{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/census/multiheadattention","entry":"MultiHeadAttention","source":"Syntology differential census (groundwork/55, run v2_2026-09-22), per sample; not an archive number","census_date":"2026-09-22","battery_sha256":["b986f7e04d794a0d88ad4c5f32cf63ec3590b5deff0192150737bc5f1c0b4677"],"runner_sha256":["5a452d0e7c0da5b80771d1be2afe3572e253568cf5c5d59a0d08ebd663d00808"],"bucket_key":"positional (rank, kind, dtype) of each array argument; the argument name is not part of the key because the harness draws the shared array from (rank, kind) and casts it to the dtype, whatever the name","bucket_fields":["rank","kind","dtype"],"claim":"Implementations sharing this entry name were each run on one shared input fixed by the positional (rank, kind, dtype) of their array arguments (the bucket). A cluster is the set whose recorded output digest (sha256 of the output rounded to 6 decimals) is identical. Identical values to six decimals on the shared input are agreement on those inputs, not a statement about the whole domain and not a substitution claim.","n_implementations_compared":9,"n_papers":7,"n_buckets":3,"n_distinct_outputs":9,"n_class_bearing":9,"not_compared":{"not_run_on_shared_input":{"n":10,"by_error":{"RuntimeError":8,"AssertionError":2}},"no_array_argument_ran_on_own_fixture_arguments_only":{"n":0,"n_papers":0,"recorded_shared_digest_equals_own_fixture_digest":0},"output_not_digested_non_numeric":{"n":0,"by_type":{}}},"withdrawn_excluded":0,"code_page":"/code/multiheadattention","buckets":[{"bucket":[[3,"float","float32"],[3,"float","float32"],[3,"float","float32"]],"n_implementations_compared":4,"n_papers":4,"n_not_digested":0,"not_digested_by_type":{},"clusters":[{"output_sha":"0cabec910b03f811","size":1,"n_papers":1,"shape":[2,4,8],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[-1.2639271020889282,1.921033263206482,-0.7299928665161133,0.608796238899231,0.4196648597717285,-0.9679499864578247,0.5958523750305176,-0.5834766626358032,-2.4991509914398193,0.0660015195608139,0.7691965103149414,-0.2573468089103699,0.7773661017417908,0.3254137337207794,0.5621915459632874,0.2563280165195465,-0.24809212982654572,-1.2082109451293945,0.7514098286628723,-0.014512820169329643,-1.0896884202957153,0.7226306796073914,-0.801014244556427,1.887477993965149,-0.6406448483467102,-0.12542279064655304,0.2675744295120239,-0.7278733253479004,-0.649242639541626,0.5185819268226624,-0.9587438702583313,2.3157711029052734,-1.415938138961792,1.5138217210769653,-1.1772407293319702,-0.8154778480529785,1.2240394353866577,0.276719331741333,0.195613294839859,0.19846272468566895,-2.2611169815063477,-0.5104866027832031,0.42372485995292664,-0.33514678478240967,1.098427176475525,0.3088294565677643,0.9688833355903625,0.30688560009002686,-1.4739748239517212,0.2583472430706024,1.8581740856170654,-0.9017640948295593,0.5714228749275208,0.6170899868011475,-0.043039415031671524,-0.8862560987472534,-1.5128473043441772,0.5878434181213379,1.3442842960357666,-0.4879070818424225,1.129837155342102,0.6576647162437439,-0.6003681421279907,-1.1185070276260376],"values_recorded":64,"members":[{"code_sha256_prefix":"abb0132a11948e66","path":"libfewshot_core/model/metric/feat.py","papers":["2605.13672"],"paper_pages":[{"arxiv_id":"2605.13672","page":"/paper/arxiv-2605-13672"}],"arg_sig_recorded":[["q",3,"float","float32"],["k",3,"float","float32"],["v",3,"float","float32"]],"scalar_args":{},"class_bearing":true}]},{"output_sha":"58fd5c1158b2ed68","size":1,"n_papers":1,"shape":[2,4,8],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[-0.0,1.3231886625289917,-1.2191145420074463,-0.0,0.18958130478858948,-0.8869925141334534,-0.3044855296611786,0.0,-2.0020804405212402,0.4054765999317169,-1.0287843942642212,-0.7244134545326233,0.04826860874891281,0.02773199789226055,-0.30769