diff --git "a/trainer_state.json" "b/trainer_state.json" new file mode 100644--- /dev/null +++ "b/trainer_state.json" @@ -0,0 +1,6426 @@ +{ + "best_metric": 4.216700077056885, + "best_model_checkpoint": "/Users/bbunzeck/Documents/lexdec-llamas/large-bpe/checkpoint-8136", + "epoch": 0.9999723214038584, + "eval_steps": 904, + "global_step": 9032, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.001107143845664148, + "grad_norm": 2.518164873123169, + "learning_rate": 1.4999999999999999e-05, + "loss": 8.2577, + "step": 10 + }, + { + "epoch": 0.002214287691328296, + "grad_norm": 2.570223569869995, + "learning_rate": 2.9999999999999997e-05, + "loss": 8.1792, + "step": 20 + }, + { + "epoch": 0.0033214315369924437, + "grad_norm": 3.2387843132019043, + "learning_rate": 4.4999999999999996e-05, + "loss": 7.8221, + "step": 30 + }, + { + "epoch": 0.004428575382656592, + "grad_norm": 2.3386170864105225, + "learning_rate": 5.9999999999999995e-05, + "loss": 7.4134, + "step": 40 + }, + { + "epoch": 0.00553571922832074, + "grad_norm": 1.8533618450164795, + "learning_rate": 7.5e-05, + "loss": 6.9659, + "step": 50 + }, + { + "epoch": 0.006642863073984887, + "grad_norm": 4.732933521270752, + "learning_rate": 8.999999999999999e-05, + "loss": 6.7172, + "step": 60 + }, + { + "epoch": 0.007750006919649035, + "grad_norm": 1.1716065406799316, + "learning_rate": 0.00010499999999999999, + "loss": 6.5705, + "step": 70 + }, + { + "epoch": 0.008857150765313184, + "grad_norm": 1.577109932899475, + "learning_rate": 0.00011999999999999999, + "loss": 6.2671, + "step": 80 + }, + { + "epoch": 0.009964294610977332, + "grad_norm": 0.8941097259521484, + "learning_rate": 0.000135, + "loss": 6.0985, + "step": 90 + }, + { + "epoch": 0.01107143845664148, + "grad_norm": 0.9588972330093384, + "learning_rate": 0.00015, + "loss": 6.0416, + "step": 100 + }, + { + "epoch": 0.012178582302305627, + "grad_norm": 1.2730671167373657, + "learning_rate": 0.000165, + "loss": 6.0525, + "step": 110 + }, + { + "epoch": 0.013285726147969775, + "grad_norm": 1.12746262550354, + "learning_rate": 0.00017999999999999998, + "loss": 5.9563, + "step": 120 + }, + { + "epoch": 0.014392869993633923, + "grad_norm": 0.8732702136039734, + "learning_rate": 0.000195, + "loss": 6.006, + "step": 130 + }, + { + "epoch": 0.01550001383929807, + "grad_norm": 0.8938935995101929, + "learning_rate": 0.00020999999999999998, + "loss": 5.9543, + "step": 140 + }, + { + "epoch": 0.01660715768496222, + "grad_norm": 1.1050419807434082, + "learning_rate": 0.000225, + "loss": 5.8664, + "step": 150 + }, + { + "epoch": 0.017714301530626368, + "grad_norm": 0.8527132868766785, + "learning_rate": 0.00023999999999999998, + "loss": 5.9658, + "step": 160 + }, + { + "epoch": 0.018821445376290515, + "grad_norm": 1.2807834148406982, + "learning_rate": 0.00025499999999999996, + "loss": 5.7008, + "step": 170 + }, + { + "epoch": 0.019928589221954663, + "grad_norm": 1.0580153465270996, + "learning_rate": 0.00027, + "loss": 5.6691, + "step": 180 + }, + { + "epoch": 0.02103573306761881, + "grad_norm": 1.0796706676483154, + "learning_rate": 0.000285, + "loss": 5.6045, + "step": 190 + }, + { + "epoch": 0.02214287691328296, + "grad_norm": 1.0151618719100952, + "learning_rate": 0.0003, + "loss": 5.4736, + "step": 200 + }, + { + "epoch": 0.023250020758947106, + "grad_norm": 1.0638478994369507, + "learning_rate": 0.0002999990510519666, + "loss": 5.4839, + "step": 210 + }, + { + "epoch": 0.024357164604611254, + "grad_norm": 1.1054071187973022, + "learning_rate": 0.0002999962042198732, + "loss": 5.4842, + "step": 220 + }, + { + "epoch": 0.025464308450275402, + "grad_norm": 0.9934037327766418, + "learning_rate": 0.0002999914595397396, + "loss": 5.4912, + "step": 230 + }, + { + "epoch": 0.02657145229593955, + "grad_norm": 0.9983835816383362, + "learning_rate": 0.00029998481707159863, + "loss": 5.3407, + "step": 240 + }, + { + "epoch": 0.027678596141603697, + "grad_norm": 0.9708263874053955, + "learning_rate": 0.000299976276899495, + "loss": 5.3992, + "step": 250 + }, + { + "epoch": 0.028785739987267845, + "grad_norm": 1.0524625778198242, + "learning_rate": 0.00029996583913148453, + "loss": 5.158, + "step": 260 + }, + { + "epoch": 0.029892883832931993, + "grad_norm": 0.950106680393219, + "learning_rate": 0.00029995350389963246, + "loss": 5.3275, + "step": 270 + }, + { + "epoch": 0.03100002767859614, + "grad_norm": 0.894207239151001, + "learning_rate": 0.0002999392713600121, + "loss": 5.2935, + "step": 280 + }, + { + "epoch": 0.03210717152426029, + "grad_norm": 1.008943796157837, + "learning_rate": 0.00029992314169270256, + "loss": 5.3003, + "step": 290 + }, + { + "epoch": 0.03321431536992444, + "grad_norm": 1.0820633172988892, + "learning_rate": 0.0002999051151017868, + "loss": 5.1637, + "step": 300 + }, + { + "epoch": 0.03432145921558859, + "grad_norm": 1.0281634330749512, + "learning_rate": 0.0002998851918153488, + "loss": 5.2656, + "step": 310 + }, + { + "epoch": 0.035428603061252735, + "grad_norm": 0.9343926310539246, + "learning_rate": 0.0002998633720854707, + "loss": 5.2241, + "step": 320 + }, + { + "epoch": 0.03653574690691688, + "grad_norm": 0.8846531510353088, + "learning_rate": 0.0002998396561882297, + "loss": 5.1459, + "step": 330 + }, + { + "epoch": 0.03764289075258103, + "grad_norm": 1.0342435836791992, + "learning_rate": 0.0002998140444236946, + "loss": 5.0918, + "step": 340 + }, + { + "epoch": 0.03875003459824518, + "grad_norm": 0.8708397746086121, + "learning_rate": 0.0002997865371159217, + "loss": 5.1038, + "step": 350 + }, + { + "epoch": 0.039857178443909326, + "grad_norm": 0.872075080871582, + "learning_rate": 0.0002997571346129512, + "loss": 5.0619, + "step": 360 + }, + { + "epoch": 0.040964322289573474, + "grad_norm": 0.9121406078338623, + "learning_rate": 0.00029972583728680237, + "loss": 4.9992, + "step": 370 + }, + { + "epoch": 0.04207146613523762, + "grad_norm": 0.8964458703994751, + "learning_rate": 0.000299692645533469, + "loss": 5.1084, + "step": 380 + }, + { + "epoch": 0.04317860998090177, + "grad_norm": 1.0498356819152832, + "learning_rate": 0.00029965755977291443, + "loss": 5.0388, + "step": 390 + }, + { + "epoch": 0.04428575382656592, + "grad_norm": 0.9154861569404602, + "learning_rate": 0.0002996205804490661, + "loss": 5.0252, + "step": 400 + }, + { + "epoch": 0.045392897672230065, + "grad_norm": 0.9080700278282166, + "learning_rate": 0.00029958170802981023, + "loss": 5.0699, + "step": 410 + }, + { + "epoch": 0.04650004151789421, + "grad_norm": 0.851718544960022, + "learning_rate": 0.0002995409430069855, + "loss": 4.9268, + "step": 420 + }, + { + "epoch": 0.04760718536355836, + "grad_norm": 1.0082451105117798, + "learning_rate": 0.00029949828589637703, + "loss": 5.0694, + "step": 430 + }, + { + "epoch": 0.04871432920922251, + "grad_norm": 0.9662661552429199, + "learning_rate": 0.00029945373723771, + "loss": 4.9221, + "step": 440 + }, + { + "epoch": 0.049821473054886656, + "grad_norm": 1.0844182968139648, + "learning_rate": 0.00029940729759464247, + "loss": 4.9185, + "step": 450 + }, + { + "epoch": 0.050928616900550804, + "grad_norm": 1.005959153175354, + "learning_rate": 0.00029935896755475865, + "loss": 4.9541, + "step": 460 + }, + { + "epoch": 0.05203576074621495, + "grad_norm": 0.8470442295074463, + "learning_rate": 0.00029930874772956113, + "loss": 4.9238, + "step": 470 + }, + { + "epoch": 0.0531429045918791, + "grad_norm": 0.9112293720245361, + "learning_rate": 0.00029925663875446325, + "loss": 4.7723, + "step": 480 + }, + { + "epoch": 0.05425004843754325, + "grad_norm": 0.931795060634613, + "learning_rate": 0.0002992026412887812, + "loss": 4.9303, + "step": 490 + }, + { + "epoch": 0.055357192283207395, + "grad_norm": 0.8564447164535522, + "learning_rate": 0.0002991467560157254, + "loss": 4.8894, + "step": 500 + }, + { + "epoch": 0.05646433612887154, + "grad_norm": 1.0363521575927734, + "learning_rate": 0.0002990889836423922, + "loss": 4.8259, + "step": 510 + }, + { + "epoch": 0.05757147997453569, + "grad_norm": 0.9189797043800354, + "learning_rate": 0.00029902932489975464, + "loss": 4.891, + "step": 520 + }, + { + "epoch": 0.05867862382019984, + "grad_norm": 0.9805159568786621, + "learning_rate": 0.0002989677805426534, + "loss": 4.8568, + "step": 530 + }, + { + "epoch": 0.059785767665863986, + "grad_norm": 0.9564447999000549, + "learning_rate": 0.000298904351349787, + "loss": 4.8496, + "step": 540 + }, + { + "epoch": 0.06089291151152813, + "grad_norm": 0.9181010127067566, + "learning_rate": 0.0002988390381237023, + "loss": 4.8451, + "step": 550 + }, + { + "epoch": 0.06200005535719228, + "grad_norm": 0.8183259963989258, + "learning_rate": 0.000298771841690784, + "loss": 4.7755, + "step": 560 + }, + { + "epoch": 0.06310719920285643, + "grad_norm": 0.9422429800033569, + "learning_rate": 0.00029870276290124443, + "loss": 4.8762, + "step": 570 + }, + { + "epoch": 0.06421434304852058, + "grad_norm": 0.9148269295692444, + "learning_rate": 0.0002986318026291127, + "loss": 4.7477, + "step": 580 + }, + { + "epoch": 0.06532148689418472, + "grad_norm": 0.9477198123931885, + "learning_rate": 0.0002985589617722236, + "loss": 4.7061, + "step": 590 + }, + { + "epoch": 0.06642863073984888, + "grad_norm": 0.8547405004501343, + "learning_rate": 0.0002984842412522063, + "loss": 4.9329, + "step": 600 + }, + { + "epoch": 0.06753577458551302, + "grad_norm": 0.9699390530586243, + "learning_rate": 0.00029840764201447266, + "loss": 4.7919, + "step": 610 + }, + { + "epoch": 0.06864291843117717, + "grad_norm": 0.9100772142410278, + "learning_rate": 0.0002983291650282053, + "loss": 4.7424, + "step": 620 + }, + { + "epoch": 0.06975006227684132, + "grad_norm": 0.8909862041473389, + "learning_rate": 0.0002982488112863453, + "loss": 4.7653, + "step": 630 + }, + { + "epoch": 0.07085720612250547, + "grad_norm": 1.0143611431121826, + "learning_rate": 0.0002981665818055797, + "loss": 4.6796, + "step": 640 + }, + { + "epoch": 0.07196434996816961, + "grad_norm": 0.8786947727203369, + "learning_rate": 0.0002980824776263286, + "loss": 4.7385, + "step": 650 + }, + { + "epoch": 0.07307149381383377, + "grad_norm": 0.9391975998878479, + "learning_rate": 0.00029799649981273186, + "loss": 4.7794, + "step": 660 + }, + { + "epoch": 0.0741786376594979, + "grad_norm": 0.9254968166351318, + "learning_rate": 0.0002979086494526358, + "loss": 4.7993, + "step": 670 + }, + { + "epoch": 0.07528578150516206, + "grad_norm": 0.9523051381111145, + "learning_rate": 0.0002978189276575795, + "loss": 4.7954, + "step": 680 + }, + { + "epoch": 0.0763929253508262, + "grad_norm": 0.897014856338501, + "learning_rate": 0.0002977273355627806, + "loss": 4.8478, + "step": 690 + }, + { + "epoch": 0.07750006919649036, + "grad_norm": 0.9511550664901733, + "learning_rate": 0.0002976338743271209, + "loss": 4.7259, + "step": 700 + }, + { + "epoch": 0.0786072130421545, + "grad_norm": 0.8601536750793457, + "learning_rate": 0.0002975385451331318, + "loss": 4.6512, + "step": 710 + }, + { + "epoch": 0.07971435688781865, + "grad_norm": 0.9739886522293091, + "learning_rate": 0.00029744134918697937, + "loss": 4.723, + "step": 720 + }, + { + "epoch": 0.08082150073348279, + "grad_norm": 0.865020215511322, + "learning_rate": 0.0002973422877184489, + "loss": 4.761, + "step": 730 + }, + { + "epoch": 0.08192864457914695, + "grad_norm": 0.8409671187400818, + "learning_rate": 0.00029724136198092967, + "loss": 4.7693, + "step": 740 + }, + { + "epoch": 0.08303578842481109, + "grad_norm": 0.8909373879432678, + "learning_rate": 0.00029713857325139857, + "loss": 4.773, + "step": 750 + }, + { + "epoch": 0.08414293227047524, + "grad_norm": 0.8702926635742188, + "learning_rate": 0.00029703392283040453, + "loss": 4.7161, + "step": 760 + }, + { + "epoch": 0.08525007611613938, + "grad_norm": 0.9433870911598206, + "learning_rate": 0.0002969274120420517, + "loss": 4.7037, + "step": 770 + }, + { + "epoch": 0.08635721996180354, + "grad_norm": 0.9102666974067688, + "learning_rate": 0.00029681904223398273, + "loss": 4.6889, + "step": 780 + }, + { + "epoch": 0.08746436380746768, + "grad_norm": 0.918194591999054, + "learning_rate": 0.0002967088147773619, + "loss": 4.701, + "step": 790 + }, + { + "epoch": 0.08857150765313183, + "grad_norm": 0.8887764811515808, + "learning_rate": 0.0002965967310668576, + "loss": 4.4922, + "step": 800 + }, + { + "epoch": 0.08967865149879597, + "grad_norm": 0.9276973009109497, + "learning_rate": 0.00029648279252062463, + "loss": 4.6985, + "step": 810 + }, + { + "epoch": 0.09078579534446013, + "grad_norm": 0.8622846603393555, + "learning_rate": 0.00029636700058028654, + "loss": 4.6268, + "step": 820 + }, + { + "epoch": 0.09189293919012427, + "grad_norm": 0.891849935054779, + "learning_rate": 0.0002962493567109171, + "loss": 4.7091, + "step": 830 + }, + { + "epoch": 0.09300008303578843, + "grad_norm": 0.8380031585693359, + "learning_rate": 0.00029612986240102186, + "loss": 4.677, + "step": 840 + }, + { + "epoch": 0.09410722688145257, + "grad_norm": 0.9445401430130005, + "learning_rate": 0.0002960085191625193, + "loss": 4.5742, + "step": 850 + }, + { + "epoch": 0.09521437072711672, + "grad_norm": 0.9989162683486938, + "learning_rate": 0.00029588532853072196, + "loss": 4.679, + "step": 860 + }, + { + "epoch": 0.09632151457278088, + "grad_norm": 0.9389657974243164, + "learning_rate": 0.0002957602920643164, + "loss": 4.6184, + "step": 870 + }, + { + "epoch": 0.09742865841844502, + "grad_norm": 0.8651659488677979, + "learning_rate": 0.00029563341134534424, + "loss": 4.6631, + "step": 880 + }, + { + "epoch": 0.09853580226410917, + "grad_norm": 0.8622980713844299, + "learning_rate": 0.0002955046879791816, + "loss": 4.7446, + "step": 890 + }, + { + "epoch": 0.09964294610977331, + "grad_norm": 0.9320902228355408, + "learning_rate": 0.0002953741235945188, + "loss": 4.6667, + "step": 900 + }, + { + "epoch": 0.10008580364803897, + "eval_loss": 4.9949951171875, + "eval_runtime": 208.3906, + "eval_samples_per_second": 239.934, + "eval_steps_per_second": 29.992, + "step": 904 + }, + { + "epoch": 0.10075008995543747, + "grad_norm": 0.903203010559082, + "learning_rate": 0.0002952417198433401, + "loss": 4.6963, + "step": 910 + }, + { + "epoch": 0.10185723380110161, + "grad_norm": 0.8986852765083313, + "learning_rate": 0.00029510747840090273, + "loss": 4.5441, + "step": 920 + }, + { + "epoch": 0.10296437764676576, + "grad_norm": 0.8018354773521423, + "learning_rate": 0.00029497140096571516, + "loss": 4.6414, + "step": 930 + }, + { + "epoch": 0.1040715214924299, + "grad_norm": 0.8711630702018738, + "learning_rate": 0.00029483348925951635, + "loss": 4.6914, + "step": 940 + }, + { + "epoch": 0.10517866533809406, + "grad_norm": 0.9504655599594116, + "learning_rate": 0.00029469374502725355, + "loss": 4.5654, + "step": 950 + }, + { + "epoch": 0.1062858091837582, + "grad_norm": 0.8903089761734009, + "learning_rate": 0.0002945521700370603, + "loss": 4.5563, + "step": 960 + }, + { + "epoch": 0.10739295302942235, + "grad_norm": 0.8594041466712952, + "learning_rate": 0.00029440876608023394, + "loss": 4.5535, + "step": 970 + }, + { + "epoch": 0.1085000968750865, + "grad_norm": 0.8852702379226685, + "learning_rate": 0.00029426353497121334, + "loss": 4.6342, + "step": 980 + }, + { + "epoch": 0.10960724072075065, + "grad_norm": 0.975365936756134, + "learning_rate": 0.00029411647854755534, + "loss": 4.5329, + "step": 990 + }, + { + "epoch": 0.11071438456641479, + "grad_norm": 0.8069102764129639, + "learning_rate": 0.0002939675986699121, + "loss": 4.549, + "step": 1000 + }, + { + "epoch": 0.11182152841207894, + "grad_norm": 0.8447535634040833, + "learning_rate": 0.0002938168972220071, + "loss": 4.6189, + "step": 1010 + }, + { + "epoch": 0.11292867225774308, + "grad_norm": 0.9278955459594727, + "learning_rate": 0.0002936643761106117, + "loss": 4.6185, + "step": 1020 + }, + { + "epoch": 0.11403581610340724, + "grad_norm": 0.8939646482467651, + "learning_rate": 0.00029351003726552056, + "loss": 4.5531, + "step": 1030 + }, + { + "epoch": 0.11514295994907138, + "grad_norm": 0.8682196736335754, + "learning_rate": 0.0002933538826395276, + "loss": 4.674, + "step": 1040 + }, + { + "epoch": 0.11625010379473553, + "grad_norm": 0.9392927885055542, + "learning_rate": 0.00029319591420840126, + "loss": 4.492, + "step": 1050 + }, + { + "epoch": 0.11735724764039968, + "grad_norm": 1.0816600322723389, + "learning_rate": 0.0002930361339708592, + "loss": 4.5446, + "step": 1060 + }, + { + "epoch": 0.11846439148606383, + "grad_norm": 0.9101948738098145, + "learning_rate": 0.00029287454394854337, + "loss": 4.6679, + "step": 1070 + }, + { + "epoch": 0.11957153533172797, + "grad_norm": 0.8949317932128906, + "learning_rate": 0.00029271114618599416, + "loss": 4.4307, + "step": 1080 + }, + { + "epoch": 0.12067867917739213, + "grad_norm": 0.8716855049133301, + "learning_rate": 0.0002925459427506248, + "loss": 4.4294, + "step": 1090 + }, + { + "epoch": 0.12178582302305627, + "grad_norm": 0.8888041377067566, + "learning_rate": 0.00029237893573269487, + "loss": 4.5163, + "step": 1100 + }, + { + "epoch": 0.12289296686872042, + "grad_norm": 0.9682092666625977, + "learning_rate": 0.0002922101272452842, + "loss": 4.5234, + "step": 1110 + }, + { + "epoch": 0.12400011071438456, + "grad_norm": 0.9267321228981018, + "learning_rate": 0.0002920395194242658, + "loss": 4.4814, + "step": 