summaryrefslogtreecommitdiff
path: root/results/cifar10
diff options
context:
space:
mode:
authorYurenHao0426 <Blackhao0426@gmail.com>2026-03-23 19:46:08 -0500
committerYurenHao0426 <Blackhao0426@gmail.com>2026-03-23 19:46:08 -0500
commit32123cb36ae9521f60c9b6f67458b931b6540ef2 (patch)
tree4731e1dc513f5b613f80c4d20fc4114044c266d3 /results/cifar10
parentbbb1a36d67f2f0c83106c1e771ea2c2fcb7fd83a (diff)
Add final report, plots, experiment guide, and complete NOTE.md
All experiments complete: - Toy LQ: credit bridge matches state bridge (~0.94 costate cosine) - CIFAR-10: credit bridge (29.6%) comparable to DFA (30.0%), both beat state bridge (18.5%) - State bridge confirms core hypothesis: perfect state prediction != useful credit - Terminal gradient matching is essential for credit bridge
Diffstat (limited to 'results/cifar10')
-rw-r--r--results/cifar10/results_cifar10.json1985
1 files changed, 1985 insertions, 0 deletions
diff --git a/results/cifar10/results_cifar10.json b/results/cifar10/results_cifar10.json
new file mode 100644
index 0000000..b4bd094
--- /dev/null
+++ b/results/cifar10/results_cifar10.json
@@ -0,0 +1,1985 @@
+{
+ "42": {
+ "bp": {
+ "log": {
+ "train_loss": [
+ 1.9159408319854736,
+ 1.6833457696151732,
+ 1.5935267736053467,
+ 1.5389775912475585,
+ 1.4867800998687744,
+ 1.4524269732284545,
+ 1.4178237664794922,
+ 1.3897597776794433,
+ 1.3664597565460206,
+ 1.342870984840393,
+ 1.3235610486221314,
+ 1.301216788482666,
+ 1.279724592475891,
+ 1.2618738833999634,
+ 1.2407353455352783,
+ 1.22462412940979,
+ 1.2109693797302246,
+ 1.1892420639038086,
+ 1.1723702849578856,
+ 1.1557361126327514,
+ 1.1423172385787963,
+ 1.1261568475341797,
+ 1.1062135691070556,
+ 1.0966129437637329,
+ 1.0756375621795655,
+ 1.0581538507270813,
+ 1.0487863679885865,
+ 1.0285031553268433,
+ 1.0112163684082032,
+ 0.9978593858337402,
+ 0.9784134501266479,
+ 0.9645692346191406,
+ 0.9500791883468628,
+ 0.9311204533195495,
+ 0.9137258143806457,
+ 0.8955737284660339,
+ 0.8760775566101074,
+ 0.8636410433959961,
+ 0.8472155108451843,
+ 0.8257175942802429,
+ 0.8081484350395203,
+ 0.7949162258720398,
+ 0.7682366105842591,
+ 0.7535309395217895,
+ 0.7374585538101196,
+ 0.7163707330703736,
+ 0.7002532124710082,
+ 0.6810121598052978,
+ 0.6590044124794007,
+ 0.6416365943050385,
+ 0.6245126323890686,
+ 0.6053193593788146,
+ 0.5843256318569183,
+ 0.5672504850196839,
+ 0.5574265190124512,
+ 0.5282080096054077,
+ 0.5124851553916931,
+ 0.49809444856643675,
+ 0.4815664218711853,
+ 0.46325312936782836,
+ 0.44530024614334107,
+ 0.4232256102848053,
+ 0.41155227093696595,
+ 0.3951142164516449,
+ 0.3779419419193268,
+ 0.36394513979434967,
+ 0.3495154238319397,
+ 0.33435829617500307,
+ 0.3247876634025574,
+ 0.31244264134407046,
+ 0.30034732496261596,
+ 0.2886586371707916,
+ 0.2739575230407715,
+ 0.26273826862812044,
+ 0.25702780318260193,
+ 0.23921729731559754,
+ 0.2356625511932373,
+ 0.23087995275020598,
+ 0.21462623807907105,
+ 0.21162452461242676,
+ 0.20137000153541565,
+ 0.1957099277973175,
+ 0.19024162811279297,
+ 0.1820225635457039,
+ 0.17860664487838745,
+ 0.17204254170417785,
+ 0.1648883514738083,
+ 0.1611987146615982,
+ 0.15891523240566255,
+ 0.15874515238285064,
+ 0.15378751336574553,
+ 0.15460485486030578,
+ 0.1495811799764633,
+ 0.14360208065509797,
+ 0.14562011506080627,
+ 0.14506465085983275,
+ 0.1431816494703293,
+ 0.1397772234392166,
+ 0.14342590034008026,
+ 0.142473215675354
+ ],
+ "train_acc": [
+ 0.31126,
+ 0.39118,
+ 0.41882,
+ 0.44388,
+ 0.4617,
+ 0.4731,
+ 0.48828,
+ 0.49882,
+ 0.50712,
+ 0.51646,
+ 0.5211,
+ 0.53098,
+ 0.53964,
+ 0.54616,
+ 0.55448,
+ 0.55806,
+ 0.56566,
+ 0.57324,
+ 0.57908,
+ 0.58582,
+ 0.58824,
+ 0.59418,
+ 0.60314,
+ 0.60678,
+ 0.61408,
+ 0.62094,
+ 0.62264,
+ 0.63056,
+ 0.63836,
+ 0.64046,
+ 0.64822,
+ 0.65186,
+ 0.65832,
+ 0.66298,
+ 0.67222,
+ 0.67798,
+ 0.686,
+ 0.688,
+ 0.695,
+ 0.7014,
+ 0.709,
+ 0.71344,
+ 0.7245,
+ 0.72892,
+ 0.73524,
+ 0.7426,
+ 0.74692,
+ 0.75382,
+ 0.76378,
+ 0.76914,
+ 0.77694,
+ 0.78392,
+ 0.78984,
+ 0.79764,
+ 0.79948,
+ 0.81182,
+ 0.81536,
+ 0.82284,
+ 0.82968,
+ 0.83584,
+ 0.84158,
+ 0.8514,
+ 0.85392,
+ 0.86052,
+ 0.86604,
+ 0.87092,
+ 0.87644,
+ 0.88364,
+ 0.88486,
+ 0.88988,
+ 0.89316,
+ 0.89914,
+ 0.9048,
+ 0.90892,
+ 0.90944,
+ 0.91828,
+ 0.9189,
+ 0.92122,
+ 0.92622,
+ 0.92784,
+ 0.93044,
+ 0.9332,
+ 0.93514,
+ 0.93836,
+ 0.93948,
+ 0.94166,
+ 0.94506,
+ 0.9452,
+ 0.94606,
+ 0.94738,
+ 0.9493,
+ 0.94928,
+ 0.94976,
+ 0.95178,
+ 0.9521,
+ 0.95234,
+ 0.95228,
+ 0.9543,
+ 0.95236,
+ 0.9534
+ ],
+ "test_acc": [
+ 0.3908,
+ 0.433,
+ 0.4419,
+ 0.4858,
+ 0.4871,
+ 0.4878,
+ 0.5141,
+ 0.5211,
+ 0.5188,
