{"id":"https://openalex.org/W3159727696","doi":"https://doi.org/10.1109/hpca51647.2021.00018","title":"SpAtten: Efficient Sparse Attention Architecture with Cascade Token and Head Pruning","display_name":"SpAtten: Efficient Sparse Attention Architecture with Cascade Token and Head Pruning","publication_year":2021,"publication_date":"2021-02-01","ids":{"openalex":"https://openalex.org/W3159727696","doi":"https://doi.org/10.1109/hpca51647.2021.00018","mag":"3159727696"},"language":"en","primary_location":{"id":"doi:10.1109/hpca51647.2021.00018","doi":"https://doi.org/10.1109/hpca51647.2021.00018","is_oa":false,"landing_page_url":"https://doi.org/10.1109/hpca51647.2021.00018","pdf_url":null,"source":{"id":"https://openalex.org/S4210184388","display_name":"IEEE International Symposium on High-Performance Computer Architecture (HPCA)","issn_l":"1530-0897","issn":["1530-0897","2378-203X"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","jufo-1","norway-2"],"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2021 IEEE International Symposium on High-Performance Computer Architecture (HPCA)","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2012.09852","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5119012009","display_name":"Hanrui Wang","orcid":"https://orcid.org/0000-0002-7229-4015","observed_orcids":["https://orcid.org/0000-0002-7229-4015","https://orcid.org/0000-0002-7498-0908"]},"institutions":[{"id":"https://openalex.org/I63966007","display_name":"Massachusetts Institute of Technology","ror":"https://ror.org/042nb2s44","country_code":"US","type":"education","lineage":["https://openalex.org/I63966007"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hanrui Wang","raw_affiliation_strings":["Massachusetts Institute of Technology,EECS,Cambridge,MA,US"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Massachusetts Institute of Technology,EECS,Cambridge,MA,US","institution_ids":["https://openalex.org/I63966007"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5065492534","display_name":"Zhekai Zhang","orcid":"https://orcid.org/0000-0002-9777-3792","observed_orcids":["https://orcid.org/0000-0002-9777-3792"]},"institutions":[{"id":"https://openalex.org/I63966007","display_name":"Massachusetts Institute of Technology","ror":"https://ror.org/042nb2s44","country_code":"US","type":"education","lineage":["https://openalex.org/I63966007"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhekai Zhang","raw_affiliation_strings":["Massachusetts Institute of Technology,EECS,Cambridge,MA,US"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Massachusetts Institute of Technology,EECS,Cambridge,MA,US","institution_ids":["https://openalex.org/I63966007"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5070926896","display_name":"Song Han","orcid":"https://orcid.org/0000-0002-4186-7618","observed_orcids":["https://orcid.org/0000-0002-4186-7618"]},"institutions":[{"id":"https://openalex.org/I63966007","display_name":"Massachusetts Institute of Technology","ror":"https://ror.org/042nb2s44","country_code":"US","type":"education","lineage":["https://openalex.org/I63966007"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Song Han","raw_affiliation_strings":["Massachusetts Institute of Technology,EECS,Cambridge,MA,US"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Massachusetts Institute of Technology,EECS,Cambridge,MA,US","institution_ids":["https://openalex.org/I63966007"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I63966007"],"apc_list":null,"apc_paid":null,"fwci":18.4417,"has_fulltext":true,"cited_by_count":427,"citation_normalized_percentile":{"value":0.99414655,"is_in_top_1_percent":true,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":88,"max":100},"biblio":{"volume":null,"issue":null,"first_page":"97","last_page":"110"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.5806000232696533,"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":0.5806000232696533,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.36059999465942383,"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/T10028","display_name":"Topic Modeling","score":0.011099999770522118,"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/sparse-attention","display_name":"sparse attention","score":0.7839999794960022},{"id":"https://openalex.org/keywords/token-pruning","display_name":"token pruning","score":0.7549999952316284},{"id":"https://openalex.org/keywords/algorithm-architecture-co-design","display_name":"algorithm-architecture co-design","score":0.8450000286102295},{"id":"https://openalex.org/keywords/dram-access-reduction","display_name":"DRAM access reduction","score":0.777999997138977},{"id":"https://openalex.org/keywords/energy-efficiency","display_name":"energy