summaryrefslogtreecommitdiff
path: root/results/cifar10_seed123
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_seed123
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_seed123')
-rw-r--r--results/cifar10_seed123/results_cifar10.json1985
1 files changed, 1985 insertions, 0 deletions
diff --git a/results/cifar10_seed123/results_cifar10.json b/results/cifar10_seed123/results_cifar10.json
new file mode 100644
index 0000000..fd8e585
--- /dev/null
+++ b/results/cifar10_seed123/results_cifar10.json
@@ -0,0 +1,1985 @@
+{
+ "123": {
+ "bp": {
+ "log": {
+ "train_loss": [
+ 1.9175124251556397,
+ 1.6847565829849243,
+ 1.5875056425857543,
+ 1.5320447119140626,
+ 1.4794927943801879,
+ 1.4405573056030274,
+ 1.4091039984893798,
+ 1.3804642078399658,
+ 1.3562547276306152,
+ 1.3386102033233642,
+ 1.3113672603607178,
+ 1.286414957962036,
+ 1.2713352434539795,
+ 1.2514976085662841,
+ 1.2332095288848877,
+ 1.2136467219924927,
+ 1.2016698885345458,
+ 1.1760129442214966,
+ 1.1615581885910033,
+ 1.1460976659965516,
+ 1.1286324452972412,
+ 1.1125253838348388,
+ 1.0944270223999024,
+ 1.0830238839530946,
+ 1.0608218252563477,
+ 1.0437702729034424,
+ 1.034650561027527,
+ 1.017014478340149,
+ 0.9988842744445801,
+ 0.9852388265800476,
+ 0.9635365026664734,
+ 0.9439176512908936,
+ 0.9298832908248902,
+ 0.9137922927856446,
+ 0.895711201210022,
+ 0.8814193804550171,
+ 0.8632927577590942,
+ 0.84351964012146,
+ 0.827347836894989,
+ 0.807900242767334,
+ 0.7877964169311523,
+ 0.7770450174331665,
+ 0.7534891272926331,
+ 0.7365407562255859,
+ 0.7147518105125428,
+ 0.6962322827911377,
+ 0.6742078248596192,
+ 0.6564239510726929,
+ 0.6403258143234253,
+ 0.6197586893653869,
+ 0.6014538905525207,
+ 0.583438274936676,
+ 0.5667518968963623,
+ 0.543176082239151,
+ 0.5258174272155761,
+ 0.5108640935707093,
+ 0.4897696580505371,
+ 0.47116634141922,
+ 0.4534935041809082,
+ 0.4354778845024109,
+ 0.4233958884239197,
+ 0.4017178301334381,
+ 0.3859938202762604,
+ 0.37423386340141296,
+ 0.365246470451355,
+ 0.3411486894798279,
+ 0.32471628838539124,
+ 0.31114961336135866,
+ 0.29552998408317566,
+ 0.28871005929946897,
+ 0.27408093184471133,
+ 0.26901549862384794,
+ 0.2596221783256531,
+ 0.24521093573570252,
+ 0.2299592537355423,
+ 0.22053928468227388,
+ 0.21593001784324645,
+ 0.1976856986284256,
+ 0.19874527307510376,
+ 0.19333338455200194,
+ 0.18259587281227113,
+ 0.17466963967323304,
+ 0.1689953342151642,
+ 0.16150579230308532,
+ 0.15866700973510742,
+ 0.15368373103141786,
+ 0.1482925960588455,
+ 0.145876254529953,
+ 0.14030234585762025,
+ 0.13952740331172944,
+ 0.1361180233001709,
+ 0.1323960340666771,
+ 0.12677059319972991,
+ 0.1287061265563965,
+ 0.12679034420013427,
+ 0.12853138402462005,
+ 0.12579439376831056,
+ 0.1280787334895134,
+ 0.12339243181705475,
+ 0.12544435278892518
+ ],
+ "train_acc": [
+ 0.31434,
+ 0.39008,
+ 0.42376,
+ 0.44654,
+ 0.46472,
+ 0.47872,
+ 0.49158,
+ 0.5009,
+ 0.51142,
+ 0.51944,
+ 0.52706,
+ 0.53622,
+ 0.54168,
+ 0.55052,
+ 0.5568,
+ 0.56322,
+ 0.56372,
+ 0.57526,
+ 0.58298,
+ 0.58918,
+ 0.59476,
+ 0.6007,
+ 0.6078,
+ 0.61058,
+ 0.61816,
+ 0.62692,
+ 0.62922,
+ 0.63428,
+ 0.64188,
+ 0.64736,
+ 0.6547,
+ 0.65824,
+ 0.6673,
+ 0.67046,
+ 0.67872,
+ 0.68352,
+ 0.68988,
+ 0.69742,
+ 0.70294,
+ 0.7099,
+ 0.71534,
+ 0.72286,
+ 0.72898,
+ 0.73446,
+ 0.74268,
+ 0.75076,
+ 0.75844,
+ 0.76398,
+ 0.76912,
+ 0.77878,
+ 0.78504,
+ 0.78992,
+ 0.79944,
+ 0.80564,
+ 0.81262,
+ 0.81906,
+ 0.82502,
+ 0.83208,
+ 0.83826,
+ 0.84468,
+ 0.8507,
+ 0.8589,
+ 0.86356,
+ 0.86656,
+ 0.87174,
+ 0.87936,
+ 0.88406,
+ 0.8908,
+ 0.89792,
+ 0.89858,
+ 0.90396,
+ 0.90662,
+ 0.90894,
+ 0.91456,
+ 0.9191,
+ 0.92516,
+ 0.92578,
+ 0.93306,
+ 0.93206,
+ 0.93384,
+ 0.9388,
+ 0.94064,
+ 0.94322,
+ 0.94482,
+ 0.9463,
+ 0.94904,
+ 0.94962,
+ 0.95136,
+ 0.95286,
+ 0.9532,
+ 0.95498,
+ 0.95586,
+ 0.9589,
+ 0.95728,
+ 0.95824,
+ 0.95676,
+ 0.95896,
+ 0.95834,
+ 0.95978,
+ 0.95964
+ ],
+ "test_acc": [
+ 0.3978,
+ 0.4369,
+ 0.4561,
+ 0.4846,
+ 0.4943,
+ 0.5129,
+ 0.5144,
+ 0.5192,
+ 0.5235,
+ 0.5332,
+ 0.53,
+ 0.5382,
+ 0.5478,
+ 0.5507,
+ 0.5534,
+ 0.5638,
+ 0.5725,
+ 0.5626,
+ 0.572,
+ 0.5735,
+ 0.5778,
+ 0.5778,
+ 0.583,
+ 0.5789,
+ 0.5885,
+ 0.589,
+ 0.5849,
+ 0.5957,
+ 0.5916,
+ 0.5872,
+ 0.5864,
+ 0.589,
+ 0.599,
+ 0.6012,
+ 0.5912,
+ 0.5967,
+ 0.5973,
+ 0.5923,
+ 0.5983,
+ 0.5947,
+ 0.5966,
+ 0.5996,
+ 0.6007,
+ 0.5935,
+ 0.6005,
+ 0.6016,
+ 0.5972,
+ 0.5976,
+ 0.596,
+ 0.5938,
+ 0.5933,
+ 0.598,
+ 0.5987,
+ 0.5998,
+ 0.5914,
+ 0.5916,
+ 0.5988,
+ 0.5984,
+ 0.5936,
+ 0.6002,
+ 0.5917,