+ 0.5279,
+ 0.5296,
+ 0.5402,
+ 0.5415,
+ 0.5584,
+ 0.5559,
+ 0.5648,
+ 0.5531,
+ 0.5686,
+ 0.5685,
+ 0.575,
+ 0.5694,
+ 0.5749,
+ 0.5711,
+ 0.585,
+ 0.5778,
+ 0.5875,
+ 0.5827,
+ 0.5876,
+ 0.5868,
+ 0.5947,
+ 0.5903,
+ 0.6001,
+ 0.5963,
+ 0.6013,
+ 0.5943,
+ 0.5994,
+ 0.5879,
+ 0.5925,
+ 0.6006,
+ 0.5978,
+ 0.6007,
+ 0.6034,
+ 0.5914,
+ 0.5958,
+ 0.5959,
+ 0.5959,
+ 0.5928,
+ 0.5961,
+ 0.6013,
+ 0.5978,
+ 0.5974,
+ 0.5998,
+ 0.5919,
+ 0.5936,
+ 0.5958,
+ 0.6019,
+ 0.5972,
+ 0.5927,
+ 0.5998,
+ 0.5945,
+ 0.5942,
+ 0.5984,
+ 0.5904,
+ 0.5907,
+ 0.595,
+ 0.5928,
+ 0.5944,
+ 0.5973,
+ 0.5924,
+ 0.5909,
+ 0.5976,
+ 0.5938,
+ 0.5943,
+ 0.5926,
+ 0.5969,
+ 0.5918,
+ 0.5943,
+ 0.5881,
+ 0.5907,
+ 0.5921,
+ 0.5934,
+ 0.5926,
+ 0.5919,
+ 0.5923,
+ 0.5912,
+ 0.5908,
+ 0.5888,
+ 0.5899,
+ 0.5904,
+ 0.5911,
+ 0.5911,
+ 0.5905,
+ 0.5894,
+ 0.5918,
+ 0.5926,
+ 0.5931,
+ 0.5935,
+ 0.5936,
+ 0.5939,
+ 0.594
+ ]
+ },
+ "diagnostics": {
+ "bp_cosine": [
+ 0.9423961043357849,
+ 0.9423063397407532,
+ 0.9428478479385376,
+ 0.9422410726547241,
+ 0.9424552917480469,
+ 0.9417660236358643,
+ 0.9408372640609741,
+ 0.9404677152633667,
+ 0.9403290748596191,
+ 0.9396632313728333,
+ 0.935971736907959,
+ 0.9338305592536926
+ ],
+ "perturbation_rho": [
+ 0.987454891204834,
+ 0.9897942543029785,
+ 0.990141749382019,
+ 0.9901803731918335,
+ 0.9920680522918701,
+ 0.9906818866729736,
+ 0.9903223514556885,
+ 0.9926531910896301,
+ 0.9914379119873047,
+ 0.9897843599319458,
+ 0.9882909059524536,
+ 0.9869540929794312
+ ],
+ "nudging": {
+ "0.001": [
+ -0.002897711703553796,
+ -0.002905686851590872,
+ -0.0029487141873687506,
+ -0.0030073150992393494,
+ -0.002988070249557495,
+ -0.0028954721055924892,
+ -0.0027900305576622486,
+ -0.002700705546885729,
+ -0.0025893584825098515,
+ -0.0024207504466176033,
+ -0.002165877493098378,
+ -0.00189895648509264
+ ],
+ "0.003": [
+ -0.008684658445417881,
+ -0.008708733133971691,
+ -0.008837150409817696,
+ -0.009012945927679539,
+ -0.008955370634794235,
+ -0.008678246289491653,
+ -0.008362163789570332,
+ -0.008094580844044685,
+ -0.007761792279779911,
+ -0.007257224526256323,
+ -0.006493166089057922,
+ -0.005693747662007809
+ ],
+ "0.01": [
+ -0.028850272297859192,
+ -0.028929298743605614,
+ -0.029354672878980637,
+ -0.029939891770482063,
+ -0.029747068881988525,
+ -0.028829604387283325,
+ -0.027784891426563263,
+ -0.026898572221398354,
+ -0.025797637179493904,
+ -0.024130169302225113,
+ -0.02159387245774269,
+ -0.018940377980470657
+ ]
+ }
+ },
+ "drift": {
+ "embed.weight": 8.833701837291958,
+ "embed.bias": 14.261815769395275,
+ "blocks.0.ln.weight": 0.34404799666331826,
+ "blocks.0.w1.weight": 4.175094808562947,
+ "blocks.0.w1.bias": 6.98398106061982,
+ "blocks.0.w2.weight": 10.592751719664884,
+ "blocks.1.ln.weight": 0.2322510556310593,
+ "blocks.1.w1.weight": 4.523413515316886,
+ "blocks.1.w1.bias": 4.3749413052218085,
+ "blocks.1.w2.weight": 11.940958034527862,
+ "blocks.2.ln.weight": 0.2150404740407862,
+ "blocks.2.w1.weight": 4.58583359520105,
+ "blocks.2.w1.bias": 4.159787661269507,
+ "blocks.2.w2.weight": 12.401201224911981,
+ "blocks.3.ln.weight": 0.22107801019329978,
+ "blocks.3.w1.weight": 4.586995215167557,
+ "blocks.3.w1.bias": 3.881613437324734,
+ "blocks.3.w2.weight": 12.64024804795375,
+ "blocks.4.ln.weight": 0.23170407382172373,
+ "blocks.4.w1.weight": 4.578125313924808,
+ "blocks.4.w1.bias": 3.7932938751482683,
+ "blocks.4.w2.weight": 12.963501907885533,
+ "blocks.5.ln.weight": 0.23628225424614754,
+ "blocks.5.w1.weight": 4.5240873092186416,
+ "blocks.5.w1.bias": 3.803760854867017,
+ "blocks.5.w2.weight": 12.912394584749915,
+ "blocks.6.ln.weight": 0.2619501911907522,
+ "blocks.6.w1.weight": 4.461384236716512,
+ "blocks.6.w1.bias": 3.799496573870174,
+ "blocks.6.w2.weight": 13.155197671866139,
+ "blocks.7.ln.weight": 0.2657175505296468,
+ "blocks.7.w1.weight": 4.422447215896801,
+ "blocks.7.w1.bias": 3.725743197114046,
+ "blocks.7.w2.weight": 13.149302405423146,
+ "blocks.8.ln.weight": 0.28028080422594653,
+ "blocks.8.w1.weight": 4.320993108213275,
+ "blocks.8.w1.bias": 3.577600797368634,
+ "blocks.8.w2.weight": 13.112473965027567,
+ "blocks.9.ln.weight": 0.3122222575319159,
+ "blocks.9.w1.weight": 4.3011661234952125,
+ "blocks.9.w1.bias": 3.5701937101673704,
+ "blocks.9.w2.weight": 13.20746990540504,
+ "blocks.10.ln.weight": 0.2907587848995931,
+ "blocks.10.w1.weight": 4.244185315341837,
+ "blocks.10.w1.bias": 3.487530574891753,
+ "blocks.10.w2.weight": 13.342063167376029,
+ "blocks.11.ln.weight": 0.30672613980463925,
+ "blocks.11.w1.weight": 4.1911870376224645,
+ "blocks.11.w1.bias": 3.626239385452188,