625306129456,0.643687903881073,1.8643274307250977,-0.5342126488685608,-0.22291339933872223,0.6711726188659668,-0.8205062747001648,1.1490241289138794,-0.7491438388824463,1.5736992359161377,0.4060533940792084,0.8335259556770325,-0.8647501468658447,-0.11851217597723007,-0.9931480884552002,1.2974340915679932,-0.8708940148353577,1.7934441566467285,-1.0854768753051758,1.7510948181152344,-0.9642760157585144,-0.8718568682670593,0.0,0.5607959032058716,-0.06493108719587326,1.2657803297042847,-1.1688705682754517,-0.18567952513694763,-0.7285125255584717,-0.8200921416282654,1.2045402526855469,0.0,-0.11998669803142548,0.9228118062019348,-1.0299136638641357,0.5869636535644531,0.6055055856704712,-0.8423100113868713,1.4819899797439575,0.8461235165596008,-0.9573739767074585,-0.6045351028442383,-1.2628992795944214,1.149863600730896,1.3458138704299927,-0.7239015698432922,1.8883963823318481,1.2090845108032227,-1.2575194835662842,-0.0],"values_recorded":64,"members":[{"code_sha256_prefix":"9c478f15d1800b2a","path":"src/nlp/layers/encoder.py","papers":["1706.03762"],"paper_pages":[{"arxiv_id":"1706.03762","page":"/paper/attention-is-all-you-need"}],"arg_sig_recorded":[["q",3,"float","float32"],["k",3,"float","float32"],["v",3,"float","float32"]],"scalar_args":{"mask":"None"},"class_bearing":true}]},{"output_sha":"b47c72b25d748899","size":1,"n_papers":1,"shape":[2,4,8],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[0.035665031522512436,-0.41347187757492065,-0.17776678502559662,-0.20891612768173218,0.23824049532413483,0.4353683888912201,0.07005111128091812,0.06412246823310852,-0.043702803552150726,-0.37056779861450195,-0.045622602105140686,-0.28043389320373535,0.26344311237335205,0.5007596611976624,0.11103163659572601,0.017366262152791023,-0.08810088038444519,-0.33968478441238403,0.025232261046767235,-0.3111000657081604,0.2627497911453247,0.40158283710479736,0.03840214014053345,0.06573568284511566,-0.2914539873600006,-0.3103500008583069,0.36562082171440125,-0.41121673583984375,0.27313169836997986,0.4466610550880432,0.08698577433824539,0.07119617611169815,0.27753540873527527,-0.2434968501329422,0.7968509793281555,0.2794754207134247,0.5437536239624023,0.7484618425369263,0.501624584197998,-0.7124384045600891,0.204190194606781,-0.33975186944007874,0.7545833587646484,0.09606796503067017,0.558253824710846,0.7708683013916016,0.5216636657714844,-0.7288520336151123,0.23968885838985443,-0.30435043573379517,0.8338120579719543,0.07210776209831238,0.5301781892776489,0.6326044201850891,0.5066598057746887,-0.7208601236343384,0.2980561852455139,-0.231140598654747,0.8879780769348145,0.17884540557861328,0.4861412048339844,0.612830400466919,0.43331024050712585,-0.6608325839042664],"values_recorded":64,"members":[{"code_sha256_prefix":"db4e1c41512bdaac","path":"summarizer_transformer.py","papers":["1704.04368"],"paper_pages":[{"arxiv_id":"1704.04368","page":"/paper/get-to-the-point-summarization-with-pointer"}],"arg_sig_recorded":[["q",3,"float","float32"],["k",3,"float","float32"],["v",3,"float","float32"]],"scalar_args":{"mask":"False","cover":"None"},"class_bearing":true}]},{"output_sha":"f1e879c2058dc97c","size":1,"n_papers":1,"shape":[2,4,8],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[0.2403973788022995,-0.0802590548992157,-0.20528855919837952,0.05419182777404785,-0.2920725345611572,0.6059292554855347,-0.41717204451560974,-0.009299322962760925,0.3049015998840332,-0.0616350919008255,-0.2653224766254425,0.07345864176750183,-0.24996031820774078,0.6187258362770081,-0.41545602679252625,0.11365693807601929,0.36760762333869934,-0.04780752956867218,-0.32353225350379944,-0.016383886337280273,-0.23996464908123016,0.647084653377533,-0.4004666805267334,0.12314118444919586,0.2664025127887726,-0.07335162162780762,-0.2244851142168045,0.15064933896064758,-0.22135065495967865,0.5932337641716003,-0.3984