{ "legacy": { "distance": { "base": { "effective_heads": { "count": 4800, "p10": 5.7840811612328125, "median": 8.93182211792415, "p90": 11.67265834734814, "mean": 8.821002725510846 }, "effective_fraction": { "count": 4800, "p10": 0.3615050725770508, "median": 0.5582388823702594, "p90": 0.7295411467092587, "mean": 0.5513126703444279 }, "concentration": { "count": 4800, "p10": 0.02471497387658298, "median": 0.05275652115976448, "p90": 0.11774752758475547, "mean": 0.06432512729180001 }, "top1": { "count": 4800, "p10": 0.1457405612874147, "median": 0.21746353579124, "p90": 0.34538740233748155, "mean": 0.23503983543721835 }, "top25": { "count": 4800, "p10": 0.46438300527426274, "median": 0.5649686672386036, "p90": 0.6741052273899216, "mean": 0.5682829621251091 }, "bottom25": { "count": 4800, "p10": 0.04389082141962781, "median": 0.06803373425718254, "p90": 0.10318453074095452, "mean": 0.07107512763171646 }, "heads90": { "count": 4800, "p10": 10.0, "median": 11.0, "p90": 13.0, "mean": 11.273125 }, "total": { "count": 4800, "p10": 0.22150190912652762, "median": 0.4924028478562832, "p90": 0.7786339074373245, "mean": 0.4935038255766646 } }, "finetuned": { "effective_heads": { "count": 4800, "p10": 5.776575675851443, "median": 8.918300551653985, "p90": 11.683939551627272, "mean": 8.807297020017815 }, "effective_fraction": { "count": 4800, "p10": 0.36103597974071516, "median": 0.5573937844783741, "p90": 0.7302462219767045, "mean": 0.5504560637511134 }, "concentration": { "count": 4800, "p10": 0.024626743482520943, "median": 0.05293758540105196, "p90": 0.11798713654455906, "mean": 0.06466812007727665 }, "top1": { "count": 4800, "p10": 0.14612856743114083, "median": 0.21806944835017916, "p90": 0.3463747132982079, "mean": 0.23575051686039453 }, "top25": { "count": 4800, "p10": 0.46398691517338536, "median": 0.5656402664666489, "p90": 0.6751455273905578, "mean": 0.5687308911331159 }, "bottom25": { "count": 4800, "p10": 0.04375078049578534, "median": 0.06814265463530741, "p90": 0.10351322156021461, "mean": 0.07115813787142872 }, "heads90": { "count": 4800, "p10": 10.0, "median": 11.0, "p90": 13.0, "mean": 11.275 }, "total": { "count": 4800, "p10": 0.22458011865383015, "median": 0.4925860376097262, "p90": 0.7803986554034055, "mean": 0.49494720498192085 } }, "paired": { "effective_heads_delta": { "count": 4800, "p10": -0.12155525346733853, "median": -0.007827372994309911, "p90": 0.08552963831211874, "mean": -0.013705705493032733 }, "increased_fraction": 0.45625, "decreased_fraction": 0.54375, "winner_agreement": 0.981875 }, "per_text": { "fresh-zh-1": { "base": { "count": 1200, "p10": 5.724006719505733, "median": 8.447270312666955, "p90": 11.400422886121929, "mean": 8.52580403089048 }, "finetuned": { "count": 1200, "p10": 5.692180523473624, "median": 8.439940185642019, "p90": 11.383613339828717, "mean": 8.519406557532788 }, "paired_delta": { "count": 1200, "p10": -0.09676644962819134, "median": -0.0003564206512196577, "p90": 0.07296984397018705, "mean": -0.0063974733576900285 } }, "fresh-zh-2": { "base": { "count": 1200, "p10": 6.0045436576758515, "median": 9.280419095307465, "p90": 11.738746210854123, "mean": 9.057065620159456 }, "finetuned": { "count": 1200, "p10": 5.968510727111234, "median": 9.252540842866502, "p90": 11.720895166646851, "mean": 