1120 + }, + { + "epoch": 0.1251072545600487, + "grad_norm": 0.9908064603805542, + "learning_rate": 0.00029186711442827915, + "loss": 4.4873, + "step": 1130 + }, + { + "epoch": 0.12621439840571286, + "grad_norm": 0.9782397747039795, + "learning_rate": 0.0002916929144387027, + "loss": 4.4187, + "step": 1140 + }, + { + "epoch": 0.127321542251377, + "grad_norm": 0.868129312992096, + "learning_rate": 0.0002915169216596262, + "loss": 4.5196, + "step": 1150 + }, + { + "epoch": 0.12842868609704117, + "grad_norm": 0.8631125688552856, + "learning_rate": 0.00029133913831782307, + "loss": 4.4316, + "step": 1160 + }, + { + "epoch": 0.1295358299427053, + "grad_norm": 0.9837842583656311, + "learning_rate": 0.00029115956666272194, + "loss": 4.4438, + "step": 1170 + }, + { + "epoch": 0.13064297378836945, + "grad_norm": 0.9683934450149536, + "learning_rate": 0.0002909782089663785, + "loss": 4.4687, + "step": 1180 + }, + { + "epoch": 0.1317501176340336, + "grad_norm": 0.9495298862457275, + "learning_rate": 0.00029079506752344636, + "loss": 4.369, + "step": 1190 + }, + { + "epoch": 0.13285726147969776, + "grad_norm": 0.8343674540519714, + "learning_rate": 0.0002906101446511484, + "loss": 4.5115, + "step": 1200 + }, + { + "epoch": 0.13396440532536188, + "grad_norm": 0.8238400816917419, + "learning_rate": 0.0002904234426892473, + "loss": 4.3912, + "step": 1210 + }, + { + "epoch": 0.13507154917102604, + "grad_norm": 0.9322042465209961, + "learning_rate": 0.00029023496400001577, + "loss": 4.4049, + "step": 1220 + }, + { + "epoch": 0.1361786930166902, + "grad_norm": 0.879301905632019, + "learning_rate": 0.0002900447109682069, + "loss": 4.4323, + "step": 1230 + }, + { + "epoch": 0.13728583686235435, + "grad_norm": 0.884285569190979, + "learning_rate": 0.0002898526860010239, + "loss": 4.4153, + "step": 1240 + }, + { + "epoch": 0.13839298070801848, + "grad_norm": 0.9528766870498657, + "learning_rate": 0.00028965889152808967, + "loss": 4.4578, + "step": 1250 + }, + { + "epoch": 0.13950012455368263, + "grad_norm": 0.8995983004570007, + "learning_rate": 0.00028946333000141595, + "loss": 4.3977, + "step": 1260 + }, + { + "epoch": 0.14060726839934679, + "grad_norm": 0.9160441160202026, + "learning_rate": 0.00028926600389537246, + "loss": 4.3376, + "step": 1270 + }, + { + "epoch": 0.14171441224501094, + "grad_norm": 0.9049811363220215, + "learning_rate": 0.00028906691570665545, + "loss": 4.5101, + "step": 1280 + }, + { + "epoch": 0.1428215560906751, + "grad_norm": 0.8610453009605408, + "learning_rate": 0.0002888660679542562, + "loss": 4.4694, + "step": 1290 + }, + { + "epoch": 0.14392869993633922, + "grad_norm": 1.0628710985183716, + "learning_rate": 0.00028866346317942915, + "loss": 4.3037, + "step": 1300 + }, + { + "epoch": 0.14503584378200338, + "grad_norm": 0.9687007665634155, + "learning_rate": 0.00028845910394565956, + "loss": 4.4664, + "step": 1310 + }, + { + "epoch": 0.14614298762766753, + "grad_norm": 0.9112701416015625, + "learning_rate": 0.00028825299283863137, + "loss": 4.4269, + "step": 1320 + }, + { + "epoch": 0.14725013147333169, + "grad_norm": 0.9069480299949646, + "learning_rate": 0.0002880451324661944, + "loss": 4.363, + "step": 1330 + }, + { + "epoch": 0.1483572753189958, + "grad_norm": 0.9578545093536377, + "learning_rate": 0.00028783552545833115, + "loss": 4.393, + "step": 1340 + }, + { + "epoch": 0.14946441916465997, + "grad_norm": 1.0073641538619995, + "learning_rate": 0.00028762417446712363, + "loss": 4.3806, + "step": 1350 + }, + { + "epoch": 0.15057156301032412, + "grad_norm": 0.8804338574409485, + "learning_rate": 0.00028741108216672015, + "loss": 4.5177, + "step": 1360 + }, + { + "epoch": 0.15167870685598828, + "grad_norm": 0.9387159943580627, + "learning_rate": 0.00028719625125330075, + "loss": 4.4155, + "step": 1370 + }, + { + "epoch": 0.1527858507016524, + "grad_norm": 0.8402112722396851, + "learning_rate": 0.00028697968444504386, + "loss": 4.3476, + "step": 1380 + }, + { + "epoch": 0.15389299454731656, + "grad_norm": 0.9387818574905396, + "learning_rate": 0.0002867613844820914, + "loss": 4.3451, + "step": 1390 + }, + { + "epoch": 0.1550001383929807, + "grad_norm": 0.9778910875320435, + "learning_rate": 0.0002865413541265143, + "loss": 4.4328, + "step": 1400 + }, + { + "epoch": 0.15610728223864487, + "grad_norm": 0.8600465655326843, + "learning_rate": 0.0002863195961622776, + "loss": 4.3858, + "step": 1410 + }, + { + "epoch": 0.157214426084309, + "grad_norm": 0.9855389595031738, + "learning_rate": 0.00028609611339520496, + "loss": 4.3716, + "step": 1420 + }, + { + "epoch": 0.15832156992997315, + "grad_norm": 0.899785578250885, + "learning_rate": 0.0002858709086529436, + "loss": 4.2872, + "step": 1430 + }, + { + "epoch": 0.1594287137756373, + "grad_norm": 0.8720322251319885, + "learning_rate": 0.0002856439847849281, + "loss": 4.3237, + "step": 1440 + }, + { + "epoch": 0.16053585762130146, + "grad_norm": 0.8622561097145081, + "learning_rate": 0.00028541534466234463, + "loss": 4.4075, + "step": 1450 + }, + { + "epoch": 0.16164300146696559, + "grad_norm": 0.9291304349899292, + "learning_rate": 0.0002851849911780943, + "loss": 4.4397, + "step": 1460 + }, + { + "epoch": 0.16275014531262974, + "grad_norm": 1.0178428888320923, + "learning_rate": 0.00028495292724675707, + "loss": 4.3703, + "step": 1470 + }, + { + "epoch": 0.1638572891582939, + "grad_norm": 0.9619563817977905, + "learning_rate": 0.0002847191558045544, + "loss": 4.3373, + "step": 1480 + }, + { + "epoch": 0.16496443300395805, + "grad_norm": 0.8943428993225098, + "learning_rate": 0.00028448367980931223, + "loss": 4.2764, + "step": 1490 + }, + { + "epoch": 0.16607157684962218, + "grad_norm": 0.9054276347160339, + "learning_rate": 0.0002842465022404237, + "loss": 4.3998, + "step": 1500 + }, + { + "epoch": 0.16717872069528633, + "grad_norm": 0.9450926184654236, + "learning_rate": 0.0002840076260988113, + "loss": 4.2571, + "step": 1510 + }, + { + "epoch": 0.16828586454095049, + "grad_norm": 0.9150902628898621, + "learning_rate": 0.0002837670544068891, + "loss": 4.3664, + "step": 1520 + }, + { + "epoch": 0.16939300838661464, + "grad_norm": 0.904039740562439, + "learning_rate": 0.00028352479020852395, + "loss": 4.3458, + "step": 1530 + }, + { + "epoch": 0.17050015223227877, + "grad_norm": 0.8239765763282776, + "learning_rate": 0.00028328083656899784, + "loss": 4.3421, + "step": 1540 + }, + { + "epoch": 0.17160729607794292, + "grad_norm": 0.9170050621032715, + "learning_rate": 0.0002830351965749684, + "loss": 4.2966, + "step": 1550 + }, + { + "epoch": 0.17271443992360708, + "grad_norm": 0.9204264283180237, + "learning_rate": 0.0002827878733344301, + "loss": 4.365, + "step": 1560 + }, + { + "epoch": 0.17382158376927123, + "grad_norm": 0.8800514340400696, + "learning_rate": 0.00028253886997667505, + "loss": 4.2917, + "step": 1570 + }, + { + "epoch": 0.17492872761493536, + "grad_norm": 0.94333815574646, + "learning_rate": 0.0002822881896522532, + "loss": 4.2403, + "step": 1580 + }, + { + "epoch": 0.1760358714605995, + "grad_norm": 0.8230079412460327, + "learning_rate": 0.0002820358355329326, + "loss": 4.3373, + "step": 1590 + }, + { + "epoch": 0.17714301530626367, + "grad_norm": 0.9081759452819824, + "learning_rate": 0.00028178181081165907, + "loss": 4.2739, + "step": 1600 + }, + { + "epoch": 0.17825015915192782, + "grad_norm": 0.9538459181785583, + "learning_rate": 0.0002815261187025162, + "loss": 4.299, + "step": 1610 + }, + { + "epoch": 0.17935730299759195, + "grad_norm": 0.9281736016273499, + "learning_rate": 0.00028126876244068415, + "loss": 4.3223, + "step": 1620 + }, + { + "epoch": 0.1804644468432561, + "grad_norm": 0.8907910585403442, + "learning_rate": 0.00028100974528239936, + "loss": 4.2654, + "step": 1630 + }, + { + "epoch": 0.18157159068892026, + "grad_norm": 0.9079713821411133, + "learning_rate": 0.0002807490705049127, + "loss": 4.2758, + "step": 1640 + }, + { + "epoch": 0.1826787345345844, + "grad_norm": 0.941882848739624, + "learning_rate": 0.0002804867414064483, + "loss": 4.2795, + "step": 1650 + }, + { + "epoch": 0.18378587838024854, + "grad_norm": 0.8275599479675293, + "learning_rate": 0.00028022276130616216, + "loss": 4.3271, + "step": 1660 + }, + { + "epoch": 0.1848930222259127, + "grad_norm": 0.9626572132110596, + "learning_rate": 0.0002799571335440994, + "loss": 4.2517, + "step": 1670 + }, + { + "epoch": 0.18600016607157685, + "grad_norm": 0.9606955051422119, + "learning_rate": 0.00027968986148115265, + "loss": 4.2689, + "step": 1680 + }, + { + "epoch": 0.187107309917241, + "grad_norm": 0.8812878131866455, + "learning_rate": 0.0002794209484990192, + "loss": 4.2801, + "step": 1690 + }, + { + "epoch": 0.18821445376290513, + "grad_norm": 0.8765847086906433, + "learning_rate": 0.0002791503980001583, + "loss": 4.2174, + "step": 1700 + }, + { + "epoch": 0.1893215976085693, + "grad_norm": 0.9191439151763916, + "learning_rate": 0.00027887821340774816, + "loss": 4.2512, + "step": 1710 + }, + { + "epoch": 0.19042874145423344, + "grad_norm": 0.9322866201400757, + "learning_rate": 0.0002786043981656426, + "loss": 4.2643, + "step": 1720 + }, + { + "epoch": 0.1915358852998976, + "grad_norm": 0.8535199761390686, + "learning_rate": 0.00027832895573832734, + "loss": 4.279, + "step": 1730 + }, + { + "epoch": 0.19264302914556175, + "grad_norm": 0.8324198126792908, + "learning_rate": 0.0002780518896108765, + "loss": 4.3407, + "step": 1740 + }, + { + "epoch": 0.19375017299122588, + "grad_norm": 0.8924321532249451, + "learning_rate": 0.0002777732032889081, + "loss": 4.1649, + "step": 1750 + }, + { + "epoch": 0.19485731683689003, + "grad_norm": 0.8889706134796143, + "learning_rate": 0.00027749290029853993, + "loss": 4.2351, + "step": 1760 + }, + { + "epoch": 0.1959644606825542, + "grad_norm": 0.8804425597190857, + "learning_rate": 0.00027721098418634495, + "loss": 4.3398, + "step": 1770 + }, + { + "epoch": 0.19707160452821834, + "grad_norm": 0.9893689155578613, + "learning_rate": 0.0002769274585193065, + "loss": 4.2616, + "step": 1780 + }, + { + "epoch": 0.19817874837388247, + "grad_norm": 1.0338963270187378, + "learning_rate": 0.0002766423268847727, + "loss": 4.2726, + "step": 1790 + }, + { + "epoch": 0.19928589221954662, + "grad_norm": 0.9166854619979858, + "learning_rate": 0.0002763555928904117, + "loss": 4.2709, + "step": 1800 + }, + { + "epoch": 0.20017160729607794, + "eval_loss": 4.702579021453857, + "eval_runtime": 207.9043, + "eval_samples_per_second": 240.495, + "eval_steps_per_second": 30.062, + "step": 1808 + }, + { + "epoch": 0.20039303606521078, + "grad_norm": 0.8702342510223389, + "learning_rate": 0.00027606726016416567, + "loss": 4.2119, + "step": 1810 + }, + { + "epoch": 0.20150017991087493, + "grad_norm": 0.9843454360961914, + "learning_rate": 0.00027577733235420487, + "loss": 4.2597, + "step": 1820 + }, + { + "epoch": 0.20260732375653906, + "grad_norm": 0.9212245345115662, + "learning_rate": 0.0002754858131288817, + "loss": 4.2382, + "step": 1830 + }, + { + "epoch": 0.20371446760220321, + "grad_norm": 1.0015140771865845, + "learning_rate": 0.000275192706176684, + "loss": 4.2796, + "step": 1840 + }, + { + "epoch": 0.20482161144786737, + "grad_norm": 0.9595922827720642, + "learning_rate": 0.00027489801520618867, + "loss": 4.2406, + "step": 1850 + }, + { + "epoch": 0.20592875529353152, + "grad_norm": 0.925445556640625, + "learning_rate": 0.0002746017439460146, + "loss": 4.2742, + "step": 1860 + }, + { + "epoch": 0.20703589913919565, + "grad_norm": 0.9547020196914673, + "learning_rate": 0.0002743038961447757, + "loss": 4.3034, + "step": 1870 + }, + { + "epoch": 0.2081430429848598, + "grad_norm": 0.9122299551963806, + "learning_rate": 0.0002740044755710328, + "loss": 4.1809, + "step": 1880 + }, + { + "epoch": 0.20925018683052396, + "grad_norm": 0.920444905757904, + "learning_rate": 0.000273703486013247, + "loss": 4.311, + "step": 1890 + }, + { + "epoch": 0.21035733067618811, + "grad_norm": 0.9135159850120544, + "learning_rate": 0.0002734009312797309, + "loss": 4.2509, + "step": 1900 + }, + { + "epoch": 0.21146447452185224, + "grad_norm": 0.9983066916465759, + "learning_rate": 0.0002730968151986007, + "loss": 4.2194, + "step": 1910 + }, + { + "epoch": 0.2125716183675164, + "grad_norm": 0.8982075452804565, + "learning_rate": 0.0002727911416177278, + "loss": 4.2059, + "step": 1920 + }, + { + "epoch": 0.21367876221318055, + "grad_norm": 0.9029157161712646, + "learning_rate": 0.0002724839144046902, + "loss": 4.2622, + "step": 1930 + }, + { + "epoch": 0.2147859060588447, + "grad_norm": 0.8970075845718384, + "learning_rate": 0.00027217513744672337, + "loss": 4.2886, + "step": 1940 + }, + { + "epoch": 0.21589304990450883, + "grad_norm": 0.9548723101615906, + "learning_rate": 0.000271864814650671, + "loss": 4.1917, + "step": 1950 + }, + { + "epoch": 0.217000193750173, + "grad_norm": 0.8849389553070068, + "learning_rate": 0.00027155294994293604, + "loss": 4.2806, + "step": 1960 + }, + { + "epoch": 0.21810733759583714, + "grad_norm": 0.8880273699760437, + "learning_rate": 0.0002712395472694304, + "loss": 4.1694, + "step": 1970 + }, + { + "epoch": 0.2192144814415013, + "grad_norm": 0.9328926205635071, + "learning_rate": 0.0002709246105955254, + "loss": 4.2009, + "step": 1980 + }, + { + "epoch": 0.22032162528716542, + "grad_norm": 0.9135597944259644, + "learning_rate": 0.0002706081439060016, + "loss": 4.1757, + "step": 1990 + }, + { + "epoch": 0.22142876913282958, + "grad_norm": 0.9954074025154114, + "learning_rate": 0.00027029015120499825, + "loss": 4.2423, + "step": 2000 + }, + { + "epoch": 0.22253591297849373, + "grad_norm": 0.8993476033210754, + "learning_rate": 0.00026997063651596257, + "loss": 4.3151, + "step": 2010 + }, + { + "epoch": 0.2236430568241579, + "grad_norm": 0.8682587742805481, + "learning_rate": 0.00026964960388159903, + "loss": 4.2915, + "step": 2020 + }, + { + "epoch": 0.22475020066982201, + "grad_norm": 0.9429246783256531, + "learning_rate": 0.00026932705736381826, + "loss": 4.2017, + "step": 2030 + }, + { + "epoch": 0.22585734451548617, + "grad_norm": 0.9352082014083862, + "learning_rate": 0.00026900300104368524, + "loss": 4.2973, + "step": 2040 + }, + { + "epoch": 0.22696448836115032, + "grad_norm": 0.983976423740387, + "learning_rate": 0.0002686774390213682, + "loss": 4.1693, + "step": 2050 + }, + { + "epoch": 0.22807163220681448, + "grad_norm": 0.8369963765144348, + "learning_rate": 0.00026835037541608623, + "loss": 4.256, + "step": 2060 + }, + { + "epoch": 0.2291787760524786, + "grad_norm": 0.9674303531646729, + "learning_rate": 0.00026802181436605757, + "loss": 4.1899, + "step": 2070 + }, + { + "epoch": 0.23028591989814276, + "grad_norm": 0.9745166301727295, + "learning_rate": 0.00026769176002844705, + "loss": 4.1351, + "step": 2080 + }, + { + "epoch": 0.23139306374380691, + "grad_norm": 0.9514241814613342, + "learning_rate": 0.0002673602165793136, + "loss": 4.2211, + "step": 2090 + }, + { + "epoch": 0.23250020758947107, + "grad_norm": 0.9059280157089233, + "learning_rate": 0.00026702718821355713, + "loss": 4.1147, + "step": 2100 + }, + { + "epoch": 0.2336073514351352, + "grad_norm": 0.9010078310966492, + "learning_rate": 0.00026669267914486596, + "loss": 4.1971, + "step": 2110 + }, + { + "epoch": 0.23471449528079935, + "grad_norm": 1.1091523170471191, + "learning_rate": 0.00026635669360566296, + "loss": 4.1756, + "step": 2120 + }, + { + "epoch": 0.2358216391264635, + "grad_norm": 0.8980200886726379, + "learning_rate": 0.0002660192358470524, + "loss": 4.1555, + "step": 2130 + }, + { + "epoch": 0.23692878297212766, + "grad_norm": 0.9868527054786682, + "learning_rate": 0.000265680310138766, + "loss": 4.1636, + "step": 2140 + }, + { + "epoch": 0.23803592681779182, + "grad_norm": 0.962148904800415, + "learning_rate": 0.00026533992076910876, + "loss": 4.1806, + "step": 2150 + }, + { + "epoch": 0.23914307066345594, + "grad_norm": 0.9093447327613831, + "learning_rate": 0.00026499807204490505, + "loss": 4.1583, + "step": 2160 + }, + { + "epoch": 0.2402502145091201, + "grad_norm": 0.8678917288780212, + "learning_rate": 0.00026465476829144394, + "loss": 4.196, + "step": 2170 + }, + { + "epoch": 0.24135735835478425, + "grad_norm": 0.9782723784446716, + "learning_rate": 0.00026431001385242433, + "loss": 4.1442, + "step": 2180 + }, + { + "epoch": 0.2424645022004484, + "grad_norm": 0.8897956609725952, + "learning_rate": 0.00026396381308990013, + "loss": 4.0803, + "step": 2190 + }, + { + "epoch": 0.24357164604611253, + "grad_norm": 0.9058307409286499, + "learning_rate": 0.00026361617038422517, + "loss": 4.1569, + "step": 2200 + }, + { + "epoch": 0.2446787898917767, + "grad_norm": 0.9241141080856323, + "learning_rate": 0.0002632670901339976, + "loss": 4.1793, + "step": 2210 + }, + { + "epoch": 0.24578593373744084, + "grad_norm": 0.8878569602966309, + "learning_rate": 0.00026291657675600426, + "loss": 4.0903, + "step": 2220 + }, + { + "epoch": 0.246893077583105, + "grad_norm": 0.9263564348220825, + "learning_rate": 0.00026256463468516495, + "loss": 4.1212, + "step": 2230 + }, + { + "epoch": 0.24800022142876912, + "grad_norm": 1.0177162885665894, + "learning_rate": 0.00026221126837447617, + "loss": 4.1407, + "step": 2240 + }, + { + "epoch": 0.24910736527443328, + "grad_norm": 0.9397019743919373, + "learning_rate": 0.00026185648229495477, + "loss": 4.2293, + "step": 2250 + }, + { + "epoch": 