{"id":"https://openalex.org/W7204479564","doi":"https://doi.org/10.48550/arxiv.2608.25380","title":"APT: Accelerating Diffusion Transformers via Attention Probability-Guided Pruning and Quantization","display_name":"APT: Accelerating Diffusion Transformers via Attention Probability-Guided Pruning and Quantization","publication_year":2026,"publication_date":"2026-08-26","ids":{"openalex":"https://openalex.org/W7204479564","doi":"https://doi.org/10.48550/arxiv.2608.25380"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.25380","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.25380","pdf_url":"https://arxiv.org/pdf/2608.25380","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","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.25380","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5048500417","display_name":"Sungyeob Yoo","orcid":"https://orcid.org/0000-0002-7783-9176","observed_orcids":["https://orcid.org/0000-0002-7783-9176"]},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Yoo, Sungyeob","raw_affiliation_strings":["KAIST Daejeon , Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"KAIST Daejeon , Republic of Korea","institution_ids":["https://openalex.org/I157485424"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5064805094","display_name":"Seeyeon Kim","orcid":"https://orcid.org/0009-0002-4587-2972","observed_orcids":["https://orcid.org/0009-0002-4587-2972"]},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Kim, Seeyeon","raw_affiliation_strings":["KAIST Daejeon , Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"KAIST Daejeon , Republic of Korea","institution_ids":["https://openalex.org/I157485424"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5109692719","display_name":"Joonyong Park","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Park, Joonyong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5072167726","display_name":"Seunghee Han","orcid":"https://orcid.org/0000-0001-8696-6823","observed_orcids":["https://orcid.org/0000-0001-8696-6823","https://orcid.org/0000-0002-8733-5537"]},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Han, Seunghee","raw_affiliation_strings":["KAIST Daejeon , Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"KAIST Daejeon , Republic of Korea","institution_ids":["https://openalex.org/I157485424"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5148446240","display_name":"Joo-Young Kim","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Kim, Joo-Young","raw_affiliation_strings":["Joonyong Park KAIST Daejeon , Republic of Korea","KAIST Daejeon , Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Joonyong Park KAIST Daejeon , Republic of Korea","institution_ids":["https://openalex.org/I157485424"]},{"raw_affiliation_string":"KAIST Daejeon , Republic of Korea","institution_ids":["https://openalex.org/I157485424"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"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":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":0.6151999831199646,"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":0.6151999831199646,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.3352000117301941,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.02370000071823597,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/diffusion-transformer","display_name":"diffusion transformer","score":0.9359999895095825},{"id":"https://openalex.org/keywords/sparse-attention","display_name":"sparse attention","score":0.48899999260902405},{"id":"https://openalex.org/keywords/model-pruning","display_name":"model pruning","score":0.5109999775886536},{"id":"https://openalex.org/keywords/adaptive-quantization","display_name":"adaptive quantization","score":0.6650000214576721},{"id":"https://openalex.org/keywords/flashattention","display_name":"FlashAttention","score":0.781000018119812},{"id":"https://openalex.org/keywords/high-resolution-image-synthesis","display_name":"high-resolution image generation","score":0.7870000004768372},{"id":"https://openalex.org/keywords/energy-efficiency","display_name":"energy efficiency","score":0.49799999594688416}],"concepts":[{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.8801000118255615},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7468000054359436},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.6154000163078308},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5853999853134155},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.550000011920929},{"id":"https://openalex.org/C3826847","wikidata":"https://www.wikidata.org/wiki/Q188768","display_name":"FLOPS","level":2,"score":0.5278000235557556},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.47110000252723694},{"id":"https://openalex.org/C2742236","wikidata":"https://www.wikidata.org/wiki/Q924713","display_name":"Efficient energy use","level":2,"score":0.40209999680519104},{"id":"https://openalex.org/C191178318","wikidata":"https://www.wikidata.org/wiki/Q2256906","display_name":"Thresholding","level":3,"score":0.4016000032424927},{"id":"https://openalex.org/C96146094","wikidata":"https://www.wikidata.org/wiki/Q609057","display_name":"Unification","level":2,"score":0.39469999074935913},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.3723999857902527},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.37229999899864197},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.3573000133037567},{"id":"https://openalex.org/C127532173","wikidata":"https://www.wikidata.org/wiki/Q179904","display_name":"Hourglass","level":2,"score":0.3391000032424927},{"id":"https://openalex.org/C3770464","wikidata":"https://www.wikidata.org/wiki/Q775963","display_name":"Smoothing","level":2,"score":0.33219999074935913},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.32420000433921814},{"id":"https://openalex.org/C63000827","wikidata":"https://www.wikidata.org/wiki/Q3080428","display_name":"Software