{"id":"https://openalex.org/W4392088024","doi":"https://doi.org/10.48550/arxiv.2402.13516","title":"ProSparse: Introducing and Enhancing Intrinsic Activation Sparsity within Large Language Models","display_name":"ProSparse: Introducing and Enhancing Intrinsic Activation Sparsity within Large Language Models","publication_year":2024,"publication_date":"2024-02-21","ids":{"openalex":"https://openalex.org/W4392088024","doi":"https://doi.org/10.48550/arxiv.2402.13516"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2402.13516","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.13516","pdf_url":"https://arxiv.org/pdf/2402.13516","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2402.13516","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5108903196","display_name":"Chenyang Song","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Song, Chenyang","raw_affiliation_strings":["Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101259627","display_name":"Xu Han","orcid":"https://orcid.org/0000-0001-6998-8254","observed_orcids":["https://orcid.org/0000-0001-6998-8254"]},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Han, Xu","raw_affiliation_strings":["Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101584823","display_name":"Zhengyan Zhang","orcid":"https://orcid.org/0000-0003-2988-0083","observed_orcids":["https://orcid.org/0000-0003-2988-0083"]},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhang, Zhengyan","raw_affiliation_strings":["Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010557815","display_name":"Shengding Hu","orcid":"https://orcid.org/0009-0000-8037-2055","observed_orcids":["https://orcid.org/0009-0000-8037-2055"]},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Hu, Shengding","raw_affiliation_strings":["Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033198789","display_name":"Xiyu Shi","orcid":"https://orcid.org/0000-0001-6174-3383","observed_orcids":["https://orcid.org/0000-0001-6174-3383"]},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210090176","display_name":"Institute of Computing Technology","ror":"https://ror.org/0090r4d87","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shi, Xiyu","raw_affiliation_strings":["SKLP , Institute of Computing Technology , Chinese Academy of Sciences , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SKLP , Institute of Computing Technology , Chinese Academy of Sciences , China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102620905","display_name":"Kuai Li","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Li, Kuai","raw_affiliation_strings":["Tencent Machine Learning Platform , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tencent Machine Learning Platform , China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100418547","display_name":"Chen Chen","orcid":"https://orcid.org/0000-0003-1957-6432","observed_orcids":["https://orcid.org/0000-0003-1957-6432"]},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chen, Chen","raw_affiliation_strings":["Tencent Machine Learning Platform , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tencent Machine Learning Platform , China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100320723","display_name":"Zhiyuan Liu","orcid":"https://orcid.org/0000-0002-7709-2543","observed_orcids":["https://orcid.org/0000-0002-7709-2543","https://orcid.org/0009-0002-0195-7114"]},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Liu, Zhiyuan","raw_affiliation_strings":["Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100607669","display_name":"Guangli Li","orcid":"https://orcid.org/0000-0002-9738-261X","observed_orcids":["https://orcid.org/0000-0002-9738-261X"]},"institutions":[{"id":"https://openalex.org/I19820366","display_name":"Chinese Academy of Sciences","ror":"https://ror.org/034t30j35","country_code":"CN","type":"government","lineage":["https://openalex.org/I19820366"]},{"id":"https://openalex.org/I4210090176","display_name":"Institute of Computing Technology","ror":"https://ror.org/0090r4d87","country_code":"CN","type":"facility","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Li, Guangli","raw_affiliation_strings":["SKLP , Institute of Computing Technology , Chinese Academy of Sciences , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SKLP , Institute of Computing Technology , Chinese Academy of Sciences , China","institution_ids":["https://openalex.org/I19820366","https://openalex.org/I4210090176"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088935322","display_name":"Tao Yang","orcid":"https://orcid.org/0000-0003-0534-1034","observed_orcids":["https://orcid.org/0000-0003-0534-1034"]},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yang, Tao","raw_affiliation_strings":["Tencent Machine Learning Platform , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tencent Machine Learning Platform , China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5046448314","display_name":"Maosong Sun","orcid":"https://orcid.org/0000-0002-6011-6115","observed_orcids":["https://orcid.org/0000-0002-6011-6115"]},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Sun, Maosong","raw_affiliation_strings":["Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Dept. of Comp. Sci. & Tech. , Institute for AI , Tsinghua University , Beijing , China","institution_ids":["https://openalex.org/I99065089"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":42,"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":0.9384999871253967,"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":0.9384999871253967,"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/T10028","display_name":"Topic Modeling","score":0.044599998742341995,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.002199999988079071,"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/activation-sparsity","display_name":"activation sparsity","score":0.9980000257492065},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9779999852180481},{"id":"https://openalex.org/keywords/relu-activation-function","display_name":"ReLU activation","score":0.8840000033378601},{"id":"https://openalex.org/keywords/inference-acceleration","display_name":"inference acceleration","score":0.9490000009536743},{"id":"https://openalex.org/keywords/llama","display_name":"llamas","score":0.6230000257492065},{"id":"https://openalex.org/keywords/model-compression","display_name":"model