0.2502145091200974, + "grad_norm": 0.915478527545929, + "learning_rate": 0.0002615002809355814, + "loss": 4.1296, + "step": 2260 + }, + { + "epoch": 0.2513216529657616, + "grad_norm": 0.9102312922477722, + "learning_rate": 0.00026114266880324387, + "loss": 4.2228, + "step": 2270 + }, + { + "epoch": 0.2524287968114257, + "grad_norm": 0.9697858095169067, + "learning_rate": 0.00026078365042267985, + "loss": 4.1101, + "step": 2280 + }, + { + "epoch": 0.2535359406570899, + "grad_norm": 0.9541416764259338, + "learning_rate": 0.0002604232303364198, + "loss": 4.1467, + "step": 2290 + }, + { + "epoch": 0.254643084502754, + "grad_norm": 0.9324455857276917, + "learning_rate": 0.00026006141310472953, + "loss": 4.1651, + "step": 2300 + }, + { + "epoch": 0.25575022834841815, + "grad_norm": 0.881929337978363, + "learning_rate": 0.00025969820330555234, + "loss": 4.1573, + "step": 2310 + }, + { + "epoch": 0.25685737219408233, + "grad_norm": 1.0510601997375488, + "learning_rate": 0.0002593336055344513, + "loss": 4.1884, + "step": 2320 + }, + { + "epoch": 0.25796451603974646, + "grad_norm": 0.9759455919265747, + "learning_rate": 0.0002589676244045509, + "loss": 4.1501, + "step": 2330 + }, + { + "epoch": 0.2590716598854106, + "grad_norm": 0.8007738590240479, + "learning_rate": 0.0002586002645464787, + "loss": 4.2503, + "step": 2340 + }, + { + "epoch": 0.26017880373107477, + "grad_norm": 0.9922360777854919, + "learning_rate": 0.00025823153060830695, + "loss": 4.1691, + "step": 2350 + }, + { + "epoch": 0.2612859475767389, + "grad_norm": 0.910003125667572, + "learning_rate": 0.0002578614272554936, + "loss": 4.0687, + "step": 2360 + }, + { + "epoch": 0.2623930914224031, + "grad_norm": 0.9508435130119324, + "learning_rate": 0.0002574899591708233, + "loss": 4.1373, + "step": 2370 + }, + { + "epoch": 0.2635002352680672, + "grad_norm": 0.9929734468460083, + "learning_rate": 0.0002571171310543481, + "loss": 4.203, + "step": 2380 + }, + { + "epoch": 0.26460737911373133, + "grad_norm": 0.9429436922073364, + "learning_rate": 0.00025674294762332824, + "loss": 4.1433, + "step": 2390 + }, + { + "epoch": 0.2657145229593955, + "grad_norm": 0.9408954977989197, + "learning_rate": 0.00025636741361217204, + "loss": 4.055, + "step": 2400 + }, + { + "epoch": 0.26682166680505964, + "grad_norm": 0.8708512783050537, + "learning_rate": 0.0002559905337723763, + "loss": 4.1626, + "step": 2410 + }, + { + "epoch": 0.26792881065072377, + "grad_norm": 0.8821848630905151, + "learning_rate": 0.00025561231287246617, + "loss": 4.2259, + "step": 2420 + }, + { + "epoch": 0.26903595449638795, + "grad_norm": 0.8656236529350281, + "learning_rate": 0.00025523275569793466, + "loss": 4.1345, + "step": 2430 + }, + { + "epoch": 0.2701430983420521, + "grad_norm": 0.9382847547531128, + "learning_rate": 0.00025485186705118227, + "loss": 4.1867, + "step": 2440 + }, + { + "epoch": 0.27125024218771626, + "grad_norm": 0.9588480591773987, + "learning_rate": 0.00025446965175145614, + "loss": 4.1348, + "step": 2450 + }, + { + "epoch": 0.2723573860333804, + "grad_norm": 0.8816649317741394, + "learning_rate": 0.00025408611463478896, + "loss": 4.0905, + "step": 2460 + }, + { + "epoch": 0.2734645298790445, + "grad_norm": 0.999750554561615, + "learning_rate": 0.0002537012605539379, + "loss": 4.076, + "step": 2470 + }, + { + "epoch": 0.2745716737247087, + "grad_norm": 0.9172374606132507, + "learning_rate": 0.0002533150943783234, + "loss": 4.0246, + "step": 2480 + }, + { + "epoch": 0.2756788175703728, + "grad_norm": 0.9856157302856445, + "learning_rate": 0.0002529276209939672, + "loss": 4.1111, + "step": 2490 + }, + { + "epoch": 0.27678596141603695, + "grad_norm": 0.9478781819343567, + "learning_rate": 0.0002525388453034307, + "loss": 4.0386, + "step": 2500 + }, + { + "epoch": 0.27789310526170113, + "grad_norm": 0.9579760432243347, + "learning_rate": 0.00025214877222575285, + "loss": 4.0911, + "step": 2510 + }, + { + "epoch": 0.27900024910736526, + "grad_norm": 0.8776463866233826, + "learning_rate": 0.0002517574066963882, + "loss": 4.202, + "step": 2520 + }, + { + "epoch": 0.28010739295302944, + "grad_norm": 0.9522616267204285, + "learning_rate": 0.00025136475366714395, + "loss": 4.0841, + "step": 2530 + }, + { + "epoch": 0.28121453679869357, + "grad_norm": 1.0109864473342896, + "learning_rate": 0.0002509708181061178, + "loss": 4.0579, + "step": 2540 + }, + { + "epoch": 0.2823216806443577, + "grad_norm": 0.9109439253807068, + "learning_rate": 0.0002505756049976346, + "loss": 4.1536, + "step": 2550 + }, + { + "epoch": 0.2834288244900219, + "grad_norm": 0.9515150189399719, + "learning_rate": 0.00025017911934218387, + "loss": 4.0545, + "step": 2560 + }, + { + "epoch": 0.284535968335686, + "grad_norm": 0.9928112626075745, + "learning_rate": 0.000249781366156356, + "loss": 4.0397, + "step": 2570 + }, + { + "epoch": 0.2856431121813502, + "grad_norm": 0.8748706579208374, + "learning_rate": 0.0002493823504727791, + "loss": 4.0935, + "step": 2580 + }, + { + "epoch": 0.2867502560270143, + "grad_norm": 0.9956920742988586, + "learning_rate": 0.00024898207734005506, + "loss": 4.1184, + "step": 2590 + }, + { + "epoch": 0.28785739987267844, + "grad_norm": 0.8847795128822327, + "learning_rate": 0.00024858055182269586, + "loss": 4.0423, + "step": 2600 + }, + { + "epoch": 0.2889645437183426, + "grad_norm": 1.0538455247879028, + "learning_rate": 0.0002481777790010596, + "loss": 4.0091, + "step": 2610 + }, + { + "epoch": 0.29007168756400675, + "grad_norm": 0.8951561450958252, + "learning_rate": 0.00024777376397128593, + "loss": 4.1042, + "step": 2620 + }, + { + "epoch": 0.2911788314096709, + "grad_norm": 0.9119685292243958, + "learning_rate": 0.00024736851184523176, + "loss": 4.0518, + "step": 2630 + }, + { + "epoch": 0.29228597525533506, + "grad_norm": 0.9019333124160767, + "learning_rate": 0.0002469620277504066, + "loss": 4.0715, + "step": 2640 + }, + { + "epoch": 0.2933931191009992, + "grad_norm": 0.9932471513748169, + "learning_rate": 0.00024655431682990743, + "loss": 3.9567, + "step": 2650 + }, + { + "epoch": 0.29450026294666337, + "grad_norm": 0.8858083486557007, + "learning_rate": 0.000246145384242354, + "loss": 4.0748, + "step": 2660 + }, + { + "epoch": 0.2956074067923275, + "grad_norm": 1.0340659618377686, + "learning_rate": 0.0002457352351618233, + "loss": 4.0869, + "step": 2670 + }, + { + "epoch": 0.2967145506379916, + "grad_norm": 0.8488369584083557, + "learning_rate": 0.00024532387477778415, + "loss": 4.1156, + "step": 2680 + }, + { + "epoch": 0.2978216944836558, + "grad_norm": 0.8649202585220337, + "learning_rate": 0.00024491130829503165, + "loss": 4.0944, + "step": 2690 + }, + { + "epoch": 0.29892883832931993, + "grad_norm": 1.0052932500839233, + "learning_rate": 0.0002444975409336211, + "loss": 4.0797, + "step": 2700 + }, + { + "epoch": 0.30003598217498406, + "grad_norm": 0.9324126243591309, + "learning_rate": 0.00024408257792880216, + "loss": 4.0284, + "step": 2710 + }, + { + "epoch": 0.3002574109441169, + "eval_loss": 4.558005332946777, + "eval_runtime": 207.9765, + "eval_samples_per_second": 240.412, + "eval_steps_per_second": 30.051, + "step": 2712 + }, + { + "epoch": 0.30114312602064824, + "grad_norm": 0.9538119435310364, + "learning_rate": 0.00024366642453095254, + "loss": 4.141, + "step": 2720 + }, + { + "epoch": 0.30225026986631237, + "grad_norm": 0.9728295803070068, + "learning_rate": 0.00024324908600551162, + "loss": 4.0557, + "step": 2730 + }, + { + "epoch": 0.30335741371197655, + "grad_norm": 1.0474536418914795, + "learning_rate": 0.00024283056763291355, + "loss": 4.0449, + "step": 2740 + }, + { + "epoch": 0.3044645575576407, + "grad_norm": 0.8877320885658264, + "learning_rate": 0.00024241087470852095, + "loss": 3.9634, + "step": 2750 + }, + { + "epoch": 0.3055717014033048, + "grad_norm": 0.882987380027771, + "learning_rate": 0.00024199001254255744, + "loss": 4.1232, + "step": 2760 + }, + { + "epoch": 0.306678845248969, + "grad_norm": 0.9097815155982971, + "learning_rate": 0.00024156798646004065, + "loss": 4.0447, + "step": 2770 + }, + { + "epoch": 0.3077859890946331, + "grad_norm": 0.9631854295730591, + "learning_rate": 0.00024114480180071495, + "loss": 4.0401, + "step": 2780 + }, + { + "epoch": 0.30889313294029724, + "grad_norm": 0.9408524036407471, + "learning_rate": 0.00024072046391898362, + "loss": 4.1069, + "step": 2790 + }, + { + "epoch": 0.3100002767859614, + "grad_norm": 0.9656551480293274, + "learning_rate": 0.0002402949781838413, + "loss": 4.0857, + "step": 2800 + }, + { + "epoch": 0.31110742063162555, + "grad_norm": 0.9847767353057861, + "learning_rate": 0.000239868349978806, + "loss": 4.119, + "step": 2810 + }, + { + "epoch": 0.31221456447728974, + "grad_norm": 0.904832124710083, + "learning_rate": 0.00023944058470185106, + "loss": 4.0388, + "step": 2820 + }, + { + "epoch": 0.31332170832295386, + "grad_norm": 0.9817295670509338, + "learning_rate": 0.0002390116877653367, + "loss": 4.0994, + "step": 2830 + }, + { + "epoch": 0.314428852168618, + "grad_norm": 0.9541131258010864, + "learning_rate": 0.0002385816645959416, + "loss": 3.9753, + "step": 2840 + }, + { + "epoch": 0.31553599601428217, + "grad_norm": 0.9465216994285583, + "learning_rate": 0.00023815052063459437, + "loss": 4.0649, + "step": 2850 + }, + { + "epoch": 0.3166431398599463, + "grad_norm": 0.8634570240974426, + "learning_rate": 0.0002377182613364045, + "loss": 4.0477, + "step": 2860 + }, + { + "epoch": 0.3177502837056104, + "grad_norm": 0.9756444096565247, + "learning_rate": 0.0002372848921705934, + "loss": 3.9778, + "step": 2870 + }, + { + "epoch": 0.3188574275512746, + "grad_norm": 0.8720880150794983, + "learning_rate": 0.00023685041862042547, + "loss": 4.1032, + "step": 2880 + }, + { + "epoch": 0.31996457139693874, + "grad_norm": 0.8879233002662659, + "learning_rate": 0.00023641484618313818, + "loss": 3.9913, + "step": 2890 + }, + { + "epoch": 0.3210717152426029, + "grad_norm": 0.9097843766212463, + "learning_rate": 0.00023597818036987306, + "loss": 4.0956, + "step": 2900 + }, + { + "epoch": 0.32217885908826704, + "grad_norm": 0.9506983757019043, + "learning_rate": 0.00023554042670560554, + "loss": 4.1592, + "step": 2910 + }, + { + "epoch": 0.32328600293393117, + "grad_norm": 0.9886361956596375, + "learning_rate": 0.00023510159072907545, + "loss": 4.0177, + "step": 2920 + }, + { + "epoch": 0.32439314677959535, + "grad_norm": 0.8696989417076111, + "learning_rate": 0.0002346616779927165, + "loss": 4.0535, + "step": 2930 + }, + { + "epoch": 0.3255002906252595, + "grad_norm": 0.9655572175979614, + "learning_rate": 0.00023422069406258647, + "loss": 4.0486, + "step": 2940 + }, + { + "epoch": 0.3266074344709236, + "grad_norm": 0.9132712483406067, + "learning_rate": 0.00023377864451829642, + "loss": 4.015, + "step": 2950 + }, + { + "epoch": 0.3277145783165878, + "grad_norm": 0.9428880214691162, + "learning_rate": 0.0002333355349529403, + "loss": 4.0056, + "step": 2960 + }, + { + "epoch": 0.3288217221622519, + "grad_norm": 0.9156978726387024, + "learning_rate": 0.00023289137097302415, + "loss": 4.09, + "step": 2970 + }, + { + "epoch": 0.3299288660079161, + "grad_norm": 0.9907706379890442, + "learning_rate": 0.00023244615819839506, + "loss": 3.998, + "step": 2980 + }, + { + "epoch": 0.3310360098535802, + "grad_norm": 0.9192037582397461, + "learning_rate": 0.0002319999022621702, + "loss": 4.0077, + "step": 2990 + }, + { + "epoch": 0.33214315369924435, + "grad_norm": 0.9165595769882202, + "learning_rate": 0.00023155260881066548, + "loss": 4.064, + "step": 3000 + }, + { + "epoch": 0.33325029754490854, + "grad_norm": 0.9900383949279785, + "learning_rate": 0.0002311042835033241, + "loss": 4.0036, + "step": 3010 + }, + { + "epoch": 0.33435744139057266, + "grad_norm": 0.9811968207359314, + "learning_rate": 0.00023065493201264504, + "loss": 4.0194, + "step": 3020 + }, + { + "epoch": 0.33546458523623685, + "grad_norm": 0.9136149287223816, + "learning_rate": 0.0002302045600241111, + "loss": 4.1261, + "step": 3030 + }, + { + "epoch": 0.33657172908190097, + "grad_norm": 0.9643986225128174, + "learning_rate": 0.00022975317323611707, + "loss": 4.0431, + "step": 3040 + }, + { + "epoch": 0.3376788729275651, + "grad_norm": 0.985832691192627, + "learning_rate": 0.00022930077735989777, + "loss": 4.0042, + "step": 3050 + }, + { + "epoch": 0.3387860167732293, + "grad_norm": 0.9624693393707275, + "learning_rate": 0.0002288473781194555, + "loss": 3.992, + "step": 3060 + }, + { + "epoch": 0.3398931606188934, + "grad_norm": 1.0505192279815674, + "learning_rate": 0.0002283929812514879, + "loss": 4.0756, + "step": 3070 + }, + { + "epoch": 0.34100030446455754, + "grad_norm": 0.9822567701339722, + "learning_rate": 0.00022793759250531502, + "loss": 3.9195, + "step": 3080 + }, + { + "epoch": 0.3421074483102217, + "grad_norm": 0.9928208589553833, + "learning_rate": 0.00022748121764280707, + "loss": 3.9676, + "step": 3090 + }, + { + "epoch": 0.34321459215588584, + "grad_norm": 0.8698714971542358, + "learning_rate": 0.000227023862438311, + "loss": 4.0586, + "step": 3100 + }, + { + "epoch": 0.34432173600155, + "grad_norm": 0.9380168914794922, + "learning_rate": 0.0002265655326785778, + "loss": 4.0748, + "step": 3110 + }, + { + "epoch": 0.34542887984721415, + "grad_norm": 0.9094550013542175, + "learning_rate": 0.0002261062341626892, + "loss": 3.9673, + "step": 3120 + }, + { + "epoch": 0.3465360236928783, + "grad_norm": 0.8980678915977478, + "learning_rate": 0.00022564597270198408, + "loss": 4.0047, + "step": 3130 + }, + { + "epoch": 0.34764316753854246, + "grad_norm": 0.9330350756645203, + "learning_rate": 0.0002251847541199852, + "loss": 4.009, + "step": 3140 + }, + { + "epoch": 0.3487503113842066, + "grad_norm": 0.9750251173973083, + "learning_rate": 0.00022472258425232554, + "loss": 4.0146, + "step": 3150 + }, + { + "epoch": 0.3498574552298707, + "grad_norm": 0.8918101191520691, + "learning_rate": 0.00022425946894667417, + "loss": 4.1213, + "step": 3160 + }, + { + "epoch": 0.3509645990755349, + "grad_norm": 0.9062853455543518, + "learning_rate": 0.0002237954140626626, + "loss": 3.9943, + "step": 3170 + }, + { + "epoch": 0.352071742921199, + "grad_norm": 0.9306754469871521, + "learning_rate": 0.0002233304254718104, + "loss": 3.9215, + "step": 3180 + }, + { + "epoch": 0.3531788867668632, + "grad_norm": 0.9058662056922913, + "learning_rate": 0.000222864509057451, + "loss": 3.998, + "step": 3190 + }, + { + "epoch": 0.35428603061252734, + "grad_norm": 0.9954458475112915, + "learning_rate": 0.00022239767071465728, + "loss": 4.0558, + "step": 3200 + }, + { + "epoch": 0.35539317445819146, + "grad_norm": 1.0210260152816772, + "learning_rate": 0.00022192991635016697, + "loss": 3.9828, + "step": 3210 + }, + { + "epoch": 0.35650031830385565, + "grad_norm": 0.9782701730728149, + "learning_rate": 0.00022146125188230782, + "loss": 4.0702, + "step": 3220 + }, + { + "epoch": 0.3576074621495198, + "grad_norm": 1.004462480545044, + "learning_rate": 0.00022099168324092288, + "loss": 3.9741, + "step": 3230 + }, + { + "epoch": 0.3587146059951839, + "grad_norm": 1.0192267894744873, + "learning_rate": 0.00022052121636729525, + "loss": 4.0024, + "step": 3240 + }, + { + "epoch": 0.3598217498408481, + "grad_norm": 0.9733679890632629, + "learning_rate": 0.0002200498572140732, + "loss": 3.8844, + "step": 3250 + }, + { + "epoch": 0.3609288936865122, + "grad_norm": 0.9823547005653381, + "learning_rate": 0.00021957761174519458, + "loss": 3.9504, + "step": 3260 + }, + { + "epoch": 0.3620360375321764, + "grad_norm": 1.0874602794647217, + "learning_rate": 0.00021910448593581155, + "loss": 4.008, + "step": 3270 + }, + { + "epoch": 0.3631431813778405, + "grad_norm": 0.9385669827461243, + "learning_rate": 0.0002186304857722148, + "loss": 3.9816, + "step": 3280 + }, + { + "epoch": 0.36425032522350465, + "grad_norm": 0.9252676367759705, + "learning_rate": 0.00021815561725175802, + "loss": 4.019, + "step": 3290 + }, + { + "epoch": 0.3653574690691688, + "grad_norm": 1.0172173976898193, + "learning_rate": 0.00021767988638278182, + "loss": 3.9916, + "step": 3300 + }, + { + "epoch": 0.36646461291483295, + "grad_norm": 1.1126984357833862, + "learning_rate": 0.00021720329918453786, + "loss": 4.027, + "step": 3310 + }, + { + "epoch": 0.3675717567604971, + "grad_norm": 0.8856159448623657, + "learning_rate": 0.0002167258616871126, + "loss": 3.9863, + "step": 3320 + }, + { + "epoch": 0.36867890060616126, + "grad_norm": 0.9167330861091614, + "learning_rate": 0.00021624757993135107, + "loss": 3.9589, + "step": 3330 + }, + { + "epoch": 0.3697860444518254, + "grad_norm": 1.0685195922851562, + "learning_rate": 0.00021576845996878024, + "loss": 3.9226, + "step": 3340 + }, + { + "epoch": 0.3708931882974896, + "grad_norm": 1.0040068626403809, + "learning_rate": 0.00021528850786153283, + "loss": 3.954, + "step": 3350 + }, + { + "epoch": 0.3720003321431537, + "grad_norm": 1.0013192892074585, + "learning_rate": 0.00021480772968227025, + "loss": 3.9225, + "step": 3360 + }, + { + "epoch": 0.3731074759888178, + "grad_norm": 1.0042545795440674, + "learning_rate": 0.00021432613151410593, + "loss": 3.9725, + "step": 3370 + }, + { + "epoch": 0.374214619834482, + "grad_norm": 0.9191873073577881, + "learning_rate": 0.00021384371945052836, + "loss": 3.8897, + "step": 3380 + }, + { + "epoch": 0.37532176368014614, + "grad_norm": 