+ "blocks.11.w2.weight": 13.53569165604796,
+ "out_ln.weight": 0.304987181878743,
+ "out_head.weight": 3.659824943443211,
+ "out_head.bias": 0.7848376848440143
+ }
+ },
+ "dfa": {
+ "log": {
+ "train_loss": [
+ 2.060509294204712,
+ 2.0283543349456785,
+ 2.0286419232940673,
+ 2.0282608281707764,
+ 2.0229165492248535,
+ 2.0251557587432862,
+ 2.0200582264709475,
+ 2.021266340560913,
+ 2.0176746366119387,
+ 2.0216268117523195,
+ 2.016007042388916,
+ 2.0150310187530516,
+ 2.0130176863098144,
+ 2.011090052871704,
+ 2.0114055884552,
+ 2.012881319503784,
+ 2.0109381372070314,
+ 2.011416566619873,
+ 2.009194619445801,
+ 2.008145341949463,
+ 2.008374558181763,
+ 2.009201191329956,
+ 2.0073699200820925,
+ 2.0068252397155764,
+ 2.006670061340332,
+ 2.006027181663513,
+ 2.0085990719604494,
+ 2.006233550338745,
+ 2.007863084945679,
+ 2.007588200149536,
+ 2.0059480542755126,
+ 2.0044747730255126,
+ 2.006523677520752,
+ 2.0035686435317994,
+ 2.0047321628189088,
+ 2.005557894821167,
+ 2.003938685531616,
+ 2.005093680458069,
+ 2.0022387979507448,
+ 2.004965279159546,
+ 2.006689316253662,
+ 2.0055594663238527,
+ 2.004364362411499,
+ 2.0020452467346193,
+ 2.005489226150513,
+ 2.003978981628418,
+ 2.0023497142791746,
+ 2.003284743270874,
+ 2.0037755869293212,
+ 2.001501432533264,
+ 2.003944476890564,
+ 2.0025060961914063,
+ 2.002634817123413,
+ 2.0017083084869385,
+ 2.0028567737579346,
+ 2.000872339820862,
+ 2.003729987487793,
+ 2.002261364898682,
+ 2.0024312939453126,
+ 2.0008276205444337,
+ 2.001199483261108,
+ 2.0006834197235106,
+ 2.000615832366943,
+ 1.9993626528930664,
+ 2.0005247971343993,
+ 1.9997058053588868,
+ 1.999853487548828,
+ 1.9997802280044557,
+ 1.9991605996704103,
+ 2.0015039880752563,
+ 1.9991352321624756,
+ 1.9994654761505126,
+ 1.99805854221344,
+ 1.9966946460723878,
+ 1.9976105904388428,
+ 1.9990063135528564,
+ 1.995104397277832,
+ 1.9979646807479858,
+ 1.996112073097229,
+ 1.999074932899475,
+ 1.997551460647583,
+ 1.9962433197784424,
+ 1.997741110534668,
+ 1.998971039352417,
+ 1.9981346895599366,
+ 1.9993204745864868,
+ 1.9973036824798585,
+ 1.9972220791625976,
+ 1.9969999575042725,
+ 1.9983418984222412,
+ 1.9977535138702394,
+ 1.9974792404174804,
+ 1.9975066395568848,
+ 1.9948017989349365,
+ 1.9976891431427002,
+ 1.9985219841003419,
+ 1.9948620611190795,
+ 1.995753077468872,
+ 1.998552144088745,
+ 1.9990386080932616
+ ],
+ "train_acc": [
+ 0.23936,
+ 0.25434,
+ 0.25346,
+ 0.25514,
+ 0.25882,
+ 0.25722,
+ 0.25998,
+ 0.25844,
+ 0.262,
+ 0.25976,
+ 0.26256,
+ 0.2639,
+ 0.26116,
+ 0.26558,
+ 0.26594,
+ 0.26562,
+ 0.26654,
+ 0.26378,
+ 0.2653,
+ 0.26746,
+ 0.26592,
+ 0.2681,
+ 0.26896,
+ 0.27006,
+ 0.26992,
+ 0.26862,
+ 0.26736,
+ 0.27126,
+ 0.26738,
+ 0.2694,
+ 0.26836,
+ 0.26744,
+ 0.27024,
+ 0.27184,
+ 0.26816,
+ 0.26806,
+ 0.27192,
+ 0.26898,
+ 0.27294,
+ 0.27124,
+ 0.2707,
+ 0.27056,
+ 0.272,
+ 0.27288,
+ 0.27096,
+ 0.27286,
+ 0.27098,
+ 0.27252,
+ 0.26964,
+ 0.27208,
+ 0.27,
+ 0.2737,
+ 0.27368,
+ 0.27134,
+ 0.27342,
+ 0.27386,
+ 0.27258,
+ 0.27398,
+ 0.27468,
+ 0.27518,
+ 0.27328,
+ 0.2752,
+ 0.27404,
+ 0.27526,
+ 0.2754,
+ 0.27588,
+ 0.27358,
+ 0.27472,
+ 0.27548,
+ 0.27356,
+ 0.27508,
+ 0.27656,
+ 0.27638,
+ 0.27784,
+ 0.27648,
+ 0.27606,
+ 0.27718,
+ 0.27632,
+ 0.27822,
+ 0.27514,
+ 0.27692,
+ 0.27566,
+ 0.27648,
+ 0.276,
+ 0.27642,
+ 0.27526,
+ 0.27678,
+ 0.2763,
+ 0.27776,
+ 0.27652,
+ 0.27578,
+ 0.27868,
+ 0.27676,
+ 0.27782,
+ 0.27764,
+ 0.27704,
+ 0.27938,
+ 0.27672,
+ 0.2781,
+ 0.2784
+ ],
+ "test_acc": [
+ 0.2781,
+ 0.2717,
+ 0.2892,
+ 0.2711,
+ 0.2577,
+ 0.2687,
+ 0.2516,
+ 0.281,
+ 0.2759,
+ 0.2822,
+ 0.2529,
+ 0.2933,
+ 0.2908,
+ 0.3048,
+ 0.2739,
+ 0.2751,
+ 0.2888,
+ 0.2869,
+ 0.3078,
+ 0.2855,
+ 0.2884,
+ 0.2932,
+ 0.2955,
+ 0.2952,
+ 0.2919,
+ 0.2886,
+ 0.2871,
+ 0.3017,
+ 0.295,
+ 0.3001,
+ 0.2957,
+ 0.2866,
+ 0.2896,
+ 0.3055,
+ 0.2928,
+ 0.3027,
+ 0.2836,
+ 0.2903,
+ 0.2958,
+ 0.2913,
+ 0.2966,
+ 0.287,
+ 0.2877,
+ 0.2879,
+ 0.303,
+ 0.3006,
+ 0.2946,
+ 0.2867,
+ 0.2973,
+ 0.2899,
+ 0.2917,
+ 0.3002,
+ 0.2971,
+ 0.2964,
+ 0.3037,
+ 0.2892,
+ 0.3038,
+ 0.2898,
+ 0.2999,
+ 0.3003,
+ 0.3047,
+ 0.2936,
+ 0.2972,
+ 0.2905,
+ 0.3016,
+ 0.2933,
+ 0.295,
+ 0.2998,
+ 0.3,
+ 0.3008,
+ 0.3028,
+ 0.3003,
+ 0.303,
+ 0.2946,
+ 0.2915,
+ 0.2974,
+ 0.2901,
+ 0.2978,
+ 0.2991,
+ 0.2973,
+ 0.2948,
+ 0.3032,
+ 0.3049,
+ 0.3017,
+ 0.2999,
+ 0.3019,
+ 0.2958,
+ 0.2981,
+ 0.3009,
+ 0.2968,
+ 0.2987,
+ 0.297,
+ 0.2988,
+ 0.2996,
+ 0.3002,
+ 0.3003,
+ 0.3001,
+ 0.3003,
+ 0.3001,
+ 0.3001
+ ]
+ },
+ "diagnostics": {
+ "bp_cosine": [
+ 0.3634233772754669,
+ 0.001164254266768694,
+ -0.0006461217999458313,
+ -0.00019622840045485646,
+ 0.0002198266884079203,