135389328003,0.1537940800189972,-0.779963493347168,-0.22800418734550476,-0.1880931556224823,0.6722904443740845,0.019690178334712982,0.042101651430130005,-0.130692258477211,-0.2761324346065521,-0.45753857493400574,-0.19429266452789307,-0.15875458717346191,0.5185954570770264,-0.042120642960071564,0.17014820873737335,-0.20949849486351013,-0.13826137781143188,-0.3505540192127228,-0.21095286309719086,-0.12087175250053406,0.5262749195098877,-0.016870401799678802,0.15104474127292633,-0.22411316633224487,-0.03558292239904404,-0.33704903721809387,-0.2509898245334625,-0.10728009045124054,0.5537426471710205,0.0069208405911922455,0.10842011868953705,-0.20152990520000458,-0.01888224482536316],"values_recorded":64,"members":[{"code_sha256_prefix":"337e597194195937","path":"bpp/model.py","papers":["2406.12454"],"paper_pages":[{"arxiv_id":"2406.12454","page":"/paper/a-neural-column-generation-approach-to-the"}],"arg_sig_recorded":[["query",3,"float","float32"],["key",3,"float","float32"],["value",3,"float","float32"]],"scalar_args":{"mask":"None"},"class_bearing":true}]}]},{"bucket":[[3,"float","float32"]],"n_implementations_compared":4,"n_papers":2,"n_not_digested":0,"not_digested_by_type":{},"clusters":[{"output_sha":"4dfb8cb5987db9c6","size":1,"n_papers":1,"shape":[2,4,8],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[0.5348867177963257,0.24734872579574585,0.4715437889099121,-0.0019393377006053925,-0.2922147214412689,-0.14613354206085205,0.29046833515167236,-0.042763106524944305,0.5425301194190979,0.3221222460269928,0.49689388275146484,-0.037338461726903915,-0.22280409932136536,-0.1325974315404892,0.3525651693344116,-0.04272214323282242,0.5411643385887146,0.3126086890697479,0.4992320239543915,-0.04531693458557129,-0.1845218539237976,-0.10542303323745728,0.36278462409973145,-0.045396748930215836,0.5466908812522888,0.30955156683921814,0.4982638955116272,-0.041718848049640656,-0.17283108830451965,-0.1009146124124527,0.35663044452667236,-0.050478722900152206,0.5214466452598572,0.5430846810340881,0.537125825881958,-0.09481288492679596,-0.34582218527793884,-0.27197957038879395,0.4616313576698303,-0.0056200213730335236,0.5642228126525879,0.5475394129753113,0.5398889780044556,-0.05550949275493622,-0.39640188217163086,-0.3139406740665436,0.4234476089477539,7.657567039132118e-05,0.5821233987808228,0.5292019248008728,0.539843738079071,-0.03474299609661102,-0.42170336842536926,-0.3234204649925232,0.40321648120880127,0.007685123942792416,0.584008514881134,0.5283559560775757,0.523288905620575,-0.02174641378223896,-0.4115653336048126,-0.3339100778102875,0.3649446368217468,-0.023033061996102333],"values_recorded":64,"members":[{"code_sha256_prefix":"f6a1ed2d7f5be2dc","path":"models/mae.py","papers":["2010.11929"],"paper_pages":[{"arxiv_id":"2010.11929","page":"/paper/an-image-is-worth-16x16-words-transformers-1"}],"arg_sig_recorded":[["x",3,"float","float32"]],"scalar_args":{},"class_bearing":true}]},{"output_sha":"57f8a2a43047b8bd","size":1,"n_papers":1,"shape":[2,4,8],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[-0.08036612719297409,-0.11756853759288788,-0.07009713351726532,0.24619026482105255,0.2367006540298462,0.3057642877101898,-0.005808241665363312,0.41126278042793274,-0.04107289016246796,-0.0694086104631424,-0.20185893774032593,0.1784011721611023,0.3104131817817688,0.3461892604827881,0.04550781100988388,0.5059683322906494,-0.07064687460660934,-0.11329323053359985,-0.2121138870716095,0.1563444435596466,0.2858644425868988,0.27540430426597595,0.033790282905101776,0.41583752632141113,-0.08171480149030685,-0.08828888833522797,-0.19703415036201477,0.1401069611310959,0.26631155610084534,0.2914623022079468,0.004211224615573883,0.41618072986602783,0.24989096820354462,-0.4027045965194702,-0.575964629650116,0.23284873366355896,0.06622323393821716,0.3274059295654297,0.06159522384405136