0.9136641025543213, + "learning_rate": 0.00021336049959532388, + "loss": 3.9414, + "step": 3390 + }, + { + "epoch": 0.37642890752581026, + "grad_norm": 0.9560743570327759, + "learning_rate": 0.0002128764780624996, + "loss": 3.9665, + "step": 3400 + }, + { + "epoch": 0.37753605137147445, + "grad_norm": 1.0171836614608765, + "learning_rate": 0.00021239166097620596, + "loss": 3.9117, + "step": 3410 + }, + { + "epoch": 0.3786431952171386, + "grad_norm": 1.0259062051773071, + "learning_rate": 0.00021190605447065917, + "loss": 3.9211, + "step": 3420 + }, + { + "epoch": 0.37975033906280276, + "grad_norm": 0.9499109983444214, + "learning_rate": 0.00021141966469006386, + "loss": 3.9968, + "step": 3430 + }, + { + "epoch": 0.3808574829084669, + "grad_norm": 0.9641938805580139, + "learning_rate": 0.00021093249778853494, + "loss": 3.9237, + "step": 3440 + }, + { + "epoch": 0.381964626754131, + "grad_norm": 0.8998026847839355, + "learning_rate": 0.00021044455993002003, + "loss": 3.9339, + "step": 3450 + }, + { + "epoch": 0.3830717705997952, + "grad_norm": 0.925930917263031, + "learning_rate": 0.0002099558572882215, + "loss": 4.0184, + "step": 3460 + }, + { + "epoch": 0.3841789144454593, + "grad_norm": 0.8539071083068848, + "learning_rate": 0.00020946639604651815, + "loss": 3.929, + "step": 3470 + }, + { + "epoch": 0.3852860582911235, + "grad_norm": 1.0048716068267822, + "learning_rate": 0.00020897618239788704, + "loss": 3.9509, + "step": 3480 + }, + { + "epoch": 0.38639320213678763, + "grad_norm": 0.9404777884483337, + "learning_rate": 0.00020848522254482527, + "loss": 3.9742, + "step": 3490 + }, + { + "epoch": 0.38750034598245175, + "grad_norm": 0.9291903376579285, + "learning_rate": 0.00020799352269927122, + "loss": 4.0053, + "step": 3500 + }, + { + "epoch": 0.38860748982811594, + "grad_norm": 1.1116231679916382, + "learning_rate": 0.0002075010890825263, + "loss": 3.9242, + "step": 3510 + }, + { + "epoch": 0.38971463367378006, + "grad_norm": 0.9709257483482361, + "learning_rate": 0.0002070079279251761, + "loss": 3.9451, + "step": 3520 + }, + { + "epoch": 0.3908217775194442, + "grad_norm": 1.0539970397949219, + "learning_rate": 0.00020651404546701127, + "loss": 3.9622, + "step": 3530 + }, + { + "epoch": 0.3919289213651084, + "grad_norm": 1.0386793613433838, + "learning_rate": 0.00020601944795694906, + "loss": 3.948, + "step": 3540 + }, + { + "epoch": 0.3930360652107725, + "grad_norm": 0.9138751029968262, + "learning_rate": 0.000205524141652954, + "loss": 4.0322, + "step": 3550 + }, + { + "epoch": 0.3941432090564367, + "grad_norm": 0.9685201644897461, + "learning_rate": 0.00020502813282195846, + "loss": 4.0154, + "step": 3560 + }, + { + "epoch": 0.3952503529021008, + "grad_norm": 0.9906127452850342, + "learning_rate": 0.000204531427739784, + "loss": 3.8987, + "step": 3570 + }, + { + "epoch": 0.39635749674776494, + "grad_norm": 0.8722761273384094, + "learning_rate": 0.0002040340326910614, + "loss": 3.9028, + "step": 3580 + }, + { + "epoch": 0.3974646405934291, + "grad_norm": 1.0145385265350342, + "learning_rate": 0.0002035359539691514, + "loss": 3.9047, + "step": 3590 + }, + { + "epoch": 0.39857178443909325, + "grad_norm": 0.9298667311668396, + "learning_rate": 0.0002030371978760649, + "loss": 3.9114, + "step": 3600 + }, + { + "epoch": 0.3996789282847574, + "grad_norm": 1.027511477470398, + "learning_rate": 0.00020253777072238349, + "loss": 4.0121, + "step": 3610 + }, + { + "epoch": 0.4003432145921559, + "eval_loss": 4.452821254730225, + "eval_runtime": 207.75, + "eval_samples_per_second": 240.674, + "eval_steps_per_second": 30.084, + "step": 3616 + }, + { + "epoch": 0.40078607213042156, + "grad_norm": 0.9677231907844543, + "learning_rate": 0.0002020376788271793, + "loss": 3.9717, + "step": 3620 + }, + { + "epoch": 0.4018932159760857, + "grad_norm": 0.9805356860160828, + "learning_rate": 0.00020153692851793535, + "loss": 3.9047, + "step": 3630 + }, + { + "epoch": 0.40300035982174987, + "grad_norm": 0.9493558406829834, + "learning_rate": 0.0002010355261304652, + "loss": 3.9196, + "step": 3640 + }, + { + "epoch": 0.404107503667414, + "grad_norm": 0.9284310936927795, + "learning_rate": 0.00020053347800883298, + "loss": 3.869, + "step": 3650 + }, + { + "epoch": 0.4052146475130781, + "grad_norm": 0.8966405391693115, + "learning_rate": 0.0002000307905052731, + "loss": 4.033, + "step": 3660 + }, + { + "epoch": 0.4063217913587423, + "grad_norm": 0.901552140712738, + "learning_rate": 0.0001995274699801098, + "loss": 3.9329, + "step": 3670 + }, + { + "epoch": 0.40742893520440643, + "grad_norm": 1.0826923847198486, + "learning_rate": 0.00019902352280167665, + "loss": 3.9318, + "step": 3680 + }, + { + "epoch": 0.40853607905007056, + "grad_norm": 0.9739042520523071, + "learning_rate": 0.0001985189553462361, + "loss": 3.9249, + "step": 3690 + }, + { + "epoch": 0.40964322289573474, + "grad_norm": 1.0352777242660522, + "learning_rate": 0.00019801377399789882, + "loss": 3.9223, + "step": 3700 + }, + { + "epoch": 0.41075036674139886, + "grad_norm": 0.9062825441360474, + "learning_rate": 0.00019750798514854274, + "loss": 3.9928, + "step": 3710 + }, + { + "epoch": 0.41185751058706305, + "grad_norm": 0.939084529876709, + "learning_rate": 0.0001970015951977323, + "loss": 3.997, + "step": 3720 + }, + { + "epoch": 0.4129646544327272, + "grad_norm": 0.998667299747467, + "learning_rate": 0.0001964946105526375, + "loss": 4.0009, + "step": 3730 + }, + { + "epoch": 0.4140717982783913, + "grad_norm": 0.8764489889144897, + "learning_rate": 0.00019598703762795273, + "loss": 3.9858, + "step": 3740 + }, + { + "epoch": 0.4151789421240555, + "grad_norm": 1.0269612073898315, + "learning_rate": 0.0001954788828458157, + "loss": 3.9669, + "step": 3750 + }, + { + "epoch": 0.4162860859697196, + "grad_norm": 1.0808948278427124, + "learning_rate": 0.00019497015263572617, + "loss": 3.8883, + "step": 3760 + }, + { + "epoch": 0.41739322981538374, + "grad_norm": 1.0472171306610107, + "learning_rate": 0.00019446085343446462, + "loss": 3.997, + "step": 3770 + }, + { + "epoch": 0.4185003736610479, + "grad_norm": 0.9223423600196838, + "learning_rate": 0.00019395099168601065, + "loss": 3.9204, + "step": 3780 + }, + { + "epoch": 0.41960751750671205, + "grad_norm": 0.9433258771896362, + "learning_rate": 0.00019344057384146167, + "loss": 4.0019, + "step": 3790 + }, + { + "epoch": 0.42071466135237623, + "grad_norm": 1.0298986434936523, + "learning_rate": 0.0001929296063589512, + "loss": 3.9422, + "step": 3800 + }, + { + "epoch": 0.42182180519804036, + "grad_norm": 0.9166738390922546, + "learning_rate": 0.00019241809570356698, + "loss": 3.9174, + "step": 3810 + }, + { + "epoch": 0.4229289490437045, + "grad_norm": 1.0258166790008545, + "learning_rate": 0.00019190604834726943, + "loss": 3.9477, + "step": 3820 + }, + { + "epoch": 0.42403609288936867, + "grad_norm": 0.9840388894081116, + "learning_rate": 0.00019139347076880972, + "loss": 3.944, + "step": 3830 + }, + { + "epoch": 0.4251432367350328, + "grad_norm": 1.0016275644302368, + "learning_rate": 0.00019088036945364754, + "loss": 3.8937, + "step": 3840 + }, + { + "epoch": 0.426250380580697, + "grad_norm": 0.9131935834884644, + "learning_rate": 0.00019036675089386944, + "loss": 3.9342, + "step": 3850 + }, + { + "epoch": 0.4273575244263611, + "grad_norm": 0.9639363884925842, + "learning_rate": 0.00018985262158810636, + "loss": 3.9072, + "step": 3860 + }, + { + "epoch": 0.42846466827202523, + "grad_norm": 0.9744885563850403, + "learning_rate": 0.00018933798804145148, + "loss": 3.9492, + "step": 3870 + }, + { + "epoch": 0.4295718121176894, + "grad_norm": 0.9315047264099121, + "learning_rate": 0.0001888228567653781, + "loss": 3.9278, + "step": 3880 + }, + { + "epoch": 0.43067895596335354, + "grad_norm": 1.003890872001648, + "learning_rate": 0.000188307234277657, + "loss": 3.871, + "step": 3890 + }, + { + "epoch": 0.43178609980901766, + "grad_norm": 1.1092760562896729, + "learning_rate": 0.00018779112710227416, + "loss": 3.8867, + "step": 3900 + }, + { + "epoch": 0.43289324365468185, + "grad_norm": 0.9577327966690063, + "learning_rate": 0.00018727454176934808, + "loss": 3.963, + "step": 3910 + }, + { + "epoch": 0.434000387500346, + "grad_norm": 0.9155477285385132, + "learning_rate": 0.00018675748481504728, + "loss": 3.8575, + "step": 3920 + }, + { + "epoch": 0.43510753134601016, + "grad_norm": 0.9019640684127808, + "learning_rate": 0.0001862399627815074, + "loss": 3.8186, + "step": 3930 + }, + { + "epoch": 0.4362146751916743, + "grad_norm": 1.0801705121994019, + "learning_rate": 0.00018572198221674868, + "loss": 3.8844, + "step": 3940 + }, + { + "epoch": 0.4373218190373384, + "grad_norm": 0.9104591012001038, + "learning_rate": 0.00018520354967459303, + "loss": 3.9359, + "step": 3950 + }, + { + "epoch": 0.4384289628830026, + "grad_norm": 0.9868025779724121, + "learning_rate": 0.00018468467171458094, + "loss": 3.8625, + "step": 3960 + }, + { + "epoch": 0.4395361067286667, + "grad_norm": 1.0598353147506714, + "learning_rate": 0.00018416535490188866, + "loss": 3.8538, + "step": 3970 + }, + { + "epoch": 0.44064325057433085, + "grad_norm": 0.9690806865692139, + "learning_rate": 0.0001836456058072452, + "loss": 3.8496, + "step": 3980 + }, + { + "epoch": 0.44175039441999503, + "grad_norm": 1.0030450820922852, + "learning_rate": 0.00018312543100684888, + "loss": 3.8707, + "step": 3990 + }, + { + "epoch": 0.44285753826565916, + "grad_norm": 1.0722200870513916, + "learning_rate": 0.00018260483708228438, + "loss": 3.8381, + "step": 4000 + }, + { + "epoch": 0.44396468211132334, + "grad_norm": 0.9143560528755188, + "learning_rate": 0.00018208383062043957, + "loss": 3.8957, + "step": 4010 + }, + { + "epoch": 0.44507182595698747, + "grad_norm": 1.0451371669769287, + "learning_rate": 0.0001815624182134218, + "loss": 3.8399, + "step": 4020 + }, + { + "epoch": 0.4461789698026516, + "grad_norm": 1.0830594301223755, + "learning_rate": 0.00018104060645847476, + "loss": 3.8282, + "step": 4030 + }, + { + "epoch": 0.4472861136483158, + "grad_norm": 0.9718319773674011, + "learning_rate": 0.00018051840195789506, + "loss": 3.9186, + "step": 4040 + }, + { + "epoch": 0.4483932574939799, + "grad_norm": 0.9501128196716309, + "learning_rate": 0.00017999581131894843, + "loss": 3.9228, + "step": 4050 + }, + { + "epoch": 0.44950040133964403, + "grad_norm": 0.9850895404815674, + "learning_rate": 0.00017947284115378626, + "loss": 3.8867, + "step": 4060 + }, + { + "epoch": 0.4506075451853082, + "grad_norm": 0.9076871275901794, + "learning_rate": 0.00017894949807936212, + "loss": 3.9035, + "step": 4070 + }, + { + "epoch": 0.45171468903097234, + "grad_norm": 0.9681016206741333, + "learning_rate": 0.0001784257887173477, + "loss": 3.9556, + "step": 4080 + }, + { + "epoch": 0.4528218328766365, + "grad_norm": 1.0886319875717163, + "learning_rate": 0.0001779017196940493, + "loss": 3.9145, + "step": 4090 + }, + { + "epoch": 0.45392897672230065, + "grad_norm": 1.0340906381607056, + "learning_rate": 0.00017737729764032384, + "loss": 3.7999, + "step": 4100 + }, + { + "epoch": 0.4550361205679648, + "grad_norm": 1.0355632305145264, + "learning_rate": 0.00017685252919149493, + "loss": 3.8415, + "step": 4110 + }, + { + "epoch": 0.45614326441362896, + "grad_norm": 1.0527942180633545, + "learning_rate": 0.0001763274209872691, + "loss": 3.8713, + "step": 4120 + }, + { + "epoch": 0.4572504082592931, + "grad_norm": 0.8717536330223083, + "learning_rate": 0.00017580197967165167, + "loss": 3.8296, + "step": 4130 + }, + { + "epoch": 0.4583575521049572, + "grad_norm": 0.9657100439071655, + "learning_rate": 0.0001752762118928627, + "loss": 3.8306, + "step": 4140 + }, + { + "epoch": 0.4594646959506214, + "grad_norm": 0.9935140013694763, + "learning_rate": 0.00017475012430325285, + "loss": 3.8601, + "step": 4150 + }, + { + "epoch": 0.4605718397962855, + "grad_norm": 0.935280978679657, + "learning_rate": 0.00017422372355921923, + "loss": 3.86, + "step": 4160 + }, + { + "epoch": 0.4616789836419497, + "grad_norm": 0.9900361895561218, + "learning_rate": 0.00017369701632112119, + "loss": 3.8358, + "step": 4170 + }, + { + "epoch": 0.46278612748761383, + "grad_norm": 0.9922558665275574, + "learning_rate": 0.00017317000925319597, + "loss": 3.8748, + "step": 4180 + }, + { + "epoch": 0.46389327133327796, + "grad_norm": 0.9190822839736938, + "learning_rate": 0.00017264270902347454, + "loss": 3.9211, + "step": 4190 + }, + { + "epoch": 0.46500041517894214, + "grad_norm": 1.0557353496551514, + "learning_rate": 0.00017211512230369715, + "loss": 3.7962, + "step": 4200 + }, + { + "epoch": 0.46610755902460627, + "grad_norm": 0.993277907371521, + "learning_rate": 0.00017158725576922888, + "loss": 3.836, + "step": 4210 + }, + { + "epoch": 0.4672147028702704, + "grad_norm": 0.9307639598846436, + "learning_rate": 0.0001710591160989752, + "loss": 3.865, + "step": 4220 + }, + { + "epoch": 0.4683218467159346, + "grad_norm": 1.012987732887268, + "learning_rate": 0.00017053070997529737, + "loss": 3.8966, + "step": 4230 + }, + { + "epoch": 0.4694289905615987, + "grad_norm": 1.0008074045181274, + "learning_rate": 0.00017000204408392815, + "loss": 3.8718, + "step": 4240 + }, + { + "epoch": 0.4705361344072629, + "grad_norm": 0.9909558892250061, + "learning_rate": 0.00016947312511388695, + "loss": 3.8429, + "step": 4250 + }, + { + "epoch": 0.471643278252927, + "grad_norm": 0.9852434992790222, + "learning_rate": 0.00016894395975739534, + "loss": 3.7804, + "step": 4260 + }, + { + "epoch": 0.47275042209859114, + "grad_norm": 0.9223081469535828, + "learning_rate": 0.00016841455470979235, + "loss": 3.9232, + "step": 4270 + }, + { + "epoch": 0.4738575659442553, + "grad_norm": 0.939605712890625, + "learning_rate": 0.00016788491666944966, + "loss": 3.8322, + "step": 4280 + }, + { + "epoch": 0.47496470978991945, + "grad_norm": 1.0110855102539062, + "learning_rate": 0.00016735505233768695, + "loss": 3.8867, + "step": 4290 + }, + { + "epoch": 0.47607185363558363, + "grad_norm": 1.0295825004577637, + "learning_rate": 0.0001668249684186871, + "loss": 3.8241, + "step": 4300 + }, + { + "epoch": 0.47717899748124776, + "grad_norm": 0.9223111867904663, + "learning_rate": 0.00016629467161941133, + "loss": 3.8291, + "step": 4310 + }, + { + "epoch": 0.4782861413269119, + "grad_norm": 1.04537832736969, + "learning_rate": 0.00016576416864951443, + "loss": 3.8541, + "step": 4320 + }, + { + "epoch": 0.47939328517257607, + "grad_norm": 0.9531323313713074, + "learning_rate": 0.00016523346622125966, + "loss": 3.8293, + "step": 4330 + }, + { + "epoch": 0.4805004290182402, + "grad_norm": 0.955443799495697, + "learning_rate": 0.0001647025710494341, + "loss": 3.8109, + "step": 4340 + }, + { + "epoch": 0.4816075728639043, + "grad_norm": 0.9814046025276184, + "learning_rate": 0.0001641714898512634, + "loss": 3.7217, + "step": 4350 + }, + { + "epoch": 0.4827147167095685, + "grad_norm": 0.9783612489700317, + "learning_rate": 0.00016364022934632706, + "loss": 3.8977, + "step": 4360 + }, + { + "epoch": 0.48382186055523263, + "grad_norm": 1.022194504737854, + "learning_rate": 0.0001631087962564732, + "loss": 3.8461, + "step": 4370 + }, + { + "epoch": 0.4849290044008968, + "grad_norm": 0.9772112369537354, + "learning_rate": 0.00016257719730573359, + "loss": 3.9034, + "step": 4380 + }, + { + "epoch": 0.48603614824656094, + "grad_norm": 0.9720386266708374, + "learning_rate": 0.0001620454392202387, + "loss": 3.8692, + "step": 4390 + }, + { + "epoch": 0.48714329209222507, + "grad_norm": 0.9353125095367432, + "learning_rate": 0.00016151352872813239, + "loss": 3.8691, + "step": 4400 + }, + { + "epoch": 0.48825043593788925, + "grad_norm": 0.9452038407325745, + "learning_rate": 0.00016098147255948683, + "loss": 3.875, + "step": 4410 + }, + { + "epoch": 0.4893575797835534, + "grad_norm": 0.9751025438308716, + "learning_rate": 0.0001604492774462174, + "loss": 3.8193, + "step": 4420 + }, + { + "epoch": 0.4904647236292175, + "grad_norm": 0.9514757394790649, + "learning_rate": 0.0001599169501219975, + "loss": 3.8297, + "step": 4430 + }, + { + "epoch": 0.4915718674748817, + "grad_norm": 0.9499305486679077, + "learning_rate": 0.0001593844973221734, + "loss": 3.8569, + "step": 4440 + }, + { + "epoch": 0.4926790113205458, + "grad_norm": 0.890999436378479, + "learning_rate": 0.00015885192578367894, + "loss": 3.8513, + "step": 4450 + }, + { + "epoch": 0.49378615516621, + "grad_norm": 0.9929040670394897, + "learning_rate": 0.00015831924224495028, + "loss": 3.8011, + "step": 4460 + }, + { + "epoch": 0.4948932990118741, + "grad_norm": 0.9493831396102905, + "learning_rate": 0.00015778645344584078, + "loss": 3.9387, + "step": 4470 + }, + { + "epoch": 0.49600044285753825, + "grad_norm": 0.9402571320533752, + "learning_rate": 0.00015725356612753541, + "loss": 3.8295, + "step": 4480 + }, + { + "epoch": 0.49710758670320243, + "grad_norm": 1.04114830493927, + "learning_rate": 0.00015672058703246596, + "loss": 3.8357, + "step": 4490 + }, + { + "epoch": 0.49821473054886656, + "grad_norm": 1.0467301607131958, + "learning_rate": 0.00015618752290422512, + "loss": 3.8921, + "step": 4500 + }, + { + "epoch": 0.4993218743945307, + "grad_norm": 0.9718907475471497, + "learning_rate": 0.00015565438048748174, + "loss": 3.8605, + "step": 4510 + }, + { + "epoch": 0.5004290182401948, + "grad_norm": 0.9635398387908936, + "learning_rate": 0.00015512116652789516, + "loss": 