9.043651292717824 }, "paired_delta": { "count": 1200, "p10": -0.122961714917267, "median": -0.012195150149180378, "p90": 0.09760457550073502, "mean": -0.01341432744163121 } }, "fresh-en-1": { "base": { "count": 1200, "p10": 5.729701873062357, "median": 8.743397719211597, "p90": 11.683026866465143, "mean": 8.773217379604489 }, "finetuned": { "count": 1200, "p10": 5.7281595085239125, "median": 8.723519639693787, "p90": 11.708927676002391, "mean": 8.7605887879079 }, "paired_delta": { "count": 1200, "p10": -0.11081218619648876, "median": -0.007005952507822322, "p90": 0.07766016926945382, "mean": -0.012628591696590086 } }, "fresh-en-2": { "base": { "count": 1200, "p10": 5.73972334577513, "median": 9.101710336434357, "p90": 11.821849799673606, "mean": 8.927923871388966 }, "finetuned": { "count": 1200, "p10": 5.750476017284456, "median": 9.082570331629224, "p90": 11.853544794055829, "mean": 8.905541441912746 }, "paired_delta": { "count": 1200, "p10": -0.16178742240734909, "median": -0.013896515617046745, "p90": 0.1005384062632225, "mean": -0.022382429476219604 } } } }, "divergence": { "base": { "effective_heads": { "count": 4800, "p10": 1.8775771328772404, "median": 3.901782998659897, "p90": 6.9489972185767925, "mean": 4.1958991748153185 }, "effective_fraction": { "count": 4800, "p10": 0.11734857080482752, "median": 0.24386143741624355, "p90": 0.43431232616104953, "mean": 0.2622436984259574 }, "concentration": { "count": 4800, "p10": 0.08683274490451195, "median": 0.20671262373235755, "p90": 0.5014413860763198, "mean": 0.25655510469396303 }, "top1": { "count": 4800, "p10": 0.24280447987972617, "median": 0.4230606060703387, "p90": 0.7145513508216859, "mean": 0.45466951196884586 }, "top25": { "count": 4800, "p10": 0.6646066107988152, "median": 0.8229818456850793, "p90": 0.9326550608746287, "mean": 0.808919570864033 }, "bottom25": { "count": 4800, "p10": 0.0034597576965423498, "median": 0.01107327468889098, "p90": 0.03202965476590303, "mean": 0.014963302718693967 }, "heads90": { "count": 4800, "p10": 3.0, "median": 6.0, "p90": 9.0, "mean": 6.310208333333334 }, "total": { "count": 4800, "p10": 0.006019792613687969, "median": 0.026595803305099253, "p90": 0.07109484240791066, "mean": 0.03422935252324174 } }, "finetuned": { "effective_heads": { "count": 4800, "p10": 1.876908117382701, "median": 3.871619853837708, "p90": 6.96989498987909, "mean": 4.184291975347008 }, "effective_fraction": { "count": 4800, "p10": 0.11730675733641881, "median": 0.24197624086485675, "p90": 0.4356184368674431, "mean": 0.261518248459188 }, "concentration": { "count": 4800, "p10": 0.08637246380916241, "median": 0.20884247610187728, "p90": 0.5016439635373461, "mean": 0.25814451030972974 }, "top1": { "count": 4800, "p10": 0.2422939636733382, "median": 0.42624464148928093, "p90": 0.7155069102079873, "mean": 0.45622446760844526 }, "top25": { "count": 4800, "p10": 0.665414679548231, "median": 0.8240052643141178, "p90": 0.9339388132525362, "mean": 0.8095483126047844 }, "bottom25": { "count": 4800, "p10": 0.003446853632797573, "median": 0.011054384074575818, "p90": 0.03206920533547495, "mean": 0.014988975788690944 }, "heads90": { "count": 4800, "p10": 3.0, "median": 6.0, "p90": 9.0, "mean": 6.302083333333333 }, "total": { "count": 4800, "p10": 0.006202359683584291, "median": 