+ -9.971270628739148e-06,
+ 0.00014979815750848502,
+ -0.0004021040804218501,
+ -5.425223207566887e-05,
+ -0.00024170493998099118,
+ -0.00025007897056639194,
+ -0.00018576104776002467
+ ],
+ "perturbation_rho": [
+ 0.061064913868904114,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.00392885273322463,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "nudging": {
+ "0.001": [
+ -3.8743019104003906e-07,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "0.003": [
+ -1.1324882507324219e-06,
+ 0.0,
+ 2.7939677238464355e-09,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "0.01": [
+ -3.606081008911133e-06,
+ 0.0,
+ 1.3504177331924438e-08,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ]
+ }
+ },
+ "drift": {
+ "embed.weight": 333.70584841410414,
+ "embed.bias": 239.49966422804454,
+ "blocks.0.ln.weight": 10.156513336148182,
+ "blocks.0.w1.weight": 293.93859688585053,
+ "blocks.0.w1.bias": 281.7468078045666,
+ "blocks.0.w2.weight": 492.64207719718667,
+ "blocks.1.ln.weight": 8.920722535237505,
+ "blocks.1.w1.weight": 311.4826752480239,
+ "blocks.1.w1.bias": 310.86370160689785,
+ "blocks.1.w2.weight": 324.5405256645745,
+ "blocks.2.ln.weight": 10.152850269187711,
+ "blocks.2.w1.weight": 422.5618024966304,
+ "blocks.2.w1.bias": 406.03916399589247,
+ "blocks.2.w2.weight": 427.4195122419038,
+ "blocks.3.ln.weight": 6.982139429700868,
+ "blocks.3.w1.weight": 239.67203375670587,
+ "blocks.3.w1.bias": 218.06238930657585,
+ "blocks.3.w2.weight": 238.92558897030935,
+ "blocks.4.ln.weight": 7.833071503056953,
+ "blocks.4.w1.weight": 298.1966678874803,
+ "blocks.4.w1.bias": 278.16504637194174,
+ "blocks.4.w2.weight": 274.11280875965394,
+ "blocks.5.ln.weight": 7.242266411414109,
+ "blocks.5.w1.weight": 264.2525555458606,
+ "blocks.5.w1.bias": 252.72325009085455,
+ "blocks.5.w2.weight": 257.616366671581,
+ "blocks.6.ln.weight": 7.548699630616584,
+ "blocks.6.w1.weight": 262.2510067819824,
+ "blocks.6.w1.bias": 251.81627023731494,
+ "blocks.6.w2.weight": 273.41300895523614,
+ "blocks.7.ln.weight": 8.476523067012732,
+ "blocks.7.w1.weight": 318.7369678667611,
+ "blocks.7.w1.bias": 299.5996411120576,
+ "blocks.7.w2.weight": 294.5553970248432,
+ "blocks.8.ln.weight": 10.089688663753533,
+ "blocks.8.w1.weight": 404.9516419003623,
+ "blocks.8.w1.bias": 379.7449486217603,
+ "blocks.8.w2.weight": 384.6524371767172,
+ "blocks.9.ln.weight": 10.376434921092248,
+ "blocks.9.w1.weight": 410.9149275889754,
+ "blocks.9.w1.bias": 383.0074711781066,
+ "blocks.9.w2.weight": 392.8539842927831,
+ "blocks.10.ln.weight": 6.445966095557191,
+ "blocks.10.w1.weight": 233.37502830905152,
+ "blocks.10.w1.bias": 219.30705311523053,
+ "blocks.10.w2.weight": 224.7808320380166,
+ "blocks.11.ln.weight": 9.439442353351934,
+ "blocks.11.w1.weight": 374.5461893118647,
+ "blocks.11.w1.bias": 354.2106427892852,
+ "blocks.11.w2.weight": 349.272673385976,
+ "out_ln.weight": 0.6481155342918145,
+ "out_head.weight": 9.174388328815116,
+ "out_head.bias": 0.47817510393040696
+ }
+ },
+ "state_bridge": {
+ "log": {
+ "train_loss": [
+ 2.249122398071289,
+ 2.143466188659668,
+ 2.182890202178955,
+ 2.316445587387085,
+ 2.3062049280548096,
+ 2.266785619735718,
+ 2.2987561087799073,
+ 2.2884853198242188,
+ 2.240994506072998,
+ 2.2790724418640136,
+ 2.2708915201568605,
+ 2.266933690185547,
+ 2.2886832971954347,
+ 2.2700843003845215,
+ 2.1964657810211183,
+ 2.129935463180542,
+ 2.1168104545593263,
+ 2.12433757270813,
+ 2.113173648223877,
+ 2.1130354350280762,
+ 2.1059150744628905,
+ 2.091750974731445,
+ 2.092220969924927,
+ 2.124610143814087,
+ 2.096196698989868,
+ 2.089525361709595,
+ 2.0865603291320802,
+ 2.0807934146118163,
+ 2.069572374725342,
+ 2.0693005324554443,
+ 2.064417702713013,
+ 2.061735754241943,
+ 2.0726303718566896,
+ 2.068593667602539,
+ 2.0655730045318603,
+ 2.0626395158386233,
+ 2.063313564758301,
+ 2.0622880627441407,
+ 2.058299614639282,
+ 2.059487857131958,
+ 2.0631128742980955,
+ 2.0610568396759033,
+ 2.0551611611938476,
+ 2.056872489852905,
+ 2.0546858750915526,
+ 2.056325934524536,
+ 2.053202699737549,
+ 2.0516857704162597,
+ 2.0511376399230956,
+ 2.052292654342651,
+ 2.0525611149597167,
+ 2.052803971786499,
+ 2.0563543517684937,
+ 2.058361290130615,
+ 2.059238384399414,
+ 2.060622958755493,
+ 2.063846826324463,
+ 2.0644858879852297,
+ 2.0641055697631834,
+ 2.062063186683655,
+ 2.0649280325317383,
+ 2.065730286026001,
+ 2.0664522916412356,
+ 2.066531411819458,
+ 2.0671734712982177,
+ 2.066738704376221,
+ 2.06800027633667,
+ 2.0698299375915528,
+ 2.0699130378723143,
+ 2.069541824951172,
+ 2.069294273452759,
+ 2.068808924255371,
+ 2.071917738342285,