+ 0.5921,
+ 0.5944,
+ 0.5968,
+ 0.5925,
+ 0.5974,
+ 0.5946,
+ 0.5979,
+ 0.5923,
+ 0.5956,
+ 0.5968,
+ 0.5956,
+ 0.5974,
+ 0.5934,
+ 0.5943,
+ 0.5954,
+ 0.5982,
+ 0.5923,
+ 0.5921,
+ 0.5937,
+ 0.5937,
+ 0.5941,
+ 0.5921,
+ 0.5957,
+ 0.5953,
+ 0.5939,
+ 0.5939,
+ 0.5965,
+ 0.5952,
+ 0.5957,
+ 0.5951,
+ 0.5976,
+ 0.5967,
+ 0.5968,
+ 0.5974,
+ 0.5965,
+ 0.5973,
+ 0.596,
+ 0.5961,
+ 0.5962
+ ]
+ },
+ "diagnostics": {
+ "bp_cosine": [
+ 0.9620110392570496,
+ 0.9621291160583496,
+ 0.9621908664703369,
+ 0.9623005986213684,
+ 0.9625044465065002,
+ 0.9625605344772339,
+ 0.9631243348121643,
+ 0.9631022214889526,
+ 0.9631184935569763,
+ 0.9626430869102478,
+ 0.9622764587402344,
+ 0.9619318246841431
+ ],
+ "perturbation_rho": [
+ 0.9879884719848633,
+ 0.9869263172149658,
+ 0.9891279935836792,
+ 0.9888099431991577,
+ 0.991609513759613,
+ 0.9902887344360352,
+ 0.9909973740577698,
+ 0.9915982484817505,
+ 0.9906468987464905,
+ 0.9890569448471069,
+ 0.9870684146881104,
+ 0.9818294644355774
+ ],
+ "nudging": {
+ "0.001": [
+ -0.003427116898819804,
+ -0.003351739142090082,
+ -0.0033903690055012703,
+ -0.003365215379744768,
+ -0.0033891985658556223,
+ -0.003362787188962102,
+ -0.0032895321492105722,
+ -0.0031716772355139256,
+ -0.0029750647954642773,
+ -0.002764828037470579,
+ -0.0024368567392230034,
+ -0.0021113622933626175
+ ],
+ "0.003": [
+ -0.010271656326949596,
+ -0.010045798495411873,
+ -0.010160606354475021,
+ -0.010086143389344215,
+ -0.01015765406191349,
+ -0.010078245773911476,
+ -0.009858867153525352,
+ -0.009506190195679665,
+ -0.008917586877942085,
+ -0.008288294076919556,
+ -0.007305148057639599,
+ -0.006330306641757488
+ ],
+ "0.01": [
+ -0.03411904349923134,
+ -0.033372972160577774,
+ -0.033751748502254486,
+ -0.03350646048784256,
+ -0.0337451696395874,
+ -0.03347958251833916,
+ -0.03275400772690773,
+ -0.03158621862530708,
+ -0.029637902975082397,
+ -0.02755369432270527,
+ -0.024292215704917908,
+ -0.02105848491191864
+ ]
+ }
+ },
+ "drift": {
+ "embed.weight": 8.763637326480993,
+ "embed.bias": 15.277378973045789,
+ "blocks.0.ln.weight": 0.34908970813559786,
+ "blocks.0.w1.weight": 4.164846741111791,
+ "blocks.0.w1.bias": 6.735275454452179,
+ "blocks.0.w2.weight": 10.512885267241659,
+ "blocks.1.ln.weight": 0.2548678557191968,
+ "blocks.1.w1.weight": 4.428344802990091,
+ "blocks.1.w1.bias": 4.431127174572938,
+ "blocks.1.w2.weight": 11.609675967743058,
+ "blocks.2.ln.weight": 0.20993716494835368,
+ "blocks.2.w1.weight": 4.563301300136555,
+ "blocks.2.w1.bias": 4.053222361596439,
+ "blocks.2.w2.weight": 12.283684374504883,
+ "blocks.3.ln.weight": 0.22639993499269975,
+ "blocks.3.w1.weight": 4.586109188420391,
+ "blocks.3.w1.bias": 3.9811075138013527,
+ "blocks.3.w2.weight": 12.651716892358182,
+ "blocks.4.ln.weight": 0.23813753747592467,
+ "blocks.4.w1.weight": 4.55512749423403,
+ "blocks.4.w1.bias": 3.7036288780635465,
+ "blocks.4.w2.weight": 12.900508521442045,
+ "blocks.5.ln.weight": 0.24851322774648468,
+ "blocks.5.w1.weight": 4.519407405633592,
+ "blocks.5.w1.bias": 3.7828748042863825,
+ "blocks.5.w2.weight": 12.877614228932648,
+ "blocks.6.ln.weight": 0.24789524956961745,
+ "blocks.6.w1.weight": 4.510086973031788,
+ "blocks.6.w1.bias": 3.7520206765756337,
+ "blocks.6.w2.weight": 13.18270185196777,
+ "blocks.7.ln.weight": 0.27277168975906585,
+ "blocks.7.w1.weight": 4.471720887903336,
+ "blocks.7.w1.bias": 3.7457909142015353,
+ "blocks.7.w2.weight": 13.30644187374482,
+ "blocks.8.ln.weight": 0.29086480529883674,
+ "blocks.8.w1.weight": 4.434570868617996,
+ "blocks.8.w1.bias": 3.7342218578126882,
+ "blocks.8.w2.weight": 13.49877934675731,
+ "blocks.9.ln.weight": 0.2919660181755758,
+ "blocks.9.w1.weight": 4.429555710823986,
+ "blocks.9.w1.bias": 3.634447923096686,
+ "blocks.9.w2.weight": 13.75719577660119,
+ "blocks.10.ln.weight": 0.32005145624529063,
+ "blocks.10.w1.weight": 4.353451334572149,
+ "blocks.10.w1.bias": 3.4614471694540865,
+ "blocks.10.w2.weight": 13.887069102514094,
+ "blocks.11.ln.weight": 0.29895213239033414,
+ "blocks.11.w1.weight": 4.374616657143753,
+ "blocks.11.w1.bias": 3.645530912656648,
+ "blocks.11.w2.weight": 14.561155498919536,
+ "out_ln.weight": 0.283104664197929,
+ "out_head.weight": 3.694886816637194,
+ "out_head.bias": 1.105623778518022
+ }
+ },
+ "dfa": {
+ "log": {
+ "train_loss": [
+ 2.0684899793243408,
+ 2.0445925352478027,
+ 2.03087772151947,
+ 2.027939418258667,
+ 2.0253660607528685,
+ 2.0250345896148683,
+ 2.022372142868042,
+ 2.019951826019287,
+ 2.017632607574463,
+ 2.015866649131775,