0.026746092644316377, "p90": 0.07105643208897167, "mean": 0.03444768158607483 } }, "paired": { "effective_heads_delta": { "count": 4800, "p10": -0.12236077231273206, "median": -0.006747826895777909, "p90": 0.09207081416677103, "mean": -0.011607199468310615 }, "increased_fraction": 0.44125, "decreased_fraction": 0.55875, "winner_agreement": 0.981875 }, "per_text": { "fresh-zh-1": { "base": { "count": 1200, "p10": 1.86102042169738, "median": 3.674821925930784, "p90": 6.496464962062872, "mean": 3.9738265052500115 }, "finetuned": { "count": 1200, "p10": 1.8695193592812367, "median": 3.653253784859141, "p90": 6.540801874780079, "mean": 3.96916619071483 }, "paired_delta": { "count": 1200, "p10": -0.0904291820005377, "median": -0.0021758020710173076, "p90": 0.07911341590862223, "mean": -0.004660314535179818 } }, "fresh-zh-2": { "base": { "count": 1200, "p10": 2.020523053994744, "median": 4.232903043923887, "p90": 7.155352410694885, "mean": 4.422540511588559 }, "finetuned": { "count": 1200, "p10": 1.989322683403461, "median": 4.1946239719499125, "p90": 7.156199883412831, "mean": 4.412144595717297 }, "paired_delta": { "count": 1200, "p10": -0.12116194167201452, "median": -0.010746777226108173, "p90": 0.10483640341883402, "mean": -0.010395915871261636 } }, "fresh-en-1": { "base": { "count": 1200, "p10": 1.8198022208743365, "median": 3.866727985636466, "p90": 7.147920033260447, "mean": 4.154808100201001 }, "finetuned": { "count": 1200, "p10": 1.8146582042199053, "median": 3.84232558313389, "p90": 7.11390324618346, "mean": 4.146511070253586 }, "paired_delta": { "count": 1200, "p10": -0.10807865550731473, "median": -0.0034545124285838558, "p90": 0.08858334435120963, "mean": -0.008297029947415074 } }, "fresh-en-2": { "base": { "count": 1200, "p10": 1.8322151674116371, "median": 3.8933109706193667, "p90": 7.112082581969417, "mean": 4.2324215822217015 }, "finetuned": { "count": 1200, "p10": 1.8021689266802543, "median": 3.8609554823402976, "p90": 7.099615165719617, "mean": 4.209346044702316 }, "paired_delta": { "count": 1200, "p10": -0.16516077036721016, "median": -0.014694033457486322, "p90": 0.10418054058914197, "mean": -0.02307553751938593 } } } }, "QKV": { "base": { "effective_heads": { "count": 4800, "p10": 7.475883895098561, "median": 10.319912213892923, "p90": 12.712833271824215, "mean": 10.202852380581907 }, "effective_fraction": { "count": 4800, "p10": 0.46724274344366007, "median": 0.6449945133683077, "p90": 0.7945520794890134, "mean": 0.6376782737863692 }, "concentration": { "count": 4800, "p10": 0.01723804948414177, "median": 0.03669338569017259, "p90": 0.07601434362115887, "mean": 0.042570735285422205 }, "top1": { "count": 4800, "p10": 0.13321341470482867, "median": 0.18747321888747073, "p90": 0.28328865301359296, "mean": 0.19983216483858976 }, "top25": { "count": 4800, "p10": 0.42474317220679464, "median": 0.510360546343396, "p90": 0.5986294325410266, "mean": 0.5111861898360973 }, "bottom25": { "count": 4800, "p10": 0.04393379865513463, "median": 0.08040363020124161, "p90": 0.12289210604930895, "mean": 0.08222739743141799 }, "heads90": { "count": 4800, "p10": 10.0, "median": 12.0, "p90": 13.0, "mean": 11.84 }, "total": { "count": 4800, "p10": 0.3599545699544251, "median": 0.8310977094806731, "p90": 1.3057362334919163, "mean": 0.832024075460137 } }, "finetuned": { "effective_heads": { "count": 4800, "p10": 7.415670339257133, "median": 10.324447147382044, "p90": 12.701222754575259, "mean": 10.189892766418488 }, "effective_fraction": { "count": 4800, "p10": 0.4634793962035708, "median": 0.6452779467113777, "p90": 0.7938264221609537, "mean": 0.6368682979011555 }, "concentration": { "count": 4800, "p10": 0.017314748942665027, "median": 0.03664798557992152, "p90": 0.07717288111557072, "mean": 0.04288492534802562 }, "top1": { "count": 4800, "p10": 0.13294356494233925, "median": 0.18756509230157237, "p90": 0.2852351774363626, "mean": 0.20058383379684908 }, "top25": { "count": 4800, "p10": 0.4247718681503831, "median": 0.5097655989756011, "p90": 0.5995470748644118, "mean": 0.5113628895373682 }, "bottom25": { "count": 4800, "p10": 0.04420502297887021, "median": 0.0806159855691112, "p90": 0.12315113314407859, "mean": 0.08238857534024695 }, "heads90": { "count": 4800, "p10": 10.0, "median": 12.0, "p90": 13.0, "mean": 11.84875 }, "total": { "count": 4800, "p10": 0.36229494377039373, "median": 0.8324168699327856, "p90": 1.3076765275676736, "mean": 0.8336380656609678 } }, "paired": { "effective_heads_delta": { "count": 4800, "p10": -0.09236974599448597, "median": -0.004347649130346021, "p90": 0.05748593739061878, "mean": -0.012959614163417956 }, "increased_fraction": 0.45875, "decreased_fraction": 0.54125, "winner_agreement": 0.9885416666666667 }, "per_text": { "fresh-zh-1": { "base": { "count": 1200, "p10": 7.355600359875638, "median": 9.903824666297758, "p90": 12.397500772986202, "mean": 9.861890605751142 }, "finetuned": { "count": 1200, "p10": 7.264880226041378, "median": 9.898966374569639, "p90": 12.389470316075128, "mean": 9.843565948927717 }, "paired_delta": { "count": 1200, "p10": -0.0807099828273353, "median": -0.01318702061738719, "p90": 0.03943376541940449, "mean": -0.01832465682342548 } }, "fresh-zh-2": { "base": { "count": 1200, "p10": 7.355118430184284, "median": 10.04562554273904, "p90": 12.489044777205017, "mean": 10.026567005234467 }, "finetuned": { "count": 1200, "p10": 7.290105841531202, "median": 10.026991319027882, "p90": 12.491937373336743, "mean": 10.021018915906389 }, "paired_delta": { "count": 1200, "p10": -0.08841258219596346, "median": -0.0007642725752314838, "p90": 0.07047669026190546, "mean": -0.005548089328078445 } }, "fresh-en-1": { "base": { "count": 1200, "p10": 7.612360292624114, "median": 10.418042154347027, "p90": 12.785435912927255, "mean": 10.306669554290066 }, "finetuned": { "count": 1200, "p10": 7.578728433573872, "median": 10.410359543867354, "p90": 12.756459217588692, "mean": 10.291179961128332 }, "paired_delta": { "count": 1200, "p10": -0.10092656919316453, "median": -0.004362167048531873, "p90": 0.05181777535002789, "mean": -0.01548959316173382 } }, "fresh-en-2": { "base": { "count": 1200, "p10": 7.701260794945824, "median": 10.84810403910145, "p90": 12.86245594691649, "mean": 10.616282357051954 }, "finetuned": { "count": 1200, "p10": 7.649559166610384, "median": 10.86194071558867, "p90": 12.880097710531432, "mean": 10.603806239711519 }, "paired_delta": { "count": 1200, "p10": -0.11578163334072719, "median": 0.0025086418786655074, "p90": 0.07020528538982156, "mean": -0.012476117340434077 } } } } }, "gsm8k": { "distance": { "base": { "effective_heads": { "count": 