{"id":"https://openalex.org/W4409637743","doi":"https://doi.org/10.3390/fi17040185","title":"Mitigating Quantization Errors Due to Activation Spikes in Gated Linear Unit-Based Large Language Models","display_name":"Mitigating Quantization Errors Due to Activation Spikes in Gated Linear Unit-Based Large Language Models","publication_year":2025,"publication_date":"2025-04-21","ids":{"openalex":"https://openalex.org/W4409637743","doi":"https://doi.org/10.3390/fi17040185"},"language":"en","primary_location":{"id":"doi:10.3390/fi17040185","is_oa":true,"landing_page_url":"https://doi.org/10.3390/fi17040185","pdf_url":"https://www.mdpi.com/1999-5903/17/4/185/pdf?version=1745240701","source":{"id":"https://openalex.org/S34838331","display_name":"Future Internet","issn_l":"1999-5903","issn":["1999-5903"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310310987","host_organization_name":"Multidisciplinary Digital Publishing Institute","host_organization_lineage":["https://openalex.org/P4310310987"],"host_organization_lineage_names":["Multidisciplinary Digital Publishing Institute"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Future Internet","raw_type":"journal-article"},"type":"article","indexed_in":["crossref","doaj"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://www.mdpi.com/1999-5903/17/4/185/pdf?version=1745240701","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5056062126","display_name":"Jae-Woo Yang","orcid":"https://orcid.org/0000-0002-7687-026X"},"institutions":[{"id":"https://openalex.org/I4575257","display_name":"Hanyang University","ror":"https://ror.org/046865y68","country_code":"KR","type":"education","lineage":["https://openalex.org/I4575257"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Jaewoo Yang","raw_affiliation_strings":["Department of Applied Artificial Intelligence, Hanyang University at Ansan, Ansan 15588, Republic of Korea"],"affiliations":[{"raw_affiliation_string":"Department of Applied Artificial Intelligence, Hanyang University at Ansan, Ansan 15588, Republic of Korea","institution_ids":["https://openalex.org/I4575257"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111958658","display_name":"Hayun Kim","orcid":null},"institutions":[{"id":"https://openalex.org/I4575257","display_name":"Hanyang University","ror":"https://ror.org/046865y68","country_code":"KR","type":"education","lineage":["https://openalex.org/I4575257"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Hayun Kim","raw_affiliation_strings":["Department of Applied Artificial Intelligence, Hanyang University at Ansan, Ansan 15588, Republic of Korea"],"affiliations":[{"raw_affiliation_string":"Department of Applied Artificial Intelligence, Hanyang University at Ansan, Ansan 15588, Republic of Korea","institution_ids":["https://openalex.org/I4575257"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5037061933","display_name":"Junyung Ji","orcid":null},"institutions":[{"id":"https://openalex.org/I4575257","display_name":"Hanyang University","ror":"https://ror.org/046865y68","country_code":"KR","type":"education","lineage":["https://openalex.org/I4575257"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Junyung Ji","raw_affiliation_strings":["Department of Applied Artificial Intelligence, Hanyang University at Ansan, Ansan 15588, Republic of Korea"],"affiliations":[{"raw_affiliation_string":"Department of Applied Artificial Intelligence, Hanyang University at Ansan, Ansan 15588, Republic of Korea","institution_ids":["https://openalex.org/I4575257"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100327650","display_name":"Younghoon Kim","orcid":"https://orcid.org/0000-0002-3049-035X"},"institutions":[{"id":"https://openalex.org/I4575257","display_name":"Hanyang University","ror":"https://ror.org/046865y68","country_code":"KR","type":"education","lineage":["https://openalex.org/I4575257"]}],"countries":["KR"],"is_corresponding":true,"raw_author_name":"Younghoon Kim","raw_affiliation_strings":["Department of Applied Artificial Intelligence, Hanyang University at Ansan, Ansan 15588, Republic of Korea"],"affiliations":[{"raw_affiliation_string":"Department of Applied Artificial Intelligence, Hanyang University at Ansan, Ansan 15588, Republic of Korea","institution_ids":["https://openalex.org/I4575257"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":["https://openalex.org/A5100327650"],"corresponding_institution_ids":["https://openalex.org/I4575257"],"apc_list":{"value":1400,"currency":"CHF","value_usd":1515},"apc_paid":{"value":1400,"currency":"CHF","value_usd":1515},"fwci":2.3431,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":{"value":0.88651059,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":91,"max":95},"biblio":{"volume":"17","issue":"4","first_page":"185","last_page":"185"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9460999965667725,"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":0.9460999965667725,"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/computer-science","display_name":"Computer