{"id":"https://openalex.org/W7138393936","doi":"https://doi.org/10.1609/aaai.v40i34.40123","title":"First-Order Error Matters: Accurate Compensation for Quantized Large Language Models","display_name":"First-Order Error Matters: Accurate Compensation for Quantized Large Language Models","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7138393936","doi":"https://doi.org/10.1609/aaai.v40i34.40123"},"language":null,"primary_location":{"id":"doi:10.1609/aaai.v40i34.40123","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i34.40123","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40123/44084","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40123/44084","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5129674049","display_name":"Xingyu Zheng","orcid":null},"institutions":[{"id":"https://openalex.org/I82880672","display_name":"Beihang University","ror":"https://ror.org/00wk2mp56","country_code":"CN","type":"education","lineage":["https://openalex.org/I82880672"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Xingyu Zheng","raw_affiliation_strings":["State Key Laboratory of Complex & Critical Software Environment, Beihang University\nSchool of Computer Science and Engineering, Beihang University"],"affiliations":[{"raw_affiliation_string":"State Key Laboratory of Complex & Critical Software Environment, Beihang University\nSchool of Computer Science and Engineering, Beihang University","institution_ids":["https://openalex.org/I82880672"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129746991","display_name":"Haotong Qin","orcid":null},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Haotong Qin","raw_affiliation_strings":["ETH Zurich"],"affiliations":[{"raw_affiliation_string":"ETH Zurich","institution_ids":["https://openalex.org/I35440088"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129710343","display_name":"Yuye Li","orcid":null},"institutions":[{"id":"https://openalex.org/I149594827","display_name":"Xidian University","ror":"https://ror.org/05s92vm98","country_code":"CN","type":"education","lineage":["https://openalex.org/I149594827"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yuye Li","raw_affiliation_strings":["Xidian University"],"affiliations":[{"raw_affiliation_string":"Xidian University","institution_ids":["https://openalex.org/I149594827"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5011216835","display_name":"Haoran Chu","orcid":"https://orcid.org/0000-0001-5907-4897"},"institutions":[{"id":"https://openalex.org/I82880672","display_name":"Beihang University","ror":"https://ror.org/00wk2mp56","country_code":"CN","type":"education","lineage":["https://openalex.org/I82880672"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Haoran Chu","raw_affiliation_strings":["School of Computer Science and Engineering, Beihang University"],"affiliations":[{"raw_affiliation_string":"School of Computer Science and Engineering, Beihang University","institution_ids":["https://openalex.org/I82880672"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100365382","display_name":"Jiaqi Wang","orcid":"https://orcid.org/0000-0003-2210-9608"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiakai Wang","raw_affiliation_strings":["Zhongguancun Laboratory"],"affiliations":[{"raw_affiliation_string":"Zhongguancun Laboratory","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129645909","display_name":"Jinyang Guo","orcid":null},"institutions":[{"id":"https://openalex.org/I82880672","display_name":"Beihang University","ror":"https://ror.org/00wk2mp56","country_code":"CN","type":"education","lineage":["https://openalex.org/I82880672"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jinyang Guo","raw_affiliation_strings":["State Key Laboratory of Complex & Critical Software Environment, Beihang University\nSchool of Artificial Intelligence, Beihang University"],"affiliations":[{"raw_affiliation_string":"State Key Laboratory of Complex & Critical Software Environment, Beihang University\nSchool of Artificial Intelligence, Beihang University","institution_ids":["https://openalex.org/I82880672"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129658128","display_name":"Michele Magno","orcid":null},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Michele Magno","raw_affiliation_strings":["ETH Zurich"],"affiliations":[{"raw_affiliation_string":"ETH Zurich","institution_ids":["https://openalex.org/I35440088"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5129686880","display_name":"Xianglong Liu","orcid":null},"institutions":[{"id":"https://openalex.org/I82880672","display_name":"Beihang University","ror":"https://ror.org/00wk2mp56","country_code":"CN","type":"education","lineage":["https://openalex.org/I82880672"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xianglong Liu","raw_affiliation_strings":["State Key Laboratory of Complex & Critical Software Environment, Beihang University\nSchool of Computer Science and Engineering, Beihang University"],"affiliations":[{"raw_affiliation_string":"State Key Laboratory of Complex & Critical Software Environment, Beihang University\nSchool of Computer Science and Engineering, Beihang University","institution_ids":["https://openalex.org/I82880672"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":8,"corresponding_author_ids":["https://openalex.org/A5129674049"],"corresponding_institution_ids":["https://openalex.org/I82880672"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.77014925,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"40","issue":"34","first_page":"28883","last_page":"28891"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.18520000576972961,"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":0.18520000576972961,"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":0.18000000715255737,"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":0.13009999692440033,"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/quantization","display_name":"Quantization (signal processing)","score":0.6728000044822693},{"id":"https://openalex.org/keywords/hessian-matrix","display_name":"Hessian matrix","score":0.5846999883651733},{"id":"https://openalex.org/keywords/perplexity","display_name":"Perplexity","score":0.5396000146865845},{"id":"https://openalex.org/keywords/taylor-series","display_name":"Taylor