+ 2.0731585116577147,
+ 2.0724312242126466,
+ 2.073472029876709,
+ 2.0752377517700196,
+ 2.0765309967041015,
+ 2.0784705491638182,
+ 2.079111976928711,
+ 2.079522602157593,
+ 2.080129363708496,
+ 2.0800775938415526,
+ 2.0806046531677245,
+ 2.080093946685791,
+ 2.082602798614502,
+ 2.0823856858062744,
+ 2.083605904006958,
+ 2.082737918624878,
+ 2.0829258354187012,
+ 2.082134012680054,
+ 2.081413818206787,
+ 2.080659938735962,
+ 2.079646938858032,
+ 2.079785990867615,
+ 2.079912822265625,
+ 2.0787852781677247,
+ 2.080009171295166,
+ 2.0796959645080566,
+ 2.07859284072876
+ ],
+ "train_acc": [
+ 0.14508,
+ 0.16856,
+ 0.15632,
+ 0.11096,
+ 0.12388,
+ 0.14108,
+ 0.12374,
+ 0.1273,
+ 0.14434,
+ 0.12372,
+ 0.13196,
+ 0.13698,
+ 0.1278,
+ 0.13406,
+ 0.15046,
+ 0.17556,
+ 0.18282,
+ 0.17838,
+ 0.18174,
+ 0.1771,
+ 0.1835,
+ 0.18812,
+ 0.1935,
+ 0.17326,
+ 0.18206,
+ 0.1833,
+ 0.18656,
+ 0.19114,
+ 0.19432,
+ 0.19874,
+ 0.19688,
+ 0.19564,
+ 0.19196,
+ 0.19548,
+ 0.1934,
+ 0.19826,
+ 0.19794,
+ 0.19896,
+ 0.20276,
+ 0.20258,
+ 0.20152,
+ 0.20098,
+ 0.2025,
+ 0.20058,
+ 0.20388,
+ 0.20206,
+ 0.2039,
+ 0.20444,
+ 0.2054,
+ 0.20728,
+ 0.2042,
+ 0.20478,
+ 0.204,
+ 0.1998,
+ 0.20252,
+ 0.2013,
+ 0.1988,
+ 0.2002,
+ 0.20158,
+ 0.2012,
+ 0.20384,
+ 0.20102,
+ 0.1987,
+ 0.20136,
+ 0.19972,
+ 0.20122,
+ 0.19876,
+ 0.19792,
+ 0.20008,
+ 0.20138,
+ 0.1992,
+ 0.19956,
+ 0.20076,
+ 0.19942,
+ 0.20054,
+ 0.20192,
+ 0.20256,
+ 0.2009,
+ 0.1972,
+ 0.20028,
+ 0.19826,
+ 0.1992,
+ 0.19678,
+ 0.19736,
+ 0.19872,
+ 0.20076,
+ 0.20126,
+ 0.20262,
+ 0.20088,
+ 0.20116,
+ 0.2009,
+ 0.20188,
+ 0.20326,
+ 0.20284,
+ 0.20404,
+ 0.20164,
+ 0.20386,
+ 0.20324,
+ 0.20414,
+ 0.2032
+ ],
+ "test_acc": [
+ 0.1701,
+ 0.1589,
+ 0.0964,
+ 0.0977,
+ 0.147,
+ 0.1388,
+ 0.1311,
+ 0.1208,
+ 0.1018,
+ 0.1576,
+ 0.117,
+ 0.1127,
+ 0.1215,
+ 0.1579,
+ 0.1648,
+ 0.1561,
+ 0.179,
+ 0.1723,
+ 0.1962,
+ 0.1791,
+ 0.1863,
+ 0.1757,
+ 0.161,
+ 0.166,
+ 0.1768,
+ 0.1795,
+ 0.18,
+ 0.1802,
+ 0.1787,
+ 0.1767,
+ 0.1754,
+ 0.1604,
+ 0.1761,
+ 0.1675,
+ 0.1741,
+ 0.171,
+ 0.1697,
+ 0.1797,
+ 0.1836,
+ 0.1869,
+ 0.19,
+ 0.1893,
+ 0.1907,
+ 0.1866,
+ 0.1877,
+ 0.1889,
+ 0.1871,
+ 0.1949,
+ 0.1817,
+ 0.1912,
+ 0.1923,
+ 0.1955,
+ 0.1958,
+ 0.1893,
+ 0.1961,
+ 0.1915,
+ 0.1966,
+ 0.2031,
+ 0.1959,
+ 0.1915,
+ 0.1753,
+ 0.1929,
+ 0.18,
+ 0.1823,
+ 0.194,
+ 0.1869,
+ 0.1969,
+ 0.1912,
+ 0.1946,
+ 0.1919,
+ 0.1971,
+ 0.1907,
+ 0.2003,
+ 0.1926,
+ 0.1998,
+ 0.1947,
+ 0.1862,
+ 0.1997,
+ 0.1874,
+ 0.1841,
+ 0.188,
+ 0.1869,
+ 0.1828,
+ 0.1856,
+ 0.1888,
+ 0.1734,
+ 0.1826,
+ 0.1813,
+ 0.1816,
+ 0.1868,
+ 0.1879,
+ 0.1846,
+ 0.1837,
+ 0.182,
+ 0.1821,
+ 0.1836,
+ 0.1838,
+ 0.1833,
+ 0.1832,
+ 0.1836
+ ],
+ "state_pred_error": [
+ 0.3305018243932724,
+ 0.3240715994644165,
+ 0.2682553328800201,
+ 0.13893559893369675,
+ 0.07450400803923607,
+ 0.05266491491317749,
+ 0.035537137892246246,
+ 0.08557363073348999,
+ 0.0663624380338192,
+ 0.05056835491657257,
+ 0.057957575828433035,
+ 0.03669406015276909,
+ 0.044723540180921555,
+ 0.05952853068828583,
+ 0.06200094776391983,
+ 0.034456856629252436,
+ 0.02363166331708431,
+ 0.030915076084136964,
+ 0.027922455305457117,
+ 0.022749129890203477,
+ 0.01869435190498829,
+ 0.016911904378533363,
+ 0.012851448996663094,
+ 0.009603057160079479,
+ 0.0068074954861402515,
+ 0.004831086874008178,
+ 0.002716479171216488,
+ 0.0017772294883430003,
+ 0.0011674952478520572,
+ 0.0009641216253489256,
+ 0.000903967960383743,
+ 0.0008178514664620161,
+ 0.0009606654725596308,
+ 0.0010464790955930948,
+ 0.0009382549912855029,
+ 0.0008025767210498452,
+ 0.0006237311275862158,
+ 0.0005590206385217607,
+ 0.0006743966669589281,
+ 0.0007392789867147803,
+ 0.0007174553440324963,
+ 0.0006993773563578725,
+ 0.000645849712882191,
+ 0.0005546821858268231,
+ 0.00046942476788535716,
+ 0.0003906826237775385,
+ 0.0003712719864305109,
+ 0.0003606485006958246,
+ 0.0003457060167193413,
+ 0.0003746068775840104,
+ 0.00038963068067096173,
+ 0.0003513397928327322,
+ 0.00032962770885787906,
+ 0.00033959761667996645,
+ 0.0003122131848614663,
+ 0.00028035666743293406,
+ 0.00026064857746940106,
+ 0.00022029355119913816,
+ 0.00019001943845301867,
+ 0.00020799835772719233,
+ 0.0002275417461199686,
+ 0.00020783223733305932,
+ 0.00018264570890925825,
+ 0.00015985633003059774,
+ 0.0001389945726841688,
+ 0.00011476213769754394,
+ 8.967853816691787e-05,
+ 7.122783363331109e-05,
+ 5.970104366540909e-05,
+ 5.121438584523275e-05,
+ 4.536606192123145e-05,
+ 4.0154456216841935e-05,
+ 3.647202713880688e-05,
+ 3.4129825913114475e-05,
+ 3.2702885311446155e-05,
+ 3.160392617806792e-05,
+ 3.0352398289833216e-05,
+ 2.940795918868389e-05,
