{"id":"https://openalex.org/W4412377829","doi":"https://doi.org/10.1145/3726302.3730225","title":"Scaling Sparse and Dense Retrieval in Decoder-Only LLMs","display_name":"Scaling Sparse and Dense Retrieval in Decoder-Only LLMs","publication_year":2025,"publication_date":"2025-07-13","ids":{"openalex":"https://openalex.org/W4412377829","doi":"https://doi.org/10.1145/3726302.3730225"},"language":"en","primary_location":{"id":"doi:10.1145/3726302.3730225","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3726302.3730225","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3726302.3730225","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 48th International ACM SIGIR Conference on Research and Development in Information Retrieval","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://dl.acm.org/doi/pdf/10.1145/3726302.3730225","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5020583822","display_name":"Hansi Zeng","orcid":"https://orcid.org/0009-0000-2699-8460"},"institutions":[{"id":"https://openalex.org/I24603500","display_name":"University of Massachusetts Amherst","ror":"https://ror.org/0072zz521","country_code":"US","type":"education","lineage":["https://openalex.org/I24603500"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Hansi Zeng","raw_affiliation_strings":["University of Massachusetts Amherst, Amherst, MA, USA"],"raw_orcid":"https://orcid.org/0009-0000-2699-8460","affiliations":[{"raw_affiliation_string":"University of Massachusetts Amherst, Amherst, MA, USA","institution_ids":["https://openalex.org/I24603500"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5066490162","display_name":"Julian Killingback","orcid":"https://orcid.org/0000-0003-2280-8759"},"institutions":[{"id":"https://openalex.org/I24603500","display_name":"University of Massachusetts Amherst","ror":"https://ror.org/0072zz521","country_code":"US","type":"education","lineage":["https://openalex.org/I24603500"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Julian Killingback","raw_affiliation_strings":["University of Massachusetts Amherst, Amherst, MA, USA"],"raw_orcid":"https://orcid.org/0000-0003-2280-8759","affiliations":[{"raw_affiliation_string":"University of Massachusetts Amherst, Amherst, MA, USA","institution_ids":["https://openalex.org/I24603500"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5101457713","display_name":"Hamed Zamani","orcid":"https://orcid.org/0000-0002-0800-3340"},"institutions":[{"id":"https://openalex.org/I24603500","display_name":"University of Massachusetts Amherst","ror":"https://ror.org/0072zz521","country_code":"US","type":"education","lineage":["https://openalex.org/I24603500"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hamed Zamani","raw_affiliation_strings":["University of Massachusetts Amherst, Amherst, MA, USA"],"raw_orcid":"https://orcid.org/0000-0002-0800-3340","affiliations":[{"raw_affiliation_string":"University of Massachusetts Amherst, Amherst, MA, USA","institution_ids":["https://openalex.org/I24603500"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":["https://openalex.org/A5020583822"],"corresponding_institution_ids":["https://openalex.org/I24603500"],"apc_list":null,"apc_paid":null,"fwci":6.7583,"has_fulltext":true,"cited_by_count":5,"citation_normalized_percentile":{"value":0.96788978,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":99,"max":100},"biblio":{"volume":null,"issue":null,"first_page":"2679","last_page":"2684"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11181","display_name":"Advanced Data Storage Technologies","score":0.9375,"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"}},"topics":[{"id":"https://openalex.org/T11181","display_name":"Advanced Data Storage Technologies","score":0.9375,"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"}},{"id":"https://openalex.org/T10049","display_name":"Magnetic properties of thin films","score":0.9333000183105469,"subfield":{"id":"https://openalex.org/subfields/3107","display_name":"Atomic and Molecular Physics, and Optics"},"field":{"id":"https://openalex.org/fields/31","display_name":"Physics and Astronomy"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10237","display_name":"Cryptography and Data