpca61900.2025.00133"},"language":"en","primary_location":{"id":"doi:10.1109/hpca61900.2025.00133","is_oa":false,"landing_page_url":"https://doi.org/10.1109/hpca61900.2025.00133","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE International Symposium on High Performance Computer Architecture (HPCA)","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5115076654","display_name":"Minsik Cho","orcid":"https://orcid.org/0000-0003-0481-2682"},"institutions":[{"id":"https://openalex.org/I4210153776","display_name":"Apple (United States)","ror":"https://ror.org/059hsda18","country_code":"US","type":"company","lineage":["https://openalex.org/I4210153776"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Minsik Cho","raw_affiliation_strings":["Apple,USA"],"affiliations":[{"raw_affiliation_string":"Apple,USA","institution_ids":["https://openalex.org/I4210153776"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5023280155","display_name":"Keivan Alizadeh Vahid","orcid":null},"institutions":[{"id":"https://openalex.org/I4210153776","display_name":"Apple (United States)","ror":"https://ror.org/059hsda18","country_code":"US","type":"company","lineage":["https://openalex.org/I4210153776"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Keivan A. Vahid","raw_affiliation_strings":["Apple,USA"],"affiliations":[{"raw_affiliation_string":"Apple,USA","institution_ids":["https://openalex.org/I4210153776"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101753713","display_name":"Qichen Fu","orcid":"https://orcid.org/0009-0008-5976-4095"},"institutions":[{"id":"https://openalex.org/I4210153776","display_name":"Apple (United States)","ror":"https://ror.org/059hsda18","country_code":"US","type":"company","lineage":["https://openalex.org/I4210153776"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Qichen Fu","raw_affiliation_strings":["Apple,USA"],"affiliations":[{"raw_affiliation_string":"Apple,USA","institution_ids":["https://openalex.org/I4210153776"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5068392026","display_name":"Saurabh Adya","orcid":"https://orcid.org/0009-0000-4533-6577"},"institutions":[{"id":"https://openalex.org/I4210153776","display_name":"Apple (United States)","ror":"https://ror.org/059hsda18","country_code":"US","type":"company","lineage":["https://openalex.org/I4210153776"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Saurabh Adya","raw_affiliation_strings":["Apple,USA"],"affiliations":[{"raw_affiliation_string":"Apple,USA","institution_ids":["https://openalex.org/I4210153776"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5073050611","display_name":"Carlo C. Del Mundo","orcid":null},"institutions":[{"id":"https://openalex.org/I4210153776","display_name":"Apple (United States)","ror":"https://ror.org/059hsda18","country_code":"US","type":"company","lineage":["https://openalex.org/I4210153776"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Carlo C. Del Mundo","raw_affiliation_strings":["Apple,USA"],"affiliations":[{"raw_affiliation_string":"Apple,USA","institution_ids":["https://openalex.org/I4210153776"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5056246621","display_name":"Mohammad Rastegari","orcid":"https://orcid.org/0000-0001-9606-3687"},"institutions":[{"id":"https://openalex.org/I4210153776","display_name":"Apple (United States)","ror":"https://ror.org/059hsda18","country_code":"US","type":"company","lineage":["https://openalex.org/I4210153776"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Mohammad Rastegari","raw_affiliation_strings":["Apple,USA"],"affiliations":[{"raw_affiliation_string":"Apple,USA","institution_ids":["https://openalex.org/I4210153776"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113880378","display_name":"Devang Naik","orcid":"https://orcid.org/0009-0007-7838-1623"},"institutions":[{"id":"https://openalex.org/I4210153776","display_name":"Apple (United States)","ror":"https://ror.org/059hsda18","country_code":"US","type":"company","lineage":["https://openalex.org/I4210153776"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Devang Naik","raw_affiliation_strings":["Apple,USA"],"affiliations":[{"raw_affiliation_string":"Apple,USA","institution_ids":["https://openalex.org/I4210153776"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5031481674","display_name":"Peter Zatloukal","orcid":"https://orcid.org/0009-0000-0451-0591"},"institutions":[{"id":"https://openalex.org/I4210153776","display_name":"Apple (United States)","ror":"https://ror.org/059hsda18","country_code":"US","type":"company","lineage":["https://openalex.org/I4210153776"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Peter