+ 2.8640556085156276e-05,
+ 2.736381722963415e-05,
+ 2.6395615809597075e-05,
+ 2.5282455111155287e-05,
+ 2.4516119782347232e-05,
+ 2.388914132665377e-05,
+ 2.2789849505061285e-05,
+ 2.1843732089619152e-05,
+ 2.0776926304679363e-05,
+ 1.9760204966296442e-05,
+ 1.8386752204969525e-05,
+ 1.7638299430254845e-05,
+ 1.702676533255726e-05,
+ 1.6189092184067705e-05,
+ 1.5407667033723554e-05,
+ 1.4915904570952988e-05,
+ 1.4520096725318581e-05,
+ 1.3764947652525734e-05,
+ 1.333884218824096e-05,
+ 1.3029495921800845e-05,
+ 1.262850838684244e-05,
+ 1.2368245947291143e-05
+ ]
+ },
+ "diagnostics": {
+ "bp_cosine": [
+ 0.15691214799880981,
+ 0.015022292733192444,
+ 0.006593926809728146,
+ 0.007205793634057045,
+ 0.008038382977247238,
+ 0.0067636389285326,
+ 0.005863174330443144,
+ 0.0026869114954024553,
+ 0.00880957767367363,
+ 0.009782599285244942,
+ 0.00875927321612835,
+ 0.015338878147304058
+ ],
+ "perturbation_rho": [
+ 0.046312831342220306,
+ 0.002848614938557148,
+ -7.617553637828678e-05,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "nudging": {
+ "0.001": [
+ -4.103407263755798e-06,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "0.003": [
+ -1.2268312275409698e-05,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "0.01": [
+ -4.0831975638866425e-05,
+ 7.450580596923828e-09,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 9.313225746154785e-10,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ]
+ }
+ },
+ "drift": {
+ "embed.weight": 20.26978490963512,
+ "embed.bias": 97.32930339256669,
+ "blocks.0.ln.weight": 1.9308135867617757,
+ "blocks.0.w1.weight": 66.3483568365869,
+ "blocks.0.w1.bias": 53.284724707801,
+ "blocks.0.w2.weight": 177.08907259952375,
+ "blocks.1.ln.weight": 3.057512831818983,
+ "blocks.1.w1.weight": 170.34177401016777,
+ "blocks.1.w1.bias": 213.60524387953882,
+ "blocks.1.w2.weight": 200.26342063170046,
+ "blocks.2.ln.weight": 2.5608747595416883,
+ "blocks.2.w1.weight": 103.95319184812183,
+ "blocks.2.w1.bias": 101.21405935076268,
+ "blocks.2.w2.weight": 134.8362490743747,
+ "blocks.3.ln.weight": 2.3109975543869266,
+ "blocks.3.w1.weight": 116.99989433662596,
+ "blocks.3.w1.bias": 120.91968382837668,
+ "blocks.3.w2.weight": 142.44744178171467,
+ "blocks.4.ln.weight": 2.718708135007822,
+ "blocks.4.w1.weight": 124.26248150423525,
+ "blocks.4.w1.bias": 121.48105264933947,
+ "blocks.4.w2.weight": 118.7892683407888,
+ "blocks.5.ln.weight": 1.8158878954501887,
+ "blocks.5.w1.weight": 56.878036794490896,
+ "blocks.5.w1.bias": 36.57930796279972,
+ "blocks.5.w2.weight": 85.47839722655326,
+ "blocks.6.ln.weight": 3.0875197025983447,
+ "blocks.6.w1.weight": 100.99150638794737,
+ "blocks.6.w1.bias": 97.42240055489651,
+ "blocks.6.w2.weight": 87.13862478741176,
+ "blocks.7.ln.weight": 7.772248541992971,
+ "blocks.7.w1.weight": 297.4998489055548,
+ "blocks.7.w1.bias": 282.6199479818741,
+ "blocks.7.w2.weight": 193.40369417394206,
+ "blocks.8.ln.weight": 3.0697438474661696,
+ "blocks.8.w1.weight": 143.40991382362168,
+ "blocks.8.w1.bias": 132.0178424400649,
+ "blocks.8.w2.weight": 126.31745589603334,
+ "blocks.9.ln.weight": 3.3171246104472094,
+ "blocks.9.w1.weight": 150.22737087992024,
+ "blocks.9.w1.bias": 147.1140425995551,
+ "blocks.9.w2.weight": 118.0665737085708,
+ "blocks.10.ln.weight": 6.064775829759772,
+ "blocks.10.w1.weight": 281.65394416567705,
+ "blocks.10.w1.bias": 267.43803892028814,
+ "blocks.10.w2.weight": 206.89673602536328,
+ "blocks.11.ln.weight": 2.0710718946854763,
+ "blocks.11.w1.weight": 93.44069995553728,
+ "blocks.11.w1.bias": 93.69984582471675,
+ "blocks.11.w2.weight": 110.16073882917532,
+ "out_ln.weight": 0.4517673564729089,
+ "out_head.weight": 5.403719578492724,
+ "out_head.bias": 2.04427795310877
+ }
+ },
+ "credit_bridge": {
+ "log": {
+ "train_loss": [
+ 2.06320516204834,
+ 2.0399101606750487,
+ 2.033929338531494,
+ 2.0309921591186524,
+ 2.0298601413726804,
+ 2.0217224433135987,
+ 2.024232417259216,
+ 2.019040475921631,
+ 2.0207128286361695,
+ 2.019753430519104,
+ 2.016178896408081,
+ 2.011948891983032,
+ 2.0130044828796385,
+ 2.0113372691345215,
+ 2.012964423980713,
+ 2.0107208766937257,
+ 2.0117168256378175,
+ 2.0097561017227172,
+ 2.0089866023254395,
+ 2.009570707740784,
+ 2.011733226318359,
+ 2.007577508773804,
+ 2.0085436083602906,
+ 2.0043470149230957,
+ 2.0070890647888184,
+ 2.0043670698547364,
+ 2.0048324042129515,
+ 2.0042362908935547,
+ 2.0050302404403686,
+ 2.004891293258667,
+ 2.005185372543335,
+ 2.003529516067505,
+ 2.0045382270050047,
+ 2.005508458251953,
+ 2.00552710067749,
+ 2.002495421791077,
+ 1.9999042555999755,
+ 1.998048825454712,
+ 1.9961405738067628,