Security","score":0.9319999814033508,"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/scaling","display_name":"Scaling","score":0.6945823431015015},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6188080906867981},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.1256903111934662}],"concepts":[{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.6945823431015015},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6188080906867981},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.1256903111934662},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3726302.3730225","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3726302.3730225","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3726302.3730225","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 48th International ACM SIGIR Conference on Research and Development in Information Retrieval","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3726302.3730225","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3726302.3730225","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3726302.3730225","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 48th International ACM SIGIR Conference on Research and Development in Information Retrieval","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320337345","display_name":"Office of Naval Research","ror":"https://ror.org/00rk2pe57"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412377829.pdf","grobid_xml":"https://content.openalex.org/works/W4412377829.grobid-xml"},"referenced_works_count":21,"referenced_works":["https://openalex.org/W2138662031","https://openalex.org/W2897754576","https://openalex.org/W2914304175","https://openalex.org/W3034912391","https://openalex.org/W3099700870","https://openalex.org/W3152562554","https://openalex.org/W3154280800","https://openalex.org/W3154670582","https://openalex.org/W3155895380","https://openalex.org/W3184918446","https://openalex.org/W3188983256","https://openalex.org/W3206455169","https://openalex.org/W3217305727","https://openalex.org/W4225156005","https://openalex.org/W4284664419","https://openalex.org/W4295885374","https://openalex.org/W4384625631","https://openalex.org/W4389520055","https://openalex.org/W4400526199","https://openalex.org/W4400528754","https://openalex.org/W4402671767"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Scaling":[0,120],"large":[1],"language":[2],"models":[3,129,136,152,182],"(LLMs)":[4],"has":[5],"shown":[6],"great":[7],"potential":[8],"for":[9],"improving":[10],"retrieval":[11,22,34,42,59,74,151,156,181],"model":[12,78],"performance;":[13],"however,":[14],"previous":[15],"studies":[16],"have":[17],"mainly":[18],"focused":[19],"on":[20,56,104],"dense":[21,155],"trained":[23],"with":[24,125,183],"contrastive":[25],"loss":[26],"(CL),":[27],"neglecting":[28],"the":[29,84,143,184],"scaling":[30],"behavior":[31],"of":[32,186],"other":[33],"paradigms":[35,60],"and":[36,43,64,94,110,146,163,167,188],"optimization":[37],"techniques,":[38],"such":[39],"as":[40,83],"sparse":[41,180],"knowledge":[44],"distillation":[45],"(KD).":[46],"In":[47],"this":[48],"work,":[49],"we":[50,99],"conduct":[51],"a":[52,95],"systematic":[53],"comparative":[54],"study":[55],"how":[57],"different":[58,77],"(sparse":[61],"vs.":[62,68,70],"dense)":[63],"fine-tuning":[65],"objectives":[66],"(CL":[67],"KD":[69,189],"their":[71],"combination)":[72],"affect":[73],"performance":[75,132],"across":[76,142,157],"scales.":[79,148],"Using":[80],"MSMARCO":[81],"passages":[82],"training":[85,102],"dataset,":[86],"decoder-only":[87],"LLMs":[88],"(Llama-3":[89],"series:":[90],"1B,":[91,144],"3B,":[92,145],"8B),":[93],"fixed":[96],"compute":[97],"budget,":[98],"evaluate":[100],"various":[101],"configurations":[103],"both":[105,158],"in-domain":[106,159],"(MSMARCO,":[107,160],"TREC":[108,161],"DL)":[109,162],"out-of-domain":[111,164],"(BEIR)":[112,165],"benchmarks.":[113],"Our":[114],"key":[115],"findings":[116],"reveal":[117],"that:":[118],"(1)":[119],"behaviors":[121],"emerge":[122],"clearly":[123],"only":[124],"CL,":[126],"where":[127],"larger":[128],"achieve":[130],"significant":[131],"gains,":[133],"whereas":[134],"KD-trained":[135],"show":[137],"minimal":[138],"improvement,":[139],"performing":[140],"similarly":[141],"8B":[147,192],"(2)":[149],"Sparse":[150],"consistently":[153],"outperform":[154],"benchmarks,":[166],"they":[168],"demonstrate":[169],"greater":[170],"robustness":[171],"to":[172],"imperfect":[173],"supervised":[174],"signals.":[175],"(3)":[176],"We":[177],"successfully":[178],"scale":[179],"combination":[185],"CL":[187],"losses":[190],"at":[191],"scale,":[193],"achieving":[194],"state-of-the-art":[195],"(SOTA)":[196],"results":[197],"in":[198],"all":[199],"evaluation":[200],"sets.":[201]},"counts_by_year":[{"year":2026,"cited_by_count":5}],"updated_date":"2025-11-06T03:46:38.306776","created_date":"2025-10-10T00:00:00"}