{"id":"https://openalex.org/W7160027343","doi":"https://doi.org/10.48550/arxiv.2605.00539","title":"AGoQ: Activation and Gradient Quantization for Memory-Efficient Distributed Training of LLMs","display_name":"AGoQ: Activation and Gradient Quantization for Memory-Efficient Distributed Training of LLMs","publication_year":2026,"publication_date":"2026-05-01","ids":{"openalex":"https://openalex.org/W7160027343","doi":"https://doi.org/10.48550/arxiv.2605.00539"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.00539","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00539","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","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.2605.00539","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5110841034","display_name":"Wenxiang Lin","orcid":"https://orcid.org/0000-0001-5147-0844"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Wenxiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135144411","display_name":"Juntao Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Juntao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135206249","display_name":"Luhan Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Luhan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135136838","display_name":"Laili Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Laili","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103001638","display_name":"Xiang Bao","orcid":"https://orcid.org/0009-0002-1343-3326"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bao, Xiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101542216","display_name":"Mengyang Zhang","orcid":"https://orcid.org/0000-0003-4267-1761"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Mengyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135185719","display_name":"Bing Wang","orcid":"https://orcid.org/0009-0009-7907-8997"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Bing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135174367","display_name":"Shaohuai Shi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shi, Shaohuai","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":0.5687000155448914,"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":0.5687000155448914,"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":0.09229999780654907,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.05079999938607216,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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":0.8858000040054321},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.6137999892234802},{"id":"https://openalex.org/keywords/convergence","display_name":"Convergence (economics)","score":0.5544000267982483},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.459199994802475},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.42410001158714294},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.31630000472068787}],"concepts":[{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.8858000040054321},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6909000277519226},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.6137999892234802},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.5544000267982483},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.459199994802475},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4530999958515167},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.42410001158714294},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.32280001044273376},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.31630000472068787},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.30070000886917114},{"id":"https://openalex.org/C199833920","wikidata":"https://www.wikidata.org/wiki/Q612536","display_name":"Vector quantization","level":2,"score":0.27869999408721924},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.2676999866962433},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.2624000012874603},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.2587999999523163},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2572000026702881},{"id":"https://openalex.org/C115680565","wikidata":"https://www.wikidata.org/wiki/Q5977448","display_name":"Gradient method","level":2,"score":0.250900000333786}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.00539","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00539","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.00539","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00539","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Quantization":[0],"is":[1],"a":[2,48,77],"key":[3],"method":[4],"for":[5,22,57],"reducing":[6],"the":[7,117,125],"GPU":[8,110],"memory":[9,83,126],"requirement":[10],"of":[11,59,106,137],"training":[12,138,143],"large":[13],"language":[14],"models":[15],"(LLMs).":[16],"Yet,":[17],"current":[18],"approaches":[19],"are":[20],"ineffective":[21],"4-bit":[23,72],"activations":[24,58],"and":[25,66,75,85,94,116,131,151,165],"8-bit":[26,91,96],"gradients,":[27],"which":[28],"would":[29],"easily":[30],"cause":[31],"slow":[32],"convergence":[33,161],"or":[34,147],"accuracy":[35,167],"loss.":[36],"To":[37],"address":[38],"this,":[39],"we":[40],"introduce":[41],"AGoQ,":[42],"incorporating":[43],"two":[44,109],"new":[45],"techniques:":[46],"1)":[47],"layer-aware":[49],"activation":[50,73],"quantization":[51,79],"algorithm":[52,80],"that":[53,81,121],"allocates":[54],"appropriate":[55],"bit-widths":[56],"various":[60],"layers":[61],"based":[62],"on":[63,108,163,168],"their":[64],"types":[65],"pipeline":[67],"stages":[68],"to":[69,113,129,134,141,155],"achieve":[70],"near":[71],"storage,":[74],"2)":[76],"gradient":[78,92],"reduces":[82,124],"usage":[84],"shortens":[86],"communication":[87],"time":[88],"by":[89,127],"employing":[90],"storage":[93],"precision-preserving":[95],"All-Reduce":[97],"communication.":[98],"We":[99],"conduct":[100],"extensive":[101],"experiments":[102],"using":[103],"different":[104],"sizes":[105],"LLMs":[107],"clusters":[111],"(up":[112],"64":[114],"GPUs),":[115],"experimental":[118],"results":[119],"show":[120],"our":[122],"AGoQ":[123],"up":[128,133],"52\\%":[130],"achieves":[132],"1.34$\\times$":[135],"improvement":[136],"speed":[139],"compared":[140],"state-of-the-art":[142],"systems":[144],"Megatron-LM":[145],"(w/":[146],"w/o":[148],"ZeRO),":[149],"COAT":[150],"DeepSpeed":[152],"with":[153,171],"8B":[154],"32B":[156],"LLaMA":[157,172],"models,":[158],"while":[159],"achieving":[160],"loss":[162],"pretraining":[164],"comparable":[166],"downstream":[169],"tasks":[170],"architectures.":[173]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-05T00:00:00"}