4800, "p10": 6.013503917962642, "median": 8.906046247944275, "p90": 11.80188183284152, "mean": 8.9017778676544 }, "effective_fraction": { "count": 4800, "p10": 0.3758439948726651, "median": 0.5566278904965172, "p90": 0.737617614552595, "mean": 0.5563611167284 }, "concentration": { "count": 4800, "p10": 0.023714400011049176, "median": 0.053102157069302644, "p90": 0.11071189502052788, "mean": 0.06226195303507482 }, "top1": { "count": 4800, "p10": 0.1451591630194614, "median": 0.21821378474655584, "p90": 0.3395296691096961, "mean": 0.23231998812868285 }, "top25": { "count": 4800, "p10": 0.46122887875718366, "median": 0.5651361494965037, "p90": 0.6673818130476021, "mean": 0.5653113780132732 }, "bottom25": { "count": 4800, "p10": 0.04589732432631252, "median": 0.0717420375207896, "p90": 0.1049109177088932, "mean": 0.07365966028923801 }, "heads90": { "count": 4800, "p10": 10.0, "median": 11.0, "p90": 13.0, "mean": 11.383333333333333 }, "total": { "count": 4800, "p10": 0.2638525074347854, "median": 0.5107462792657316, "p90": 0.8229535373859109, "mean": 0.5267872787490235 } }, "finetuned": { "effective_heads": { "count": 4800, "p10": 5.982592644976027, "median": 8.904116953413451, "p90": 11.818114629158611, "mean": 8.894597740069383 }, "effective_fraction": { "count": 4800, "p10": 0.37391204031100167, "median": 0.5565073095883407, "p90": 0.7386321643224132, "mean": 0.5559123587543364 }, "concentration": { "count": 4800, "p10": 0.023590256743970684, "median": 0.05312810605871639, "p90": 0.11162838720929225, "mean": 0.06248960088996427 }, "top1": { "count": 4800, "p10": 0.1452759810059557, "median": 0.21887845306221598, "p90": 0.3411258772849672, "mean": 0.23272433431827313 }, "top25": { "count": 4800, "p10": 0.46133812096720794, "median": 0.5656479653815689, "p90": 0.6690157446461986, "mean": 0.5657888012738282 }, "bottom25": { "count": 4800, "p10": 0.046235325509099874, "median": 0.0717223750715486, "p90": 0.10515117365332462, "mean": 0.07383858903445469 }, "heads90": { "count": 4800, "p10": 10.0, "median": 11.0, "p90": 13.0, "mean": 11.387708333333334 }, "total": { "count": 4800, "p10": 0.2662267918465659, "median": 0.5119005234446377, "p90": 0.8266748990863564, "mean": 0.5289106785868004 } }, "paired": { "effective_heads_delta": { "count": 4800, "p10": -0.1591509935536532, "median": -0.006747152142687618, "p90": 0.13966946329318328, "mean": -0.007180127585017427 }, "increased_fraction": 0.4725, "decreased_fraction": 0.5275, "winner_agreement": 0.970625 }, "per_text": { "gsm8k-test-0000": { "base": { "count": 1200, "p10": 6.029266196155418, "median": 8.851735983796345, "p90": 11.718544652857876, "mean": 8.853039265973237 }, "finetuned": { "count": 1200, "p10": 6.0185025710306865, "median": 8.817902858881267, "p90": 11.702200456136074, "mean": 8.830522479570654 }, "paired_delta": { "count": 1200, "p10": -0.16698674746796555, "median": -0.013576864730019622, "p90": 0.10594365593799096, "mean": -0.022516786402582326 } }, "gsm8k-test-0002": { "base": { "count": 1200, "p10": 5.902382519453832, "median": 8.814916405222583, "p90": 11.684873874581289, "mean": 8.77475561713854 }, "finetuned": { "count": 1200, "p10": 5.8262902839172686, "median": 8.775601442382403, "p90": 11.697385346787549, "mean": 8.757974314967932 }, "paired_delta": { "count": 