+ 2.017579344139099,
+ 2.01485851020813,
+ 2.012323605117798,
+ 2.009420676612854,
+ 2.0132182874298095,
+ 2.007821100616455,
+ 2.0097865366363528,
+ 2.005040230484009,
+ 2.004450204849243,
+ 2.0059849935150145,
+ 2.005384602432251,
+ 2.001610414581299,
+ 2.002629147567749,
+ 2.0022453118896486,
+ 1.9994831539154052,
+ 1.9995977224349974,
+ 2.0003068398284913,
+ 2.0012258628082273,
+ 1.9992249011611938,
+ 1.9970151906967164,
+ 1.9978694770812988,
+ 1.997133782272339,
+ 1.9935067112731935,
+ 1.9952505539321899,
+ 1.992918910446167,
+ 1.9958944741058349,
+ 1.9945134730529785,
+ 1.993255687599182,
+ 1.9926469972991943,
+ 1.9950547982406617,
+ 1.9904610315322877,
+ 1.9907950922393798,
+ 1.991843765411377,
+ 1.9908243858337402,
+ 1.9923807875823976,
+ 1.9917906607055664,
+ 1.992043903465271,
+ 1.9904995797348022,
+ 1.9916813136291505,
+ 1.98937647026062,
+ 1.9889434642791748,
+ 1.9899712393188476,
+ 1.9896988627624512,
+ 1.9895934002304076,
+ 1.9901937147521973,
+ 1.9906840317535401,
+ 1.9852267303466797,
+ 1.989434090652466,
+ 1.9889313174438477,
+ 1.9886722135925292,
+ 1.9877896743392944,
+ 1.9861792268371583,
+ 1.986029208984375,
+ 1.9876557461547852,
+ 1.985245605545044,
+ 1.9858075296401978,
+ 1.9867199410247802,
+ 1.9870598261260985,
+ 1.9854085816955567,
+ 1.9867712688827515,
+ 1.9855045788955688,
+ 1.9881172052764893,
+ 1.9837462771987915,
+ 1.98582817527771,
+ 1.9842846584320069,
+ 1.9848911337280273,
+ 1.9850436270904541,
+ 1.9835600254821777,
+ 1.9840570623779297,
+ 1.9839281851196289,
+ 1.984888575820923,
+ 1.9857918256378173,
+ 1.9845330112075805,
+ 1.984420718612671,
+ 1.9853396697998047,
+ 1.9839463347625732,
+ 1.9833207806396485,
+ 1.9841535271453858,
+ 1.9810106130218506,
+ 1.9846471352386474,
+ 1.9849586167144775,
+ 1.984921664199829,
+ 1.984304642715454,
+ 1.9863978981781005,
+ 1.9833829919052124,
+ 1.984193635787964,
+ 1.983270346107483,
+ 1.9827691938018799,
+ 1.9846319453430177,
+ 1.9829401163101197
+ ],
+ "train_acc": [
+ 0.2366,
+ 0.24504,
+ 0.25212,
+ 0.25468,
+ 0.25604,
+ 0.2569,
+ 0.25714,
+ 0.2588,
+ 0.25558,
+ 0.2631,
+ 0.2608,
+ 0.26104,
+ 0.26492,
+ 0.26478,
+ 0.26264,
+ 0.26596,
+ 0.26544,
+ 0.267,
+ 0.26848,
+ 0.26756,
+ 0.26702,
+ 0.26954,
+ 0.27278,
+ 0.26886,
+ 0.26902,
+ 0.27146,
+ 0.272,
+ 0.2698,
+ 0.2728,
+ 0.27384,
+ 0.2713,
+ 0.27288,
+ 0.27422,
+ 0.2767,
+ 0.2742,
+ 0.2748,
+ 0.27392,
+ 0.27568,
+ 0.27816,
+ 0.27558,
+ 0.27752,
+ 0.27658,
+ 0.27714,
+ 0.277,
+ 0.27766,
+ 0.27532,
+ 0.27622,
+ 0.27806,
+ 0.27688,
+ 0.27908,
+ 0.27896,
+ 0.27908,
+ 0.27972,
+ 0.28032,
+ 0.27848,
+ 0.27914,
+ 0.2816,
+ 0.27856,
+ 0.27972,
+ 0.28028,
+ 0.28068,
+ 0.27928,
+ 0.28124,
+ 0.2814,
+ 0.28118,
+ 0.28066,
+ 0.2815,
+ 0.28218,
+ 0.28352,
+ 0.2822,
+ 0.28114,
+ 0.28234,
+ 0.28204,
+ 0.28104,
+ 0.28,
+ 0.2824,
+ 0.28114,
+ 0.28372,
+ 0.282,
+ 0.28204,
+ 0.28208,
+ 0.28162,
+ 0.28212,
+ 0.28526,
+ 0.28178,
+ 0.28234,
+ 0.28358,
+ 0.28328,
+ 0.28346,
+ 0.28396,
+ 0.28374,
+ 0.28334,
+ 0.28204,
+ 0.28406,
+ 0.28382,
+ 0.28466,
+ 0.28292,
+ 0.28436,
+ 0.28206,
+ 0.28276
+ ],
+ "test_acc": [
+ 0.2589,
+ 0.2617,
+ 0.2631,
+ 0.2816,
+ 0.2556,
+ 0.2607,
+ 0.2876,
+ 0.2818,
+ 0.2763,
+ 0.2847,
+ 0.3005,
+ 0.2738,
+ 0.283,
+ 0.2729,
+ 0.2859,
+ 0.2782,
+ 0.3055,
+ 0.2874,
+ 0.2796,
+ 0.3027,
+ 0.2977,
+ 0.3013,
+ 0.2869,
+ 0.2663,
+ 0.2917,
+ 0.2707,
+ 0.2784,
+ 0.3006,
+ 0.2812,
+ 0.2801,
+ 0.2755,
+ 0.2935,
+ 0.3091,
+ 0.3041,
+ 0.2967,
+ 0.3054,
+ 0.3107,
+ 0.2963,
+ 0.2956,
+ 0.2927,
+ 0.2941,
+ 0.2943,
+ 0.2948,
+ 0.3057,
+ 0.3015,
+ 0.3116,
+ 0.2928,
+ 0.301,
+ 0.288,
+ 0.296,
+ 0.2909,
+ 0.2957,
+ 0.3145,
+ 0.3124,
+ 0.2994,
+ 0.307,
+ 0.3047,
+ 0.3026,
+ 0.2989,
+ 0.2978,
+ 0.3023,
+ 0.2971,
+ 0.3062,
+ 0.3025,
+ 0.2919,
+ 0.309,
+ 0.301,
+ 0.2947,
+ 0.3086,
+ 0.2976,
+ 0.3106,
+ 0.3013,
+ 0.3109,
+ 0.3012,
+ 0.3019,
+ 0.2969,
+ 0.3022,
+ 0.3038,
+ 0.3035,
+ 0.3087,
+ 0.2972,
+ 0.2974,
+ 0.2952,
+ 0.3032,
+ 0.3034,
+ 0.3027,
+ 0.3024,
+ 0.3055,
+ 0.3031,
+ 0.3003,
+ 0.3044,
+ 0.3046,
+ 0.3032,
+ 0.3029,
+ 0.303,
+ 0.3028,
+ 0.3038,
+ 0.3032,
+ 0.3035,
+ 0.3034
+ ]
+ },
+ "diagnostics": {
+ "bp_cosine": [
+ 0.40911683440208435,
+ 0.0002619648876134306,
+ -0.00015980367606971413,
+ 0.0005440536187961698,
+ 0.0002772972220554948,
+ 7.781770545989275e-05,
+ -0.00020515912910923362,
+ -0.00044000157504342496,
+ -0.0008366729598492384,
+ 0.00015155941946431994,
+ 0.00032340025063604116,
+ -0.00015023337618913502
+ ],
+ "perturbation_rho": [
+ 0.0528201088309288,
+ 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": [