Zatloukal","raw_affiliation_strings":["Apple,USA"],"affiliations":[{"raw_affiliation_string":"Apple,USA","institution_ids":["https://openalex.org/I4210153776"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":8,"corresponding_author_ids":["https://openalex.org/A5115076654"],"corresponding_institution_ids":["https://openalex.org/I4210153776"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.03355375,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"323","last_page":"323"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9891999959945679,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9891999959945679,"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/T10028","display_name":"Topic Modeling","score":0.9857000112533569,"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/T10203","display_name":"Recommender Systems and Techniques","score":0.9394000172615051,"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/computer-science","display_name":"Computer science","score":0.7901099324226379},{"id":"https://openalex.org/keywords/cluster-analysis","display_name":"Cluster analysis","score":0.6862651109695435},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4383648633956909},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.3621693551540375}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7901099324226379},{"id":"https://openalex.org/C73555534","wikidata":"https://www.wikidata.org/wiki/Q622825","display_name":"Cluster analysis","level":2,"score":0.6862651109695435},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4383648633956909},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3621693551540375}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/hpca61900.2025.00133","is_oa":false,"landing_page_url":"https://doi.org/10.1109/hpca61900.2025.00133","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE International Symposium on High Performance Computer Architecture (HPCA)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Since":[0],"Large":[1],"Language":[2],"Models":[3],"or":[4,90],"LLMs":[5,23,38],"have":[6],"demonstrated":[7],"high-quality":[8],"performance":[9],"on":[10,150,224],"many":[11,53],"complex":[12],"language":[13],"tasks,":[14],"there":[15,170],"is":[16,62,80,171],"a":[17,57,123,144,189,214],"great":[18],"interest":[19],"in":[20],"bringing":[21],"these":[22],"to":[24,47,112,114,132,147,177,197],"mobile":[25],"devices":[26],"for":[27,68,83,152,231,235],"faster":[28],"responses":[29],"and":[30,71,100,165,187,237],"better":[31],"privacy":[32],"protection.":[33],"However,":[34],"the":[35,65,94,134,153,160,204,209],"size":[36],"of":[37,41,59,64,137,141,156,213],"(i.e.,":[39,228],"billions":[40],"parameters)":[42],"requires":[43],"highly":[44],"effective":[45],"compression":[46,54,98],"fit":[48],"into":[49],"storage-limited":[50],"devices.":[51],"Among":[52],"techniques,":[55],"weight-clustering,":[56],"form":[58],"non-linear":[60],"quantization,":[61],"one":[63],"leading":[66],"candidates":[67],"LLM":[69,84,116,226],"compression,":[70],"supported":[72],"by":[73,129,139,162,207,217],"modern":[74],"smartphones.":[75],"Yet,":[76],"its":[77,104],"training":[78],"overhead":[79],"prohibitively":[81],"significant":[82],"fine-tuning.":[85],"Especially,":[86],"Differentiable":[87],"KMeans":[88],"Clustering,":[89],"DKM,":[91,157],"has":[92],"shown":[93],"state-of-the-art":[95],"trade-off":[96],"between":[97],"ratio":[99],"accuracy":[101,223],"regression,":[102],"but":[103],"large":[105],"memory":[106,135,211],"complexity":[107],"makes":[108],"it":[109],"nearly":[110],"impossible":[111],"apply":[113],"train-time":[115,210],"compression.":[117],"In":[118],"this":[119],"letter,":[120],"we":[121,158],"propose":[122],"memory-efficient":[124],"DKM":[125,138],"implementation,":[126],"eDKM":[127,184],"powered":[128],"novel":[130],"techniques":[131],"reduce":[133],"footprint":[136,212],"orders":[140],"magnitudes.":[142],"For":[143],"given":[145],"tensor":[146,161,174],"be":[148],"saved":[149],"CPU":[151],"backward":[154],"pass":[155],"compressed":[159],"applying":[163],"uniquification":[164],"sharding":[166],"after":[167],"checking":[168],"if":[169],"no":[172],"duplicated":[173],"previously":[175],"copied":[176],"CPU.":[178],"Our":[179],"experimental":[180],"results":[181],"demonstrate":[182],"that":[183],"can":[185],"fine-tune":[186],"compress":[188],"pretrained":[190],"LLaMA":[191],"7B":[192],"model":[193],"from":[194],"12.6":[195],"GB":[196],"$2.5":[198],"\\mathrm{~GB}(3":[199],"\\mathrm{~b}":[200],"/$":[201],"weight)":[202],"with":[203],"Alpaca":[205],"dataset":[206],"reducing":[208],"decoder":[215],"layer":[216],"$130":[218],"\\times$,":[219],"while":[220],"delivering":[221],"good":[222],"broader":[225],"benchmarks":[227],"$77.7":[229],"\\%$":[230,234],"PIQA,":[232],"$66.1":[233],"Winograde,":[236],"so":[238],"on).":[239]},"counts_by_year":[],"updated_date":"2025-11-06T03:46:38.306776","created_date":"2025-10-10T00:00:00"}