+ 1.989240593185425,
+ 1.9859777431488037,
+ 1.9825943405532838,
+ 1.9781343281555175,
+ 1.9761270376968383,
+ 1.9706906524658203,
+ 1.9723773523712158,
+ 1.9700259783172607,
+ 1.9686059323883056,
+ 1.9655450037384032,
+ 1.965288708114624,
+ 1.9665511059570313,
+ 1.9662485723114014,
+ 1.9659714967346191,
+ 1.9650201892852783,
+ 1.962706196937561,
+ 1.9626981174468994,
+ 1.962720913772583,
+ 1.9627574170684814,
+ 1.9608846031951905,
+ 1.9610804913711548,
+ 1.9616359865570068,
+ 1.9621461741638184,
+ 1.9603107889556886,
+ 1.9571026586914062,
+ 1.9596751629257203,
+ 1.9583080463409424,
+ 1.9566787594604491,
+ 1.9575193979644776,
+ 1.9559337323760986,
+ 1.9559677182006836,
+ 1.9548421646499634,
+ 1.9558928606414796,
+ 1.9595830995941161,
+ 1.9542909908676147,
+ 1.9583268686676025,
+ 1.9573874398040771,
+ 1.9573734212875367,
+ 1.956038475189209,
+ 1.9572998455047608,
+ 1.95643936958313,
+ 1.9555049303436278,
+ 1.9551573079681397,
+ 1.9559287934112548,
+ 1.9554652740859986,
+ 1.9546309524536132,
+ 1.9547708898925782,
+ 1.9570305030059814,
+ 1.9546662924194336,
+ 1.953575323562622,
+ 1.9553636624526978,
+ 1.954565380935669,
+ 1.9541568119812012,
+ 1.9543512627410888,
+ 1.954592315673828,
+ 1.9545559118652345,
+ 1.9531384017562867,
+ 1.9550698448944093,
+ 1.955762207107544,
+ 1.9552061297988892,
+ 1.957028095779419
+ ],
+ "train_acc": [
+ 0.23984,
+ 0.24842,
+ 0.2523,
+ 0.25414,
+ 0.25272,
+ 0.25882,
+ 0.257,
+ 0.26044,
+ 0.25976,
+ 0.26,
+ 0.26116,
+ 0.2649,
+ 0.26124,
+ 0.2642,
+ 0.26532,
+ 0.26592,
+ 0.26512,
+ 0.26658,
+ 0.26512,
+ 0.2671,
+ 0.2687,
+ 0.26848,
+ 0.26662,
+ 0.26946,
+ 0.26794,
+ 0.26952,
+ 0.26786,
+ 0.26836,
+ 0.26706,
+ 0.27072,
+ 0.26694,
+ 0.26872,
+ 0.2661,
+ 0.26742,
+ 0.26802,
+ 0.26896,
+ 0.26998,
+ 0.27168,
+ 0.2705,
+ 0.27562,
+ 0.27362,
+ 0.27702,
+ 0.27414,
+ 0.2807,
+ 0.28288,
+ 0.2806,
+ 0.28036,
+ 0.28098,
+ 0.28132,
+ 0.28288,
+ 0.28464,
+ 0.28428,
+ 0.28284,
+ 0.28638,
+ 0.28584,
+ 0.28538,
+ 0.28668,
+ 0.28688,
+ 0.28718,
+ 0.2852,
+ 0.28592,
+ 0.28522,
+ 0.28774,
+ 0.29002,
+ 0.28946,
+ 0.28842,
+ 0.28964,
+ 0.29052,
+ 0.29124,
+ 0.29164,
+ 0.29108,
+ 0.29114,
+ 0.28798,
+ 0.28992,
+ 0.28808,
+ 0.29178,
+ 0.28762,
+ 0.28974,
+ 0.29016,
+ 0.28958,
+ 0.29256,
+ 0.29012,
+ 0.29266,
+ 0.29038,
+ 0.2914,
+ 0.2924,
+ 0.2919,
+ 0.2951,
+ 0.2932,
+ 0.29244,
+ 0.29184,
+ 0.2924,
+ 0.29446,
+ 0.2932,
+ 0.29322,
+ 0.29452,
+ 0.29264,
+ 0.29084,
+ 0.29366,
+ 0.2917
+ ],
+ "test_acc": [
+ 0.2379,
+ 0.2678,
+ 0.2756,
+ 0.2987,
+ 0.2573,
+ 0.2947,
+ 0.2621,
+ 0.2983,
+ 0.2971,
+ 0.2616,
+ 0.278,
+ 0.2983,
+ 0.2803,
+ 0.2915,
+ 0.2714,
+ 0.2835,
+ 0.2837,
+ 0.2794,
+ 0.2915,
+ 0.2982,
+ 0.2854,
+ 0.2997,
+ 0.3019,
+ 0.2841,
+ 0.2873,
+ 0.2874,
+ 0.2676,
+ 0.3023,
+ 0.3033,
+ 0.2965,
+ 0.2888,
+ 0.279,
+ 0.3008,
+ 0.2752,
+ 0.2953,
+ 0.3039,
+ 0.3075,
+ 0.3057,
+ 0.2836,
+ 0.2859,
+ 0.303,
+ 0.2871,
+ 0.3052,
+ 0.3119,
+ 0.2982,
+ 0.2954,
+ 0.3026,
+ 0.303,
+ 0.3007,
+ 0.3087,
+ 0.3026,
+ 0.301,
+ 0.3082,
+ 0.3071,
+ 0.3137,
+ 0.3006,
+ 0.3102,
+ 0.3033,
+ 0.3079,
+ 0.3136,
+ 0.3092,
+ 0.3083,
+ 0.3181,
+ 0.3098,
+ 0.312,
+ 0.312,
+ 0.3052,
+ 0.3092,
+ 0.3124,
+ 0.3069,
+ 0.3038,
+ 0.3079,
+ 0.3117,
+ 0.3054,
+ 0.3065,
+ 0.3056,
+ 0.3086,
+ 0.3103,
+ 0.3038,
+ 0.3132,
+ 0.3157,
+ 0.3131,
+ 0.3092,
+ 0.313,
+ 0.3106,
+ 0.3118,
+ 0.3085,
+ 0.3115,
+ 0.3113,
+ 0.3129,
+ 0.3098,
+ 0.3146,
+ 0.3102,
+ 0.3109,
+ 0.3104,
+ 0.3117,
+ 0.3113,
+ 0.3106,
+ 0.3104,
+ 0.3104
+ ],
+ "value_loss": [
+ 0.4835468523311615,
+ 0.17319115218162537,
+ 0.12466768116235732,
+ 0.1044780854511261,
+ 0.11361225022792816,
+ 0.09506864285826683,
+ 0.08982749298095703,
+ 0.06943095141410828,
+ 0.06597960061311722,
+ 0.0689908292388916,
+ 0.05427408220410347,
+ 0.048282792123556136,
+ 0.04676814959049225,
+ 0.04863041600465774,
+ 0.0471925854742527,
+ 0.043705004994869234,
+ 0.0460531382226944,
+ 0.03692107924818993,
+ 0.03888409217596054,
+ 0.03788371595621109,
+ 0.03733079287290573,
+ 0.0377624527490139,
+ 0.036395574208498004,
+ 0.03193316010415554,
+ 0.03243833178460598,
+ 0.027760677633285522,
+ 0.029560993740558626,
+ 0.029699042382240297,
+ 0.02779862470269203,
+ 0.02897036738872528,
+ 0.025522839539051056,
+ 0.02311282460749149,
+ 0.020680676515102386,
+ 0.023304677747488022,
+ 0.022652746698856353,
+ 0.019273730003237725,
+ 0.018258913558125495,
+ 0.02010319753885269,
+ 0.01653825303465128,
+ 0.02135470446884632,
+ 0.018013720120489598,
+ 0.016730768181085588,
+ 0.01824077806353569,
+ 0.013809284281134605,
+ 0.01404231985926628,
+ 0.016477530472278595,
+ 0.01449635619878769,
+ 0.013006495736837388,
+ 0.013415186036229133,
+ 0.011523299672305583,