+ -4.898756742477417e-07,
+ 1.862645149230957e-09,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ -1.862645149230957e-09,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "0.003": [
+ -1.2246891856193542e-06,
+ 1.862645149230957e-09,
+ -9.313225746154785e-10,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ -1.862645149230957e-09,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "0.01": [
+ -4.1155144572257996e-06,
+ -1.862645149230957e-09,
+ -9.313225746154785e-10,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ -2.7939677238464355e-09,
+ 0.0,
+ -9.313225746154785e-10,
+ 0.0
+ ]
+ }
+ },
+ "drift": {
+ "embed.weight": 323.8982252804048,
+ "embed.bias": 232.81638531211948,
+ "blocks.0.ln.weight": 10.852100889778994,
+ "blocks.0.w1.weight": 272.36679039121384,
+ "blocks.0.w1.bias": 238.87995630640614,
+ "blocks.0.w2.weight": 445.2406120934057,
+ "blocks.1.ln.weight": 9.73650143893558,
+ "blocks.1.w1.weight": 390.9712215089673,
+ "blocks.1.w1.bias": 374.17715191606453,
+ "blocks.1.w2.weight": 403.2488589095519,
+ "blocks.2.ln.weight": 6.710448026907898,
+ "blocks.2.w1.weight": 227.28731670207446,
+ "blocks.2.w1.bias": 202.7493867156122,
+ "blocks.2.w2.weight": 223.99250691076898,
+ "blocks.3.ln.weight": 7.215269500019514,
+ "blocks.3.w1.weight": 231.09319791624182,
+ "blocks.3.w1.bias": 206.9745974350102,
+ "blocks.3.w2.weight": 238.74614111179253,
+ "blocks.4.ln.weight": 9.257100922232066,
+ "blocks.4.w1.weight": 374.01823924044805,
+ "blocks.4.w1.bias": 343.71540244431617,
+ "blocks.4.w2.weight": 350.8111852358107,
+ "blocks.5.ln.weight": 6.914568252312619,
+ "blocks.5.w1.weight": 266.9118711467473,
+ "blocks.5.w1.bias": 256.9588103323234,
+ "blocks.5.w2.weight": 254.31906752172955,
+ "blocks.6.ln.weight": 8.82549425820829,
+ "blocks.6.w1.weight": 354.28665902336076,
+ "blocks.6.w1.bias": 332.44091471694884,
+ "blocks.6.w2.weight": 326.29279154408147,
+ "blocks.7.ln.weight": 10.262108389389345,
+ "blocks.7.w1.weight": 411.01500781893094,
+ "blocks.7.w1.bias": 383.25852137267697,
+ "blocks.7.w2.weight": 398.26834704743106,
+ "blocks.8.ln.weight": 9.486443170916516,
+ "blocks.8.w1.weight": 379.7970569993528,
+ "blocks.8.w1.bias": 353.44640446436694,
+ "blocks.8.w2.weight": 354.75408312717866,
+ "blocks.9.ln.weight": 7.225265383958218,
+ "blocks.9.w1.weight": 280.98274227898526,
+ "blocks.9.w1.bias": 264.21754537134655,
+ "blocks.9.w2.weight": 250.42097452359232,
+ "blocks.10.ln.weight": 9.12118778587681,
+ "blocks.10.w1.weight": 363.4177719662398,
+ "blocks.10.w1.bias": 333.2020904512465,
+ "blocks.10.w2.weight": 343.55519901127,
+ "blocks.11.ln.weight": 9.146084772655259,
+ "blocks.11.w1.weight": 367.21625652465485,
+ "blocks.11.w1.bias": 343.39432049385726,
+ "blocks.11.w2.weight": 338.5639684764078,
+ "out_ln.weight": 0.5993398286123665,
+ "out_head.weight": 9.522011867407477,
+ "out_head.bias": 0.44134062139959057
+ }
+ },
+ "state_bridge": {
+ "log": {
+ "train_loss": [
+ 2.27535658454895,
+ 2.2637348059844973,
+ 2.23140262840271,
+ 2.13100928276062,
+ 2.1194422898101806,
+ 2.113937710952759,
+ 2.123109260482788,
+ 2.123372416381836,
+ 2.118385257644653,
+ 2.123976169281006,
+ 2.143634383087158,
+ 2.15997088432312,
+ 2.1503283737182617,
+ 2.2361232402801514,
+ 2.31340713142395,
+ 2.312125205230713,
+ 2.3117453997802735,
+ 2.3094616239929198,
+ 2.3084144255065917,
+ 2.3107836222076417,
+ 2.3105288831329345,
+ 2.312015315246582,
+ 2.311362181777954,
+ 2.3104846384429933,
+ 2.3100794078826903,
+ 2.308620408859253,
+ 2.2244751792907715,
+ 2.1188239234924318,
+ 2.099667515487671,
+ 2.1012451152038576,
+ 2.101906224517822,
+ 2.1015794395446776,
+ 2.102679093170166,
+ 2.104404850845337,
+ 2.105138458328247,
+ 2.1025040253448486,
+ 2.09926293258667,
+ 2.0996690463256837,
+ 2.0955256491851806,
+ 2.0939024459075926,
+ 2.0932679664611817,
+ 2.0943772303771975,
+ 2.087596756668091,
+ 2.0909996266937254,
+ 2.087642866668701,
+ 2.089086704559326,
+ 2.0858034885406496,
+ 2.084843352355957,
+ 2.081013270111084,
+ 2.0763836125183106,
+ 2.068111530685425,
+ 2.0633422270965576,
+ 2.059932788619995,
+ 2.0572856521987917,
+ 2.0567763134765626,
+ 2.055301329345703,
+ 2.0559660177230836,
+ 2.0547456675720217,
+ 2.0552883127593993,
+ 2.052630202102661,
+ 2.050533991241455,
+ 2.0526472858428955,
+ 2.051843546676636,
+ 2.0524809063720704,
+ 2.052277666931152,
+ 2.049621379356384,
+ 2.0498316369628906,
+ 2.049651178665161,
+ 2.0472222846221926,
+ 2.0483968115997313,
+ 2.04511710357666,
+ 2.0473428889465333,
+ 2.0471922763061525,
