{"id":"https://openalex.org/W4416146831","doi":"https://doi.org/10.1145/3712285.3759852","title":"LiquidGEMM: Hardware-Efficient W4A8 GEMM Kernel for High-Performance LLM Serving","display_name":"LiquidGEMM: Hardware-Efficient W4A8 GEMM Kernel for High-Performance LLM Serving","publication_year":2025,"publication_date":"2025-11-12","ids":{"openalex":"https://openalex.org/W4416146831","doi":"https://doi.org/10.1145/3712285.3759852"},"language":null,"primary_location":{"id":"doi:10.1145/3712285.3759852","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3712285.3759852","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3712285.3759852","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5120429713","display_name":"Huanqi Hu","orcid":"https://orcid.org/0009-0007-4392-4712"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Huanqi Hu","raw_affiliation_strings":["Shanghai Jiao Tong University, Shanghai, China"],"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University, Shanghai, China","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5045174060","display_name":"Bowen Xiao","orcid":"https://orcid.org/0009-0001-0357-913X"},"institutions":[{"id":"https://openalex.org/I4210128111","display_name":"Clover Seed (China)","ror":"https://ror.org/037jwt041","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210128111"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Bowen Xiao","raw_affiliation_strings":["ByteDance Seed, Shanghai, China"],"affiliations":[{"raw_affiliation_string":"ByteDance Seed, Shanghai, China","institution_ids":["https://openalex.org/I4210128111"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5043164368","display_name":"Shixuan Sun","orcid":"https://orcid.org/0000-0003-4060-9438"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shixuan Sun","raw_affiliation_strings":["Shanghai Jiao Tong University, Shanghai, China"],"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University, Shanghai, China","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113943551","display_name":"Jingran Yin","orcid":"https://orcid.org/0009-0004-5728-6182"},"institutions":[{"id":"https://openalex.org/I4210128111","display_name":"Clover Seed (China)","ror":"https://ror.org/037jwt041","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210128111"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jianian Yin","raw_affiliation_strings":["ByteDance Seed, Shanghai, China"],"affiliations":[{"raw_affiliation_string":"ByteDance Seed, Shanghai, China","institution_ids":["https://openalex.org/I4210128111"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5003989046","display_name":"Zhexi Zhang","orcid":"https://orcid.org/0000-0003-1272-632X"},"institutions":[{"id":"https://openalex.org/I4210128111","display_name":"Clover Seed (China)","ror":"https://ror.org/037jwt041","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210128111"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhexi Zhang","raw_affiliation_strings":["ByteDance Seed, Shanghai, China"],"affiliations":[{"raw_affiliation_string":"ByteDance Seed, Shanghai, China","institution_ids":["https://openalex.org/I4210128111"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Xiang Luo","orcid":"https://orcid.org/0009-0003-0504-9559"},"institutions":[{"id":"https://openalex.org/I4210128111","display_name":"Clover Seed (China)","ror":"https://ror.org/037jwt041","country_code":"CN","type":"company","lineage":["https://openalex.org/I4210128111"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiang Luo","raw_affiliation_strings":["ByteDance Seed, Shanghai, China"],"affiliations":[{"raw_affiliation_string":"ByteDance Seed, Shanghai, China","institution_ids":["https://openalex.org/I4210128111"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5021566476","display_name":"Chengquan Jiang","orcid":"https://orcid.org/0009-0004-9356-6034"},"institutions":[{"id":"https://openalex.org/I1316868585","display_name":"Fidal","ror":"https://ror.org/04nwdyy60","country_code":"FR","type":"other","lineage":["https://openalex.org/I1316868585"]}],"countries":["FR"],"is_corresponding":false,"raw_author_name":"Chengquan Jiang","raw_affiliation_strings":["ByteDance Seed, Seattle, USA"],"affiliations":[{"raw_affiliation_string":"ByteDance Seed, Seattle, USA","institution_ids":["https://openalex.org/I1316868585"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5017498662","display_name":"Weixiao Xu","orcid":"https://orcid.org/0009-0003-6915-4032"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Weiqi Xu","raw_affiliation_strings":["ByteDance Seed, Beijing, China"],"affiliations":[{"raw_affiliation_string":"ByteDance Seed, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5027646341","display_name":"Xiaoying Jia","orcid":"https://orcid.org/0009-0009-2201-4765"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiaoying Jia","raw_affiliation_strings":["ByteDance Seed, Beijing, China"],"affiliations":[{"raw_affiliation_string":"ByteDance Seed, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113388165","display_name":"Xin Liu","orcid":"https://orcid.org/0009-0007-3988-5239"},"institutions":[{"id":"https://openalex.org/I1316868585","display_name":"Fidal","ror":"https://ror.org/04nwdyy60","country_code":"FR","type":"other","lineage":["https://openalex.org/I1316868585"]}],"countries":["FR"],"is_corresponding":false,"raw_author_name":"Xin Liu","raw_affiliation_strings":["ByteDance Seed, Seattle, USA"],"affiliations":[{"raw_affiliation_string":"ByteDance Seed, Seattle, USA","institution_ids":["https://openalex.org/I1316868585"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5039318240","display_name":"Minyi Guo","orcid":"https://orcid.org/0000-0003-0034-2302"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Minyi Guo","raw_affiliation_strings":["Shanghai Jiao Tong University, Shanghai, China"],"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University, Shanghai, China","institution_ids":["https://openalex.org/I183067930"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":11,"corresponding_author_ids":["https://openalex.org/A5120429713"],"corresponding_institution_ids":["https://openalex.org/I183067930"],"apc_list":null,"apc_paid":null,"fwci":1.2784,"has_fulltext":false,"cited_by_count":1,"citation_normalized_percentile":{"value":0.85715889,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":91,"max":95},"biblio":{"volume":null,"issue":null,"first_page":"1619","last_page":"1630"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.4708999991416931,"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":0.4708999991416931,"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/T12326","display_name":"Network