,-0.08403538912534714,0.15911142528057098,-0.2341276854276657,-0.5505369901657104,0.14732612669467926,0.08757905662059784,0.42002058029174805,0.03472066670656204,-0.02729213982820511,0.22230535745620728,-0.3146391212940216,-0.6032475233078003,0.1907680332660675,0.11419866979122162,0.4057597815990448,0.11687880009412766,-0.04148293286561966,0.023966379463672638,-0.2100507616996765,-0.46114927530288696,0.3123460114002228,0.15096351504325867,0.37115928530693054,0.036373622715473175,0.10993728041648865],"values_recorded":64,"members":[{"code_sha256_prefix":"88f8bcf4b76c2844","path":"mobileclip/clip.py","papers":["2303.15343"],"paper_pages":[{"arxiv_id":"2303.15343","page":"/paper/sigmoid-loss-for-language-image-pre-training"}],"arg_sig_recorded":[["x_q",3,"float","float32"]],"scalar_args":{"x_kv":"None","key_padding_mask":"None","attn_mask":"None"},"class_bearing":true}]},{"output_sha":"a8d86eb7e123c141","size":1,"n_papers":1,"shape":[2,4,8],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[-0.17151297628879547,0.5429428815841675,-0.09838581830263138,-0.3150128722190857,0.13220036029815674,0.5762832760810852,-0.3502090275287628,1.1374166011810303,0.28693217039108276,0.7531036734580994,-0.16080474853515625,-0.1992957890033722,0.46169647574424744,0.20359262824058533,-0.48761457204818726,1.3478370904922485,0.28693222999572754,0.7531036734580994,-0.16080473363399506,-0.19929581880569458,0.46169647574424744,0.2035927176475525,-0.48761457204818726,1.347837209701538,0.28693222999572754,0.7531036734580994,-0.16080473363399506,-0.19929581880569458,0.0,0.2035927176475525,-0.48761457204818726,1.347837209701538,-0.22655951976776123,0.4440954923629761,-0.5406329035758972,1.1019426584243774,-0.9011152386665344,-0.0,-0.4305015802383423,0.2914547324180603,-0.16203321516513824,-0.0,-0.9293537735939026,0.0,-0.2820465564727783,-1.1893041133880615,-0.42275291681289673,0.38456565141677856,-0.10548959672451019,0.059321559965610504,-1.046144962310791,1.1611435413360596,-0.535818338394165,-1.5501372814178467,-0.6502372622489929,0.3789480924606323,-0.19554084539413452,-0.10901913791894913,-0.8601436018943787,0.3725699186325073,-0.13166196644306183,-0.9754754304885864,-0.28794607520103455,0.3878946602344513],"values_recorded":64,"members":[{"code_sha256_prefix":"6082f915e0ac604c","path":"vision_transformer.py","papers":["2010.11929"],"paper_pages":[{"arxiv_id":"2010.11929","page":"/paper/an-image-is-worth-16x16-words-transformers-1"}],"arg_sig_recorded":[["x",3,"float","float32"]],"scalar_args":{},"class_bearing":true}]},{"output_sha":"c60576142d8a60f9","size":1,"n_papers":1,"shape":[2,4,8],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[0.181170254945755,-0.036605775356292725,-0.3648962080478668,0.004036068916320801,-0.2538270950317383,0.6264243721961975,-0.3798733055591583,-0.019954651594161987,0.3375280201435089,-0.04205012321472168,-0.2630488872528076,0.050308361649513245,-0.28413063287734985,0.6548824310302734,-0.4484483301639557,0.11471426486968994,0.4072014391422272,-0.02668619155883789,-0.3277263939380646,-0.04951667785644531,-0.27302438020706177,0.6863922476768494,-0.43179354071617126,0.12525230646133423,0.29475125670433044,-0.0550684928894043,-0.21767401695251465,0.13607582449913025,-0.2523421049118042,0.6265579462051392,-0.42951223254203796,0.15931108593940735,-0.2972275912761688,-0.30445677042007446,-0.09585186839103699,0.6100132465362549,0.10762833058834076,0.07204516232013702,-0.12452927231788635,0.11171582341194153,-0.4949036240577698,-0.20207272469997406,-0.17384746670722961,0.490462064743042,-0.062374092638492584,0.15450449287891388,-0.2153474986553192,-0.22309526801109314,-0.3907559812068939,-0.20795878767967224,-0.10254804790019989,0.5534375905990601,-0.025141805410385132,0.1352367401123047,-0.235845148563385,-0.051107801496982574,-0.2803289592266083,-0.2502627372