efficiency","score":0.5870000123977661}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8281580805778503},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5774999856948853},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.5517199635505676},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.5411161780357361},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.539264976978302},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5171051025390625},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.5123916268348694},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.47023364901542664},{"id":"https://openalex.org/C7366592","wikidata":"https://www.wikidata.org/wiki/Q1255620","display_name":"Dram","level":2,"score":0.46479031443595886},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.4118240475654602},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.36941084265708923},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.23926222324371338},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.16356870532035828},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.0},{"id":"https://openalex.org/C6557445","wikidata":"https://www.wikidata.org/wiki/Q173113","display_name":"Agronomy","level":1,"score":0.0},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.0},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.0}],"mesh":[],"locations_count":4,"locations":[{"id":"doi:10.1109/hpca51647.2021.00018","doi":"https://doi.org/10.1109/hpca51647.2021.00018","is_oa":false,"landing_page_url":"https://doi.org/10.1109/hpca51647.2021.00018","pdf_url":null,"source":{"id":"https://openalex.org/S4210184388","display_name":"IEEE International Symposium on High-Performance Computer Architecture (HPCA)","issn_l":"1530-0897","issn":["1530-0897","2378-203X"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","jufo-1","norway-2"],"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2021 IEEE International Symposium on High-Performance Computer Architecture (HPCA)","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2012.09852","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2012.09852","pdf_url":"https://arxiv.org/pdf/2012.09852","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":"pmh:oai:dspace.mit.edu:1721.1/143674","doi":null,"is_oa":true,"landing_page_url":"https://hdl.handle.net/1721.1/143674","pdf_url":"https://dspace.mit.edu/bitstream/1721.1/143674/2/2012.09852.pdf","source":{"id":"https://openalex.org/S4306400425","display_name":"DSpace@MIT (Massachusetts Institute of Technology)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I63966007","host_organization_name":"Massachusetts Institute of Technology","host_organization_lineage":["https://openalex.org/I63966007"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-nc-sa","license_id":"https://openalex.org/licenses/cc-by-nc-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"arXiv","raw_type":"http://purl.org/eprint/type/ConferencePaper"},{"id":"doi:10.48550/arxiv.2012.09852","doi":"https://doi.org/10.48550/arxiv.2012.09852","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2012.09852","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":"article-journal"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2012.09852","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2012.09852","pdf_url":"https://arxiv.org/pdf/2012.09852","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":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G6671297155","display_name":null,"funder_award_id":"CAREER","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G8619146197","display_name":"CAREER: Efficient Algorithms and Hardware for Accelerated Machine Learning","funder_award_id":"1943349","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320332180","display_name":"Defense Advanced Research Projects Agency","ror":"https://ror.org/02caytj08"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":90,"referenced_works":["https://openalex.org/W1539309091","https://openalex.org/W1585048358","https://openalex.org/W1632114991","https://openalex.org/W1793121960","https://openalex.org/W2034861439","https://openalex.org/W2042519306","https://openalex.org/W2119144962","https://openalex.org/W2144578941","https://openalex.org/W2251939518","https://openalex.org/W2285660444","https://openalex.org/W2464177207","https://openalex.org/W2525332836","https://openalex.org/W2563587242","https://openalex.org/W2565851976","https://openalex.org/W2606722458","https://openalex.org/W2617247391","https://openalex.org/W2625457103","https://openalex.org/W2626778328","https://openalex.org/W2719597717","https://openalex.org/W2734763043","https://openalex.org/W2761132374","https://openalex.org/W2767206889","https://openalex.org/W2794141774","https://openalex.org/W2794952988","https://openalex.org/W2810704618","https://openalex.org/W2886851211","https://openalex.org/W2891911973","https://openalex.org/W2896457183","https://openalex.org/W2904529841","https://openalex.org/W2904902077","https://openalex.org/W2913431662","https://openalex.org/W2923014074","https://openalex.org/W2931118404","https://openalex.org/W2945767825","https://openalex.org/W2946794439","https://openalex.org/W2949591530","https://openalex.org/W2949674408","https://openalex.org/W2949870694","https://openalex.org/W2951008357","https://openalex.org/W2951714314","https://openalex.org/W2952857977","https://openalex.org/W2954711779","https://openalex.org/W2963310665","https://openalex.org/W2963341956","https://openalex.org/W2963363373","https://openalex.org/W2963367920","https://openalex.org/W2963748441","https://openalex.org/W2964299589","