3.8311, + "step": 4520 + }, + { + "epoch": 0.5004290182401948, + "eval_loss": 4.376184463500977, + "eval_runtime": 208.0482, + "eval_samples_per_second": 240.329, + "eval_steps_per_second": 30.041, + "step": 4520 + }, + { + "epoch": 0.501536162085859, + "grad_norm": 1.0902246236801147, + "learning_rate": 0.00015458788777202986, + "loss": 3.8679, + "step": 4530 + }, + { + "epoch": 0.5026433059315232, + "grad_norm": 1.0641001462936401, + "learning_rate": 0.00015405455096727014, + "loss": 3.8778, + "step": 4540 + }, + { + "epoch": 0.5037504497771873, + "grad_norm": 1.089313268661499, + "learning_rate": 0.00015352116286173486, + "loss": 3.8224, + "step": 4550 + }, + { + "epoch": 0.5048575936228514, + "grad_norm": 0.9489058256149292, + "learning_rate": 0.00015298773020419192, + "loss": 3.8234, + "step": 4560 + }, + { + "epoch": 0.5059647374685156, + "grad_norm": 1.0030834674835205, + "learning_rate": 0.000152454259743973, + "loss": 3.8101, + "step": 4570 + }, + { + "epoch": 0.5070718813141798, + "grad_norm": 1.0822967290878296, + "learning_rate": 0.00015192075823088796, + "loss": 3.8355, + "step": 4580 + }, + { + "epoch": 0.5081790251598439, + "grad_norm": 0.9892169833183289, + "learning_rate": 0.00015138723241513968, + "loss": 3.9134, + "step": 4590 + }, + { + "epoch": 0.509286169005508, + "grad_norm": 1.0702983140945435, + "learning_rate": 0.0001508536890472384, + "loss": 3.8784, + "step": 4600 + }, + { + "epoch": 0.5103933128511722, + "grad_norm": 0.9505767822265625, + "learning_rate": 0.0001503201348779166, + "loss": 3.8978, + "step": 4610 + }, + { + "epoch": 0.5115004566968363, + "grad_norm": 0.9748587608337402, + "learning_rate": 0.00014978657665804323, + "loss": 3.7873, + "step": 4620 + }, + { + "epoch": 0.5126076005425004, + "grad_norm": 1.0062190294265747, + "learning_rate": 0.00014925302113853866, + "loss": 3.77, + "step": 4630 + }, + { + "epoch": 0.5137147443881647, + "grad_norm": 1.081398844718933, + "learning_rate": 0.00014871947507028913, + "loss": 3.8228, + "step": 4640 + }, + { + "epoch": 0.5148218882338288, + "grad_norm": 0.9957180619239807, + "learning_rate": 0.00014818594520406105, + "loss": 3.8186, + "step": 4650 + }, + { + "epoch": 0.5159290320794929, + "grad_norm": 0.9224368333816528, + "learning_rate": 0.00014765243829041605, + "loss": 3.8198, + "step": 4660 + }, + { + "epoch": 0.517036175925157, + "grad_norm": 0.8943349719047546, + "learning_rate": 0.00014711896107962534, + "loss": 3.7956, + "step": 4670 + }, + { + "epoch": 0.5181433197708212, + "grad_norm": 1.2254526615142822, + "learning_rate": 0.00014658552032158412, + "loss": 3.7921, + "step": 4680 + }, + { + "epoch": 0.5192504636164854, + "grad_norm": 0.9659713506698608, + "learning_rate": 0.00014605212276572665, + "loss": 3.9634, + "step": 4690 + }, + { + "epoch": 0.5203576074621495, + "grad_norm": 0.9081024527549744, + "learning_rate": 0.0001455187751609403, + "loss": 3.7661, + "step": 4700 + }, + { + "epoch": 0.5214647513078137, + "grad_norm": 0.9679198265075684, + "learning_rate": 0.0001449854842554806, + "loss": 3.8149, + "step": 4710 + }, + { + "epoch": 0.5225718951534778, + "grad_norm": 1.01339852809906, + "learning_rate": 0.00014445225679688564, + "loss": 3.8154, + "step": 4720 + }, + { + "epoch": 0.5236790389991419, + "grad_norm": 1.0070582628250122, + "learning_rate": 0.00014391909953189068, + "loss": 3.8584, + "step": 4730 + }, + { + "epoch": 0.5247861828448062, + "grad_norm": 1.055215835571289, + "learning_rate": 0.00014338601920634296, + "loss": 3.7664, + "step": 4740 + }, + { + "epoch": 0.5258933266904703, + "grad_norm": 0.9933527112007141, + "learning_rate": 0.0001428530225651161, + "loss": 3.7796, + "step": 4750 + }, + { + "epoch": 0.5270004705361344, + "grad_norm": 0.9595291614532471, + "learning_rate": 0.00014232011635202505, + "loss": 3.7712, + "step": 4760 + }, + { + "epoch": 0.5281076143817985, + "grad_norm": 1.058785080909729, + "learning_rate": 0.00014178730730974042, + "loss": 3.7558, + "step": 4770 + }, + { + "epoch": 0.5292147582274627, + "grad_norm": 0.9517855048179626, + "learning_rate": 0.00014125460217970345, + "loss": 3.8083, + "step": 4780 + }, + { + "epoch": 0.5303219020731269, + "grad_norm": 1.0455268621444702, + "learning_rate": 0.00014072200770204072, + "loss": 3.847, + "step": 4790 + }, + { + "epoch": 0.531429045918791, + "grad_norm": 0.9897139072418213, + "learning_rate": 0.0001401895306154785, + "loss": 3.744, + "step": 4800 + }, + { + "epoch": 0.5325361897644552, + "grad_norm": 1.0442320108413696, + "learning_rate": 0.0001396571776572581, + "loss": 3.8683, + "step": 4810 + }, + { + "epoch": 0.5336433336101193, + "grad_norm": 0.9899879097938538, + "learning_rate": 0.00013912495556304993, + "loss": 3.763, + "step": 4820 + }, + { + "epoch": 0.5347504774557834, + "grad_norm": 0.9290923476219177, + "learning_rate": 0.00013859287106686888, + "loss": 3.858, + "step": 4830 + }, + { + "epoch": 0.5358576213014475, + "grad_norm": 1.0133739709854126, + "learning_rate": 0.0001380609309009887, + "loss": 3.8757, + "step": 4840 + }, + { + "epoch": 0.5369647651471118, + "grad_norm": 1.0935516357421875, + "learning_rate": 0.00013752914179585707, + "loss": 3.8552, + "step": 4850 + }, + { + "epoch": 0.5380719089927759, + "grad_norm": 0.9974127411842346, + "learning_rate": 0.00013699751048001037, + "loss": 3.8158, + "step": 4860 + }, + { + "epoch": 0.53917905283844, + "grad_norm": 1.0367913246154785, + "learning_rate": 0.0001364660436799884, + "loss": 3.9148, + "step": 4870 + }, + { + "epoch": 0.5402861966841042, + "grad_norm": 1.0697216987609863, + "learning_rate": 0.00013593474812024966, + "loss": 3.7587, + "step": 4880 + }, + { + "epoch": 0.5413933405297683, + "grad_norm": 0.971474826335907, + "learning_rate": 0.00013540363052308562, + "loss": 3.8424, + "step": 4890 + }, + { + "epoch": 0.5425004843754325, + "grad_norm": 0.9516545534133911, + "learning_rate": 0.0001348726976085364, + "loss": 3.783, + "step": 4900 + }, + { + "epoch": 0.5436076282210967, + "grad_norm": 0.9858337640762329, + "learning_rate": 0.00013434195609430527, + "loss": 3.7905, + "step": 4910 + }, + { + "epoch": 0.5447147720667608, + "grad_norm": 1.041650652885437, + "learning_rate": 0.0001338114126956737, + "loss": 3.8258, + "step": 4920 + }, + { + "epoch": 0.5458219159124249, + "grad_norm": 1.0418676137924194, + "learning_rate": 0.0001332810741254167, + "loss": 3.8448, + "step": 4930 + }, + { + "epoch": 0.546929059758089, + "grad_norm": 1.0340111255645752, + "learning_rate": 0.0001327509470937173, + "loss": 3.7752, + "step": 4940 + }, + { + "epoch": 0.5480362036037533, + "grad_norm": 0.9103614687919617, + "learning_rate": 0.00013222103830808245, + "loss": 3.8245, + "step": 4950 + }, + { + "epoch": 0.5491433474494174, + "grad_norm": 1.1148477792739868, + "learning_rate": 0.00013169135447325733, + "loss": 3.7546, + "step": 4960 + }, + { + "epoch": 0.5502504912950815, + "grad_norm": 0.9774050712585449, + "learning_rate": 0.00013116190229114105, + "loss": 3.8423, + "step": 4970 + }, + { + "epoch": 0.5513576351407456, + "grad_norm": 0.9648851752281189, + "learning_rate": 0.00013063268846070182, + "loss": 3.8319, + "step": 4980 + }, + { + "epoch": 0.5524647789864098, + "grad_norm": 0.9702296853065491, + "learning_rate": 0.00013010371967789184, + "loss": 3.8478, + "step": 4990 + }, + { + "epoch": 0.5535719228320739, + "grad_norm": 0.9151732325553894, + "learning_rate": 0.00012957500263556304, + "loss": 3.8208, + "step": 5000 + }, + { + "epoch": 0.5546790666777381, + "grad_norm": 1.0036016702651978, + "learning_rate": 0.00012904654402338188, + "loss": 3.8151, + "step": 5010 + }, + { + "epoch": 0.5557862105234023, + "grad_norm": 0.9555637836456299, + "learning_rate": 0.00012851835052774536, + "loss": 3.7445, + "step": 5020 + }, + { + "epoch": 0.5568933543690664, + "grad_norm": 0.9780600666999817, + "learning_rate": 0.00012799042883169574, + "loss": 3.7619, + "step": 5030 + }, + { + "epoch": 0.5580004982147305, + "grad_norm": 0.9166061282157898, + "learning_rate": 0.00012746278561483643, + "loss": 3.8112, + "step": 5040 + }, + { + "epoch": 0.5591076420603946, + "grad_norm": 0.9202462434768677, + "learning_rate": 0.00012693542755324745, + "loss": 3.784, + "step": 5050 + }, + { + "epoch": 0.5602147859060589, + "grad_norm": 0.943366289138794, + "learning_rate": 0.0001264083613194006, + "loss": 3.7174, + "step": 5060 + }, + { + "epoch": 0.561321929751723, + "grad_norm": 1.0439735651016235, + "learning_rate": 0.00012588159358207552, + "loss": 3.8562, + "step": 5070 + }, + { + "epoch": 0.5624290735973871, + "grad_norm": 1.0018138885498047, + "learning_rate": 0.00012535513100627493, + "loss": 3.8369, + "step": 5080 + }, + { + "epoch": 0.5635362174430513, + "grad_norm": 1.0322473049163818, + "learning_rate": 0.0001248289802531406, + "loss": 3.8229, + "step": 5090 + }, + { + "epoch": 0.5646433612887154, + "grad_norm": 1.3071224689483643, + "learning_rate": 0.00012430314797986868, + "loss": 3.7767, + "step": 5100 + }, + { + "epoch": 0.5657505051343796, + "grad_norm": 0.8921655416488647, + "learning_rate": 0.00012377764083962593, + "loss": 3.8729, + "step": 5110 + }, + { + "epoch": 0.5668576489800438, + "grad_norm": 0.9934424161911011, + "learning_rate": 0.0001232524654814653, + "loss": 3.7913, + "step": 5120 + }, + { + "epoch": 0.5679647928257079, + "grad_norm": 0.9575469493865967, + "learning_rate": 0.0001227276285502417, + "loss": 3.7724, + "step": 5130 + }, + { + "epoch": 0.569071936671372, + "grad_norm": 1.0097877979278564, + "learning_rate": 0.0001222031366865282, + "loss": 3.775, + "step": 5140 + }, + { + "epoch": 0.5701790805170361, + "grad_norm": 0.9770298600196838, + "learning_rate": 0.00012167899652653166, + "loss": 3.8522, + "step": 5150 + }, + { + "epoch": 0.5712862243627004, + "grad_norm": 1.0553005933761597, + "learning_rate": 0.00012115521470200909, + "loss": 3.7464, + "step": 5160 + }, + { + "epoch": 0.5723933682083645, + "grad_norm": 0.9398956298828125, + "learning_rate": 0.00012063179784018367, + "loss": 3.7815, + "step": 5170 + }, + { + "epoch": 0.5735005120540286, + "grad_norm": 0.9554306268692017, + "learning_rate": 0.00012010875256366065, + "loss": 3.7807, + "step": 5180 + }, + { + "epoch": 0.5746076558996928, + "grad_norm": 0.924888551235199, + "learning_rate": 0.00011958608549034393, + "loss": 3.7586, + "step": 5190 + }, + { + "epoch": 0.5757147997453569, + "grad_norm": 1.1002237796783447, + "learning_rate": 0.00011906380323335202, + "loss": 3.799, + "step": 5200 + }, + { + "epoch": 0.576821943591021, + "grad_norm": 0.9993613362312317, + "learning_rate": 0.00011854191240093454, + "loss": 3.7364, + "step": 5210 + }, + { + "epoch": 0.5779290874366853, + "grad_norm": 0.9640263915061951, + "learning_rate": 0.00011802041959638852, + "loss": 3.8022, + "step": 5220 + }, + { + "epoch": 0.5790362312823494, + "grad_norm": 0.8847324252128601, + "learning_rate": 0.00011749933141797491, + "loss": 3.8197, + "step": 5230 + }, + { + "epoch": 0.5801433751280135, + "grad_norm": 1.0666438341140747, + "learning_rate": 0.00011697865445883518, + "loss": 3.8027, + "step": 5240 + }, + { + "epoch": 0.5812505189736776, + "grad_norm": 0.9423792362213135, + "learning_rate": 0.0001164583953069075, + "loss": 3.8125, + "step": 5250 + }, + { + "epoch": 0.5823576628193418, + "grad_norm": 1.0213701725006104, + "learning_rate": 0.00011593856054484402, + "loss": 3.7968, + "step": 5260 + }, + { + "epoch": 0.583464806665006, + "grad_norm": 0.9245741963386536, + "learning_rate": 0.00011541915674992698, + "loss": 3.8163, + "step": 5270 + }, + { + "epoch": 0.5845719505106701, + "grad_norm": 0.9317986369132996, + "learning_rate": 0.00011490019049398592, + "loss": 3.7613, + "step": 5280 + }, + { + "epoch": 0.5856790943563343, + "grad_norm": 1.0499074459075928, + "learning_rate": 0.00011438166834331415, + "loss": 3.7187, + "step": 5290 + }, + { + "epoch": 0.5867862382019984, + "grad_norm": 1.0535813570022583, + "learning_rate": 0.00011386359685858608, + "loss": 3.8065, + "step": 5300 + }, + { + "epoch": 0.5878933820476625, + "grad_norm": 0.969600260257721, + "learning_rate": 0.00011334598259477403, + "loss": 3.883, + "step": 5310 + }, + { + "epoch": 0.5890005258933267, + "grad_norm": 0.997873067855835, + "learning_rate": 0.00011282883210106502, + "loss": 3.7765, + "step": 5320 + }, + { + "epoch": 0.5901076697389909, + "grad_norm": 1.055496096611023, + "learning_rate": 0.00011231215192077837, + "loss": 3.7698, + "step": 5330 + }, + { + "epoch": 0.591214813584655, + "grad_norm": 0.9975765347480774, + "learning_rate": 0.00011179594859128261, + "loss": 3.7522, + "step": 5340 + }, + { + "epoch": 0.5923219574303191, + "grad_norm": 1.0061254501342773, + "learning_rate": 0.00011128022864391282, + "loss": 3.7933, + "step": 5350 + }, + { + "epoch": 0.5934291012759833, + "grad_norm": 0.9184295535087585, + "learning_rate": 0.00011076499860388822, + "loss": 3.778, + "step": 5360 + }, + { + "epoch": 0.5945362451216474, + "grad_norm": 1.034346580505371, + "learning_rate": 0.00011025026499022907, + "loss": 3.8745, + "step": 5370 + }, + { + "epoch": 0.5956433889673116, + "grad_norm": 1.0777544975280762, + "learning_rate": 0.00010973603431567483, + "loss": 3.8131, + "step": 5380 + }, + { + "epoch": 0.5967505328129757, + "grad_norm": 1.0810277462005615, + "learning_rate": 0.00010922231308660126, + "loss": 3.7126, + "step": 5390 + }, + { + "epoch": 0.5978576766586399, + "grad_norm": 1.0664763450622559, + "learning_rate": 0.00010870910780293843, + "loss": 3.79, + "step": 5400 + }, + { + "epoch": 0.598964820504304, + "grad_norm": 0.9999032020568848, + "learning_rate": 0.00010819642495808816, + "loss": 3.7428, + "step": 5410 + }, + { + "epoch": 0.6000719643499681, + "grad_norm": 0.9453679919242859, + "learning_rate": 0.00010768427103884222, + "loss": 3.7442, + "step": 5420 + }, + { + "epoch": 0.6005148218882338, + "eval_loss": 4.307396411895752, + "eval_runtime": 207.875, + "eval_samples_per_second": 240.529, + "eval_steps_per_second": 30.066, + "step": 5424 + }, + { + "epoch": 0.6011791081956324, + "grad_norm": 0.932433545589447, + "learning_rate": 0.00010717265252530005, + "loss": 3.7816, + "step": 5430 + }, + { + "epoch": 0.6022862520412965, + "grad_norm": 1.0366543531417847, + "learning_rate": 0.00010666157589078666, + "loss": 3.6868, + "step": 5440 + }, + { + "epoch": 0.6033933958869606, + "grad_norm": 1.1073342561721802, + "learning_rate": 0.00010615104760177099, + "loss": 3.7804, + "step": 5450 + }, + { + "epoch": 0.6045005397326247, + "grad_norm": 1.066766619682312, + "learning_rate": 0.00010564107411778389, + "loss": 3.818, + "step": 5460 + }, + { + "epoch": 0.6056076835782889, + "grad_norm": 1.057884693145752, + "learning_rate": 0.00010513166189133652, + "loss": 3.7558, + "step": 5470 + }, + { + "epoch": 0.6067148274239531, + "grad_norm": 0.9807105660438538, + "learning_rate": 0.00010462281736783863, + "loss": 3.736, + "step": 5480 + }, + { + "epoch": 0.6078219712696172, + "grad_norm": 1.0097297430038452, + "learning_rate": 0.00010411454698551695, + "loss": 3.8311, + "step": 5490 + }, + { + "epoch": 0.6089291151152814, + "grad_norm": 0.9198644757270813, + "learning_rate": 0.00010360685717533399, + "loss": 3.726, + "step": 5500 + }, + { + "epoch": 0.6100362589609455, + "grad_norm": 0.9404564499855042, + "learning_rate": 0.00010309975436090626, + "loss": 3.7543, + "step": 5510 + }, + { + "epoch": 0.6111434028066096, + "grad_norm": 0.9767935872077942, + "learning_rate": 0.00010259324495842343, + "loss": 3.7479, + "step": 5520 + }, + { + "epoch": 0.6122505466522737, + "grad_norm": 0.9311277866363525, + "learning_rate": 0.00010208733537656674, + "loss": 3.7773, + "step": 5530 + }, + { + "epoch": 0.613357690497938, + "grad_norm": 1.1350255012512207, + "learning_rate": 0.00010158203201642832, + "loss": 3.7339, + "step": 5540 + }, + { + "epoch": 0.6144648343436021, + "grad_norm": 0.9760239720344543, + "learning_rate": 0.0001010773412714299, + "loss": 3.7157, + "step": 5550 + }, + { + "epoch": 0.6155719781892662, + "grad_norm": 0.9780420660972595, + "learning_rate": 0.00010057326952724198, + "loss": 3.7762, + "step": 5560 + }, + { + "epoch": 0.6166791220349304, + "grad_norm": 1.0622854232788086, + "learning_rate": 0.00010006982316170313, + "loss": 3.7895, + "step": 5570 + }, + { + "epoch": 0.6177862658805945, + "grad_norm": 1.030948281288147, + "learning_rate": 9.956700854473915e-05, + "loss": 3.7788, + "step": 5580 + }, + { + "epoch": 0.6188934097262587, + "grad_norm": 0.9838833212852478, + "learning_rate": 9.906483203828267e-05, + "loss": 3.7424, + "step": 5590 + }, + { + "epoch": 0.6200005535719229, + "grad_norm": 0.9516918659210205, + "learning_rate": 9.856329999619234e-05, + "loss": 3.8105, + "step": 5600 + }, + { + "epoch": 0.621107697417587, + "grad_norm": 0.9829215407371521, + "learning_rate": 9.806241876417283e-05, + "loss": 3.7303, + "step": 5610 + }, + { + "epoch": 0.6222148412632511, + "grad_norm": 1.0436818599700928, + "learning_rate": 9.756219467969439e-05, + "loss": 3.6888, + "step": 5620 + }, + { + "epoch": 0.6233219851089152, + "grad_norm": 1.025931477546692, + "learning_rate": 9.706263407191233e-05, + "loss": 3.7063, + "step": 5630 + }, + { + "epoch": 0.6244291289545795, + "grad_norm": 1.0877312421798706, + "learning_rate": 9.656374326158751e-05, + "loss": 3.7192, + "step": 5640 + }, + { + "epoch": 