{"id":"https://openalex.org/W4400720190","doi":"https://doi.org/10.48550/arxiv.2407.10969","title":"Q-Sparse: All Large Language Models can be Fully Sparsely-Activated","display_name":"Q-Sparse: All Large Language Models can be Fully Sparsely-Activated","publication_year":2024,"publication_date":"2024-07-15","ids":{"openalex":"https://openalex.org/W4400720190","doi":"https://doi.org/10.48550/arxiv.2407.10969"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2407.10969","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.10969","pdf_url":"https://arxiv.org/pdf/2407.10969","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"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2407.10969","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5104631605","display_name":"Hongyu Wang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Hongyu","raw_affiliation_strings":["https ://aka. / GeneralAI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"https ://aka. / GeneralAI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104631606","display_name":"Shuming Ma","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Shuming","raw_affiliation_strings":["https ://aka. / GeneralAI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"https ://aka. / GeneralAI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104631607","display_name":"Ruiping Wang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Ruiping","raw_affiliation_strings":["https ://aka. / GeneralAI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"https ://aka. / GeneralAI","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5014662947","display_name":"Furu Wei","orcid":"https://orcid.org/0000-0002-7810-5852","observed_orcids":["https://orcid.org/0000-0002-7810-5852"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wei, Furu","raw_affiliation_strings":["https ://aka. / GeneralAI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"https ://aka. / GeneralAI","institution_ids":[]}]}],"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":true,"cited_by_count":2,"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":0.9208999872207642,"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":0.9208999872207642,"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":0.055399999022483826,"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":0.003599999938160181,"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-activation","display_name":"sparse activation","score":0.5730000138282776},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9700000286102295},{"id":"https://openalex.org/keywords/top-k-sparsification","display_name":"top-k sparsification","score":0.953000009059906},{"id":"https://openalex.org/keywords/straight-through-estimator","display_name":"straight-through estimator","score":0.875},{"id":"https://openalex.org/keywords/inference-efficiency","display_name":"inference efficiency","score":0.9039999842643738},{"id":"https://openalex.org/keywords/one-bit-quantization","display_name":"1-bit quantization","score":0.7829999923706055},{"id":"https://openalex.org/keywords/bitnet","display_name":"BITNET","score":0.9549999833106995},{"id":"https://openalex.org/keywords/batch-training","display_name":"batch training","score":0.34299999475479126},{"id":"https://openalex.org/keywords/inference-latency-reduction","display_name":"inference latency reduction","score":0.3240000009536743},{"id":"https://openalex.org/keywords/energy-efficiency","display_name":"energy efficiency","score":0.328000009059906}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5160457491874695}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2407.10969","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.10969","pdf_url":"https://arxiv.org/pdf/2407.10969","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":"doi:10.48550/arxiv.2407.10969","doi":"https://doi.org/10.48550/arxiv.2407.10969","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2407.10969","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2407.10969","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.10969","pdf_url":"https://arxiv.org/pdf/2407.10969","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":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4400720190.pdf","grobid_xml":"https://content.openalex.org/works/W4400720190.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052","https://openalex.org/W2382290278","https://openalex.org/W4395014643"],"abstract_inverted_index":{"We":[0,47,84],"introduce,":[1],"Q-Sparse,":[2],"a":[3,137],"simple":[4],"yet":[5],"effective":[6,96],"approach":[7],"to":[8,38,44,70,140],"training":[9,54],"sparsely-activated":[10,91],"large":[11],"language":[12],"models":[13],"(LLMs).":[14],"Q-Sparse":[15,51,65,94,109,127],"enables":[16],"full":[17],"sparsity":[18],"of":[19,72,103,123,149],"activations":[20,40],"in":[21,29,97],"LLMs":[22,74,116],"which":[23],"can":[24,66],"bring":[25],"significant":[26],"efficiency":[27],"gains":[28],"inference.":[30,56],"This":[31],"is":[32,95],"achieved":[33],"by":[34],"applying":[35],"top-K":[36],"sparsification":[37],"the":[39,42,45,121,134,142],"and":[41,55,106,114,126,136,146],"straight-through-estimator":[43],"training.":[46],"also":[48],"introduce":[49],"Block":[50],"for":[52,90,111],"batch":[53],"The":[57],"key":[58],"results":[59,68],"from":[60],"this":[61],"work":[62],"are,":[63],"(1)":[64],"achieve":[67],"comparable":[69],"those":[71],"baseline":[73],"while":[75],"being":[76],"much":[77],"more":[78],"efficient":[79],"at":[80],"inference":[81],"time;":[82],"(2)":[83],"present":[85],"an":[86],"inference-optimal":[87],"scaling":[88],"law":[89],"LLMs;":[92],"(3)":[93],"different":[98],"settings,":[99],"including":[100,144],"training-from-scratch,":[101],"continue-training":[102],"off-the-shelf":[104],"LLMs,":[105],"finetuning;":[107],"(4)":[108],"works":[110],"both":[112],"full-precision":[113],"1-bit":[115],"(e.g.,":[117],"BitNet":[118,124],"b1.58).":[119],"Particularly,":[120],"synergy":[122],"b1.58":[125],"(can":[128],"be":[129],"equipped":[130],"with":[131],"MoE)":[132],"provides":[133],"cornerstone":[135],"clear":[138],"path":[139],"revolutionize":[141],"efficiency,":[143],"cost":[145],"energy":[147],"consumption,":[148],"future":[150],"LLMs.":[151]},"counts_by_year":[{"year":2025,"cited_by_count":2}],"updated_date":"2026-09-30T09:27:27.532063","created_date":"2025-10-10T00:00:00"}