{"id":"https://openalex.org/W7166018208","doi":"https://doi.org/10.48550/arxiv.2606.26587","title":"SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference","display_name":"SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference","publication_year":2026,"publication_date":"2026-06-25","ids":{"openalex":"https://openalex.org/W7166018208","doi":"https://doi.org/10.48550/arxiv.2606.26587"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.26587","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.26587","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.2606.26587","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139444621","display_name":"Haoqian Meng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Meng, Haoqian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139427971","display_name":"Yilun Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luo, Yilun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139394317","display_name":"Yafei Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Yafei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139458691","display_name":"Wenyuan Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Wenyuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139441340","display_name":"Huaqing Zheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng, Huaqing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5110658926","display_name":"Xindian Ma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Xindian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5056795737","display_name":"P Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Peng","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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.19859999418258667,"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"}},"topics":[{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.19859999418258667,"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"}},{"id":"https://openalex.org/T11181","display_name":"Advanced Data Storage Technologies","score":0.06759999692440033,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.06440000236034393,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.6636999845504761},{"id":"https://openalex.org/keywords/residual","display_name":"Residual","score":0.5770999789237976},{"id":"https://openalex.org/keywords/sparse-matrix","display_name":"Sparse matrix","score":0.550599992275238},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5338000059127808},{"id":"https://openalex.org/keywords/vector-quantization","display_name":"Vector quantization","score":0.43560001254081726},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.43149998784065247},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.41990000009536743},{"id":"https://openalex.org/keywords/bridging","display_name":"Bridging (networking)","score":0.41449999809265137},{"id":"https://openalex.org/keywords/outlier","display_name":"Outlier","score":0.35920000076293945}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6705999970436096},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.6636999845504761},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.6208999752998352},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.5770999789237976},{"id":"https://openalex.org/C56372850","wikidata":"https://www.wikidata.org/wiki/Q1050404","display_name":"Sparse matrix","level":3,"score":0.550599992275238},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5338000059127808},{"id":"https://openalex.org/C199833920","wikidata":"https://www.wikidata.org/wiki/Q612536","display_name":"Vector quantization","level":2,"score":0.43560001254081726},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.43149998784065247},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.41990000009536743},{"id":"https://openalex.org/C174348530","wikidata":"https://www.wikidata.org/wiki/Q188635","display_name":"Bridging (networking)","level":2,"score":0.41449999809265137},{"id":"https://openalex.org/C79337645","wikidata":"https://www.wikidata.org/wiki/Q779824","display_name":"Outlier","level":2,"score":0.35920000076293945},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3587000072002411},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.35580000281333923},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.3384999930858612},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.3116999864578247},{"id":"https://openalex.org/C136886441","wikidata":"https://www.wikidata.org/wiki/Q926129","display_name":"Normalization (sociology)","level":2,"score":0.3107999861240387},{"id":"https://openalex.org/C74193536","wikidata":"https://www.wikidata.org/wiki/Q574844","display_name":"Kernel (algebra)","level":2,"score":0.30869999527931213},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2922999858856201},{"id":"https://openalex.org/C2776639698","wikidata":"https://www.wikidata.org/wiki/Q4927140","display_name":"Block Error Rate","level":3,"score":0.29109999537467957},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.28439998626708984},{"id":"https://openalex.org/C57654395","wikidata":"https://www.wikidata.org/wiki/Q1097775","display_name":"Compression artifact","level":5,"score":0.28380000591278076},{"id":"https://openalex.org/C77637269","wikidata":"https://www.wikidata.org/wiki/Q7002051","display_name":"Neural coding","level":2,"score":0.28130000829696655},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2809000015258789},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.2806999981403351},{"id":"https://openalex.org/C77246614","wikidata":"https://www.wikidata.org/wiki/Q1409400","display_name":"Gramian matrix","level":3,"score":0.2754000127315521},{"id":"https://openalex.org/C103088060","wikidata":"https://www.wikidata.org/wiki/Q1062839","display_name":"Error detection and correction","level":2,"score":0.2615000009536743},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.2542000114917755},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.25369998812675476},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.25060001015663147}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.26587","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.26587","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.2606.26587","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.26587","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":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Low-bit":[0],"floating-point":[1],"formats":[2],"and":[3,32,58,86,121,138,148,168,180,183,188,202,212],"semi-structured":[4],"sparsity":[5,36,57],"are":[6],"increasingly":[7],"supported":[8],"by":[9],"modern":[10],"accelerators,":[11],"yet":[12],"combining":[13],"them":[14],"for":[15,116],"LLM":[16],"activation":[17,56,68,119],"compression":[18],"remains":[19],"challenging:":[20],"activations":[21],"contain":[22],"input-dependent":[23],"outliers":[24],"that":[25,54],"dominate":[26],"block":[27],"scales":[28],"in":[29,208],"FP4":[30,59,105,113,131],"quantization,":[31],"directly":[33],"applying":[34],"N:M":[35,74],"masks":[37],"discards":[38],"moderate":[39],"values,":[40],"coupling":[41],"sparsification":[42],"loss":[43,120],"with":[44,134,223],"quantization":[45,60,123],"error.":[46,124],"We":[47],"introduce":[48],"SharQ,":[49],"a":[50,88,111,129,139],"training-free":[51],"inference":[52],"method":[53],"bridges":[55],"through":[61],"an":[62,72,78,192],"online":[63],"sparse--dense":[64],"decomposition.":[65],"For":[66],"each":[67],"tensor,":[69],"SharQ":[70,154,170,195],"generates":[71],"input-adaptive":[73],"mask":[75,144],"to":[76,84,92,214],"extract":[77],"outlier-dominated":[79],"sparse":[80,95,101,104],"backbone,":[81],"quantizes":[82],"it":[83],"FP4,":[85],"defines":[87],"dense":[89,112],"residual":[90,146],"relative":[91],"the":[93,99,108,174],"quantized":[94],"backbone":[96,109],"rather":[97],"than":[98],"unquantized":[100],"values.":[102],"A":[103],"GEMM":[106,114],"processes":[107],"while":[110],"compensates":[115],"both":[117],"mask-induced":[118],"sparse-path":[122],"The":[125],"two":[126],"paths":[127],"share":[128],"single":[130],"weight":[132],"payload":[133],"path-specific":[135],"scale":[136],"views,":[137],"fused":[140],"preparation":[141],"kernel":[142],"absorbs":[143],"generation,":[145],"construction,":[147],"layer":[149],"normalization":[150],"into":[151],"one":[152],"operator.":[153],"requires":[155],"no":[156],"calibration":[157],"data,":[158],"retraining,":[159],"or":[160],"model-specific":[161],"tuning.":[162],"Evaluated":[163],"on":[164,217],"Llama-3.1-8B,":[165],"Qwen2.5-7B,":[166],"Qwen3-30B-A3B,":[167],"Qwen3-VL-8B,":[169],"recovers":[171],"43--63%":[172],"of":[173],"NVFP4-to-FP16":[175],"accuracy":[176],"gap":[177],"across":[178,185],"language":[179,209],"vision-language":[181],"tasks,":[182],"generalizes":[184],"NVFP4,":[186],"HiF4,":[187],"MXFP4":[189],"formats.":[190],"On":[191],"RTX":[193],"5090,":[194],"delivers":[196],"2.2--2.4$\\times$":[197],"latency":[198],"reduction":[199],"over":[200,206],"FP16":[201],"1.2--1.4$\\times$":[203],"throughput":[204],"improvement":[205],"FP8":[207],"model":[210],"serving,":[211],"up":[213],"1.58$\\times$":[215],"speedup":[216],"Wan2.2-T2V-A14B":[218],"video":[219],"generation":[220],"when":[221],"combined":[222],"SageAttention.":[224],"Our":[225],"code":[226],"is":[227],"available":[228],"at":[229],"https://github.com/actypedef/SharQ.":[230]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-27T00:00:00"}