portability","level":2,"score":0.320499986410141},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3077999949455261},{"id":"https://openalex.org/C186633575","wikidata":"https://www.wikidata.org/wiki/Q845060","display_name":"Maxima and minima","level":2,"score":0.28209999203681946},{"id":"https://openalex.org/C3017489831","wikidata":"https://www.wikidata.org/wiki/Q2393193","display_name":"Running time","level":2,"score":0.27570000290870667},{"id":"https://openalex.org/C205711294","wikidata":"https://www.wikidata.org/wiki/Q176953","display_name":"Rendering (computer graphics)","level":2,"score":0.27090001106262207},{"id":"https://openalex.org/C2776135515","wikidata":"https://www.wikidata.org/wiki/Q17143721","display_name":"Regularization (linguistics)","level":2,"score":0.27079999446868896},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.26649999618530273}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.25380","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.25380","pdf_url":"https://arxiv.org/pdf/2608.25380","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.2608.25380","doi":"https://doi.org/10.48550/arxiv.2608.25380","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.25380","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:2608.25380","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.25380","pdf_url":"https://arxiv.org/pdf/2608.25380","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":[],"funders":[{"id":"https://openalex.org/F4320328359","display_name":"Ministry of Science and ICT, South Korea","ror":"https://ror.org/01wpjm123"},{"id":"https://openalex.org/F4320332195","display_name":"Samsung","ror":"https://ror.org/04w3jy968"},{"id":"https://openalex.org/F4320335489","display_name":"Institute for Information and Communications Technology Promotion","ror":"https://ror.org/01g0hqq23"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204479564.pdf","grobid_xml":"https://content.openalex.org/works/W7204479564.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4416429522","https://openalex.org/W4414198959","https://openalex.org/W7221122824","https://openalex.org/W7162445054","https://openalex.org/W7169792106","https://openalex.org/W7129216374","https://openalex.org/W4415799237","https://openalex.org/W7166774909","https://openalex.org/W7137137800","https://openalex.org/W4413440951"],"abstract_inverted_index":{"Recent":[0],"advances":[1],"in":[2,44,90],"generative":[3],"AI":[4],"have":[5,28,70],"significantly":[6],"increased":[7],"the":[8,31,56,60,64,87,127,171],"demand":[9],"for":[10,103],"high-resolution":[11,91,104],"image":[12],"and":[13,39,78,122,143,184,196,213,221,229,234],"video":[14],"generation,":[15],"positioning":[16],"diffusion":[17,243],"models":[18,34],"as":[19,30,55,110],"a":[20,99,111,176,197,241],"core":[21],"technology.":[22],"Among":[23],"them,":[24],"Diffusion":[25,211],"Transformers":[26],"(DiTs)":[27],"emerged":[29],"state-of-the-art":[32],"(SOTA)":[33],"due":[35],"to":[36,51,72,115,218,231],"their":[37],"scalability":[38],"output":[40,65],"quality.":[41],"However,":[42],"self-attention":[43],"DiTs":[45],"incurs":[46],"significant":[47],"computational":[48,88],"overhead,":[49],"leading":[50],"excessively":[52],"long":[53],"latency":[54],"complexity":[57],"grows":[58],"with":[59,152],"fourth":[61],"power":[62],"of":[63,84],"resolution.":[66],"While":[67],"prior":[68],"works":[69],"attempted":[71],"mitigate":[73],"this":[74,94],"cost":[75,89],"using":[76,146],"sparsity":[77,183],"quantization":[79],"techniques,":[80],"they":[81],"fall":[82],"short":[83],"effectively":[85],"reducing":[86],"DiTs.":[92,105],"In":[93],"paper,":[95],"we":[96,130,155,174,201],"present":[97],"APT,":[98],"software-hardware":[100],"co-designed":[101],"accelerator":[102,178],"APT":[106,203,215],"leverages":[107],"attention":[108,162],"probabilities":[109,163],"unified":[112],"importance":[113],"metric":[114],"jointly":[116],"optimize":[117],"computation":[118],"through":[119],"fine-grained":[120],"pruning":[121],"adaptive":[123],"precision":[124,144],"scaling.":[125],"At":[126,170],"algorithm":[128],"level,":[129,173],"propose":[131],"Attention":[132],"Probability-guided":[133],"Adaptive":[134],"Dual":[135],"Thresholding":[136],"(APDT),":[137],"which":[138,160],"dynamically":[139],"performs":[140],"element":[141],"selection":[142],"assignment":[145],"dual":[147],"thresholds.":[148],"To":[149],"ensure":[150],"compatibility":[151],"memory-efficient":[153],"FlashAttention,":[154],"introduce":[156],"Timestep-Aware":[157],"FlashAttention":[158],"(TAFA),":[159],"predicts":[161],"across":[164],"timesteps":[165],"by":[166],"exploiting":[167],"temporal":[168],"similarity.":[169],"architecture":[172],"co-design":[175],"specialized":[177],"that":[179],"efficiently":[180],"supports":[181],"irregular":[182],"dual-precision":[185,193],"execution,":[186],"featuring":[187],"dynamic":[188],"mask":[189],"management,":[190],"address":[191],"translation,":[192],"compute":[194],"units,":[195],"tile-based":[198],"dataflow.":[199],"Finally,":[200],"evaluate":[202],"on":[204],"SOTA":[205,242],"DiT":[206],"models,":[207],"including":[208],"PixArt-$\u03b1$,":[209],"Stable":[210],"3,":[212],"FLUX.":[214],"achieves":[216],"up":[217,230],"8.16$\\times$":[219],"speedup":[220,233],"14.98$\\times$":[222],"higher":[223,236],"energy":[224,237],"efficiency":[225,238],"over":[226,239],"NVIDIA":[227],"A100,":[228],"3.01$\\times$":[232],"2.04$\\times$":[235],"EXION,":[240],"model":[244],"accelerator.":[245]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-08-28T00:00:00"}