741699,-0.15373560786247253,0.4665229320526123,-0.038075707852840424,0.15893566608428955,-0.19740089774131775,-0.09978032857179642],"values_recorded":64,"members":[{"code_sha256_prefix":"6bad9911ea340c54","path":"src/ViT.py","papers":["2010.11929"],"paper_pages":[{"arxiv_id":"2010.11929","page":"/paper/an-image-is-worth-16x16-words-transformers-1"}],"arg_sig_recorded":[["x",3,"float","float32"]],"scalar_args":{},"class_bearing":true}]}]},{"bucket":[[3,"float","float32"],[3,"float","float32"]],"n_implementations_compared":1,"n_papers":1,"n_not_digested":0,"not_digested_by_type":{},"clusters":[{"output_sha":"8132fe143af39200","size":1,"n_papers":1,"shape":[2,4,16],"dtype":"float32","type":"Tensor","finite":true,"max_difference_among_recorded_values":null,"members_with_recorded_values":1,"torch_reference_conventions_with_this_digest":[],"values":[0.14948414266109467,0.17225395143032074,-0.10339341312646866,0.017099136486649513,-0.14438001811504364,-0.10501884669065475,0.1021457388997078,-0.14145976305007935,-0.11019617319107056,-0.28591978549957275,-0.06804672628641129,-0.04399704560637474,0.1684390902519226,0.10478177666664124,-0.11451125890016556,0.21894849836826324,0.09608074277639389,0.1328832060098648,-0.01385557558387518,0.053985126316547394,-0.17171968519687653,-0.17464609444141388,0.16750231385231018,-0.022560380399227142,-0.17986920475959778,-0.3168364465236664,-0.08485275506973267,-0.008389417082071304,0.16347533464431763,0.14089950919151306,-0.12286701798439026,0.11630566418170929,-0.056779395788908005,0.056772857904434204,0.07466788589954376,0.11798769235610962,-0.1642218679189682,-0.24508509039878845,0.18025141954421997,0.09692508727312088,-0.22860941290855408,-0.35946720838546753,-0.02872702293097973,0.12803798913955688,0.1376887559890747,0.22836342453956604,-0.11164631694555283,-0.04761097952723503,0.09108777344226837,0.12262646108865738,-0.0039609018713235855,0.062012068927288055,-0.13017073273658752,-0.19451701641082764,0.1306992471218109,-0.05476873740553856,-0.21137845516204834,-0.32545554637908936,-0.09050581604242325,0.027655869722366333,0.16204296052455902,0.15464821457862854,-0.1178174838423729,0.08241598308086395,0.11459491401910782,0.05017484351992607,-0.03453956916928291,0.08157143741846085,0.2429245263338089,0.037750497460365295,-0.19624535739421844,-0.15949572622776031,-0.18607652187347412,0.08391018211841583,-0.2004357874393463,0.009990117512643337,0.16148927807807922,-0.13362622261047363,0.11019088327884674,-0.14925731718540192,0.10198897868394852,0.07519691437482834,0.01193350087851286,0.07184500992298126,0.22432073950767517,0.03278600797057152,-0.1753832995891571,-0.13173241913318634,-0.23114335536956787,0.048398490995168686,-0.21635106205940247,0.03627330809831619,0.12328500300645828,-0.11716210097074509,0.11751110851764679,-0.1813894510269165,0.1578773558139801,0.17392702400684357,-0.04845451936125755,0.04794659465551376,0.15796500444412231,-0.026605840772390366,-0.12640631198883057,-0.24223452806472778,-0.29454505443573,-0.12857477366924286,-0.2221338301897049,0.08475029468536377,0.08195538073778152,-0.036335963755846024,0.03841662034392357,-0.04904438927769661,0.1806328296661377,0.19945189356803894,-0.08027605712413788,0.06342341750860214,0.10788799077272415,-0.07636082917451859,-0.09167790412902832,-0.2986653447151184,-0.33965936303138733,-0.22690223157405853,-0.22720150649547577,0.11287730932235718,0.0938904657959938,0.004004702437669039,-0.013774421997368336,0.022954808548092842],"values_recorded":128,"members":[{"code_sha256_prefix":"3e9c7f43e77628ea","path":"nets/attention_model.py","papers":["2110.02544"],"paper_pages":[{"arxiv_id":"2110.02544","page":"/paper/learning-to-iteratively-solve-routing"}],"arg_sig_recorded":[["q",3,"float","float32"],["h",3,"float","float32"]],"scalar_args":{"mask":"None"},"class_bearing":true}]}]}]}