+ 2.050198796348572,
+ 2.0515007726287844,
+ 2.051603550682068,
+ 2.0523804613494874,
+ 2.054355347518921,
+ 2.055121799697876,
+ 2.0543520391082764,
+ 2.0561765496063233,
+ 2.056192929840088,
+ 2.054548904418945,
+ 2.05620756149292,
+ 2.055558791809082,
+ 2.056474009552002,
+ 2.055143079071045,
+ 2.054688854446411,
+ 2.0569355255126953,
+ 2.0542632417297364,
+ 2.054791463470459,
+ 2.054189739532471,
+ 2.0533234220123293,
+ 2.0529643350982667,
+ 2.053456902542114,
+ 2.0533332372283937,
+ 2.0545412349700927,
+ 2.053972573928833,
+ 2.0536360316467284,
+ 2.053008867416382
+ ],
+ "train_acc": [
+ 0.13698,
+ 0.13998,
+ 0.1447,
+ 0.17058,
+ 0.17036,
+ 0.1725,
+ 0.1677,
+ 0.16766,
+ 0.1698,
+ 0.1709,
+ 0.16482,
+ 0.1638,
+ 0.16334,
+ 0.14304,
+ 0.10524,
+ 0.11206,
+ 0.10998,
+ 0.11288,
+ 0.11528,
+ 0.10822,
+ 0.10712,
+ 0.10462,
+ 0.1037,
+ 0.10238,
+ 0.10366,
+ 0.10334,
+ 0.15168,
+ 0.17212,
+ 0.17798,
+ 0.17688,
+ 0.17846,
+ 0.17562,
+ 0.17644,
+ 0.17718,
+ 0.175,
+ 0.17492,
+ 0.1766,
+ 0.17688,
+ 0.17674,
+ 0.18138,
+ 0.18304,
+ 0.17868,
+ 0.18296,
+ 0.18172,
+ 0.18454,
+ 0.18428,
+ 0.18226,
+ 0.18568,
+ 0.18754,
+ 0.19132,
+ 0.2004,
+ 0.20496,
+ 0.207,
+ 0.20808,
+ 0.2054,
+ 0.20596,
+ 0.20646,
+ 0.20616,
+ 0.20466,
+ 0.20608,
+ 0.20848,
+ 0.20616,
+ 0.20746,
+ 0.20588,
+ 0.20528,
+ 0.20778,
+ 0.20856,
+ 0.2078,
+ 0.2101,
+ 0.20888,
+ 0.21296,
+ 0.211,
+ 0.20934,
+ 0.20746,
+ 0.20674,
+ 0.20832,
+ 0.20954,
+ 0.20736,
+ 0.20522,
+ 0.20496,
+ 0.20606,
+ 0.20302,
+ 0.20594,
+ 0.20828,
+ 0.2064,
+ 0.20466,
+ 0.20378,
+ 0.20496,
+ 0.20534,
+ 0.20862,
+ 0.20546,
+ 0.20594,
+ 0.20688,
+ 0.20658,
+ 0.20712,
+ 0.20672,
+ 0.20568,
+ 0.20664,
+ 0.20646,
+ 0.20888
+ ],
+ "test_acc": [
+ 0.1514,
+ 0.1391,
+ 0.1828,
+ 0.1457,
+ 0.1765,
+ 0.186,
+ 0.1602,
+ 0.1723,
+ 0.1754,
+ 0.1714,
+ 0.1724,
+ 0.1634,
+ 0.1615,
+ 0.0959,
+ 0.1252,
+ 0.1087,
+ 0.135,
+ 0.1366,
+ 0.1028,
+ 0.1085,
+ 0.1257,
+ 0.114,
+ 0.119,
+ 0.1,
+ 0.0848,
+ 0.1194,
+ 0.1607,
+ 0.1795,
+ 0.1859,
+ 0.1635,
+ 0.1571,
+ 0.1682,
+ 0.1748,
+ 0.1717,
+ 0.1789,
+ 0.1658,
+ 0.1673,
+ 0.1783,
+ 0.1799,
+ 0.1761,
+ 0.1812,
+ 0.173,
+ 0.1716,
+ 0.164,
+ 0.1657,
+ 0.1769,
+ 0.1728,
+ 0.1726,
+ 0.1801,
+ 0.1758,
+ 0.1873,
+ 0.1862,
+ 0.1948,
+ 0.1935,
+ 0.2009,
+ 0.1933,
+ 0.2102,
+ 0.2094,
+ 0.2153,
+ 0.2097,
+ 0.2054,
+ 0.211,
+ 0.2109,
+ 0.1975,
+ 0.2107,
+ 0.2099,
+ 0.2109,
+ 0.202,
+ 0.2055,
+ 0.2103,
+ 0.2068,
+ 0.2068,
+ 0.2085,
+ 0.2085,
+ 0.206,
+ 0.203,
+ 0.1998,
+ 0.2021,
+ 0.2092,
+ 0.2039,
+ 0.2034,
+ 0.1976,
+ 0.2064,
+ 0.2098,
+ 0.2078,
+ 0.2045,
+ 0.2078,
+ 0.2055,
+ 0.2068,
+ 0.2065,
+ 0.2075,
+ 0.2084,
+ 0.206,
+ 0.2056,
+ 0.2061,
+ 0.2063,
+ 0.2066,
+ 0.2067,
+ 0.2068,
+ 0.2069
+ ],
+ "state_pred_error": [
+ 0.3455641258764267,
+ 0.1890164827811718,
+ 0.06542741327524185,
+ 0.15099643572807311,
+ 0.15767600553512573,
+ 0.14595494849205018,
+ 0.09619720283508301,
+ 0.05848943448066712,
+ 0.04476255656838417,
+ 0.03768930072784424,
+ 0.034250264310240745,
+ 0.02679455811023712,
+ 0.024154067809581757,
+ 0.020689564958810806,
+ 0.015427032306194306,
+ 0.01494347603648901,
+ 0.012853515672385693,
+ 0.00926094017356634,
+ 0.00883663154706359,
+ 0.00866620466351509,
+ 0.008630412181615829,
+ 0.009436789927184582,
+ 0.010587773686945438,
+ 0.006365314861536026,
+ 0.0058888214336335655,
+ 0.004059859301149845,
+ 0.0059570299243927,
+ 0.005231297912150622,
+ 0.004334697146564722,
+ 0.005451656972020865,
+ 0.005537090619504452,
+ 0.006261329069137574,
+ 0.006663460256308317,
+ 0.0068712751638889315,
+ 0.006620877954363823,
+ 0.0059924150297045705,
+ 0.004774981212168932,
+ 0.0035280540695041417,
+ 0.003671990767121315,
+ 0.0035207420617341997,
+ 0.0030331813983619214,
+ 0.0031412731628119943,
+ 0.0028461132158339023,
+ 0.002806654215082526,
+ 0.0028670129170268776,
+ 0.0030720302928984164,
+ 0.0032383741122484207,
+ 0.002952979260608554,
+ 0.002582101986259222,
+ 0.0024391376911848785,
+ 0.002423995024561882,
+ 0.002365549064427614,
+ 0.002312273128144443,
+ 0.0020981324925273656,
+ 0.0018509870688617229,
+ 0.001701984341070056,
+ 0.001557768930643797,
+ 0.001504055576622486,
+ 0.0015004488579928876,
+ 0.0016905955363065005,
+ 0.001851444918513298,
+ 0.002072783012315631,
+ 0.0021090806840360164,
+ 0.0020937418549507856,
+ 0.001958419643007219,
+ 0.0018501618901640177,
+ 0.0017675739562511444,
+ 0.001684609704464674,
+ 0.0016207635071873664,
+ 0.0015579476598650216,
+ 0.0014976035876572132,
+ 0.0014524117913842201,
+ 0.0013800759248435498,
+ 0.001322413182966411,
+ 0.0012982203678414226,