science","score":0.8791015148162842},{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.6506805419921875},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.4626288115978241},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4532895088195801},{"id":"https://openalex.org/keywords/speech-recognition","display_name":"Speech recognition","score":0.40814638137817383},{"id":"https://openalex.org/keywords/real-time-computing","display_name":"Real-time computing","score":0.34072357416152954}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8791015148162842},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.6506805419921875},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4626288115978241},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4532895088195801},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.40814638137817383},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.34072357416152954}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.3390/fi17040185","is_oa":true,"landing_page_url":"https://doi.org/10.3390/fi17040185","pdf_url":"https://www.mdpi.com/1999-5903/17/4/185/pdf?version=1745240701","source":{"id":"https://openalex.org/S34838331","display_name":"Future Internet","issn_l":"1999-5903","issn":["1999-5903"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310310987","host_organization_name":"Multidisciplinary Digital Publishing Institute","host_organization_lineage":["https://openalex.org/P4310310987"],"host_organization_lineage_names":["Multidisciplinary Digital Publishing Institute"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Future Internet","raw_type":"journal-article"},{"id":"pmh:oai:RePEc:gam:jftint:v:17:y:2025:i:4:p:185-:d:1639317","is_oa":false,"landing_page_url":"https://www.mdpi.com/1999-5903/17/4/185/","pdf_url":null,"source":{"id":"https://openalex.org/S4306401271","display_name":"RePEc: Research Papers in Economics","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I77793887","host_organization_name":"Federal Reserve Bank of St. Louis","host_organization_lineage":["https://openalex.org/I77793887"],"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":"article"},{"id":"pmh:oai:doaj.org/article:d2c4dfd3622b4535bb87127c9d1b37ff","is_oa":true,"landing_page_url":"https://doaj.org/article/d2c4dfd3622b4535bb87127c9d1b37ff","pdf_url":null,"source":{"id":"https://openalex.org/S112646816","display_name":"SHILAP Revista de lepidopterolog\u00eda","issn_l":"0300-5267","issn":["0300-5267","2340-4078"],"is_oa":true,"is_in_doaj":true,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Future Internet, Vol 17, Iss 4, p 185 (2025)","raw_type":"article"}],"best_oa_location":{"id":"doi:10.3390/fi17040185","is_oa":true,"landing_page_url":"https://doi.org/10.3390/fi17040185","pdf_url":"https://www.mdpi.com/1999-5903/17/4/185/pdf?version=1745240701","source":{"id":"https://openalex.org/S34838331","display_name":"Future Internet","issn_l":"1999-5903","issn":["1999-5903"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310310987","host_organization_name":"Multidisciplinary Digital Publishing Institute","host_organization_lineage":["https://openalex.org/P4310310987"],"host_organization_lineage_names":["Multidisciplinary Digital Publishing Institute"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Future Internet","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G131110657","display_name":null,"funder_award_id":"P0025661","funder_id":"https://openalex.org/F4320322064","funder_display_name":"Korea Institute for Advancement of Technology"},{"id":"https://openalex.org/G4026382463","display_name":null,"funder_award_id":"RS-2022-00155885","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G4386727674","display_name":null,"funder_award_id":"RS-2022-00155885","funder_id":"https://openalex.org/F4320321142","funder_display_name":"Hanyang University"},{"id":"https://openalex.org/G4700831490","display_name":null,"funder_award_id":"2022-","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G6072120315","display_name":null,"funder_award_id":"funded","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G826808851","display_name":null,"funder_award_id":"RS-2022-00155885","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"},{"id":"https://openalex.org/G889103765","display_name":null,"funder_award_id":"MOTIE","funder_id":"https://openalex.org/F4320322064","funder_display_name":"Korea Institute for Advancement of Technology"},{"id":"https://openalex.org/G992484961","display_name":null,"funder_award_id":"Korea","funder_id":"https://openalex.org/F4320321681","funder_display_name":"Ministry of Trade, Industry and Energy"}],"funders":[{"id":"https://openalex.org/F4320321142","display_name":"Hanyang University","ror":"https://ror.org/046865y68"},{"id":"https://openalex.org/F4320321681","display_name":"Ministry of