{"id":"https://openalex.org/W4382322807","doi":"https://doi.org/10.1145/3580305.3599284","title":"Constraint-aware and Ranking-distilled Token Pruning for Efficient Transformer Inference","display_name":"Constraint-aware and Ranking-distilled Token Pruning for Efficient Transformer Inference","publication_year":2023,"publication_date":"2023-08-04","ids":{"openalex":"https://openalex.org/W4382322807","doi":"https://doi.org/10.1145/3580305.3599284"},"language":"en","primary_location":{"id":"doi:10.1145/3580305.3599284","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3580305.3599284","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["arxiv","crossref"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2306.14393","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100621054","display_name":"Junyan Li","orcid":"https://orcid.org/0009-0005-3557-4922"},"institutions":[{"id":"https://openalex.org/I76130692","display_name":"Zhejiang University","ror":"https://ror.org/00a2xv884","country_code":"CN","type":"education","lineage":["https://openalex.org/I76130692"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Junyan Li","raw_affiliation_strings":["Zhejiang University, Hangzhou, China"],"raw_orcid":"https://orcid.org/0009-0005-3557-4922","affiliations":[{"raw_affiliation_string":"Zhejiang University, Hangzhou, China","institution_ids":["https://openalex.org/I76130692"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103007399","display_name":"Li Lyna Zhang","orcid":"https://orcid.org/0000-0002-4465-1628"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Li Lyna Zhang","raw_affiliation_strings":["Microsoft Research, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-4465-1628","affiliations":[{"raw_affiliation_string":"Microsoft Research, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5065385916","display_name":"Jiahang Xu","orcid":"https://orcid.org/0000-0001-9186-619X"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jiahang Xu","raw_affiliation_strings":["Microsoft Research, Shanghai, China"],"raw_orcid":"https://orcid.org/0000-0001-9186-619X","affiliations":[{"raw_affiliation_string":"Microsoft Research, Shanghai, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100776576","display_name":"Yujing Wang","orcid":"https://orcid.org/0000-0002-7940-5216"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yujing Wang","raw_affiliation_strings":["Microsoft, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-7940-5216","affiliations":[{"raw_affiliation_string":"Microsoft, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5026114186","display_name":"Shaoguang Yan","orcid":"https://orcid.org/0009-0009-1990-5743"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shaoguang Yan","raw_affiliation_strings":["Microsoft, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0009-1990-5743","affiliations":[{"raw_affiliation_string":"Microsoft, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5069555295","display_name":"Yunqing Xia","orcid":"https://orcid.org/0009-0005-8608-574X"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yunqing Xia","raw_affiliation_strings":["Microsoft, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0005-8608-574X","affiliations":[{"raw_affiliation_string":"Microsoft, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101421201","display_name":"Yuqing Yang","orcid":"https://orcid.org/0000-0003-3518-5212"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yuqing Yang","raw_affiliation_strings":["Microsoft Research, Shanghai, China"],"raw_orcid":"https://orcid.org/0000-0003-3518-5212","affiliations":[{"raw_affiliation_string":"Microsoft Research, Shanghai, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101534161","display_name":"Ting Cao","orcid":"https://orcid.org/0000-0002-9107-013X"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Ting Cao","raw_affiliation_strings":["Microsoft Research, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-9107-013X","affiliations":[{"raw_affiliation_string":"Microsoft Research, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101539732","display_name":"Hao Sun","orcid":"https://orcid.org/0009-0004-5027-7478"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Hao Sun","raw_affiliation_strings":["Microsoft, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0004-5027-7478","affiliations":[{"raw_affiliation_string":"Microsoft, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5016794246","display_name":"Weiwei Deng","orcid":"https://orcid.org/0009-0001-4793-9715"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Weiwei Deng","raw_affiliation_strings":["Microsoft, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0001-4793-9715","affiliations":[{"raw_affiliation_string":"Microsoft, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5022403899","display_name":"Qi Zhang","orcid":"https://orcid.org/0009-0009-7438-7248"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qi Zhang","raw_affiliation_strings":["Microsoft, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0009-7438-7248","affiliations":[{"raw_affiliation_string":"Microsoft, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100438310","display_name":"Mao Yang","orcid":"https://orcid.org/0009-0009-6455-3898"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Mao Yang","raw_affiliation_strings":["Microsoft Research, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0009-6455-3898","affiliations":[{"raw_affiliation_string":"Microsoft Research, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":13,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1280","last_page":"1290"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9855999946594238,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9855999946594238,"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/T14276","display_name":"Power Systems and Technologies","score":0.9726999998092651,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11269","display_name":"Algorithms and Data Compression","score":0.9682000279426575,"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/security-token","display_name":"Security token","score":0.7272507548332214},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6124860048294067},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6099553108215332},{"id":"https://openalex.org/keywords/constraint","display_name":"Constraint (computer-aided