+ 0.0012594390138238667,
+ 0.0012120740934088827,
+ 0.0011591399506293237,
+ 0.001117924986295402,
+ 0.0010595228926092387,
+ 0.0010102490696310997,
+ 0.0009533731169998645,
+ 0.0009156427671574056,
+ 0.0008670865286141634,
+ 0.0008243481399863959,
+ 0.0007827390401065349,
+ 0.0007445751608535648,
+ 0.0007058267378434539,
+ 0.0006572319846227765,
+ 0.0006192173832282424,
+ 0.0005905027001723647,
+ 0.0005582181675359607,
+ 0.0005225332551077009,
+ 0.0004948109107278287,
+ 0.0004635604086332023,
+ 0.0004460584534611553,
+ 0.00041995075650513175,
+ 0.00039428566612303257,
+ 0.0003732312531303614,
+ 0.0003609425350651145
+ ]
+ },
+ "diagnostics": {
+ "bp_cosine": [
+ 0.13039325177669525,
+ -0.0001226231688633561,
+ -6.530806422233582e-05,
+ 0.0008607044583186507,
+ 0.00019595431513153017,
+ 0.0003264691913500428,
+ 0.0007734490791335702,
+ 0.00040342434658668935,
+ 0.0006691448506899178,
+ 0.0006843262817710638,
+ 0.0009011077927425504,
+ 0.0007364039774984121
+ ],
+ "perturbation_rho": [
+ 0.06105021387338638,
+ 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": [
+ -8.485279977321625e-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": [
+ -2.5329645723104477e-05,
+ 9.313225746154785e-09,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ],
+ "0.01": [
+ -8.417712524533272e-05,
+ -4.190951585769653e-09,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0,
+ 0.0
+ ]
+ }
+ },
+ "drift": {
+ "embed.weight": 18.244614908467394,
+ "embed.bias": 91.4065704680882,
+ "blocks.0.ln.weight": 2.7002334851153766,
+ "blocks.0.w1.weight": 134.61543991177393,
+ "blocks.0.w1.bias": 122.71752984768428,
+ "blocks.0.w2.weight": 276.02165008885817,
+ "blocks.1.ln.weight": 6.996325721977635,
+ "blocks.1.w1.weight": 316.6573873211569,
+ "blocks.1.w1.bias": 288.7266985364082,
+ "blocks.1.w2.weight": 232.96436903686606,
+ "blocks.2.ln.weight": 3.005190384482946,
+ "blocks.2.w1.weight": 113.28816941109099,
+ "blocks.2.w1.bias": 102.79639141280903,
+ "blocks.2.w2.weight": 154.79558550064107,
+ "blocks.3.ln.weight": 3.854659793583277,
+ "blocks.3.w1.weight": 136.45977626032953,
+ "blocks.3.w1.bias": 125.02248154351406,
+ "blocks.3.w2.weight": 160.12595576311708,
+ "blocks.4.ln.weight": 5.321574474789146,
+ "blocks.4.w1.weight": 200.03582097403066,
+ "blocks.4.w1.bias": 185.1705216854366,
+ "blocks.4.w2.weight": 197.8192680660012,
+ "blocks.5.ln.weight": 3.689939791539998,
+ "blocks.5.w1.weight": 121.91986650696485,
+ "blocks.5.w1.bias": 112.80564875251632,
+ "blocks.5.w2.weight": 175.23886752618034,
+ "blocks.6.ln.weight": 5.257637367329094,
+ "blocks.6.w1.weight": 159.5448554465857,
+ "blocks.6.w1.bias": 148.9053783966513,
+ "blocks.6.w2.weight": 146.5770282072606,
+ "blocks.7.ln.weight": 4.684853931243148,
+ "blocks.7.w1.weight": 133.95190141354294,
+ "blocks.7.w1.bias": 120.52600626182299,
+ "blocks.7.w2.weight": 137.015420718898,
+ "blocks.8.ln.weight": 4.553687541635819,
+ "blocks.8.w1.weight": 134.4598258206248,
+ "blocks.8.w1.bias": 124.54115675222299,
+ "blocks.8.w2.weight": 138.93702924685354,
+ "blocks.9.ln.weight": 4.188852950738847,
+ "blocks.9.w1.weight": 128.35707494719534,
+ "blocks.9.w1.bias": 120.498165463794,
+ "blocks.9.w2.weight": 149.134498305582,
+ "blocks.10.ln.weight": 3.3037616990170426,
+ "blocks.10.w1.weight": 108.25092049032806,
+ "blocks.10.w1.bias": 101.78620102240454,
+ "blocks.10.w2.weight": 127.47031512506157,
+ "blocks.11.ln.weight": 1.6054786857904342,
+ "blocks.11.w1.weight": 50.83956015087797,
+ "blocks.11.w1.bias": 49.1835893514986,
+ "blocks.11.w2.weight": 90.71530778227628,
+ "out_ln.weight": 0.5849072301486865,
+ "out_head.weight": 3.217267238682831,
+ "out_head.bias": 0.8024946363335762
+ }
+ },
+ "credit_bridge": {
+ "log": {
+ "train_loss": [
+ 2.07422056892395,
+ 2.055305141906738,
+ 2.0377359949493408,
+ 2.0334118033599853,
+ 2.034562978172302,
+ 2.0315364552307127,
+ 2.0276041582489013,
+ 2.029994778060913,
+ 2.025310488204956,
+ 2.0206511740875244,
+ 2.024435912742615,
+ 2.017680696372986,
+ 2.014923624572754,
+ 2.015591604614258,
+ 2.0193257177734374,
+ 2.0126227207946776,
+ 2.013033943634033,
+ 2.0076499520111084,
+ 2.0094668504333497,
+ 2.0096006536865234,
+ 2.0050730683135987,
+ 2.006946638031006,
+ 2.0071442542266844,
+ 2.004831351776123,
+ 2.0035909980010986,
+ 2.004108156204224,
+ 2.0006491820907595,
+ 2.0016160436248778,
+ 1.9999982108306884,
+ 1.9992566723251344,
+ 1.9979173023986816,
+ 1.996204802093506,
+ 1.9980539007568359,
+ 1.995016636199951,
+ 1.9975717867279053,
+ 1.9963324966812135,