1200, "p10": -0.17076082866715633, "median": -0.01759333235505922, "p90": 0.1360708722019009, "mean": -0.01678130217060961 } }, "gsm8k-test-0003": { "base": { "count": 1200, "p10": 6.009438240780767, "median": 8.777000895915759, "p90": 11.955121550413566, "mean": 8.882677112308222 }, "finetuned": { "count": 1200, "p10": 5.917225855217569, "median": 8.809163836018584, "p90": 11.993881916855772, "mean": 8.891379637659044 }, "paired_delta": { "count": 1200, "p10": -0.15862776051872815, "median": 0.004761271301123582, "p90": 0.17976640500609026, "mean": 0.008702525350821659 } }, "gsm8k-test-0004": { "base": { "count": 1200, "p10": 6.202833613253674, "median": 9.134206344734961, "p90": 11.802513491297152, "mean": 9.096639475197604 }, "finetuned": { "count": 1200, "p10": 6.200667188305828, "median": 9.152546172625186, "p90": 11.8442730026282, "mean": 9.098514528079903 }, "paired_delta": { "count": 1200, "p10": -0.13349573481076807, "median": -6.686357594176684e-05, "p90": 0.135442900421867, "mean": 0.0018750528823005652 } } } }, "divergence": { "base": { "effective_heads": { "count": 4800, "p10": 1.875457108203813, "median": 3.8759650883020083, "p90": 7.058800921305472, "mean": 4.23391325232939 }, "effective_fraction": { "count": 4800, "p10": 0.11721606926273831, "median": 0.24224781801887552, "p90": 0.441175057581592, "mean": 0.26461957827058685 }, "concentration": { "count": 4800, "p10": 0.08444493087745249, "median": 0.2085336117901452, "p90": 0.5020835674830071, "mean": 0.253319472363892 }, "top1": { "count": 4800, "p10": 0.24145052956037794, "median": 0.42031807846570973, "p90": 0.7179985045869327, "mean": 0.45173368725639157 }, "top25": { "count": 4800, "p10": 0.6594778658704781, "median": 0.8245290381303709, "p90": 0.9284529929105115, "mean": 0.8070190913253547 }, "bottom25": { "count": 4800, "p10": 0.0038457043002112366, "median": 0.011943733868154319, "p90": 0.033109714130431796, "mean": 0.015698414569523155 }, "heads90": { "count": 4800, "p10": 4.0, "median": 6.0, "p90": 9.0, "mean": 6.355625 }, "total": { "count": 4800, "p10": 0.008542956798601153, "median": 0.02885237614827929, "p90": 0.07744390014122474, "mean": 0.03753269894324168 } }, "finetuned": { "effective_heads": { "count": 4800, "p10": 1.8724184389765561, "median": 3.866394622099988, "p90": 7.027904627662534, "mean": 4.227338292664017 }, "effective_fraction": { "count": 4800, "p10": 0.11702615243603476, "median": 0.24164966388124925, "p90": 0.43924403922890837, "mean": 0.26420864329150107 }, "concentration": { "count": 4800, "p10": 0.08510925478091586, "median": 0.20921482882495243, "p90": 0.5030065669595358, "mean": 0.25406259246563967 }, "top1": { "count": 4800, "p10": 0.2428326290560375, "median": 0.4203880973993296, "p90": 0.7175810034689571, "mean": 0.45244932037391994 }, "top25": { "count": 4800, "p10": 0.6583606497231608, "median": 0.8246663787782131, "p90": 0.9290253485256015, "mean": 0.8076610638985393 }, "bottom25": { "count": 4800, "p10": 0.0038046385599517733, "median": 0.011966184570507691, "p90": 0.033259522931901334, "mean": 0.015794616049737992 }, "heads90": { "count": 4800, "p10": 4.0, "median": 6.0, "p90": 9.0, "mean": 6.350208333333334 }, "total": { "count": 4800, "p10": 0.008829576032712793, "median": 0.029070507331198314, "p90": 