Trade, Industry and Energy","ror":"https://ror.org/008nkqk13"},{"id":"https://openalex.org/F4320322064","display_name":"Korea Institute for Advancement of Technology","ror":"https://ror.org/015w1qa96"},{"id":"https://openalex.org/F4320328359","display_name":"Ministry of Science and ICT, South Korea","ror":"https://ror.org/01wpjm123"},{"id":"https://openalex.org/F4320335489","display_name":"Institute for Information and Communications Technology Promotion","ror":"https://ror.org/01g0hqq23"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4409637743.pdf","grobid_xml":"https://content.openalex.org/works/W4409637743.grobid-xml"},"referenced_works_count":34,"referenced_works":["https://openalex.org/W2894175714","https://openalex.org/W2933138175","https://openalex.org/W2946609015","https://openalex.org/W2963015836","https://openalex.org/W2963122961","https://openalex.org/W2996908057","https://openalex.org/W2998617917","https://openalex.org/W3133264589","https://openalex.org/W3137147200","https://openalex.org/W3164771864","https://openalex.org/W3175752238","https://openalex.org/W3196813608","https://openalex.org/W3202028501","https://openalex.org/W4309591680","https://openalex.org/W4362655426","https://openalex.org/W4385574241","https://openalex.org/W4388979610","https://openalex.org/W4389518760","https://openalex.org/W4389524393","https://openalex.org/W4401044057","https://openalex.org/W4402683788","https://openalex.org/W4404782932","https://openalex.org/W6698183232","https://openalex.org/W6739901393","https://openalex.org/W6769627184","https://openalex.org/W6772381481","https://openalex.org/W6778883912","https://openalex.org/W6838633097","https://openalex.org/W6846659131","https://openalex.org/W6852905540","https://openalex.org/W6853667026","https://openalex.org/W6855007554","https://openalex.org/W6857799723","https://openalex.org/W6893640197"],"related_works":["https://openalex.org/W2051487156","https://openalex.org/W2073681303","https://openalex.org/W2053286651","https://openalex.org/W2181743346","https://openalex.org/W2187401768","https://openalex.org/W2181413294","https://openalex.org/W2052122378","https://openalex.org/W2544423928","https://openalex.org/W2062023542","https://openalex.org/W4313731559"],"abstract_inverted_index":{"Modern":[0],"large":[1,79],"language":[2],"models":[3],"(LLMs)":[4],"achieve":[5],"state-of-the-art":[6],"performance":[7,180],"through":[8],"architectural":[9],"advancements":[10],"but":[11],"require":[12],"high":[13],"computational":[14],"costs":[15,28],"for":[16,49,196],"inference.":[17],"Post-training":[18],"quantization":[19,48,73,176,192],"is":[20,204],"a":[21,43,99,127,138],"widely":[22],"adopted":[23],"approach":[24],"to":[25,34,77,85,90],"reduce":[26],"these":[27,103],"by":[29],"quantizing":[30],"weights":[31],"and":[32,117,123,153,186,200],"activations":[33],"lower":[35],"precision,":[36],"such":[37],"as":[38,86],"INT8.":[39],"However,":[40],"we":[41,83,145],"identify":[42],"critical":[44],"challenge":[45],"in":[46,59,93,108,174],"activation":[47,80,87,159,171],"GLU":[50,184],"(Gated":[51],"Linear":[52],"Unit)":[53],"variants,":[54],"which":[55,82,157],"are":[56,124],"commonly":[57],"used":[58],"the":[60,67,109,115,121,179,188,202],"feed-forward":[61],"networks":[62],"of":[63,102,120,130,181,190],"modern":[64],"LLMs":[65,182],"like":[66],"LLaMA":[68],"family.":[69],"Specifically,":[70],"severe":[71],"local":[72],"errors":[74],"arise":[75],"due":[76],"excessively":[78],"magnitudes,":[81],"refer":[84],"spikes,":[88],"leading":[89],"significant":[91],"degradation":[92],"model":[94,122],"performance.":[95],"Our":[96],"analysis":[97],"reveals":[98],"systematic":[100],"pattern":[101],"spikes:":[104],"they":[105],"predominantly":[106],"occur":[107],"FFN":[110],"(feed-forward":[111],"network)":[112],"layers":[113,119],"at":[114],"early":[116],"late":[118],"concentrated":[125],"on":[126],"small":[128],"subset":[129],"tokens":[131],"rather":[132],"than":[133],"being":[134],"uniformly":[135],"distributed":[136],"across":[137],"token":[139],"sequence.":[140],"To":[141],"mitigate":[142],"this":[143],"issue,":[144],"propose":[146],"two":[147],"empirical":[148],"methods:":[149],"Quantization-free":[150,154],"Module":[151],"(QFeM)":[152],"Prefix":[155],"(QFeP),":[156],"isolate":[158],"spikes":[160],"during":[161],"quantization.":[162],"Extensive":[163],"experiments":[164,203],"demonstrated":[165],"that":[166],"our":[167,198,207],"methods":[168,199],"effectively":[169],"improve":[170],"quantization,":[172],"particularly":[173],"coarse-grained":[175],"schemes,":[177],"enhancing":[178],"with":[183],"variants":[185],"addressing":[187],"limitations":[189],"existing":[191],"techniques.":[193],"The":[194],"code":[195],"implementing":[197],"reproducing":[201],"publicly":[205],"available":[206],"GitHub":[208],"repository.":[209]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2026-04-10T15:06:20.359241","created_date":"2025-10-10T00:00:00"}