series","score":0.5175999999046326},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.42100000381469727},{"id":"https://openalex.org/keywords/generalization","display_name":"Generalization","score":0.4138999879360199},{"id":"https://openalex.org/keywords/range","display_name":"Range (aeronautics)","score":0.38940000534057617}],"concepts":[{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.6728000044822693},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6276999711990356},{"id":"https://openalex.org/C203616005","wikidata":"https://www.wikidata.org/wiki/Q620495","display_name":"Hessian matrix","level":2,"score":0.5846999883651733},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5644999742507935},{"id":"https://openalex.org/C100279451","wikidata":"https://www.wikidata.org/wiki/Q372193","display_name":"Perplexity","level":3,"score":0.5396000146865845},{"id":"https://openalex.org/C158946198","wikidata":"https://www.wikidata.org/wiki/Q131187","display_name":"Taylor series","level":2,"score":0.5175999999046326},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.42100000381469727},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.4138999879360199},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.38940000534057617},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.3449000120162964},{"id":"https://openalex.org/C129844170","wikidata":"https://www.wikidata.org/wiki/Q41299","display_name":"Quadratic equation","level":2,"score":0.3418999910354614},{"id":"https://openalex.org/C205711294","wikidata":"https://www.wikidata.org/wiki/Q176953","display_name":"Rendering (computer graphics)","level":2,"score":0.3353999853134155},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3278999924659729},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3107999861240387},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3000999987125397},{"id":"https://openalex.org/C2780023022","wikidata":"https://www.wikidata.org/wiki/Q1338171","display_name":"Compensation (psychology)","level":2,"score":0.2786000072956085},{"id":"https://openalex.org/C103088060","wikidata":"https://www.wikidata.org/wiki/Q1062839","display_name":"Error detection and correction","level":2,"score":0.2703000009059906},{"id":"https://openalex.org/C39927690","wikidata":"https://www.wikidata.org/wiki/Q11197","display_name":"Logarithm","level":2,"score":0.2669000029563904},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.2597000002861023},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.25130000710487366}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1609/aaai.v40i34.40123","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i34.40123","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40123/44084","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i34.40123","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i34.40123","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40123/44084","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7138393936.pdf","grobid_xml":"https://content.openalex.org/works/W7138393936.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Post-training":[0],"quantization":[1,34,91],"(PTQ)":[2],"offers":[3],"an":[4,109],"efficient":[5],"approach":[6],"to":[7,32,89,137,205],"compressing":[8],"large":[9],"language":[10],"models":[11,170],"(LLMs),":[12],"significantly":[13],"reducing":[14],"memory":[15],"access":[16],"and":[17,64,117,148,171,194,227],"computational":[18,146,161],"costs.":[19],"Existing":[20],"compensation-based":[21],"weight":[22],"calibration":[23],"methods":[24],"often":[25],"rely":[26],"on":[27,112],"a":[28,79,99,166],"second-order":[29],"Taylor":[30,101],"expansion":[31,102],"model":[33],"error,":[35],"under":[36,222],"the":[37,40,53,104,113,123,129,132,135,140,144,178,188,196,223,230],"assumption":[38,70],"that":[39,52,83,174],"first-order":[41,59,86,100],"term":[42],"is":[43],"negligible":[44],"in":[45],"well-trained":[46],"full-precision":[47,66,118,234],"models.":[48],"However,":[49],"we":[50,76],"reveal":[51],"progressive":[54],"compensation":[55],"process":[56],"introduces":[57,157],"accumulated":[58],"deviations":[60],"between":[61,115],"latent":[62,116],"weights":[63,119],"their":[65],"counterparts,":[67],"making":[68],"this":[69],"fundamentally":[71],"flawed.":[72],"To":[73],"address":[74],"this,":[75],"propose":[77],"FOEM,":[78],"novel":[80],"PTQ":[81],"method":[82],"explicitly":[84,138],"incorporates":[85],"gradient":[87,153],"terms":[88],"improve":[90],"error":[92],"compensation.":[93],"FOEM":[94,175,186,208],"approximates":[95],"gradients":[96],"by":[97,192,203],"performing":[98],"around":[103],"pre-quantization":[105],"weights.":[106],"This":[107,155],"yields":[108],"approximation":[110],"based":[111],"difference":[114],"as":[120,122,217],"well":[121],"Hessian":[124],"matrix.":[125],"When":[126],"substituted":[127],"into":[128],"theoretical":[130],"solution,":[131],"formulation":[133],"eliminates":[134],"need":[136],"compute":[139],"Hessian,":[141],"thereby":[142],"avoiding":[143],"high":[145],"cost":[147],"limited":[149],"generalization":[150],"of":[151,169,190],"backpropagation-based":[152],"methods.":[154,239],"design":[156],"only":[158],"minimal":[159],"additional":[160,220],"overhead.":[162],"Extensive":[163],"experiments":[164],"across":[165],"wide":[167],"range":[168],"benchmarks":[172],"demonstrate":[173],"consistently":[176],"outperforms":[177],"classical":[179],"GPTQ":[180],"method.":[181],"In":[182],"3-bit":[183],"weight-only":[184],"quantization,":[185],"reduces":[187],"perplexity":[189],"Llama3-8B":[191],"17.3%":[193],"increases":[195],"5-shot":[197],"MMLU":[198],"accuracy":[199],"from":[200],"53.8%":[201],"achieved":[202],"GPTAQ":[204],"56.1%.":[206],"Moreover,":[207],"can":[209],"be":[210],"seamlessly":[211],"combined":[212],"with":[213,233],"advanced":[214],"techniques":[215],"such":[216],"SpinQuant,":[218],"delivering":[219],"gains":[221],"challenging":[224],"W4A4KV4":[225],"setting":[226],"further":[228],"narrowing":[229],"performance":[231],"gap":[232],"baselines,":[235],"surpassing":[236],"existing":[237],"state-of-the-art":[238]},"counts_by_year":[],"updated_date":"2026-03-20T20:47:17.329874","created_date":"2026-03-18T00:00:00"}