0.0773094397965906, "mean": 0.03776397203446997 } }, "paired": { "effective_heads_delta": { "count": 4800, "p10": -0.16764575951403446, "median": -0.005232527666755082, "p90": 0.147146854618983, "mean": -0.006574959665373348 }, "increased_fraction": 0.4708333333333333, "decreased_fraction": 0.5291666666666667, "winner_agreement": 0.970625 }, "per_text": { "gsm8k-test-0000": { "base": { "count": 1200, "p10": 1.8431839972639215, "median": 3.780522276551954, "p90": 7.099145196184062, "mean": 4.201646544541655 }, "finetuned": { "count": 1200, "p10": 1.8178500400391873, "median": 3.763288357122958, "p90": 7.07356609840848, "mean": 4.1789342231976345 }, "paired_delta": { "count": 1200, "p10": -0.17308023435001257, "median": -0.010169449696791233, "p90": 0.10548838266201924, "mean": -0.022712321344020476 } }, "gsm8k-test-0002": { "base": { "count": 1200, "p10": 1.8908285003131953, "median": 3.833913849764531, "p90": 6.885469709475229, "mean": 4.147627032640852 }, "finetuned": { "count": 1200, "p10": 1.882329448871885, "median": 3.8167847379661057, "p90": 6.858497363265328, "mean": 4.134744784084032 }, "paired_delta": { "count": 1200, "p10": -0.1809234445013397, "median": -0.008109501325470725, "p90": 0.14323401996659646, "mean": -0.012882248556819976 } }, "gsm8k-test-0003": { "base": { "count": 1200, "p10": 1.8375673511907138, "median": 3.71674632002963, "p90": 7.103448318892024, "mean": 4.206191412664545 }, "finetuned": { "count": 1200, "p10": 1.8527289652676626, "median": 3.706038772453823, "p90": 7.179711007377434, "mean": 4.214699774936081 }, "paired_delta": { "count": 1200, "p10": -0.16685274354222296, "median": -0.0006958865930237756, "p90": 0.1928958134565214, "mean": 0.008508362271537395 } }, "gsm8k-test-0004": { "base": { "count": 1200, "p10": 1.9721590562878277, "median": 4.12739017447695, "p90": 7.14039115383894, "mean": 4.380188019470508 }, "finetuned": { "count": 1200, "p10": 1.9503733783669992, "median": 4.112801742052209, "p90": 7.108045377865233, "mean": 4.380974388438318 }, "paired_delta": { "count": 1200, "p10": -0.15822896650350238, "median": -0.00089962387563014, "p90": 0.14725001676604996, "mean": 0.0007863689678096667 } } } }, "QKV": { "base": { "effective_heads": { "count": 4800, "p10": 7.625732433101848, "median": 10.387572837790255, "p90": 12.80026994261185, "mean": 10.270078901238158 }, "effective_fraction": { "count": 4800, "p10": 0.4766082770688655, "median": 0.649223302361891, "p90": 0.8000168714132406, "mean": 0.6418799313273849 }, "concentration": { "count": 4800, "p10": 0.01666490928122742, "median": 0.03602013846457927, "p90": 0.07321060761354471, "mean": 0.04147005556576124 }, "top1": { "count": 4800, "p10": 0.12950340585115858, "median": 0.18680523070086608, "p90": 0.2757132983869958, "mean": 0.19655336416070232 }, "top25": { "count": 4800, "p10": 0.42480706934875345, "median": 0.5080353266733936, "p90": 0.5975627753401668, "mean": 0.5104096028921625 }, "bottom25": { "count": 4800, "p10": 0.04665453901718664, "median": 0.08203003363299655, "p90": 0.12238641633949394, "mean": 0.08348550233229636 }, "heads90": { "count": 4800, "p10": 11.0, "median": 12.0, "p90": 13.0, "mean": 11.844375 }, "total": { "count": 4800, "p10": 0.4073887151665986, "median": 0.8801846825517714, "p90": 1.4507116448599853, "mean": 0.9032721801065388 } }, "finetuned": { "effective_heads": { "count": 4800, "p10": 7.576624449214384, "median": 10.374178847146444, "p90": 12.821177123485356, "mean": 10.264366715676784 }, "effective_fraction": { "count": 4800, "p10": 0.473539028075899, "median": 0.6483861779466528, "p90": 0.8013235702178347, "mean": 0.641522919729799 }, "concentration": { "count": 4800, "p10": 0.01652902257201789, "median": 0.03615271705068857, "p90": 0.07411723202733041, "mean": 0.04167572867201576 }, "top1": { "count": 4800, "p10": 0.12935968415437357, "median": 0.18691265969017154, "p90": 0.27822878056333283, "mean": 0.19706959048869968 }, "top25": { "count": 4800, "p10": 0.4240190586277136, "median": 0.5083111063316925, "p90": 0.5988581244801308, "mean": 0.5105091226254993 }, "bottom25": { "count": 4800, "p10": 0.0467784088506235, "median": 0.08242644414525736, "p90": 0.12263218106221838, "mean": 0.08378531660409673 }, "heads90": { "count": 4800, "p10": 11.0, "median": 12.0, "p90": 13.0, "mean": 11.856041666666666 }, "total": { "count": 4800, "p10": 0.4107431098818779, "median": 0.8829612950794399, "p90": 1.4518962780828588, "mean": 0.9056679941666214 } }, "paired": { "effective_heads_delta": { "count": 4800, "p10": -0.12757638238450753, "median": 0.0037255281277426278, "p90": 0.10112692014932954, "mean": -0.00571218556137241 }, "increased_fraction": 0.5208333333333334, "decreased_fraction": 0.4791666666666667, "winner_agreement": 0.98125 }, "per_text": { "gsm8k-test-0000": { "base": { "count": 1200, "p10": 7.7989315507884305, "median": 10.521758099510357, "p90": 12.900443813085287, "mean": 10.413248549181015 }, "finetuned": { "count": 1200, "p10": 7.705277296646966, "median": 10.515329005859204, "p90": 12.93002657004937, "mean": 10.395004783500097 }, "paired_delta": { "count": 1200, "p10": -0.14676809507078134, "median": 0.004664857327336236, "p90": 0.09345818785291339, "mean": -0.01824376568091725 } }, "gsm8k-test-0002": { "base": { "count": 1200, "p10": 7.421168885724907, "median": 10.089098930849946, "p90": 12.800938492049042, "mean": 10.070913598602885 }, "finetuned": { "count": 1200, "p10": 7.428372247200956, "median": 10.104854688534296, "p90": 12.80511577506368, "mean": 10.069668789579037 }, "paired_delta": { "count": 1200, "p10": -0.12106868600811112, "median": 0.00750254328635247, "p90": 0.09535499714872045, "mean": -0.0012448090238489822 } }, "gsm8k-test-0003": { "base": { "count": 1200, "p10": 7.503553341013432, "median": 10.018123459522165, "p90": 12.583518311707772, "mean": 10.084090114274893 }, "finetuned": { "count": 1200, "p10": 7.459289228486346, "median": 10.039510574599497, "p90": 12.60156308027857, "mean": 10.082087289747665 }, "paired_delta": { "count": 1200, "p10": -0.13171799446835042, "median": -0.002012921035841053, "p90": 0.12220187488991908, "mean": -0.0020028245272282457 } }, "gsm8k-test-0004": { "base": { "count": 1200, "p10": 7.85864274660863, "median": 10.79465641409404, "p90": 12.796374361069722, "mean": 10.512063342893834 }, "finetuned": { "count": 1200, "p10": 7.781806995703267, "median": 10.808550667269296, "p90": 12.815653984573563, "mean": 10.510705999880338 }, "paired_delta": { "count": 1200, "p10": -0.10302150633895835, "median": 0.004346750055341353, "p90": 0.09226896386400388, "mean": -0.0013573430134951603 } } } } } }