https://openalex.org/W2965643269","https://openalex.org/W2970565456","https://openalex.org/W2979310060","https://openalex.org/W2979439447","https://openalex.org/W2979719709","https://openalex.org/W2980113464","https://openalex.org/W2980186997","https://openalex.org/W3003074159","https://openalex.org/W3007303655","https://openalex.org/W3011554625","https://openalex.org/W3016542674","https://openalex.org/W3016832937","https://openalex.org/W3022810465","https://openalex.org/W3025572566","https://openalex.org/W3035251378","https://openalex.org/W3035332806","https://openalex.org/W3042495273","https://openalex.org/W3042501387","https://openalex.org/W3042803431","https://openalex.org/W3043023836","https://openalex.org/W3043406639","https://openalex.org/W3043433718","https://openalex.org/W3043495022","https://openalex.org/W3043504674","https://openalex.org/W3043802286","https://openalex.org/W3092618035","https://openalex.org/W3099513244","https://openalex.org/W3099971873","https://openalex.org/W3100023286","https://openalex.org/W3101262113","https://openalex.org/W3102337469","https://openalex.org/W3102587717","https://openalex.org/W3103168911","https://openalex.org/W3103835917","https://openalex.org/W3103837983","https://openalex.org/W3105802176","https://openalex.org/W4236868170","https://openalex.org/W4240168186","https://openalex.org/W4247198796","https://openalex.org/W4285719527","https://openalex.org/W6719768283","https://openalex.org/W6966426965"],"related_works":["https://openalex.org/W4360831803","https://openalex.org/W7133504784","https://openalex.org/W4414198425","https://openalex.org/W4280582299","https://openalex.org/W7167698792","https://openalex.org/W7167088804","https://openalex.org/W4402599357","https://openalex.org/W4308083827","https://openalex.org/W4407784386","https://openalex.org/W4414500226"],"abstract_inverted_index":{"The":[0],"attention":[1,33,83,137,197],"mechanism":[2],"is":[3,123,131,189],"becoming":[4],"increasingly":[5],"popular":[6],"in":[7,108,135],"Natural":[8],"Language":[9],"Processing":[10],"(NLP)":[11],"applications,":[12],"showing":[13],"superior":[14],"performance":[15],"than":[16],"convolutional":[17,52],"and":[18,27,40,56,77,85,139,143,165,182,194,222,229],"recurrent":[19,54],"architectures.":[20],"However,":[21],"general-purpose":[22],"platforms":[23],"such":[24],"as":[25],"CPUs":[26],"GPUs":[28],"are":[29,145],"inefficient":[30],"when":[31],"performing":[32],"inference":[34],"due":[35],"to":[36,80,103,117,162],"complicated":[37],"data":[38],"movement":[39],"low":[41],"arithmetic":[42],"intensity.":[43],"Moreover,":[44],"existing":[45],"NN":[46],"accelerators":[47],"mainly":[48],"focus":[49],"on":[50,147,154,205,210],"optimizing":[51],"or":[53],"models,":[55],"cannot":[57],"efficiently":[58,151],"support":[59,152],"attention.":[60],"In":[61],"this":[62],"paper,":[63],"we":[64,96,156,173],"present":[65],"SpAtten,":[66],"an":[67],"efficient":[68],"algorithm-architecture":[69],"co-design":[70],"that":[71,177],"leverages":[72],"token":[73,101,164],"sparsity,":[74,76],"head":[75,115,166],"quantization":[78,176],"opportunities":[79],"reduce":[81],"the":[82,90,98,109,136,140,148,184,187,196],"computation":[84,200],"memory":[86,202],"access.":[87],"Inspired":[88],"by":[89,216],"high":[91,170],"redundancy":[92],"of":[93],"human":[94],"languages,":[95],"propose":[97,113,174],"novel":[99,159],"cascade":[100,114],"pruning":[102,116,122,128],"prune":[104],"away":[105],"unimportant":[106],"tokens":[107,142],"sentence.":[110],"We":[111],"also":[112],"remove":[118],"unessential":[119],"heads.":[120],"Cascade":[121],"fundamentally":[124],"different":[125],"from":[126],"weight":[127,134],"since":[129],"there":[130],"no":[132,219],"trainable":[133],"mechanism,":[138],"pruned":[141],"heads":[144],"selected":[146],"fly.":[149],"To":[150],"them":[153],"hardware,":[155],"design":[157],"a":[158],"top-k":[160],"engine":[161],"rank":[163],"importance":[167],"scores":[168],"with":[169,218],"throughput.":[171],"Furthermore,":[172],"progressive":[175],"first":[178],"fetches":[179,192],"MSBs":[180],"only":[181],"performs":[183],"computation;":[185],"if":[186],"confidence":[188],"low,":[190],"it":[191],"LSBs":[193],"recomputes":[195],"outputs,":[198],"trading":[199],"for":[201],"reduction.Extensive":[203],"experiments":[204],"30":[206],"benchmarks":[207],"show":[208],"that,":[209],"average,":[211],"SpAtten":[212],"reduces":[213],"DRAM":[214],"access":[215],"10.0\u00d7":[217],"accuracy":[220],"loss,":[221],"achieves":[223],"1.6\u00d7,":[224],"3.0\u00d7,":[225],"162\u00d7,":[226],"347\u00d7":[227],"speedup,":[228],"1.4\u00d7,":[230],"3.2\u00d7,":[231],"1193\u00d7,":[232],"4059\u00d7":[233],"energy":[234],"savings":[235],"over":[236],"A3accelerator,":[237],"MNNFast":[238],"accelerator,":[239],"TITAN":[240],"Xp":[241],"GPU,":[242],"Xeon":[243],"CPU,":[244],"respectively.":[245]},"counts_by_year":[{"year":2026,"cited_by_count":62},{"year":2025,"cited_by_count":138},{"year":2024,"cited_by_count":107},{"year":2023,"cited_by_count":62},{"year":2022,"cited_by_count":43},{"year":2021,"cited_by_count":14},{"year":2020,"cited_by_count":1}],"updated_date":"2026-10-31T12:09:11.096373","created_date":"2021-05-10T00:00:00"}