+ 1.9947675510406495,
+ 1.9942235395812988,
+ 1.995222140235901,
+ 1.9917342311096191,
+ 1.9899134133148193,
+ 1.988487755279541,
+ 1.9872248657608031,
+ 1.9865933574676513,
+ 1.9852813260650635,
+ 1.9867496614837648,
+ 1.9831518542480469,
+ 1.9813233506011962,
+ 1.9796981923675536,
+ 1.9797288468933105,
+ 1.9774669593811036,
+ 1.9774521997833252,
+ 1.9714781398773193,
+ 1.9715750045013427,
+ 1.9715392060089112,
+ 1.970840647201538,
+ 1.9684521988296508,
+ 1.9670102074813842,
+ 1.9633512356948852,
+ 1.9662449013900758,
+ 1.9640194422149657,
+ 1.9607671911239624,
+ 1.9630818260955811,
+ 1.9602737934112549,
+ 1.960026322631836,
+ 1.961152883758545,
+ 1.9590496487426758,
+ 1.9572670712280273,
+ 1.9577932699966432,
+ 1.957121894493103,
+ 1.9570679434585572,
+ 1.954615853881836,
+ 1.9549290727996826,
+ 1.9529239632415771,
+ 1.9538047943115235,
+ 1.9561965101623535,
+ 1.9530966287994385,
+ 1.9537969312286376,
+ 1.9531251779174805,
+ 1.9546874990463257,
+ 1.9529049239730836,
+ 1.9550184294891357,
+ 1.9533232469940185,
+ 1.9538247448730468,
+ 1.9541607789611817,
+ 1.954350821533203,
+ 1.9529511869049072,
+ 1.9544397570037841,
+ 1.9555162839126587,
+ 1.9529233142089844,
+ 1.9550124865722656,
+ 1.954115503501892,
+ 1.9515360607910157,
+ 1.9521209592056274,
+ 1.9523113599014281,
+ 1.9537422107696534,
+ 1.953715394668579,
+ 1.952323772354126,
+ 1.9519077111434937,
+ 1.9538698761367799
+ ],
+ "train_acc": [
+ 0.2349,
+ 0.23648,
+ 0.24714,
+ 0.25066,
+ 0.24922,
+ 0.2532,
+ 0.25626,
+ 0.25148,
+ 0.2532,
+ 0.26078,
+ 0.25636,
+ 0.2586,
+ 0.26336,
+ 0.26146,
+ 0.25874,
+ 0.26418,
+ 0.2629,
+ 0.26778,
+ 0.26784,
+ 0.26732,
+ 0.26768,
+ 0.26668,
+ 0.2681,
+ 0.26688,
+ 0.26936,
+ 0.26936,
+ 0.2677,
+ 0.2705,
+ 0.2677,
+ 0.2709,
+ 0.27164,
+ 0.27174,
+ 0.27098,
+ 0.27192,
+ 0.27154,
+ 0.26932,
+ 0.27164,
+ 0.26854,
+ 0.26818,
+ 0.27076,
+ 0.26892,
+ 0.26664,
+ 0.27098,
+ 0.2697,
+ 0.2714,
+ 0.2725,
+ 0.27024,
+ 0.27174,
+ 0.27278,
+ 0.26878,
+ 0.27302,
+ 0.27324,
+ 0.27562,
+ 0.2744,
+ 0.27374,
+ 0.2738,
+ 0.27576,
+ 0.27626,
+ 0.2782,
+ 0.27834,
+ 0.27942,
+ 0.27984,
+ 0.28104,
+ 0.27978,
+ 0.28036,
+ 0.281,
+ 0.28012,
+ 0.28318,
+ 0.28276,
+ 0.28262,
+ 0.28174,
+ 0.28244,
+ 0.28384,
+ 0.28378,
+ 0.28154,
+ 0.2809,
+ 0.28316,
+ 0.2827,
+ 0.2864,
+ 0.2831,
+ 0.2849,
+ 0.28526,
+ 0.28446,
+ 0.28448,
+ 0.28456,
+ 0.28378,
+ 0.28358,
+ 0.28354,
+ 0.2834,
+ 0.2829,
+ 0.28564,
+ 0.28302,
+ 0.28616,
+ 0.28478,
+ 0.2849,
+ 0.28396,
+ 0.28816,
+ 0.28346,
+ 0.28456,
+ 0.28572
+ ],
+ "test_acc": [
+ 0.2422,
+ 0.2524,
+ 0.2511,
+ 0.2709,
+ 0.2522,
+ 0.2695,
+ 0.2654,
+ 0.2622,
+ 0.282,
+ 0.2861,
+ 0.259,
+ 0.2868,
+ 0.2668,
+ 0.2736,
+ 0.2902,
+ 0.2962,
+ 0.3035,
+ 0.2747,
+ 0.296,
+ 0.2894,
+ 0.2905,
+ 0.2902,
+ 0.3018,
+ 0.2788,
+ 0.2989,
+ 0.2832,
+ 0.3054,
+ 0.2928,
+ 0.2921,
+ 0.2723,
+ 0.29,
+ 0.281,
+ 0.2819,
+ 0.2924,
+ 0.3021,
+ 0.291,
+ 0.2721,
+ 0.2887,
+ 0.2827,
+ 0.2766,
+ 0.2938,
+ 0.2976,
+ 0.2883,
+ 0.2888,
+ 0.3015,
+ 0.2923,
+ 0.2893,
+ 0.2859,
+ 0.2876,
+ 0.2915,
+ 0.2857,
+ 0.2848,
+ 0.2936,
+ 0.2842,
+ 0.2815,
+ 0.2766,
+ 0.2905,
+ 0.2933,
+ 0.2846,
+ 0.2876,
+ 0.2872,
+ 0.2876,
+ 0.293,
+ 0.2942,
+ 0.2917,
+ 0.2816,
+ 0.2931,
+ 0.2874,
+ 0.2906,
+ 0.2935,
+ 0.2918,
+ 0.2838,
+ 0.2872,
+ 0.29,
+ 0.2891,
+ 0.2893,
+ 0.2908,
+ 0.2873,
+ 0.293,
+ 0.292,
+ 0.286,
+ 0.286,
+ 0.2872,
+ 0.2867,
+ 0.2876,
+ 0.2865,
+ 0.2841,
+ 0.2826,
+ 0.2844,
+ 0.2853,
+ 0.2866,
+ 0.2887,
+ 0.2856,
+ 0.2867,
+ 0.2882,
+ 0.2895,
+ 0.2885,
+ 0.2885,
+ 0.2885,
+ 0.2885
+ ],
+ "value_loss": [
+ 0.46219210285663603,
+ 0.16240292426109315,
+ 0.11675834127902984,
+ 0.10173126588344573,
+ 0.10192921797513962,
+ 0.08491676461458206,
+ 0.08081110213756561,
+ 0.10098561144590378,
+ 0.086660576441288,
+ 0.07095537901163101,
+ 0.07209414057731628,
+ 0.06507166877388953,
+ 0.05680496859550476,
+ 0.051555730621814726,
+ 0.05897924602746964,
+ 0.0477625340873003,
+ 0.050076086919903756,
+ 0.04791341652750969,
+ 0.043811181032657624,
+ 0.048990424308776856,
+ 0.041765395416021346,
+ 0.04788978707790375,
+ 0.038938788844347,
+ 0.038173083332777025,
+ 0.037605566675663,
+ 0.03472667615830898,
+ 0.029743025472164154,
+ 0.03321148899495602,
+ 0.031504894320964816,
+ 0.028381420345902443,
+ 0.02973802535891533,
+ 0.025553143047094346,
+ 0.028812745112776757,
+ 0.026098545892238617,
+ 0.02787227361112833,
+ 0.022305913635492326,
+ 0.02363199061870575,
+ 0.021854550403952597,
+ 0.02323986495792866,
+ 0.018420420945882796,
+ 0.018242734627723693,
+ 0.019795424582660197,
+ 0.01764130767583847,
