{"id":"https://openalex.org/W4412888762","doi":"https://doi.org/10.18653/v1/2025.findings-acl.154","title":"Position-Aware Depth Decay Decoding (D3): Boosting Large Language Model Inference Efficiency","display_name":"Position-Aware Depth Decay Decoding (D3): Boosting Large Language Model Inference Efficiency","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412888762","doi":"https://doi.org/10.18653/v1/2025.findings-acl.154"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.154","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.154","pdf_url":"https://aclanthology.org/2025.findings-acl.154.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-acl.154.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102684384","display_name":"Siqi Fan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Siqi Fan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5108640082","display_name":"Xuezhi Fang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xuezhi Fang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5050972774","display_name":"Xingrun Xing","orcid":"https://orcid.org/0000-0002-0716-859X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xingrun Xing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5011200770","display_name":"Han Peng","orcid":"https://orcid.org/0000-0002-1828-1782"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Peng Han","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102754146","display_name":"Shuo Shang","orcid":"https://orcid.org/0000-0002-1117-2890"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shuo Shang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5029640741","display_name":"Yequan Wang","orcid":"https://orcid.org/0000-0001-7530-6125"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yequan Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"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":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"2990","last_page":"3001"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11269","display_name":"Algorithms and Data Compression","score":0.9840999841690063,"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/T11269","display_name":"Algorithms and Data Compression","score":0.9840999841690063,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.948199987411499,"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/T12029","display_name":"DNA and Biological Computing","score":0.9434000253677368,"subfield":{"id":"https://openalex.org/subfields/1312","display_name":"Molecular Biology"},"field":{"id":"https://openalex.org/fields/13","display_name":"Biochemistry, Genetics and Molecular Biology"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/boosting","display_name":"Boosting (machine learning)","score":0.7422438859939575},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.7324416041374207},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.6743351817131042},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.667141318321228},{"id":"https://openalex.org/keywords/position","display_name":"Position (finance)","score":0.489018052816391},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.45584267377853394},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4416621923446655},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.23809388279914856}],"concepts":[{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.7422438859939575},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7324416041374207},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.6743351817131042},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.667141318321228},{"id":"https://openalex.org/C198082294","wikidata":"https://www.wikidata.org/wiki/Q3399648","display_name":"Position (finance)","level":2,"score":0.489018052816391},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.45584267377853394},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4416621923446655},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.23809388279914856},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.0},{"id":"https://openalex.org/C10138342","wikidata":"https://www.wikidata.org/wiki/Q43015","display_name":"Finance","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.154","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.154","pdf_url":"https://aclanthology.org/2025.findings-acl.154.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.154","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.154","pdf_url":"https://aclanthology.org/2025.findings-acl.154.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G8622132457","display_name":null,"funder_award_id":"62106249","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320329860","display_name":"National Science and Technology Major Project","ror":null}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412888762.pdf","grobid_xml":"https://content.openalex.org/works/W4412888762.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W2125652721","https://openalex.org/W1540371141","https://openalex.org/W1549363203","https://openalex.org/W2147697413","https://openalex.org/W2154063878","https://openalex.org/W4231274751","https://openalex.org/W2556012038","https://openalex.org/W1489772951","https://openalex.org/W1538046993","https://openalex.org/W2571255492"],"abstract_inverted_index":{"Due":[0],"to":[1,58,105,111],"the":[2,7,45,107,122,135,143,162,177],"large":[3,139],"number":[4,108],"of":[5,10,48,109,131],"parameters,":[6],"inference":[8],"phase":[9],"Large":[11],"Language":[12],"Models":[13],"(LLMs)":[14],"is":[15,56],"resource-intensive.Unlike":[16],"traditional":[17],"model":[18],"compression,":[19],"which":[20,94],"needs":[21],"retraining,":[22,121],"recent":[23],"dynamic":[24,46],"computation":[25],"methods":[26],"show":[27,150],"that":[28,69,151],"not":[29],"all":[30],"components":[31],"are":[32],"required":[33],"for":[34,134],"inference,":[35],"enabling":[36],"a":[37,83,96,128],"training-free":[38,84],"pipeline.In":[39],"this":[40],"paper,":[41],"we":[42,81],"focus":[43],"on":[44,138,176],"depth":[47],"LLM":[49],"generation.A":[50],"token-position":[51],"aware":[52],"layer":[53],"skipping":[54],"framework":[55],"proposed":[57],"save":[59],"1.5x":[60,158],"times":[61],"operations":[62],"efficiently":[63],"while":[64,165],"maintaining":[65,166],"performance.We":[66],"first":[67,136],"observed":[68],"tokens":[70],"predicted":[71],"later":[72],"have":[73],"lower":[74],"perplexity":[75],"and":[76,179],"thus":[77],"require":[78],"less":[79],"computation.Then,":[80],"propose":[82],"algorithm":[85],"called":[86],"Position-Aware":[87],"Depth":[88],"Decay":[89],"Decoding":[90],"(D":[91],"3":[92,124,153],"),":[93],"leverages":[95],"power-law":[97],"decay":[98],"function,":[99],"L":[100],"(":[101],"i":[102,117],")":[103],",":[104],"determine":[106],"layers":[110],"retain":[112],"when":[113],"generating":[114],"token":[115],"T":[116],".Remarkably,":[118],"without":[119],"any":[120],"D":[123,152],"achieves":[125],"success":[126],"across":[127],"wide":[129],"range":[130],"generation":[132],"tasks":[133],"time.Experiments":[137],"language":[140],"models":[141],"(i.e.,":[142],"Llama)":[144],"with":[145,161,169],"7":[146],"70":[147],"billion":[148],"parameters":[149],"can":[154],"achieve":[155],"an":[156],"average":[157],"speedup":[159],"compared":[160],"full-inference":[163],"pipeline":[164],"comparable":[167],"performance":[168,172],"nearly":[170],"no":[171],"drop":[173],"(<":[174],"1%)":[175],"GSM8K":[178],"BBH":[180],"benchmarks.":[181]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}