Packet Processing and Optimization","score":0.19900000095367432,"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/T10714","display_name":"Software-Defined Networks and 5G","score":0.043699998408555984,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.6261000037193298},{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.6234999895095825},{"id":"https://openalex.org/keywords/memory-footprint","display_name":"Memory footprint","score":0.5809000134468079},{"id":"https://openalex.org/keywords/kernel","display_name":"Kernel (algebra)","score":0.5288000106811523},{"id":"https://openalex.org/keywords/cuda","display_name":"CUDA","score":0.41530001163482666},{"id":"https://openalex.org/keywords/software","display_name":"Software","score":0.37959998846054077},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.3587000072002411}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7233999967575073},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.6376000046730042},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.6261000037193298},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.6234999895095825},{"id":"https://openalex.org/C74912251","wikidata":"https://www.wikidata.org/wiki/Q6815727","display_name":"Memory footprint","level":2,"score":0.5809000134468079},{"id":"https://openalex.org/C74193536","wikidata":"https://www.wikidata.org/wiki/Q574844","display_name":"Kernel (algebra)","level":2,"score":0.5288000106811523},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.49129998683929443},{"id":"https://openalex.org/C2778119891","wikidata":"https://www.wikidata.org/wiki/Q477690","display_name":"CUDA","level":2,"score":0.41530001163482666},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.37959998846054077},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.3587000072002411},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.329800009727478},{"id":"https://openalex.org/C149810388","wikidata":"https://www.wikidata.org/wiki/Q5374873","display_name":"Emulation","level":2,"score":0.3296000063419342},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.32679998874664307},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2962999939918518},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.2962999939918518},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.2897999882698059},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2872999906539917},{"id":"https://openalex.org/C2778562939","wikidata":"https://www.wikidata.org/wiki/Q1298791","display_name":"Synchronization (alternating current)","level":3,"score":0.28540000319480896},{"id":"https://openalex.org/C202491316","wikidata":"https://www.wikidata.org/wiki/Q272683","display_name":"Instruction set","level":2,"score":0.28299999237060547},{"id":"https://openalex.org/C92757383","wikidata":"https://www.wikidata.org/wiki/Q382497","display_name":"Affine transformation","level":2,"score":0.2757999897003174},{"id":"https://openalex.org/C459310","wikidata":"https://www.wikidata.org/wiki/Q117801","display_name":"Computational science","level":1,"score":0.2606000006198883},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.25029999017715454}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3712285.3759852","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3712285.3759852","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3712285.3759852","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3712285.3759852","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":7,"referenced_works":["https://openalex.org/W2946609015","https://openalex.org/W2998617917","https://openalex.org/W3194676777","https://openalex.org/W4387321091","https://openalex.org/W4408902464","https://openalex.org/W4415796640","https://openalex.org/W4415797154"],"related_works":[],"abstract_inverted_index":{"Quantization":[0],"is":[1],"a":[2,28,68,83],"critical":[3],"technique":[4],"for":[5,73],"accelerating":[6],"LLM":[7,75],"inference":[8],"by":[9],"reducing":[10],"memory":[11,121],"footprint":[12],"and":[13,22,33,100,111,137,158],"improving":[14],"computational":[15],"efficiency.":[16],"Among":[17],"various":[18,146],"schemes,":[19],"4-bit":[20],"weight":[21,108],"8-bit":[23],"activation":[24],"quantization":[25,85],"(W4A8)":[26],"offers":[27],"strong":[29],"balance":[30],"between":[31],"accuracy":[32],"performance.":[34],"However,":[35],"existing":[36],"W4A8":[37,70,135],"GEMM":[38,71,148],"kernels":[39,136,149],"fall":[40],"short":[41],"in":[42,150],"practice":[43],"due":[44],"to":[45,130,139,145,161],"inefficient":[46],"dequantization":[47,91],"on":[48],"CUDA":[49],"Cores,":[50],"which":[51],"cannot":[52],"keep":[53],"pace":[54],"with":[55],"the":[56],"high":[57],"throughput":[58],"of":[59],"Tensor":[60],"Cores.":[61],"In":[62],"this":[63],"paper,":[64],"we":[65],"present":[66],"LiquidGEMM,":[67],"hardware-efficient":[69,84],"kernel":[72],"efficient":[74],"serving.":[76],"LiquidGEMM":[77,127,153],"designs":[78],"two":[79,94],"key":[80],"techniques:":[81],"LiquidQuant,":[82],"method":[86],"that":[87,105,126],"enables":[88],"fast,":[89],"overflow-safe":[90],"using":[92],"just":[93],"arithmetic":[95],"instructions":[96],"per":[97],"four":[98],"elements;":[99],"an":[101],"implicit":[102],"fine-grained":[103],"pipeline":[104],"fully":[106],"overlaps":[107],"loading,":[109],"dequantization,":[110],"MMA":[112],"across":[113],"warp":[114],"groups":[115],"without":[116],"software":[117],"synchronization":[118],"or":[119],"redundant":[120],"traffic.":[122],"Experimental":[123],"results":[124],"show":[125],"achieves":[128,159],"up":[129,138,160],"2.90x":[131],"speedup":[132],"over":[133],"state-of-the-art":[134],"4.94x":[140],"end-to-end":[141],"system-level":[142,163],"speedup.":[143,164],"Compared":[144],"quantized":[147],"NVIDIA":[151],"TensorRT-LLM,":[152],"delivers":[154],"1.12-1.63x":[155],"performance":[156],"gains,":[157],"1.63x":[162]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2026-03-27T05:58:40.876381","created_date":"2025-11-12T00:00:00"}