{"id":"https://openalex.org/W7127097004","doi":"https://doi.org/10.48550/arxiv.2601.22716","title":"Breaking the Blocks: Continuous Low-Rank Decomposed Scaling for Unified LLM Quantization and Adaptation","display_name":"Breaking the Blocks: Continuous Low-Rank Decomposed Scaling for Unified LLM Quantization and Adaptation","publication_year":2026,"publication_date":"2026-01-30","ids":{"openalex":"https://openalex.org/W7127097004","doi":"https://doi.org/10.48550/arxiv.2601.22716"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2601.22716","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2601.22716","pdf_url":"https://arxiv.org/pdf/2601.22716","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2601.22716","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5124795473","display_name":"Pingzhi Tang","orcid":"https://orcid.org/0009-0001-7958-7144","observed_orcids":["https://orcid.org/0009-0001-7958-7144"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tang, Pingzhi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124832654","display_name":"Ruijie Zhou","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Ruijie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124774779","display_name":"Fanxu Meng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Meng, Fanxu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5078487642","display_name":"Wenjie Pei","orcid":"https://orcid.org/0000-0001-8117-2696","observed_orcids":["https://orcid.org/0000-0001-8117-2696","https://orcid.org/0000-0002-2019-2295"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pei, Wenjie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5123113317","display_name":"Muhan Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Muhan","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":0.9502000212669373,"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":0.9502000212669373,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.007400000002235174,"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":0.005799999926239252,"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/low-rank-decomposition","display_name":"low-rank decomposition","score":0.9240000247955322},{"id":"https://openalex.org/keywords/post-training-quantization","display_name":"post-training quantization","score":0.7110000252723694},{"id":"https://openalex.org/keywords/quantization-aware-training","display_name":"quantization-aware training","score":0.8220000267028809},{"id":"https://openalex.org/keywords/parameter-efficient-fine-tuning","display_name":"parameter-efficient fine-tuning","score":0.8519999980926514},{"id":"https://openalex.org/keywords/lords","display_name":"lords","score":0.8199999928474426},{"id":"https://openalex.org/keywords/triton-kernels","display_name":"Triton kernels","score":0.7049999833106995},{"id":"https://openalex.org/keywords/inference-acceleration","display_name":"inference acceleration","score":0.6019999980926514},{"id":"https://openalex.org/keywords/llama-3-8b","display_name":"Llama-3-8B","score":0.7360000014305115}],"concepts":[{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.7609999775886536},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.7263000011444092},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6222000122070312},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.5981000065803528},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5813000202178955},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5577999949455261},{"id":"https://openalex.org/C177774035","wikidata":"https://www.wikidata.org/wiki/Q1246948","display_name":"Granularity","level":2,"score":0.5224000215530396},{"id":"https://openalex.org/C199833920","wikidata":"https://www.wikidata.org/wiki/Q612536","display_name":"Vector quantization","level":2,"score":0.5095000267028809},{"id":"https://openalex.org/C45493050","wikidata":"https://www.wikidata.org/wiki/Q7884934","display_name":"Unified Model","level":2,"score":0.4837000072002411},{"id":"https://openalex.org/C114466953","wikidata":"https://www.wikidata.org/wiki/Q6034165","display_name":"Initialization","level":2,"score":0.42809998989105225},{"id":"https://openalex.org/C42747912","wikidata":"https://www.wikidata.org/wiki/Q1048447","display_name":"Multiplicative