+ 0.01087947401329875,
+ 0.0120885151720047,
+ 0.011659734555482864,
+ 0.009783919822871684,
+ 0.009495561037063598,
+ 0.01004114615738392,
+ 0.009555040076076984,
+ 0.009653655390739441,
+ 0.009292597230225802,
+ 0.009270159545093774,
+ 0.008550122949182988,
+ 0.00753715006917715,
+ 0.007719495616853237,
+ 0.006912159933894873,
+ 0.006747392172813415,
+ 0.007531150514781475,
+ 0.006493680753260851,
+ 0.006225158845484257,
+ 0.006196837233901024,
+ 0.006545155131220818,
+ 0.005813424973487854,
+ 0.0052813203772902485,
+ 0.0056783805817365645,
+ 0.005296767835021019,
+ 0.005148515439927578,
+ 0.004331958843171596,
+ 0.00367335773833096,
+ 0.003545896329060197,
+ 0.0034597759203612802,
+ 0.003181714376732707,
+ 0.0032860505221784116,
+ 0.002376726396679878,
+ 0.002331429560147226,
+ 0.002127379362359643,
+ 0.0018355507389456033,
+ 0.0019746782282367347,
+ 0.0013750521056354046,
+ 0.0016369795854389668,
+ 0.0013883132124319672,
+ 0.0014942838795110583,
+ 0.0011334093083441258,
+ 0.0013581191193312407,
+ 0.0012651530192419886,
+ 0.0013948661609459669,
+ 0.00119701892176643,
+ 0.0009705750445276499,
+ 0.0009873379099927843,
+ 0.0013789990733936428,
+ 0.0015264763329923153,
+ 0.0007223523693159222
+ ]
+ },
+ "diagnostics": {
+ "bp_cosine": [
+ 0.442160040140152,
+ 0.011405838653445244,
+ 0.021783415228128433,
+ 0.02257826365530491,
+ 0.022579727694392204,
+ 0.022490674629807472,
+ 0.022444019094109535,
+ 0.02243717387318611,
+ 0.02238454297184944,
+ 0.021799881011247635,
+ 0.021308548748493195,
+ 0.021358368918299675
+ ],
+ "perturbation_rho": [
+ -0.0024002324789762497,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "nudging": {
+ "0.001": [
+ -9.359791874885559e-07,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "0.003": [
+ -3.080349415540695e-06,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "0.01": [
+ -1.0516028851270676e-05,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ]
+ }
+ },
+ "drift": {
+ "embed.weight": 185.7504766018605,
+ "embed.bias": 113.97655387234977,
+ "blocks.0.ln.weight": 8.006394857140304,
+ "blocks.0.w1.weight": 201.9426375330318,
+ "blocks.0.w1.bias": 163.53267733285352,
+ "blocks.0.w2.weight": 338.84776840881347,
+ "blocks.1.ln.weight": 6.295072957460427,
+ "blocks.1.w1.weight": 219.06553999907572,
+ "blocks.1.w1.bias": 226.32772829898,
+ "blocks.1.w2.weight": 242.1925329344591,
+ "blocks.2.ln.weight": 7.320912432081406,
+ "blocks.2.w1.weight": 311.3035934820304,
+ "blocks.2.w1.bias": 298.6852088590304,
+ "blocks.2.w2.weight": 316.80472460197296,
+ "blocks.3.ln.weight": 4.940329586675122,
+ "blocks.3.w1.weight": 172.30509352305455,
+ "blocks.3.w1.bias": 152.69976843471107,
+ "blocks.3.w2.weight": 171.54613400302898,
+ "blocks.4.ln.weight": 5.398528552340865,
+ "blocks.4.w1.weight": 194.3066729117384,
+ "blocks.4.w1.bias": 177.2111282722944,
+ "blocks.4.w2.weight": 181.75464322675703,
+ "blocks.5.ln.weight": 5.0186360723250045,
+ "blocks.5.w1.weight": 177.28809608368013,
+ "blocks.5.w1.bias": 166.1128575027424,
+ "blocks.5.w2.weight": 177.87652026549083,
+ "blocks.6.ln.weight": 5.11718383519975,
+ "blocks.6.w1.weight": 175.68155404770292,
+ "blocks.6.w1.bias": 165.18102396382997,
+ "blocks.6.w2.weight": 188.26739446960036,
+ "blocks.7.ln.weight": 5.749714981932067,
+ "blocks.7.w1.weight": 211.62845167555633,
+ "blocks.7.w1.bias": 194.987439334531,
+ "blocks.7.w2.weight": 197.40821780414132,
+ "blocks.8.ln.weight": 7.020310313763905,
+ "blocks.8.w1.weight": 281.07720524059584,
+ "blocks.8.w1.bias": 259.36087916311686,
+ "blocks.8.w2.weight": 264.53573649142305,
+ "blocks.9.ln.weight": 7.277383166194383,
+ "blocks.9.w1.weight": 285.7959615127006,
+ "blocks.9.w1.bias": 261.89392533705353,
+ "blocks.9.w2.weight": 274.12202590162974,
+ "blocks.10.ln.weight": 4.304915932630116,
+ "blocks.10.w1.weight": 146.38313888426148,
+ "blocks.10.w1.bias": 134.29068175155572,
+ "blocks.10.w2.weight": 146.4223780813692,
+ "blocks.11.ln.weight": 6.661402581393363,
+ "blocks.11.w1.weight": 265.5729059266825,
+ "blocks.11.w1.bias": 247.27216580071305,
+ "blocks.11.w2.weight": 242.98119602092984,
+ "out_ln.weight": 0.5446617517258924,
+ "out_head.weight": 7.799632034433011,
+ "out_head.bias": 1.2214285543387151
+ }
+ }
+ },
+ "config": {
+ "dataset": "cifar10",
+ "d_hidden": 512,
+ "num_blocks": 12,
+ "batch_size": 128,
+ "epochs": 100,
+ "lr": 0.001,
+ "lr_fb": 0.001,
+ "wd": 0.01,
+ "lam": 0.1,
+ "K": 4,
+ "sigma_bridge": 0.05,
+ "ema_momentum": 0.995,
+ "term_grad_weight": 1.0,
+ "seeds": [
+ 42
+ ],
+ "gpu": 0,
+ "output_dir": "results/cifar10",
+ "num_classes": 10
+ }
+} \ No newline at end of file