compression","score":0.38199999928474426}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4386480450630188},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.36718541383743286},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.3322867155075073},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.12437421083450317}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2402.13516","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.13516","pdf_url":"https://arxiv.org/pdf/2402.13516","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2402.13516","doi":"https://doi.org/10.48550/arxiv.2402.13516","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2402.13516","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:2402.13516","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.13516","pdf_url":"https://arxiv.org/pdf/2402.13516","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G1681167796","display_name":null,"funder_award_id":"62236011","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G1989434931","display_name":null,"funder_award_id":"2023M733566","funder_id":"https://openalex.org/F4320321543","funder_display_name":"China Postdoctoral Science Foundation"},{"id":"https://openalex.org/G2565984771","display_name":null,"funder_award_id":"62236004","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G7223888118","display_name":null,"funder_award_id":"62302479","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320321543","display_name":"China Postdoctoral Science Foundation","ror":"https://ror.org/0426zh255"},{"id":"https://openalex.org/F4320322392","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4392088024.pdf","grobid_xml":"https://content.openalex.org/works/W4392088024.grobid-xml"},"referenced_works_count":41,"referenced_works":["https://openalex.org/W2766839578","https://openalex.org/W2896457183","https://openalex.org/W2899663614","https://openalex.org/W2924902521","https://openalex.org/W3001279689","https://openalex.org/W3006439205","https://openalex.org/W3083410900","https://openalex.org/W3118781290","https://openalex.org/W3177813494","https://openalex.org/W3194959296","https://openalex.org/W3195140297","https://openalex.org/W3198002980","https://openalex.org/W3210904070","https://openalex.org/W4229005866","https://openalex.org/W4306808908","https://openalex.org/W4319166707","https://openalex.org/W4320854821","https://openalex.org/W4322718191","https://openalex.org/W4327810158","https://openalex.org/W4362515116","https://openalex.org/W4368755730","https://openalex.org/W4376167329","https://openalex.org/W4380993498","https://openalex.org/W4381586827","https://openalex.org/W4384918448","https://openalex.org/W4385570984","https://openalex.org/W4385653792","https://openalex.org/W4386839902","https://openalex.org/W4387559178","https://openalex.org/W4387561365","https://openalex.org/W4389157038","https://openalex.org/W4389519291","https://openalex.org/W4391631824","https://openalex.org/W4393146966","https://openalex.org/W4394657409","https://openalex.org/W4394709684","https://openalex.org/W4395112660","https://openalex.org/W4401042689","https://openalex.org/W4404401018","https://openalex.org/W4404781868","https://openalex.org/W7133245307"],"related_works":["https://openalex.org/W4399596941","https://openalex.org/W4417451801","https://openalex.org/W4402704384","https://openalex.org/W4417182790","https://openalex.org/W4410583234","https://openalex.org/W4416895716","https://openalex.org/W4391631824","https://openalex.org/W4404355295","https://openalex.org/W4416037140","https://openalex.org/W4415796073"],"abstract_inverted_index":{"Activation":[0],"sparsity":[1,26,50,75,85,107,125,141,158],"refers":[2],"to":[3,33,70,101,175,221],"the":[4,18,21,66,115,133,182,208],"existence":[5],"of":[6,17,118,159,213],"considerable":[7],"weakly-contributed":[8],"elements":[9],"among":[10,187],"activation":[11,23,25,45,49,68,74,106,116,140,151,217],"outputs.":[12],"As":[13],"a":[14,30,93,128],"prevalent":[15],"property":[16],"models":[19,42,186],"using":[20],"ReLU":[22,61],"function,":[24],"has":[27],"been":[28],"proven":[29],"promising":[31],"paradigm":[32],"boost":[34],"model":[35,88],"inference":[36,77,203,223],"efficiency.":[37],"Nevertheless,":[38],"most":[39,183],"large":[40],"language":[41],"(LLMs)":[43],"adopt":[44],"functions":[46],"without":[47],"intrinsic":[48],"(e.g.,":[51],"GELU":[52],"and":[53,76,86,95,142,166,191,199],"Swish).":[54],"Some":[55],"recent":[56],"efforts":[57],"have":[58],"explored":[59],"introducing":[60],"or":[62],"its":[63],"variants":[64],"as":[65],"substitutive":[67],"function":[69,117],"help":[71],"LLMs":[72,103,119,214],"achieve":[73],"acceleration,":[78],"but":[79],"few":[80],"can":[81,138],"simultaneously":[82],"obtain":[83,156],"high":[84,157],"comparable":[87,110,173],"performance.":[89,111],"This":[90,137],"paper":[91],"introduces":[92],"simple":[94],"effective":[96],"sparsification":[97],"method":[98],"named":[99],"\"ProSparse\"":[100],"push":[102],"for":[104,161,164,168],"higher":[105,216],"while":[108],"maintaining":[109],"Specifically,":[112],"after":[113],"substituting":[114],"with":[120,127,215],"ReLU,":[121],"ProSparse":[122],"adopts":[123],"progressive":[124],"regularization":[126],"factor":[129],"smoothly":[130],"increasing":[131],"along":[132],"multi-stage":[134],"sine":[135],"curves.":[136],"enhance":[139],"mitigate":[143],"performance":[144,174],"degradation":[145],"by":[146],"avoiding":[147],"radical":[148],"shifts":[149],"in":[150],"distributions.":[152],"With":[153],"ProSparse,":[154],"we":[155],"89.32%":[160],"LLaMA2-7B,":[162],"88.80%":[163],"LLaMA2-13B,":[165],"87.89%":[167],"end-size":[169,193],"MiniCPM-1B,":[170],"respectively,":[171],"achieving":[172],"their":[176],"original":[177],"Swish-activated":[178],"versions.":[179],"These":[180],"present":[181],"sparsely":[184],"activated":[185],"open-source":[188],"LLaMA":[189],"versions":[190],"competitive":[192],"models,":[194],"considerably":[195],"surpassing":[196],"ReluLLaMA-7B":[197],"(66.98%)":[198],"ReluLLaMA-13B":[200],"(71.56%).":[201],"Our":[202],"acceleration":[204,211],"experiments":[205],"further":[206],"demonstrate":[207],"significant":[209],"practical":[210],"potential":[212],"sparsity,":[218],"obtaining":[219],"up":[220],"4.52$\\times$":[222],"speedup.":[224]},"counts_by_year":[{"year":2026,"cited_by_count":9},{"year":2025,"cited_by_count":22},{"year":2024,"cited_by_count":11}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2024-02-23T00:00:00"}