0.6255362728002436, + "grad_norm": 1.0679664611816406, + "learning_rate": 9.606552856100587e-05, + "loss": 3.7071, + "step": 5650 + }, + { + "epoch": 0.6266434166459077, + "grad_norm": 1.053995966911316, + "learning_rate": 9.556799627389895e-05, + "loss": 3.7224, + "step": 5660 + }, + { + "epoch": 0.6277505604915719, + "grad_norm": 1.0624290704727173, + "learning_rate": 9.507115269536393e-05, + "loss": 3.7319, + "step": 5670 + }, + { + "epoch": 0.628857704337236, + "grad_norm": 1.114103078842163, + "learning_rate": 9.457500411178387e-05, + "loss": 3.6996, + "step": 5680 + }, + { + "epoch": 0.6299648481829002, + "grad_norm": 1.0226870775222778, + "learning_rate": 9.407955680074851e-05, + "loss": 3.8006, + "step": 5690 + }, + { + "epoch": 0.6310719920285643, + "grad_norm": 1.0463651418685913, + "learning_rate": 9.358481703097444e-05, + "loss": 3.7138, + "step": 5700 + }, + { + "epoch": 0.6321791358742285, + "grad_norm": 0.9294445514678955, + "learning_rate": 9.309079106222617e-05, + "loss": 3.6723, + "step": 5710 + }, + { + "epoch": 0.6332862797198926, + "grad_norm": 0.9535320997238159, + "learning_rate": 9.259748514523653e-05, + "loss": 3.7882, + "step": 5720 + }, + { + "epoch": 0.6343934235655567, + "grad_norm": 1.0739271640777588, + "learning_rate": 9.210490552162802e-05, + "loss": 3.7159, + "step": 5730 + }, + { + "epoch": 0.6355005674112209, + "grad_norm": 1.0555413961410522, + "learning_rate": 9.16130584238335e-05, + "loss": 3.8402, + "step": 5740 + }, + { + "epoch": 0.6366077112568851, + "grad_norm": 1.1149111986160278, + "learning_rate": 9.112195007501741e-05, + "loss": 3.7407, + "step": 5750 + }, + { + "epoch": 0.6377148551025492, + "grad_norm": 0.922713577747345, + "learning_rate": 9.063158668899718e-05, + "loss": 3.722, + "step": 5760 + }, + { + "epoch": 0.6388219989482133, + "grad_norm": 1.0295876264572144, + "learning_rate": 9.014197447016429e-05, + "loss": 3.7612, + "step": 5770 + }, + { + "epoch": 0.6399291427938775, + "grad_norm": 1.0831807851791382, + "learning_rate": 8.965311961340625e-05, + "loss": 3.7426, + "step": 5780 + }, + { + "epoch": 0.6410362866395416, + "grad_norm": 0.9424430727958679, + "learning_rate": 8.916502830402766e-05, + "loss": 3.8186, + "step": 5790 + }, + { + "epoch": 0.6421434304852058, + "grad_norm": 0.9513488411903381, + "learning_rate": 8.867770671767245e-05, + "loss": 3.7378, + "step": 5800 + }, + { + "epoch": 0.64325057433087, + "grad_norm": 1.1490607261657715, + "learning_rate": 8.819116102024543e-05, + "loss": 3.723, + "step": 5810 + }, + { + "epoch": 0.6443577181765341, + "grad_norm": 1.0015475749969482, + "learning_rate": 8.770539736783434e-05, + "loss": 3.7859, + "step": 5820 + }, + { + "epoch": 0.6454648620221982, + "grad_norm": 1.0109034776687622, + "learning_rate": 8.722042190663202e-05, + "loss": 3.7789, + "step": 5830 + }, + { + "epoch": 0.6465720058678623, + "grad_norm": 1.0395208597183228, + "learning_rate": 8.673624077285862e-05, + "loss": 3.7096, + "step": 5840 + }, + { + "epoch": 0.6476791497135266, + "grad_norm": 1.0548279285430908, + "learning_rate": 8.625286009268392e-05, + "loss": 3.8512, + "step": 5850 + }, + { + "epoch": 0.6487862935591907, + "grad_norm": 1.0581004619598389, + "learning_rate": 8.577028598214987e-05, + "loss": 3.721, + "step": 5860 + }, + { + "epoch": 0.6498934374048548, + "grad_norm": 0.9412915110588074, + "learning_rate": 8.528852454709317e-05, + "loss": 3.7625, + "step": 5870 + }, + { + "epoch": 0.651000581250519, + "grad_norm": 1.0232173204421997, + "learning_rate": 8.480758188306806e-05, + "loss": 3.8293, + "step": 5880 + }, + { + "epoch": 0.6521077250961831, + "grad_norm": 1.0306179523468018, + "learning_rate": 8.432746407526912e-05, + "loss": 3.7437, + "step": 5890 + }, + { + "epoch": 0.6532148689418472, + "grad_norm": 1.0161283016204834, + "learning_rate": 8.384817719845437e-05, + "loss": 3.7166, + "step": 5900 + }, + { + "epoch": 0.6543220127875115, + "grad_norm": 0.999501645565033, + "learning_rate": 8.336972731686829e-05, + "loss": 3.755, + "step": 5910 + }, + { + "epoch": 0.6554291566331756, + "grad_norm": 1.040917158126831, + "learning_rate": 8.289212048416525e-05, + "loss": 3.6961, + "step": 5920 + }, + { + "epoch": 0.6565363004788397, + "grad_norm": 0.9548594355583191, + "learning_rate": 8.241536274333274e-05, + "loss": 3.7875, + "step": 5930 + }, + { + "epoch": 0.6576434443245038, + "grad_norm": 1.1434271335601807, + "learning_rate": 8.193946012661506e-05, + "loss": 3.7004, + "step": 5940 + }, + { + "epoch": 0.658750588170168, + "grad_norm": 0.9456180334091187, + "learning_rate": 8.146441865543689e-05, + "loss": 3.7524, + "step": 5950 + }, + { + "epoch": 0.6598577320158322, + "grad_norm": 0.976205050945282, + "learning_rate": 8.099024434032717e-05, + "loss": 3.7097, + "step": 5960 + }, + { + "epoch": 0.6609648758614963, + "grad_norm": 0.9842648506164551, + "learning_rate": 8.051694318084302e-05, + "loss": 3.7636, + "step": 5970 + }, + { + "epoch": 0.6620720197071605, + "grad_norm": 0.9501277804374695, + "learning_rate": 8.004452116549383e-05, + "loss": 3.7765, + "step": 5980 + }, + { + "epoch": 0.6631791635528246, + "grad_norm": 1.0266845226287842, + "learning_rate": 7.95729842716655e-05, + "loss": 3.7146, + "step": 5990 + }, + { + "epoch": 0.6642863073984887, + "grad_norm": 1.0417708158493042, + "learning_rate": 7.910233846554479e-05, + "loss": 3.6865, + "step": 6000 + }, + { + "epoch": 0.665393451244153, + "grad_norm": 1.0642472505569458, + "learning_rate": 7.863258970204388e-05, + "loss": 3.6825, + "step": 6010 + }, + { + "epoch": 0.6665005950898171, + "grad_norm": 0.9697334170341492, + "learning_rate": 7.816374392472497e-05, + "loss": 3.748, + "step": 6020 + }, + { + "epoch": 0.6676077389354812, + "grad_norm": 0.9792755842208862, + "learning_rate": 7.76958070657251e-05, + "loss": 3.7177, + "step": 6030 + }, + { + "epoch": 0.6687148827811453, + "grad_norm": 1.0015872716903687, + "learning_rate": 7.722878504568109e-05, + "loss": 3.6781, + "step": 6040 + }, + { + "epoch": 0.6698220266268095, + "grad_norm": 1.003909945487976, + "learning_rate": 7.676268377365466e-05, + "loss": 3.7518, + "step": 6050 + }, + { + "epoch": 0.6709291704724737, + "grad_norm": 1.093845248222351, + "learning_rate": 7.629750914705761e-05, + "loss": 3.7164, + "step": 6060 + }, + { + "epoch": 0.6720363143181378, + "grad_norm": 1.010558843612671, + "learning_rate": 7.583326705157721e-05, + "loss": 3.7515, + "step": 6070 + }, + { + "epoch": 0.6731434581638019, + "grad_norm": 0.9907075762748718, + "learning_rate": 7.53699633611018e-05, + "loss": 3.7339, + "step": 6080 + }, + { + "epoch": 0.6742506020094661, + "grad_norm": 1.017988681793213, + "learning_rate": 7.49076039376464e-05, + "loss": 3.7288, + "step": 6090 + }, + { + "epoch": 0.6753577458551302, + "grad_norm": 1.0418598651885986, + "learning_rate": 7.444619463127853e-05, + "loss": 3.8575, + "step": 6100 + }, + { + "epoch": 0.6764648897007943, + "grad_norm": 1.084040641784668, + "learning_rate": 7.398574128004426e-05, + "loss": 3.7484, + "step": 6110 + }, + { + "epoch": 0.6775720335464586, + "grad_norm": 1.0427488088607788, + "learning_rate": 7.352624970989424e-05, + "loss": 3.7441, + "step": 6120 + }, + { + "epoch": 0.6786791773921227, + "grad_norm": 0.9982630610466003, + "learning_rate": 7.306772573461017e-05, + "loss": 3.6737, + "step": 6130 + }, + { + "epoch": 0.6797863212377868, + "grad_norm": 1.0794692039489746, + "learning_rate": 7.261017515573098e-05, + "loss": 3.7014, + "step": 6140 + }, + { + "epoch": 0.6808934650834509, + "grad_norm": 0.9856328368186951, + "learning_rate": 7.215360376247964e-05, + "loss": 3.738, + "step": 6150 + }, + { + "epoch": 0.6820006089291151, + "grad_norm": 1.0592615604400635, + "learning_rate": 7.169801733168987e-05, + "loss": 3.8042, + "step": 6160 + }, + { + "epoch": 0.6831077527747793, + "grad_norm": 0.9542872905731201, + "learning_rate": 7.124342162773293e-05, + "loss": 3.7479, + "step": 6170 + }, + { + "epoch": 0.6842148966204434, + "grad_norm": 1.1304407119750977, + "learning_rate": 7.07898224024448e-05, + "loss": 3.7512, + "step": 6180 + }, + { + "epoch": 0.6853220404661076, + "grad_norm": 0.9870150685310364, + "learning_rate": 7.033722539505341e-05, + "loss": 3.7327, + "step": 6190 + }, + { + "epoch": 0.6864291843117717, + "grad_norm": 1.041024923324585, + "learning_rate": 6.988563633210597e-05, + "loss": 3.6614, + "step": 6200 + }, + { + "epoch": 0.6875363281574358, + "grad_norm": 1.0168806314468384, + "learning_rate": 6.94350609273965e-05, + "loss": 3.7174, + "step": 6210 + }, + { + "epoch": 0.6886434720031, + "grad_norm": 1.0131466388702393, + "learning_rate": 6.89855048818936e-05, + "loss": 3.7817, + "step": 6220 + }, + { + "epoch": 0.6897506158487642, + "grad_norm": 0.9758023619651794, + "learning_rate": 6.853697388366826e-05, + "loss": 3.7546, + "step": 6230 + }, + { + "epoch": 0.6908577596944283, + "grad_norm": 0.9726858139038086, + "learning_rate": 6.808947360782194e-05, + "loss": 3.787, + "step": 6240 + }, + { + "epoch": 0.6919649035400924, + "grad_norm": 0.9504683613777161, + "learning_rate": 6.764300971641474e-05, + "loss": 3.7665, + "step": 6250 + }, + { + "epoch": 0.6930720473857566, + "grad_norm": 1.1621665954589844, + "learning_rate": 6.719758785839373e-05, + "loss": 3.7421, + "step": 6260 + }, + { + "epoch": 0.6941791912314207, + "grad_norm": 1.0132784843444824, + "learning_rate": 6.675321366952153e-05, + "loss": 3.7578, + "step": 6270 + }, + { + "epoch": 0.6952863350770849, + "grad_norm": 0.9619897603988647, + "learning_rate": 6.6309892772305e-05, + "loss": 3.798, + "step": 6280 + }, + { + "epoch": 0.696393478922749, + "grad_norm": 1.0339223146438599, + "learning_rate": 6.5867630775924e-05, + "loss": 3.6826, + "step": 6290 + }, + { + "epoch": 0.6975006227684132, + "grad_norm": 1.0287562608718872, + "learning_rate": 6.542643327616062e-05, + "loss": 3.7703, + "step": 6300 + }, + { + "epoch": 0.6986077666140773, + "grad_norm": 1.0636241436004639, + "learning_rate": 6.498630585532813e-05, + "loss": 3.6555, + "step": 6310 + }, + { + "epoch": 0.6997149104597414, + "grad_norm": 0.8900817632675171, + "learning_rate": 6.454725408220057e-05, + "loss": 3.7343, + "step": 6320 + }, + { + "epoch": 0.7006006255362728, + "eval_loss": 4.260047912597656, + "eval_runtime": 207.9965, + "eval_samples_per_second": 240.389, + "eval_steps_per_second": 30.049, + "step": 6328 + }, + { + "epoch": 0.7008220543054057, + "grad_norm": 1.082946538925171, + "learning_rate": 6.410928351194216e-05, + "loss": 3.6329, + "step": 6330 + }, + { + "epoch": 0.7019291981510698, + "grad_norm": 1.0826716423034668, + "learning_rate": 6.367239968603704e-05, + "loss": 3.7056, + "step": 6340 + }, + { + "epoch": 0.7030363419967339, + "grad_norm": 1.1230672597885132, + "learning_rate": 6.323660813221919e-05, + "loss": 3.7472, + "step": 6350 + }, + { + "epoch": 0.704143485842398, + "grad_norm": 1.0083682537078857, + "learning_rate": 6.280191436440243e-05, + "loss": 3.775, + "step": 6360 + }, + { + "epoch": 0.7052506296880622, + "grad_norm": 1.0960787534713745, + "learning_rate": 6.236832388261071e-05, + "loss": 3.7019, + "step": 6370 + }, + { + "epoch": 0.7063577735337264, + "grad_norm": 1.1135135889053345, + "learning_rate": 6.19358421729085e-05, + "loss": 3.7374, + "step": 6380 + }, + { + "epoch": 0.7074649173793905, + "grad_norm": 0.9962683916091919, + "learning_rate": 6.150447470733139e-05, + "loss": 3.6669, + "step": 6390 + }, + { + "epoch": 0.7085720612250547, + "grad_norm": 1.0689204931259155, + "learning_rate": 6.10742269438168e-05, + "loss": 3.7288, + "step": 6400 + }, + { + "epoch": 0.7096792050707188, + "grad_norm": 1.0299161672592163, + "learning_rate": 6.064510432613499e-05, + "loss": 3.7152, + "step": 6410 + }, + { + "epoch": 0.7107863489163829, + "grad_norm": 1.0302820205688477, + "learning_rate": 6.0217112283820164e-05, + "loss": 3.7506, + "step": 6420 + }, + { + "epoch": 0.7118934927620472, + "grad_norm": 0.9921202659606934, + "learning_rate": 5.979025623210173e-05, + "loss": 3.733, + "step": 6430 + }, + { + "epoch": 0.7130006366077113, + "grad_norm": 1.0907381772994995, + "learning_rate": 5.936454157183582e-05, + "loss": 3.723, + "step": 6440 + }, + { + "epoch": 0.7141077804533754, + "grad_norm": 1.0985629558563232, + "learning_rate": 5.8939973689436986e-05, + "loss": 3.6942, + "step": 6450 + }, + { + "epoch": 0.7152149242990395, + "grad_norm": 0.9944116473197937, + "learning_rate": 5.851655795680997e-05, + "loss": 3.6958, + "step": 6460 + }, + { + "epoch": 0.7163220681447037, + "grad_norm": 1.031113624572754, + "learning_rate": 5.809429973128188e-05, + "loss": 3.6456, + "step": 6470 + }, + { + "epoch": 0.7174292119903678, + "grad_norm": 1.0297033786773682, + "learning_rate": 5.767320435553402e-05, + "loss": 3.6857, + "step": 6480 + }, + { + "epoch": 0.718536355836032, + "grad_norm": 1.1770987510681152, + "learning_rate": 5.725327715753496e-05, + "loss": 3.6317, + "step": 6490 + }, + { + "epoch": 0.7196434996816962, + "grad_norm": 1.0155738592147827, + "learning_rate": 5.683452345047249e-05, + "loss": 3.734, + "step": 6500 + }, + { + "epoch": 0.7207506435273603, + "grad_norm": 1.1366873979568481, + "learning_rate": 5.641694853268668e-05, + "loss": 3.6604, + "step": 6510 + }, + { + "epoch": 0.7218577873730244, + "grad_norm": 1.0000633001327515, + "learning_rate": 5.600055768760286e-05, + "loss": 3.7297, + "step": 6520 + }, + { + "epoch": 0.7229649312186885, + "grad_norm": 1.1106007099151611, + "learning_rate": 5.5585356183664645e-05, + "loss": 3.7134, + "step": 6530 + }, + { + "epoch": 0.7240720750643528, + "grad_norm": 0.9138331413269043, + "learning_rate": 5.51713492742674e-05, + "loss": 3.6558, + "step": 6540 + }, + { + "epoch": 0.7251792189100169, + "grad_norm": 1.0178135633468628, + "learning_rate": 5.475854219769166e-05, + "loss": 3.7602, + "step": 6550 + }, + { + "epoch": 0.726286362755681, + "grad_norm": 1.0721853971481323, + "learning_rate": 5.434694017703697e-05, + "loss": 3.7504, + "step": 6560 + }, + { + "epoch": 0.7273935066013452, + "grad_norm": 1.131288766860962, + "learning_rate": 5.3936548420155697e-05, + "loss": 3.7644, + "step": 6570 + }, + { + "epoch": 0.7285006504470093, + "grad_norm": 1.0158541202545166, + "learning_rate": 5.352737211958718e-05, + "loss": 3.6988, + "step": 6580 + }, + { + "epoch": 0.7296077942926735, + "grad_norm": 0.9620969891548157, + "learning_rate": 5.311941645249202e-05, + "loss": 3.6741, + "step": 6590 + }, + { + "epoch": 0.7307149381383377, + "grad_norm": 1.0111806392669678, + "learning_rate": 5.27126865805866e-05, + "loss": 3.6139, + "step": 6600 + }, + { + "epoch": 0.7318220819840018, + "grad_norm": 1.0611088275909424, + "learning_rate": 5.2307187650077754e-05, + "loss": 3.6492, + "step": 6610 + }, + { + "epoch": 0.7329292258296659, + "grad_norm": 1.1051170825958252, + "learning_rate": 5.190292479159762e-05, + "loss": 3.6957, + "step": 6620 + }, + { + "epoch": 0.73403636967533, + "grad_norm": 1.127403736114502, + "learning_rate": 5.1499903120138836e-05, + "loss": 3.7367, + "step": 6630 + }, + { + "epoch": 0.7351435135209942, + "grad_norm": 1.0560911893844604, + "learning_rate": 5.109812773498967e-05, + "loss": 3.616, + "step": 6640 + }, + { + "epoch": 0.7362506573666584, + "grad_norm": 1.04979407787323, + "learning_rate": 5.069760371966961e-05, + "loss": 3.8031, + "step": 6650 + }, + { + "epoch": 0.7373578012123225, + "grad_norm": 1.068346619606018, + "learning_rate": 5.029833614186508e-05, + "loss": 3.6949, + "step": 6660 + }, + { + "epoch": 0.7384649450579867, + "grad_norm": 1.064867615699768, + "learning_rate": 4.9900330053365016e-05, + "loss": 3.6225, + "step": 6670 + }, + { + "epoch": 0.7395720889036508, + "grad_norm": 1.0293861627578735, + "learning_rate": 4.9503590489997526e-05, + "loss": 3.6972, + "step": 6680 + }, + { + "epoch": 0.7406792327493149, + "grad_norm": 1.023012638092041, + "learning_rate": 4.9108122471565575e-05, + "loss": 3.6976, + "step": 6690 + }, + { + "epoch": 0.7417863765949791, + "grad_norm": 0.9692345261573792, + "learning_rate": 4.871393100178383e-05, + "loss": 3.5928, + "step": 6700 + }, + { + "epoch": 0.7428935204406433, + "grad_norm": 0.9640194773674011, + "learning_rate": 4.8321021068215213e-05, + "loss": 3.6253, + "step": 6710 + }, + { + "epoch": 0.7440006642863074, + "grad_norm": 1.0556492805480957, + "learning_rate": 4.792939764220783e-05, + "loss": 3.7034, + "step": 6720 + }, + { + "epoch": 0.7451078081319715, + "grad_norm": 0.9859820008277893, + "learning_rate": 4.7539065678832124e-05, + "loss": 3.7282, + "step": 6730 + }, + { + "epoch": 0.7462149519776357, + "grad_norm": 0.9723745584487915, + "learning_rate": 4.715003011681802e-05, + "loss": 3.6685, + "step": 6740 + }, + { + "epoch": 0.7473220958232999, + "grad_norm": 1.0325344800949097, + "learning_rate": 4.676229587849266e-05, + "loss": 3.6618, + "step": 6750 + }, + { + "epoch": 0.748429239668964, + "grad_norm": 1.0485987663269043, + "learning_rate": 4.6375867869717907e-05, + "loss": 3.6496, + "step": 6760 + }, + { + "epoch": 0.7495363835146281, + "grad_norm": 1.019169807434082, + "learning_rate": 4.599075097982845e-05, + "loss": 3.7998, + "step": 6770 + }, + { + "epoch": 0.7506435273602923, + "grad_norm": 1.0943533182144165, + "learning_rate": 4.5606950081569844e-05, + "loss": 3.6559, + "step": 6780 + }, + { + "epoch": 0.7517506712059564, + "grad_norm": 