+ 0.01667629898786545,
+ 0.016429091787040235,
+ 0.015772414491176605,
+ 0.014359855057001113,
+ 0.012779874252080917,
+ 0.012927391214668751,
+ 0.014349255759716033,
+ 0.013130034263134003,
+ 0.013914088915586471,
+ 0.010768152612149716,
+ 0.012323264547288418,
+ 0.012266730073690415,
+ 0.012296598086953162,
+ 0.011092334405183792,
+ 0.008943616542220116,
+ 0.009557433066666126,
+ 0.008033960376381873,
+ 0.008238149175047874,
+ 0.0075173489648103714,
+ 0.009243258455693721,
+ 0.008397357228696346,
+ 0.008250873034894466,
+ 0.009054240943193436,
+ 0.0061908810329437255,
+ 0.007204210565388202,
+ 0.006787694528698921,
+ 0.006576583880782127,
+ 0.0068760585397481915,
+ 0.00587797952502966,
+ 0.006058499993383884,
+ 0.004864574173986912,
+ 0.005336316513419152,
+ 0.004476593678146601,
+ 0.00504198596701026,
+ 0.00414283695474267,
+ 0.003692763058245182,
+ 0.0036398935145139696,
+ 0.0031052616691589355,
+ 0.003147630747780204,
+ 0.002547671852298081,
+ 0.0023914932864904405,
+ 0.0022162304655462504,
+ 0.0017571020238101483,
+ 0.0018844479297846555,
+ 0.0015003561322391033,
+ 0.001468994250819087,
+ 0.0015521602664142846,
+ 0.0010236410408467054,
+ 0.000977659215182066,
+ 0.001276740662381053,
+ 0.0008685083091259003,
+ 0.0012216014919430019,
+ 0.001092172961011529,
+ 0.0008087569988891483,
+ 0.0009069699543341994,
+ 0.0013558788930997253,
+ 0.0006715716527029872
+ ]
+ },
+ "diagnostics": {
+ "bp_cosine": [
+ 0.37664300203323364,
+ 0.00668198149651289,
+ 0.02027757838368416,
+ 0.02049040049314499,
+ 0.020580384880304337,
+ 0.020434698089957237,
+ 0.020465239882469177,
+ 0.021316751837730408,
+ 0.021797576919198036,
+ 0.021945470944046974,
+ 0.021979298442602158,
+ 0.021852022036910057
+ ],
+ "perturbation_rho": [
+ 0.009359830990433693,
+ 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": [
+ -9.862706065177917e-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": [
+ -2.905726432800293e-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": [
+ -9.72812995314598e-06,
+ 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": 179.58500900515028,
+ "embed.bias": 134.71300218044718,
+ "blocks.0.ln.weight": 9.192845521766614,
+ "blocks.0.w1.weight": 201.0064270139136,
+ "blocks.0.w1.bias": 136.56099913744478,
+ "blocks.0.w2.weight": 299.7118104200996,
+ "blocks.1.ln.weight": 6.775393287001583,
+ "blocks.1.w1.weight": 298.99303772053656,
+ "blocks.1.w1.bias": 288.6803183648879,
+ "blocks.1.w2.weight": 299.96832262513726,
+ "blocks.2.ln.weight": 4.442030759950682,
+ "blocks.2.w1.weight": 156.3224513959878,
+ "blocks.2.w1.bias": 134.77200050699193,
+ "blocks.2.w2.weight": 164.77485249426002,
+ "blocks.3.ln.weight": 4.854857462306176,
+ "blocks.3.w1.weight": 165.45425459739124,
+ "blocks.3.w1.bias": 143.7040311765167,
+ "blocks.3.w2.weight": 178.2893036667054,
+ "blocks.4.ln.weight": 6.433765762816246,
+ "blocks.4.w1.weight": 262.9945171233052,
+ "blocks.4.w1.bias": 237.2497554637253,
+ "blocks.4.w2.weight": 247.70202715443386,
+ "blocks.5.ln.weight": 4.669370190359616,
+ "blocks.5.w1.weight": 179.65430058156437,
+ "blocks.5.w1.bias": 167.17352257399838,
+ "blocks.5.w2.weight": 176.56809805245783,
+ "blocks.6.ln.weight": 5.836574580577738,
+ "blocks.6.w1.weight": 236.71076201812485,
+ "blocks.6.w1.bias": 216.59863578081908,
+ "blocks.6.w2.weight": 222.1449202786447,
+ "blocks.7.ln.weight": 7.017034323466784,
+ "blocks.7.w1.weight": 283.45578137085835,
+ "blocks.7.w1.bias": 258.2162159056869,
+ "blocks.7.w2.weight": 275.54620177391195,
+ "blocks.8.ln.weight": 6.320927520653432,
+ "blocks.8.w1.weight": 252.95263320055057,
+ "blocks.8.w1.bias": 230.26130735012416,
+ "blocks.8.w2.weight": 239.07712197948655,
+ "blocks.9.ln.weight": 4.771737132124388,
+ "blocks.9.w1.weight": 184.2226928948631,
+ "blocks.9.w1.bias": 168.12677483943122,
+ "blocks.9.w2.weight": 169.7918496964596,
+ "blocks.10.ln.weight": 6.193554347477001,
+ "blocks.10.w1.weight": 248.40930737157998,
+ "blocks.10.w1.bias": 221.7518821584096,
+ "blocks.10.w2.weight": 238.1444968107213,
+ "blocks.11.ln.weight": 5.976649802588373,
+ "blocks.11.w1.weight": 240.99115354219526,
+ "blocks.11.w1.bias": 220.01049332622438,
+ "blocks.11.w2.weight": 226.72197924893874,
+ "out_ln.weight": 0.47882108182026845,
+ "out_head.weight": 7.4188430237353336,
+ "out_head.bias": 1.3782394013786212
+ }
+ }
+ },
+ "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": [
+ 123
+ ],
+ "gpu": 0,
+ "output_dir": "results/cifar10_seed123",
+ "num_classes": 10
+ }
+} \ No newline at end of file