design)","score":0.5872999429702759},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.5455571413040161},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.5078058838844299},{"id":"https://openalex.org/keywords/ranking","display_name":"Ranking (information retrieval)","score":0.4634411633014679},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4367567002773285},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.23297721147537231},{"id":"https://openalex.org/keywords/computer-security","display_name":"Computer security","score":0.15564125776290894},{"id":"https://openalex.org/keywords/engineering","display_name":"Engineering","score":0.09042161703109741},{"id":"https://openalex.org/keywords/biology","display_name":"Biology","score":0.07068473100662231}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.7272507548332214},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6124860048294067},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6099553108215332},{"id":"https://openalex.org/C2776036281","wikidata":"https://www.wikidata.org/wiki/Q48769818","display_name":"Constraint (computer-aided design)","level":2,"score":0.5872999429702759},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5455571413040161},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.5078058838844299},{"id":"https://openalex.org/C189430467","wikidata":"https://www.wikidata.org/wiki/Q7293293","display_name":"Ranking (information retrieval)","level":2,"score":0.4634411633014679},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4367567002773285},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.23297721147537231},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.15564125776290894},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.09042161703109741},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.07068473100662231},{"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/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.0},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1145/3580305.3599284","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3580305.3599284","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2306.14393","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2306.14393","pdf_url":"https://arxiv.org/pdf/2306.14393","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":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2306.14393","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2306.14393","pdf_url":"https://arxiv.org/pdf/2306.14393","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":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[{"score":0.4300000071525574,"display_name":"Decent work and economic growth","id":"https://metadata.un.org/sdg/8"}],"awards":[],"funders":[{"id":"https://openalex.org/F4320308943","display_name":"Microsoft Research","ror":"https://ror.org/00d0nc645"}],"has_content":{"pdf":true,"grobid_xml":false},"content_urls":{"pdf":"https://content.openalex.org/works/W4382322807.pdf"},"referenced_works_count":21,"referenced_works":["https://openalex.org/W1493526108","https://openalex.org/W2898073868","https://openalex.org/W2948947170","https://openalex.org/W2963323070","https://openalex.org/W2979691890","https://openalex.org/W2981852735","https://openalex.org/W2998183051","https://openalex.org/W3038012435","https://openalex.org/W3105966348","https://openalex.org/W3159727696","https://openalex.org/W3167266074","https://openalex.org/W3171750540","https://openalex.org/W3174488167","https://openalex.org/W3174708387","https://openalex.org/W3180037928","https://openalex.org/W4226126941","https://openalex.org/W4245255589","https://openalex.org/W4287704453","https://openalex.org/W4288089799","https://openalex.org/W4306317920","https://openalex.org/W6781533629"],"related_works":["https://openalex.org/W4388335561","https://openalex.org/W2970530566","https://openalex.org/W4288261899","https://openalex.org/W4307309205","https://openalex.org/W2967478618","https://openalex.org/W4385009901","https://openalex.org/W4385572700","https://openalex.org/W2997152889","https://openalex.org/W4387768015","https://openalex.org/W4304700937"],"abstract_inverted_index":{"Deploying":[0],"pre-trained":[1],"transformer":[2,114],"models":[3,93],"like":[4],"BERT":[5,158],"on":[6,130,165,177],"downstream":[7],"tasks":[8,135],"in":[9,71],"resource-constrained":[10],"scenarios":[11],"is":[12,182],"challenging":[13],"due":[14],"to":[15,54,94,175],"their":[16],"high":[17],"inference":[18,57],"cost,":[19],"which":[20,40,82],"grows":[21],"rapidly":[22],"with":[23,147],"input":[24,46],"sequence":[25,47],"length.":[26],"In":[27],"this":[28],"work,":[29],"we":[30],"propose":[31],"a":[32,77,103,169],"constraint-aware":[33],"and":[34,116,133,143,150,167],"ranking-distilled":[35,78],"token":[36,68,79,85,118,141],"pruning":[37,105,119,142],"method":[38],"ToP,":[39],"selectively":[41],"removes":[42],"unnecessary":[43],"tokens":[44],"as":[45],"passes":[48],"through":[49,76,124],"layers,":[50],"allowing":[51],"the":[52,64,72,88,110,154],"model":[53,144],"improve":[55],"online":[56],"speed":[58],"while":[59,161],"preserving":[60],"accuracy.":[61],"ToP":[62,101,138,152],"overcomes":[63],"limitation":[65],"of":[66,91,97,113,157,173],"inaccurate":[67],"importance":[69],"ranking":[70],"conventional":[73],"self-attention":[74],"mechanism":[75],"distillation":[80],"technique,":[81],"distills":[83],"effective":[84],"rankings":[86],"from":[87],"final":[89],"layer":[90],"unpruned":[92],"early":[95],"layers":[96,115,123],"pruned":[98],"models.":[99],"Then,":[100],"introduces":[102],"coarse-to-fine":[104],"approach":[106],"that":[107,137],"automatically":[108],"selects":[109],"optimal":[111],"subset":[112],"optimizes":[117],"decisions":[120],"within":[121],"these":[122],"improved":[125,148],"L0":[126],"regularization.":[127],"Extensive":[128],"experiments":[129],"GLUE":[131],"benchmark":[132],"SQuAD":[134],"demonstrate":[136],"outperforms":[139],"state-of-the-art":[140],"compression":[145],"methods":[146],"accuracy":[149,164],"speedups.":[151],"reduces":[153],"average":[155],"FLOPs":[156],"by":[159],"8.1X":[160],"achieving":[162],"competitive":[163],"GLUE,":[166],"provides":[168],"real":[170],"latency":[171],"speedup":[172],"up":[174],"7.4X":[176],"an":[178],"Intel":[179],"CPU.":[180],"Code":[181],"available":[183],"at":[184],"https://github.com/microsoft/Moonlit/tree/main/ToP":[185]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":7},{"year":2024,"cited_by_count":3}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2023-06-28T00:00:00"}