0.9578700065612793, + "learning_rate": 4.522447003103673e-05, + "loss": 3.6334, + "step": 6790 + }, + { + "epoch": 0.7528578150516205, + "grad_norm": 1.0839344263076782, + "learning_rate": 4.4843315667611805e-05, + "loss": 3.7387, + "step": 6800 + }, + { + "epoch": 0.7539649588972848, + "grad_norm": 1.0185452699661255, + "learning_rate": 4.446349181390415e-05, + "loss": 3.7342, + "step": 6810 + }, + { + "epoch": 0.7550721027429489, + "grad_norm": 0.9841724634170532, + "learning_rate": 4.408500327568839e-05, + "loss": 3.6741, + "step": 6820 + }, + { + "epoch": 0.756179246588613, + "grad_norm": 1.1085660457611084, + "learning_rate": 4.370785484184392e-05, + "loss": 3.6902, + "step": 6830 + }, + { + "epoch": 0.7572863904342771, + "grad_norm": 1.0940853357315063, + "learning_rate": 4.3332051284294294e-05, + "loss": 3.7004, + "step": 6840 + }, + { + "epoch": 0.7583935342799413, + "grad_norm": 1.032070517539978, + "learning_rate": 4.295759735794677e-05, + "loss": 3.7579, + "step": 6850 + }, + { + "epoch": 0.7595006781256055, + "grad_norm": 1.1027837991714478, + "learning_rate": 4.258449780063224e-05, + "loss": 3.6729, + "step": 6860 + }, + { + "epoch": 0.7606078219712696, + "grad_norm": 1.0637125968933105, + "learning_rate": 4.2212757333045283e-05, + "loss": 3.6168, + "step": 6870 + }, + { + "epoch": 0.7617149658169338, + "grad_norm": 0.9541471600532532, + "learning_rate": 4.184238065868437e-05, + "loss": 3.7032, + "step": 6880 + }, + { + "epoch": 0.7628221096625979, + "grad_norm": 0.9935281276702881, + "learning_rate": 4.147337246379235e-05, + "loss": 3.6974, + "step": 6890 + }, + { + "epoch": 0.763929253508262, + "grad_norm": 1.025411605834961, + "learning_rate": 4.1105737417297344e-05, + "loss": 3.7179, + "step": 6900 + }, + { + "epoch": 0.7650363973539263, + "grad_norm": 1.1179848909378052, + "learning_rate": 4.0739480170753206e-05, + "loss": 3.6828, + "step": 6910 + }, + { + "epoch": 0.7661435411995904, + "grad_norm": 1.090013861656189, + "learning_rate": 4.0374605358281376e-05, + "loss": 3.6958, + "step": 6920 + }, + { + "epoch": 0.7672506850452545, + "grad_norm": 1.0558772087097168, + "learning_rate": 4.00111175965116e-05, + "loss": 3.7651, + "step": 6930 + }, + { + "epoch": 0.7683578288909186, + "grad_norm": 1.135996699333191, + "learning_rate": 3.964902148452383e-05, + "loss": 3.6412, + "step": 6940 + }, + { + "epoch": 0.7694649727365828, + "grad_norm": 1.0703589916229248, + "learning_rate": 3.9288321603789986e-05, + "loss": 3.6319, + "step": 6950 + }, + { + "epoch": 0.770572116582247, + "grad_norm": 1.1619700193405151, + "learning_rate": 3.892902251811595e-05, + "loss": 3.7381, + "step": 6960 + }, + { + "epoch": 0.7716792604279111, + "grad_norm": 1.0365691184997559, + "learning_rate": 3.857112877358391e-05, + "loss": 3.7131, + "step": 6970 + }, + { + "epoch": 0.7727864042735753, + "grad_norm": 0.9841147661209106, + "learning_rate": 3.8214644898494636e-05, + "loss": 3.7535, + "step": 6980 + }, + { + "epoch": 0.7738935481192394, + "grad_norm": 1.025385856628418, + "learning_rate": 3.7859575403310534e-05, + "loss": 3.6783, + "step": 6990 + }, + { + "epoch": 0.7750006919649035, + "grad_norm": 1.1125547885894775, + "learning_rate": 3.75059247805982e-05, + "loss": 3.6559, + "step": 7000 + }, + { + "epoch": 0.7761078358105676, + "grad_norm": 1.0286872386932373, + "learning_rate": 3.715369750497182e-05, + "loss": 3.6792, + "step": 7010 + }, + { + "epoch": 0.7772149796562319, + "grad_norm": 1.1428214311599731, + "learning_rate": 3.68028980330365e-05, + "loss": 3.6337, + "step": 7020 + }, + { + "epoch": 0.778322123501896, + "grad_norm": 1.0305466651916504, + "learning_rate": 3.645353080333174e-05, + "loss": 3.7038, + "step": 7030 + }, + { + "epoch": 0.7794292673475601, + "grad_norm": 1.098001480102539, + "learning_rate": 3.610560023627555e-05, + "loss": 3.705, + "step": 7040 + }, + { + "epoch": 0.7805364111932243, + "grad_norm": 1.0811655521392822, + "learning_rate": 3.57591107341083e-05, + "loss": 3.6624, + "step": 7050 + }, + { + "epoch": 0.7816435550388884, + "grad_norm": 1.0603693723678589, + "learning_rate": 3.541406668083704e-05, + "loss": 3.7823, + "step": 7060 + }, + { + "epoch": 0.7827506988845526, + "grad_norm": 1.0910733938217163, + "learning_rate": 3.507047244218014e-05, + "loss": 3.641, + "step": 7070 + }, + { + "epoch": 0.7838578427302167, + "grad_norm": 0.9448225498199463, + "learning_rate": 3.4728332365511954e-05, + "loss": 3.7483, + "step": 7080 + }, + { + "epoch": 0.7849649865758809, + "grad_norm": 1.0241824388504028, + "learning_rate": 3.43876507798079e-05, + "loss": 3.6946, + "step": 7090 + }, + { + "epoch": 0.786072130421545, + "grad_norm": 0.9677013158798218, + "learning_rate": 3.404843199558945e-05, + "loss": 3.695, + "step": 7100 + }, + { + "epoch": 0.7871792742672091, + "grad_norm": 1.1355061531066895, + "learning_rate": 3.3710680304870056e-05, + "loss": 3.7133, + "step": 7110 + }, + { + "epoch": 0.7882864181128734, + "grad_norm": 1.012036681175232, + "learning_rate": 3.3374399981100354e-05, + "loss": 3.6428, + "step": 7120 + }, + { + "epoch": 0.7893935619585375, + "grad_norm": 0.9935883283615112, + "learning_rate": 3.30395952791144e-05, + "loss": 3.6946, + "step": 7130 + }, + { + "epoch": 0.7905007058042016, + "grad_norm": 1.1006075143814087, + "learning_rate": 3.2706270435075716e-05, + "loss": 3.7463, + "step": 7140 + }, + { + "epoch": 0.7916078496498657, + "grad_norm": 1.0247091054916382, + "learning_rate": 3.2374429666423614e-05, + "loss": 3.7416, + "step": 7150 + }, + { + "epoch": 0.7927149934955299, + "grad_norm": 1.0335090160369873, + "learning_rate": 3.2044077171820156e-05, + "loss": 3.7086, + "step": 7160 + }, + { + "epoch": 0.793822137341194, + "grad_norm": 1.0554313659667969, + "learning_rate": 3.1715217131096554e-05, + "loss": 3.6972, + "step": 7170 + }, + { + "epoch": 0.7949292811868582, + "grad_norm": 1.0465693473815918, + "learning_rate": 3.138785370520073e-05, + "loss": 3.6397, + "step": 7180 + }, + { + "epoch": 0.7960364250325224, + "grad_norm": 1.0831694602966309, + "learning_rate": 3.106199103614441e-05, + "loss": 3.5906, + "step": 7190 + }, + { + "epoch": 0.7971435688781865, + "grad_norm": 1.0017293691635132, + "learning_rate": 3.073763324695078e-05, + "loss": 3.6436, + "step": 7200 + }, + { + "epoch": 0.7982507127238506, + "grad_norm": 1.0355520248413086, + "learning_rate": 3.041478444160233e-05, + "loss": 3.6523, + "step": 7210 + }, + { + "epoch": 0.7993578565695147, + "grad_norm": 1.1570299863815308, + "learning_rate": 3.0093448704988815e-05, + "loss": 3.6559, + "step": 7220 + }, + { + "epoch": 0.800465000415179, + "grad_norm": 1.1363317966461182, + "learning_rate": 2.977363010285588e-05, + "loss": 3.7604, + "step": 7230 + }, + { + "epoch": 0.8006864291843118, + "eval_loss": 4.2339019775390625, + "eval_runtime": 207.9823, + "eval_samples_per_second": 240.405, + "eval_steps_per_second": 30.051, + "step": 7232 + }, + { + "epoch": 0.8015721442608431, + "grad_norm": 1.0776280164718628, + "learning_rate": 2.945533268175326e-05, + "loss": 3.6494, + "step": 7240 + }, + { + "epoch": 0.8026792881065072, + "grad_norm": 1.0561859607696533, + "learning_rate": 2.9138560468983813e-05, + "loss": 3.757, + "step": 7250 + }, + { + "epoch": 0.8037864319521714, + "grad_norm": 1.100935697555542, + "learning_rate": 2.882331747255242e-05, + "loss": 3.6322, + "step": 7260 + }, + { + "epoch": 0.8048935757978355, + "grad_norm": 1.063999891281128, + "learning_rate": 2.8509607681115376e-05, + "loss": 3.6916, + "step": 7270 + }, + { + "epoch": 0.8060007196434997, + "grad_norm": 1.0940972566604614, + "learning_rate": 2.8197435063929918e-05, + "loss": 3.7358, + "step": 7280 + }, + { + "epoch": 0.8071078634891639, + "grad_norm": 1.0041556358337402, + "learning_rate": 2.7886803570803794e-05, + "loss": 3.7316, + "step": 7290 + }, + { + "epoch": 0.808215007334828, + "grad_norm": 1.0293116569519043, + "learning_rate": 2.757771713204575e-05, + "loss": 3.7122, + "step": 7300 + }, + { + "epoch": 0.8093221511804921, + "grad_norm": 1.0600935220718384, + "learning_rate": 2.7270179658415286e-05, + "loss": 3.6158, + "step": 7310 + }, + { + "epoch": 0.8104292950261562, + "grad_norm": 1.1017769575119019, + "learning_rate": 2.6964195041073493e-05, + "loss": 3.704, + "step": 7320 + }, + { + "epoch": 0.8115364388718205, + "grad_norm": 1.0898927450180054, + "learning_rate": 2.665976715153377e-05, + "loss": 3.6594, + "step": 7330 + }, + { + "epoch": 0.8126435827174846, + "grad_norm": 1.1266783475875854, + "learning_rate": 2.6356899841612606e-05, + "loss": 3.6819, + "step": 7340 + }, + { + "epoch": 0.8137507265631487, + "grad_norm": 1.0668282508850098, + "learning_rate": 2.6055596943381336e-05, + "loss": 3.6972, + "step": 7350 + }, + { + "epoch": 0.8148578704088129, + "grad_norm": 1.0482220649719238, + "learning_rate": 2.575586226911705e-05, + "loss": 3.6429, + "step": 7360 + }, + { + "epoch": 0.815965014254477, + "grad_norm": 1.081392765045166, + "learning_rate": 2.5457699611254932e-05, + "loss": 3.6739, + "step": 7370 + }, + { + "epoch": 0.8170721581001411, + "grad_norm": 0.9860684275627136, + "learning_rate": 2.516111274233983e-05, + "loss": 3.7527, + "step": 7380 + }, + { + "epoch": 0.8181793019458053, + "grad_norm": 1.0662422180175781, + "learning_rate": 2.486610541497876e-05, + "loss": 3.751, + "step": 7390 + }, + { + "epoch": 0.8192864457914695, + "grad_norm": 1.1191221475601196, + "learning_rate": 2.4572681361793407e-05, + "loss": 3.6454, + "step": 7400 + }, + { + "epoch": 0.8203935896371336, + "grad_norm": 0.9559420347213745, + "learning_rate": 2.4280844295372688e-05, + "loss": 3.7063, + "step": 7410 + }, + { + "epoch": 0.8215007334827977, + "grad_norm": 1.122487187385559, + "learning_rate": 2.3990597908226218e-05, + "loss": 3.6159, + "step": 7420 + }, + { + "epoch": 0.8226078773284619, + "grad_norm": 1.032131552696228, + "learning_rate": 2.370194587273711e-05, + "loss": 3.71, + "step": 7430 + }, + { + "epoch": 0.8237150211741261, + "grad_norm": 1.0027601718902588, + "learning_rate": 2.3414891841115786e-05, + "loss": 3.6323, + "step": 7440 + }, + { + "epoch": 0.8248221650197902, + "grad_norm": 1.0453424453735352, + "learning_rate": 2.312943944535373e-05, + "loss": 3.6308, + "step": 7450 + }, + { + "epoch": 0.8259293088654543, + "grad_norm": 1.0495673418045044, + "learning_rate": 2.2845592297177385e-05, + "loss": 3.7073, + "step": 7460 + }, + { + "epoch": 0.8270364527111185, + "grad_norm": 1.1159603595733643, + "learning_rate": 2.2563353988002756e-05, + "loss": 3.648, + "step": 7470 + }, + { + "epoch": 0.8281435965567826, + "grad_norm": 0.9944765567779541, + "learning_rate": 2.2282728088889602e-05, + "loss": 3.6275, + "step": 7480 + }, + { + "epoch": 0.8292507404024468, + "grad_norm": 1.0571640729904175, + "learning_rate": 2.2003718150496596e-05, + "loss": 3.6975, + "step": 7490 + }, + { + "epoch": 0.830357884248111, + "grad_norm": 1.1020572185516357, + "learning_rate": 2.1726327703036133e-05, + "loss": 3.7093, + "step": 7500 + }, + { + "epoch": 0.8314650280937751, + "grad_norm": 0.9671516418457031, + "learning_rate": 2.145056025622981e-05, + "loss": 3.785, + "step": 7510 + }, + { + "epoch": 0.8325721719394392, + "grad_norm": 1.0117616653442383, + "learning_rate": 2.117641929926403e-05, + "loss": 3.6757, + "step": 7520 + }, + { + "epoch": 0.8336793157851033, + "grad_norm": 1.0444458723068237, + "learning_rate": 2.090390830074563e-05, + "loss": 3.7839, + "step": 7530 + }, + { + "epoch": 0.8347864596307675, + "grad_norm": 1.097193717956543, + "learning_rate": 2.0633030708658395e-05, + "loss": 3.6356, + "step": 7540 + }, + { + "epoch": 0.8358936034764317, + "grad_norm": 1.1735577583312988, + "learning_rate": 2.036378995031908e-05, + "loss": 3.618, + "step": 7550 + }, + { + "epoch": 0.8370007473220958, + "grad_norm": 1.1020944118499756, + "learning_rate": 2.009618943233419e-05, + "loss": 3.6484, + "step": 7560 + }, + { + "epoch": 0.83810789116776, + "grad_norm": 1.0746279954910278, + "learning_rate": 1.9830232540556907e-05, + "loss": 3.7195, + "step": 7570 + }, + { + "epoch": 0.8392150350134241, + "grad_norm": 1.2530303001403809, + "learning_rate": 1.9565922640044023e-05, + "loss": 3.6372, + "step": 7580 + }, + { + "epoch": 0.8403221788590882, + "grad_norm": 1.1061261892318726, + "learning_rate": 1.930326307501383e-05, + "loss": 3.6608, + "step": 7590 + }, + { + "epoch": 0.8414293227047525, + "grad_norm": 0.9524762034416199, + "learning_rate": 1.9042257168803232e-05, + "loss": 3.6441, + "step": 7600 + }, + { + "epoch": 0.8425364665504166, + "grad_norm": 1.0636398792266846, + "learning_rate": 1.878290822382622e-05, + "loss": 3.6464, + "step": 7610 + }, + { + "epoch": 0.8436436103960807, + "grad_norm": 1.0637277364730835, + "learning_rate": 1.8525219521531693e-05, + "loss": 3.7109, + "step": 7620 + }, + { + "epoch": 0.8447507542417448, + "grad_norm": 1.0510637760162354, + "learning_rate": 1.8269194322362164e-05, + "loss": 3.6934, + "step": 7630 + }, + { + "epoch": 0.845857898087409, + "grad_norm": 1.0641145706176758, + "learning_rate": 1.8014835865712434e-05, + "loss": 3.7412, + "step": 7640 + }, + { + "epoch": 0.8469650419330732, + "grad_norm": 1.0800358057022095, + "learning_rate": 1.776214736988853e-05, + "loss": 3.5522, + "step": 7650 + }, + { + "epoch": 0.8480721857787373, + "grad_norm": 0.9840835332870483, + "learning_rate": 1.7511132032067238e-05, + "loss": 3.6749, + "step": 7660 + }, + { + "epoch": 0.8491793296244015, + "grad_norm": 1.1203852891921997, + "learning_rate": 1.7261793028255266e-05, + "loss": 3.6151, + "step": 7670 + }, + { + "epoch": 0.8502864734700656, + "grad_norm": 1.0806456804275513, + "learning_rate": 1.7014133513249474e-05, + "loss": 3.6835, + "step": 7680 + }, + { + "epoch": 0.8513936173157297, + "grad_norm": 1.1195396184921265, + "learning_rate": 1.676815662059663e-05, + "loss": 3.6201, + "step": 7690 + }, + { + "epoch": 0.852500761161394, + "grad_norm": 1.0522040128707886, + "learning_rate": 1.6523865462553854e-05, + "loss": 3.6611, + "step": 7700 + }, + { + "epoch": 0.8536079050070581, + "grad_norm": 1.006016731262207, + "learning_rate": 1.6281263130049442e-05, + "loss": 3.671, + "step": 7710 + }, + { + "epoch": 0.8547150488527222, + "grad_norm": 1.0625911951065063, + "learning_rate": 1.6040352692643356e-05, + "loss": 3.6621, + "step": 7720 + }, + { + "epoch": 0.8558221926983863, + "grad_norm": 1.0699069499969482, + "learning_rate": 1.5801137198488817e-05, + "loss": 3.734, + "step": 7730 + }, + { + "epoch": 0.8569293365440505, + "grad_norm": 1.0966960191726685, + "learning_rate": 1.5563619674293466e-05, + "loss": 3.7387, + "step": 7740 + }, + { + "epoch": 0.8580364803897146, + "grad_norm": 1.0438299179077148, + "learning_rate": 1.5327803125281103e-05, + "loss": 3.7075, + "step": 7750 + }, + { + "epoch": 0.8591436242353788, + "grad_norm": 1.0032527446746826, + "learning_rate": 1.509369053515378e-05, + "loss": 3.6073, + "step": 7760 + }, + { + "epoch": 0.860250768081043, + "grad_norm": 1.0899827480316162, + "learning_rate": 1.486128486605383e-05, + "loss": 3.6489, + "step": 7770 + }, + { + "epoch": 0.8613579119267071, + "grad_norm": 1.0680930614471436, + "learning_rate": 1.4630589058526743e-05, + "loss": 3.6152, + "step": 7780 + }, + { + "epoch": 0.8624650557723712, + "grad_norm": 1.0146864652633667, + "learning_rate": 1.4401606031483497e-05, + "loss": 3.801, + "step": 7790 + }, + { + "epoch": 0.8635721996180353, + "grad_norm": 1.014323353767395, + "learning_rate": 1.4174338682164094e-05, + "loss": 3.6855, + "step": 7800 + }, + { + "epoch": 0.8646793434636996, + "grad_norm": 1.147574543952942, + "learning_rate": 1.3948789886100591e-05, + "loss": 3.6981, + "step": 7810 + }, + { + "epoch": 0.8657864873093637, + "grad_norm": 1.0883703231811523, + "learning_rate": 1.3724962497080732e-05, + "loss": 3.6546, + "step": 7820 + }, + { + "epoch": 0.8668936311550278, + "grad_norm": 1.1061820983886719, + "learning_rate": 1.3502859347112095e-05, + "loss": 3.6553, + "step": 7830 + }, + { + "epoch": 0.868000775000692, + "grad_norm": 1.0940380096435547, + "learning_rate": 1.3282483246385856e-05, + "loss": 3.6678, + "step": 7840 + }, + { + "epoch": 0.8691079188463561, + "grad_norm": 1.0558832883834839, + "learning_rate": 1.306383698324171e-05, + "loss": 3.648, + "step": 7850 + }, + { + "epoch": 0.8702150626920203, + "grad_norm": 1.1125390529632568, + "learning_rate": 1.284692332413209e-05, + "loss": 3.5852, + "step": 7860 + }, + { + "epoch": 0.8713222065376844, + "grad_norm": 1.1055091619491577, + "learning_rate": 1.2631745013587602e-05, + "loss": 3.7159, + "step": 7870 + }, + { + "epoch": 0.8724293503833486, + "grad_norm": 1.4560612440109253, + "learning_rate": 1.2418304774182075e-05, + "loss": 3.6914, + "step": 7880 + }, + { + "epoch": 0.8735364942290127, + "grad_norm": 1.111989140510559, + "learning_rate": 1.2206605306498003e-05, + "loss": 3.6762, + "step": 7890 + }, + { + "epoch": 0.8746436380746768, + "grad_norm": 1.0178395509719849, + "learning_rate": 1.1996649289092742e-05, + "loss": 3.7498, + "step": 7900 + }, + { + "epoch": 0.875750781920341, + "grad_norm": 1.0071059465408325, + "learning_rate": 1.1788439378464198e-05, + "loss": 3.6331, + "step": 7910 + }, + { + "epoch": 0.8768579257660052, + "grad_norm": 1.0470887422561646, + "learning_rate": 1.1581978209017567e-05, + "loss": 3.6654, + "step": 7920 + }, + { + "epoch": 