{"id":"https://openalex.org/W7153094953","doi":"https://doi.org/10.48550/arxiv.2604.07808","title":"GRASS: Gradient-based Adaptive Layer-wise Importance Sampling for Memory-efficient Large Language Model Fine-tuning","display_name":"GRASS: Gradient-based Adaptive Layer-wise Importance Sampling for Memory-efficient Large Language Model Fine-tuning","publication_year":2026,"publication_date":"2026-04-09","ids":{"openalex":"https://openalex.org/W7153094953","doi":"https://doi.org/10.48550/arxiv.2604.07808"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.07808","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.07808","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":null,"license_id":null,"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.2604.07808","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5133364630","display_name":"Kaiyuan Tian","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tian, Kaiyuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133335893","display_name":"Yu Tang","orcid":"https://orcid.org/0009-0002-1715-7429"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tang, Yu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126410597","display_name":"Gongqingjian Jiang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiang, Gongqingjian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5036008442","display_name":"Baihui Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Baihui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5011016539","display_name":"Yifu Gao","orcid":"https://orcid.org/0000-0002-1743-8055"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Yifu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5025427937","display_name":"Xialin Su","orcid":"https://orcid.org/0000-0003-3617-0856"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Su, Xialin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5043045076","display_name":"Linbo Qiao","orcid":"https://orcid.org/0000-0002-8285-2738"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qiao, Linbo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5133339505","display_name":"Dongsheng Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Dongsheng","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/T10028","display_name":"Topic Modeling","score":0.2696000039577484,"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":0.2696000039577484,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.10740000009536743,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.09669999778270721,"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/sampling","display_name":"Sampling (signal processing)","score":0.6532999873161316},{"id":"https://openalex.org/keywords/adaptive-sampling","display_name":"Adaptive sampling","score":0.6327999830245972},{"id":"https://openalex.org/keywords/layer","display_name":"Layer (electronics)","score":0.5828999876976013},{"id":"https://openalex.org/keywords/metric","display_name":"Metric (unit)","score":0.5788000226020813},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.5575000047683716},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.5546000003814697},{"id":"https://openalex.org/keywords/adaptation","display_name":"Adaptation (eye)","score":0.5203999876976013},{"id":"https://openalex.org/keywords/performance-metric","display_name":"Performance metric","score":0.4429999887943268},{"id":"https://openalex.org/keywords/limit","display_name":"Limit (mathematics)","score":0.43619999289512634}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8001999855041504},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.6532999873161316},{"id":"https://openalex.org/C2781395549","wikidata":"https://www.wikidata.org/wiki/Q4680762","display_name":"Adaptive sampling","level":3,"score":0.6327999830245972},{"id":"https://openalex.org/C2779227376","wikidata":"https://www.wikidata.org/wiki/Q6505497","display_name":"Layer (electronics)","level":2,"score":0.5828999876976013},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.5788000226020813},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.5575000047683716},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5546000003814697},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.5203999876976013},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4634000062942505},{"id":"https://openalex.org/C2780898871","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Performance metric","level":2,"score":0.4429999887943268},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4410000145435333},{"id":"https://openalex.org/C151201525","wikidata":"https://www.wikidata.org/wiki/Q177239","display_name":"Limit (mathematics)","level":2,"score":0.43619999289512634},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.4187999963760376},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.4074999988079071},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.33869999647140503},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.3287999927997589},{"id":"https://openalex.org/C19768560","wikidata":"https://www.wikidata.org/wiki/Q320727","display_name":"Dependency (UML)","level":2,"score":0.3176000118255615},{"id":"https://openalex.org/C2778915421","wikidata":"https://www.wikidata.org/wiki/Q3643177","display_name":"Performance improvement","level":2,"score":0.3140999972820282},{"id":"https://openalex.org/C48103436","wikidata":"https://www.wikidata.org/wiki/Q599031","display_name":"State (computer science)","level":2,"score":0.29989999532699585},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.29980000853538513},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2985999882221222},{"id":"https://openalex.org/C2778334786","wikidata":"https://www.wikidata.org/wiki/Q1586270","display_name":"Variation (astronomy)","level":2,"score":0.2808000147342682},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.2750000059604645},{"id":"https://openalex.org/C12186640","wikidata":"https://www.wikidata.org/wiki/Q6815743","display_name":"Memory model","level":3,"score":0.2639000117778778},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.26350000500679016},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.25690001249313354}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.07808","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.07808","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.07808","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.07808","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education","score":0.4917546212673187}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Full-parameter":[0],"fine-tuning":[1,41],"of":[2,24,161],"large":[3],"language":[4],"models":[5,146],"is":[6],"constrained":[7],"by":[8,19,170],"substantial":[9],"GPU":[10],"memory":[11,135,168],"requirements.":[12],"Low-rank":[13],"adaptation":[14],"methods":[15,42,59],"mitigate":[16],"this":[17],"challenge":[18],"updating":[20],"only":[21],"a":[22,84,97,121],"subset":[23],"parameters.":[25],"However,":[26,57],"these":[27,58,79],"approaches":[28],"often":[29],"limit":[30],"model":[31],"expressiveness":[32],"and":[33,67,99,130,147,166],"yield":[34],"lower":[35],"performance":[36,73],"than":[37],"full-parameter":[38],"fine-tuning.":[39],"Layer-wise":[40],"have":[43],"emerged":[44],"as":[45,96],"an":[46,114,157],"alternative,":[47],"enabling":[48],"memory-efficient":[49],"training":[50,68,116,140],"through":[51,113],"static":[52],"layer":[53,63,104,110],"importance":[54,64,88],"sampling":[55,89,111],"strategies.":[56],"overlook":[60],"variations":[61],"in":[62,71],"across":[65,144],"tasks":[66],"stages,":[69],"resulting":[70],"suboptimal":[72],"on":[74],"downstream":[75],"tasks.":[76],"To":[77],"address":[78],"limitations,":[80],"we":[81],"propose":[82],"GRASS,":[83],"gradient-based":[85],"adaptive":[86,115],"layer-wise":[87,122],"framework.":[90],"GRASS":[91,107,151],"utilizes":[92],"mean":[93],"gradient":[94],"norms":[95],"task-aware":[98],"training-stage-aware":[100],"metric":[101],"for":[102],"estimating":[103],"importance.":[105],"Furthermore,":[106],"adaptively":[108],"adjusts":[109],"probabilities":[112],"strategy.":[117],"We":[118],"also":[119],"introduce":[120],"optimizer":[123],"state":[124],"offloading":[125],"mechanism":[126],"that":[127,150],"overlaps":[128],"computation":[129],"communication":[131],"to":[132,163,172],"further":[133],"reduce":[134],"usage":[136,169],"while":[137],"maintaining":[138],"comparable":[139],"throughput.":[141],"Extensive":[142],"experiments":[143],"multiple":[145],"benchmarks":[148],"demonstrate":[149],"consistently":[152],"outperforms":[153],"state-of-the-art":[154],"methods,":[155],"achieving":[156],"average":[158],"accuracy":[159],"improvement":[160],"up":[162,171],"4.38":[164],"points":[165],"reducing":[167],"19.97\\%.":[173]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-11T00:00:00"}