function","level":2,"score":0.3671000003814697},{"id":"https://openalex.org/C70451592","wikidata":"https://www.wikidata.org/wiki/Q376493","display_name":"Dither","level":3,"score":0.34450000524520874},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.34130001068115234},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3310999870300293},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.295199990272522},{"id":"https://openalex.org/C111030470","wikidata":"https://www.wikidata.org/wiki/Q1430460","display_name":"Curse of dimensionality","level":2,"score":0.28679999709129333},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2808000147342682}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2601.22716","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2601.22716","pdf_url":"https://arxiv.org/pdf/2601.22716","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2601.22716","doi":"https://doi.org/10.48550/arxiv.2601.22716","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.22716","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2601.22716","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2601.22716","pdf_url":"https://arxiv.org/pdf/2601.22716","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4403754519","https://openalex.org/W4387156572","https://openalex.org/W4399554365","https://openalex.org/W7155067614","https://openalex.org/W7214496050","https://openalex.org/W4391591083","https://openalex.org/W7103170887","https://openalex.org/W7163189905","https://openalex.org/W4387686975","https://openalex.org/W4415792811"],"abstract_inverted_index":{"Current":[0],"quantization":[1,27,65,146,168],"methods":[2],"for":[3,197],"LLMs":[4],"predominantly":[5],"rely":[6],"on":[7,152,175,185],"block-wise":[8,34],"structures":[9],"to":[10,158],"maintain":[11],"efficiency,":[12],"often":[13],"at":[14,163],"the":[15,44,73],"cost":[16],"of":[17,75,96,202],"representational":[18],"flexibility.":[19],"In":[20],"this":[21,68],"work,":[22],"we":[23],"demonstrate":[24],"that":[25,63],"element-wise":[26],"can":[28],"be":[29],"made":[30],"as":[31,33,47,112],"efficient":[32],"scaling":[35,45,99],"while":[36,123,179],"providing":[37],"strictly":[38],"superior":[39],"expressive":[40],"power":[41],"by":[42,130,183],"modeling":[43],"manifold":[46],"continuous":[48],"low-rank":[49,69,121],"matrices":[50],"($S":[51],"=":[52],"BA$).":[53],"We":[54],"propose":[55],"Low-Rank":[56],"Decomposed":[57],"Scaling":[58],"(LoRDS),":[59],"a":[60,80,120,159,171,192],"unified":[61,198],"framework":[62],"rethinks":[64],"granularity":[66],"through":[67],"decomposition.":[70],"By":[71],"\"breaking":[72],"blocks\"":[74],"spatial":[76],"constraints,":[77],"LoRDS":[78,114,135],"establishes":[79],"seamless":[81],"efficiency":[82],"lifecycle:":[83],"it":[84],"provides":[85],"high-fidelity":[86],"PTQ":[87],"initialization":[88],"refined":[89],"via":[90],"iterative":[91],"optimization,":[92],"enables":[93,115],"joint":[94],"QAT":[95],"weights":[97],"and":[98,101,147,169,194,200],"factors,":[100],"facilitates":[102],"high-rank":[103,116],"multiplicative":[104],"PEFT":[105,109,181],"adaptation.":[106],"Unlike":[107],"additive":[108],"approaches":[110],"such":[111],"QLoRA,":[113,190],"weight":[117],"updates":[118],"within":[119],"budget":[122],"incurring":[124],"no":[125],"additional":[126],"inference":[127,173],"overhead.":[128],"Supported":[129],"highly":[131],"optimized":[132],"Triton":[133],"kernels,":[134],"consistently":[136],"outperforms":[137],"state-of-the-art":[138],"baselines":[139],"across":[140],"various":[141],"model":[142],"families":[143],"in":[144],"both":[145],"downstream":[148,186],"fine-tuning":[149],"tasks.":[150],"Notably,":[151],"Llama3-8B,":[153],"our":[154],"method":[155],"achieves":[156],"up":[157],"27.0%":[160],"accuracy":[161],"improvement":[162],"3":[164],"bits":[165],"over":[166,188],"NormalFloat":[167],"delivers":[170],"1.5x":[172],"speedup":[174],"NVIDIA":[176],"RTX":[177],"4090":[178],"enhancing":[180],"performance":[182],"9.6%":[184],"tasks":[187],"4bit":[189],"offering":[191],"robust":[193],"integrated":[195],"solution":[196],"compression":[199],"adaptation":[201],"LLMs.":[203]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-02-03T00:00:00"}