0.8779650696116693, + "grad_norm": 1.0238986015319824, + "learning_rate": 1.1377268393031791e-05, + "loss": 3.6159, + "step": 7930 + }, + { + "epoch": 0.8790722134573334, + "grad_norm": 1.124292016029358, + "learning_rate": 1.1174312520626555e-05, + "loss": 3.6193, + "step": 7940 + }, + { + "epoch": 0.8801793573029976, + "grad_norm": 1.012042760848999, + "learning_rate": 1.0973113159729562e-05, + "loss": 3.6962, + "step": 7950 + }, + { + "epoch": 0.8812865011486617, + "grad_norm": 1.1950116157531738, + "learning_rate": 1.0773672856043914e-05, + "loss": 3.7116, + "step": 7960 + }, + { + "epoch": 0.8823936449943259, + "grad_norm": 1.0466302633285522, + "learning_rate": 1.0575994133016142e-05, + "loss": 3.6718, + "step": 7970 + }, + { + "epoch": 0.8835007888399901, + "grad_norm": 1.0536062717437744, + "learning_rate": 1.0380079491803994e-05, + "loss": 3.6996, + "step": 7980 + }, + { + "epoch": 0.8846079326856542, + "grad_norm": 1.0780149698257446, + "learning_rate": 1.0185931411245008e-05, + "loss": 3.6488, + "step": 7990 + }, + { + "epoch": 0.8857150765313183, + "grad_norm": 1.0425546169281006, + "learning_rate": 9.993552347825078e-06, + "loss": 3.6289, + "step": 8000 + }, + { + "epoch": 0.8868222203769824, + "grad_norm": 1.050801396369934, + "learning_rate": 9.802944735647245e-06, + "loss": 3.6609, + "step": 8010 + }, + { + "epoch": 0.8879293642226467, + "grad_norm": 1.0203200578689575, + "learning_rate": 9.614110986401169e-06, + "loss": 3.6997, + "step": 8020 + }, + { + "epoch": 0.8890365080683108, + "grad_norm": 1.1377030611038208, + "learning_rate": 9.427053489332304e-06, + "loss": 3.5994, + "step": 8030 + }, + { + "epoch": 0.8901436519139749, + "grad_norm": 0.9475767016410828, + "learning_rate": 9.241774611211977e-06, + "loss": 3.657, + "step": 8040 + }, + { + "epoch": 0.8912507957596391, + "grad_norm": 1.0409544706344604, + "learning_rate": 9.058276696307155e-06, + "loss": 3.6654, + "step": 8050 + }, + { + "epoch": 0.8923579396053032, + "grad_norm": 1.0468426942825317, + "learning_rate": 8.876562066351023e-06, + "loss": 3.6205, + "step": 8060 + }, + { + "epoch": 0.8934650834509673, + "grad_norm": 1.050571322441101, + "learning_rate": 8.696633020513455e-06, + "loss": 3.6525, + "step": 8070 + }, + { + "epoch": 0.8945722272966316, + "grad_norm": 1.1455354690551758, + "learning_rate": 8.518491835371926e-06, + "loss": 3.6012, + "step": 8080 + }, + { + "epoch": 0.8956793711422957, + "grad_norm": 1.0796661376953125, + "learning_rate": 8.342140764882876e-06, + "loss": 3.6813, + "step": 8090 + }, + { + "epoch": 0.8967865149879598, + "grad_norm": 1.141502022743225, + "learning_rate": 8.16758204035291e-06, + "loss": 3.6405, + "step": 8100 + }, + { + "epoch": 0.8978936588336239, + "grad_norm": 0.9693289399147034, + "learning_rate": 7.994817870410858e-06, + "loss": 3.7405, + "step": 8110 + }, + { + "epoch": 0.8990008026792881, + "grad_norm": 1.0907636880874634, + "learning_rate": 7.823850440979645e-06, + "loss": 3.6703, + "step": 8120 + }, + { + "epoch": 0.9001079465249523, + "grad_norm": 1.1713448762893677, + "learning_rate": 7.65468191524865e-06, + "loss": 3.6358, + "step": 8130 + }, + { + "epoch": 0.9007722328323507, + "eval_loss": 4.216700077056885, + "eval_runtime": 207.9891, + "eval_samples_per_second": 240.397, + "eval_steps_per_second": 30.05, + "step": 8136 + }, + { + "epoch": 0.9012150903706164, + "grad_norm": 1.0672483444213867, + "learning_rate": 7.487314433646441e-06, + "loss": 3.6018, + "step": 8140 + }, + { + "epoch": 0.9023222342162805, + "grad_norm": 0.9663351774215698, + "learning_rate": 7.321750113813546e-06, + "loss": 3.6877, + "step": 8150 + }, + { + "epoch": 0.9034293780619447, + "grad_norm": 1.083055019378662, + "learning_rate": 7.1579910505758285e-06, + "loss": 3.6316, + "step": 8160 + }, + { + "epoch": 0.9045365219076088, + "grad_norm": 1.0686644315719604, + "learning_rate": 6.996039315917756e-06, + "loss": 3.7178, + "step": 8170 + }, + { + "epoch": 0.905643665753273, + "grad_norm": 1.0282578468322754, + "learning_rate": 6.8358969589564715e-06, + "loss": 3.6704, + "step": 8180 + }, + { + "epoch": 0.9067508095989372, + "grad_norm": 0.9950483441352844, + "learning_rate": 6.6775660059156295e-06, + "loss": 3.6609, + "step": 8190 + }, + { + "epoch": 0.9078579534446013, + "grad_norm": 1.0958738327026367, + "learning_rate": 6.521048460099787e-06, + "loss": 3.6537, + "step": 8200 + }, + { + "epoch": 0.9089650972902654, + "grad_norm": 1.1011713743209839, + "learning_rate": 6.366346301869252e-06, + "loss": 3.633, + "step": 8210 + }, + { + "epoch": 0.9100722411359295, + "grad_norm": 1.0458924770355225, + "learning_rate": 6.213461488614741e-06, + "loss": 3.646, + "step": 8220 + }, + { + "epoch": 0.9111793849815938, + "grad_norm": 1.1526204347610474, + "learning_rate": 6.062395954732879e-06, + "loss": 3.6598, + "step": 8230 + }, + { + "epoch": 0.9122865288272579, + "grad_norm": 1.0499478578567505, + "learning_rate": 5.913151611601541e-06, + "loss": 3.6439, + "step": 8240 + }, + { + "epoch": 0.913393672672922, + "grad_norm": 1.1182613372802734, + "learning_rate": 5.7657303475556974e-06, + "loss": 3.6399, + "step": 8250 + }, + { + "epoch": 0.9145008165185862, + "grad_norm": 1.031019926071167, + "learning_rate": 5.620134027863671e-06, + "loss": 3.6228, + "step": 8260 + }, + { + "epoch": 0.9156079603642503, + "grad_norm": 1.0122407674789429, + "learning_rate": 5.476364494703256e-06, + "loss": 3.6294, + "step": 8270 + }, + { + "epoch": 0.9167151042099144, + "grad_norm": 1.0196013450622559, + "learning_rate": 5.334423567138734e-06, + "loss": 3.6208, + "step": 8280 + }, + { + "epoch": 0.9178222480555787, + "grad_norm": 1.1157269477844238, + "learning_rate": 5.1943130410975805e-06, + "loss": 3.6156, + "step": 8290 + }, + { + "epoch": 0.9189293919012428, + "grad_norm": 1.015960931777954, + "learning_rate": 5.056034689347927e-06, + "loss": 3.7179, + "step": 8300 + }, + { + "epoch": 0.9200365357469069, + "grad_norm": 1.0555146932601929, + "learning_rate": 4.919590261476047e-06, + "loss": 3.6462, + "step": 8310 + }, + { + "epoch": 0.921143679592571, + "grad_norm": 0.9901772737503052, + "learning_rate": 4.784981483864198e-06, + "loss": 3.6395, + "step": 8320 + }, + { + "epoch": 0.9222508234382352, + "grad_norm": 1.1814167499542236, + "learning_rate": 4.652210059668893e-06, + "loss": 3.6363, + "step": 8330 + }, + { + "epoch": 0.9233579672838994, + "grad_norm": 1.0374478101730347, + "learning_rate": 4.521277668799144e-06, + "loss": 3.6644, + "step": 8340 + }, + { + "epoch": 0.9244651111295635, + "grad_norm": 1.1175576448440552, + "learning_rate": 4.392185967895495e-06, + "loss": 3.6488, + "step": 8350 + }, + { + "epoch": 0.9255722549752277, + "grad_norm": 1.0541727542877197, + "learning_rate": 4.264936590308782e-06, + "loss": 3.6246, + "step": 8360 + }, + { + "epoch": 0.9266793988208918, + "grad_norm": 1.0559375286102295, + "learning_rate": 4.139531146079622e-06, + "loss": 3.6704, + "step": 8370 + }, + { + "epoch": 0.9277865426665559, + "grad_norm": 1.1315760612487793, + "learning_rate": 4.015971221918046e-06, + "loss": 3.5962, + "step": 8380 + }, + { + "epoch": 0.9288936865122202, + "grad_norm": 0.9821998476982117, + "learning_rate": 3.894258381183313e-06, + "loss": 3.7408, + "step": 8390 + }, + { + "epoch": 0.9300008303578843, + "grad_norm": 1.2824774980545044, + "learning_rate": 3.774394163864258e-06, + "loss": 3.6566, + "step": 8400 + }, + { + "epoch": 0.9311079742035484, + "grad_norm": 1.0640493631362915, + "learning_rate": 3.6563800865596794e-06, + "loss": 3.6931, + "step": 8410 + }, + { + "epoch": 0.9322151180492125, + "grad_norm": 1.0892513990402222, + "learning_rate": 3.540217642459331e-06, + "loss": 3.6628, + "step": 8420 + }, + { + "epoch": 0.9333222618948767, + "grad_norm": 1.083923578262329, + "learning_rate": 3.4259083013248257e-06, + "loss": 3.6209, + "step": 8430 + }, + { + "epoch": 0.9344294057405408, + "grad_norm": 1.1324032545089722, + "learning_rate": 3.3134535094711146e-06, + "loss": 3.6517, + "step": 8440 + }, + { + "epoch": 0.935536549586205, + "grad_norm": 1.0142607688903809, + "learning_rate": 3.2028546897483186e-06, + "loss": 3.6968, + "step": 8450 + }, + { + "epoch": 0.9366436934318692, + "grad_norm": 1.0392403602600098, + "learning_rate": 3.0941132415234602e-06, + "loss": 3.6975, + "step": 8460 + }, + { + "epoch": 0.9377508372775333, + "grad_norm": 0.9889246225357056, + "learning_rate": 2.98723054066306e-06, + "loss": 3.6797, + "step": 8470 + }, + { + "epoch": 0.9388579811231974, + "grad_norm": 1.1555813550949097, + "learning_rate": 2.882207939515435e-06, + "loss": 3.5963, + "step": 8480 + }, + { + "epoch": 0.9399651249688615, + "grad_norm": 1.1098711490631104, + "learning_rate": 2.7790467668938277e-06, + "loss": 3.7014, + "step": 8490 + }, + { + "epoch": 0.9410722688145258, + "grad_norm": 1.175168514251709, + "learning_rate": 2.677748328059487e-06, + "loss": 3.7329, + "step": 8500 + }, + { + "epoch": 0.9421794126601899, + "grad_norm": 0.9285978078842163, + "learning_rate": 2.578313904705098e-06, + "loss": 3.7243, + "step": 8510 + }, + { + "epoch": 0.943286556505854, + "grad_norm": 0.9846440553665161, + "learning_rate": 2.480744754938696e-06, + "loss": 3.5868, + "step": 8520 + }, + { + "epoch": 0.9443937003515182, + "grad_norm": 1.0419180393218994, + "learning_rate": 2.38504211326761e-06, + "loss": 3.7148, + "step": 8530 + }, + { + "epoch": 0.9455008441971823, + "grad_norm": 1.0027834177017212, + "learning_rate": 2.291207190583011e-06, + "loss": 3.6636, + "step": 8540 + }, + { + "epoch": 0.9466079880428465, + "grad_norm": 1.0876221656799316, + "learning_rate": 2.1992411741443896e-06, + "loss": 3.6787, + "step": 8550 + }, + { + "epoch": 0.9477151318885106, + "grad_norm": 1.0277339220046997, + "learning_rate": 2.1091452275647013e-06, + "loss": 3.6853, + "step": 8560 + }, + { + "epoch": 0.9488222757341748, + "grad_norm": 1.1001217365264893, + "learning_rate": 2.02092049079558e-06, + "loss": 3.7571, + "step": 8570 + }, + { + "epoch": 0.9499294195798389, + "grad_norm": 1.083074927330017, + "learning_rate": 1.9345680801128806e-06, + "loss": 3.6392, + "step": 8580 + }, + { + "epoch": 0.951036563425503, + "grad_norm": 0.9950634837150574, + "learning_rate": 1.8500890881026076e-06, + "loss": 3.7111, + "step": 8590 + }, + { + "epoch": 0.9521437072711673, + "grad_norm": 1.0303641557693481, + "learning_rate": 1.7674845836470609e-06, + "loss": 3.7491, + "step": 8600 + }, + { + "epoch": 0.9532508511168314, + "grad_norm": 1.0995064973831177, + "learning_rate": 1.6867556119113612e-06, + "loss": 3.6798, + "step": 8610 + }, + { + "epoch": 0.9543579949624955, + "grad_norm": 0.9651690721511841, + "learning_rate": 1.6079031943301623e-06, + "loss": 3.5852, + "step": 8620 + }, + { + "epoch": 0.9554651388081596, + "grad_norm": 1.0190938711166382, + "learning_rate": 1.5309283285947271e-06, + "loss": 3.6699, + "step": 8630 + }, + { + "epoch": 0.9565722826538238, + "grad_norm": 1.0279735326766968, + "learning_rate": 1.4558319886403714e-06, + "loss": 3.5868, + "step": 8640 + }, + { + "epoch": 0.9576794264994879, + "grad_norm": 1.0262198448181152, + "learning_rate": 1.3826151246340566e-06, + "loss": 3.6984, + "step": 8650 + }, + { + "epoch": 0.9587865703451521, + "grad_norm": 1.1557973623275757, + "learning_rate": 1.311278662962484e-06, + "loss": 3.7045, + "step": 8660 + }, + { + "epoch": 0.9598937141908163, + "grad_norm": 1.0873647928237915, + "learning_rate": 1.2418235062202187e-06, + "loss": 3.6725, + "step": 8670 + }, + { + "epoch": 0.9610008580364804, + "grad_norm": 1.0803563594818115, + "learning_rate": 1.174250533198401e-06, + "loss": 3.6623, + "step": 8680 + }, + { + "epoch": 0.9621080018821445, + "grad_norm": 1.0678378343582153, + "learning_rate": 1.1085605988735536e-06, + "loss": 3.6275, + "step": 8690 + }, + { + "epoch": 0.9632151457278086, + "grad_norm": 1.1217657327651978, + "learning_rate": 1.0447545343968078e-06, + "loss": 3.6539, + "step": 8700 + }, + { + "epoch": 0.9643222895734729, + "grad_norm": 1.0530879497528076, + "learning_rate": 9.82833147083345e-07, + "loss": 3.6443, + "step": 8710 + }, + { + "epoch": 0.965429433419137, + "grad_norm": 0.9867540001869202, + "learning_rate": 9.22797220402205e-07, + "loss": 3.7205, + "step": 8720 + }, + { + "epoch": 0.9665365772648011, + "grad_norm": 1.0800204277038574, + "learning_rate": 8.646475139664266e-07, + "loss": 3.6829, + "step": 8730 + }, + { + "epoch": 0.9676437211104653, + "grad_norm": 1.105226993560791, + "learning_rate": 8.083847635232732e-07, + "loss": 3.6317, + "step": 8740 + }, + { + "epoch": 0.9687508649561294, + "grad_norm": 1.0623337030410767, + "learning_rate": 7.54009680945139e-07, + "loss": 3.6105, + "step": 8750 + }, + { + "epoch": 0.9698580088017936, + "grad_norm": 1.0684354305267334, + "learning_rate": 7.015229542203904e-07, + "loss": 3.6479, + "step": 8760 + }, + { + "epoch": 0.9709651526474578, + "grad_norm": 1.1195191144943237, + "learning_rate": 6.509252474446891e-07, + "loss": 3.7102, + "step": 8770 + }, + { + "epoch": 0.9720722964931219, + "grad_norm": 1.028835415840149, + "learning_rate": 6.022172008126658e-07, + "loss": 3.7225, + "step": 8780 + }, + { + "epoch": 0.973179440338786, + "grad_norm": 1.0451563596725464, + "learning_rate": 5.553994306096765e-07, + "loss": 3.6487, + "step": 8790 + }, + { + "epoch": 0.9742865841844501, + "grad_norm": 1.0458861589431763, + "learning_rate": 5.104725292041423e-07, + "loss": 3.6472, + "step": 8800 + }, + { + "epoch": 0.9753937280301143, + "grad_norm": 1.0627021789550781, + "learning_rate": 4.674370650400217e-07, + "loss": 3.7469, + "step": 8810 + }, + { + "epoch": 0.9765008718757785, + "grad_norm": 1.0259780883789062, + "learning_rate": 4.2629358262953327e-07, + "loss": 3.6384, + "step": 8820 + }, + { + "epoch": 0.9776080157214426, + "grad_norm": 1.0855342149734497, + "learning_rate": 3.870426025463946e-07, + "loss": 3.6849, + "step": 8830 + }, + { + "epoch": 0.9787151595671068, + "grad_norm": 0.9996906518936157, + "learning_rate": 3.496846214191107e-07, + "loss": 3.6816, + "step": 8840 + }, + { + "epoch": 0.9798223034127709, + "grad_norm": 1.040379285812378, + "learning_rate": 3.142201119248289e-07, + "loss": 3.7141, + "step": 8850 + }, + { + "epoch": 0.980929447258435, + "grad_norm": 1.1035263538360596, + "learning_rate": 2.8064952278321063e-07, + "loss": 3.6134, + "step": 8860 + }, + { + "epoch": 0.9820365911040992, + "grad_norm": 1.020015001296997, + "learning_rate": 2.489732787508525e-07, + "loss": 3.6357, + "step": 8870 + }, + { + "epoch": 0.9831437349497634, + "grad_norm": 1.088901400566101, + "learning_rate": 2.1919178061589048e-07, + "loss": 3.6801, + "step": 8880 + }, + { + "epoch": 0.9842508787954275, + "grad_norm": 0.9592273235321045, + "learning_rate": 1.913054051928875e-07, + "loss": 3.7028, + "step": 8890 + }, + { + "epoch": 0.9853580226410916, + "grad_norm": 1.0482735633850098, + "learning_rate": 1.653145053181537e-07, + "loss": 3.6337, + "step": 8900 + }, + { + "epoch": 0.9864651664867558, + "grad_norm": 1.109691858291626, + "learning_rate": 1.412194098451669e-07, + "loss": 3.6586, + "step": 8910 + }, + { + "epoch": 0.98757231033242, + "grad_norm": 1.0465811491012573, + "learning_rate": 1.1902042364054232e-07, + "loss": 3.6988, + "step": 8920 + }, + { + "epoch": 0.9886794541780841, + "grad_norm": 1.025052547454834, + "learning_rate": 9.871782758005264e-08, + "loss": 3.7138, + "step": 8930 + }, + { + "epoch": 0.9897865980237482, + "grad_norm": 1.1050868034362793, + "learning_rate": 8.031187854514731e-08, + "loss": 3.7036, + "step": 8940 + }, + { + "epoch": 0.9908937418694124, + "grad_norm": 1.0677896738052368, + "learning_rate": 6.380280941968852e-08, + "loss": 3.6513, + "step": 8950 + }, + { + "epoch": 0.9920008857150765, + "grad_norm": 1.1986464262008667, + "learning_rate": 4.9190829086986904e-08, + "loss": 3.6899, + "step": 8960 + }, + { + "epoch": 0.9931080295607406, + "grad_norm": 1.0726470947265625, + "learning_rate": 3.647612242717035e-08, + "loss": 3.694, + "step": 8970 + }, + { + "epoch": 0.9942151734064049, + "grad_norm": 1.1176546812057495, + "learning_rate": 2.5658850314852486e-08, + "loss": 3.7426, + "step": 8980 + }, + { + "epoch": 0.995322317252069, + "grad_norm": 1.0686414241790771, + "learning_rate": 1.673914961710099e-08, + "loss": 3.6311, + "step": 8990 + }, + { + "epoch": 0.9964294610977331, + "grad_norm": 0.9893191456794739, + "learning_rate": 9.717133191672332e-09, + "loss": 3.6464, + "step": 9000 + }, + { + "epoch": 0.9975366049433972, + "grad_norm": 0.9576684832572937, + "learning_rate": 4.592889885596252e-09, + "loss": 3.6427, + "step": 9010 + }, + { + "epoch": 0.9986437487890614, + "grad_norm": 1.1582540273666382, + "learning_rate": 1.366484534109924e-09, + "loss": 3.4482, + "step": 9020 + }, + { + "epoch": 0.9997508926347256, + "grad_norm": 0.99860680103302, + "learning_rate": 3.795795975869431e-11, + "loss": 3.6223, + "step": 9030 + } + ], + "logging_steps": 10, + "max_steps": 9032, + "num_input_tokens_seen": 0, + "num_train_epochs": 1, + "save_steps": 904, + "stateful_callbacks": { + "TrainerControl": { + "args": { + "should_epoch_stop": false, + "should_evaluate": false, + "should_log": false, + "should_save": true, + "should_training_stop": true + }, + "attributes": {} + } + }, + "total_flos": 2.216699427692544e+16, + "train_batch_size": 16, + "trial_name": null, + "trial_params": null +}