diff --git "a/trainer_state.json" "b/trainer_state.json" new file mode 100644--- /dev/null +++ "b/trainer_state.json" @@ -0,0 +1,4543 @@ +{ + "best_metric": null, + "best_model_checkpoint": null, + "epoch": 2.9976711690731253, + "eval_steps": 500, + "global_step": 3219, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.004657661853749418, + "grad_norm": 6.8637231643996905, + "learning_rate": 7.763975155279503e-07, + "loss": 0.8093, + "step": 5 + }, + { + "epoch": 0.009315323707498836, + "grad_norm": 4.729727723722731, + "learning_rate": 1.5527950310559006e-06, + "loss": 0.7776, + "step": 10 + }, + { + "epoch": 0.013972985561248253, + "grad_norm": 1.8424610027102168, + "learning_rate": 2.329192546583851e-06, + "loss": 0.6952, + "step": 15 + }, + { + "epoch": 0.018630647414997672, + "grad_norm": 3.025250026849208, + "learning_rate": 3.1055900621118013e-06, + "loss": 0.6745, + "step": 20 + }, + { + "epoch": 0.02328830926874709, + "grad_norm": 1.2524031096906298, + "learning_rate": 3.881987577639752e-06, + "loss": 0.6287, + "step": 25 + }, + { + "epoch": 0.027945971122496506, + "grad_norm": 0.9007115948452182, + "learning_rate": 4.658385093167702e-06, + "loss": 0.6003, + "step": 30 + }, + { + "epoch": 0.032603632976245925, + "grad_norm": 0.73097263427958, + "learning_rate": 5.4347826086956525e-06, + "loss": 0.5803, + "step": 35 + }, + { + "epoch": 0.037261294829995344, + "grad_norm": 0.594517021000242, + "learning_rate": 6.2111801242236025e-06, + "loss": 0.56, + "step": 40 + }, + { + "epoch": 0.04191895668374476, + "grad_norm": 0.6204146287775623, + "learning_rate": 6.9875776397515525e-06, + "loss": 0.5224, + "step": 45 + }, + { + "epoch": 0.04657661853749418, + "grad_norm": 0.49713032953276337, + "learning_rate": 7.763975155279503e-06, + "loss": 0.524, + "step": 50 + }, + { + "epoch": 0.05123428039124359, + "grad_norm": 0.47323409551062934, + "learning_rate": 8.540372670807453e-06, + "loss": 0.5123, + "step": 55 + }, + { + "epoch": 0.05589194224499301, + "grad_norm": 0.5140578006747811, + "learning_rate": 9.316770186335403e-06, + "loss": 0.5081, + "step": 60 + }, + { + "epoch": 0.06054960409874243, + "grad_norm": 0.42753521836981706, + "learning_rate": 1.0093167701863353e-05, + "loss": 0.5013, + "step": 65 + }, + { + "epoch": 0.06520726595249185, + "grad_norm": 0.5311644721427882, + "learning_rate": 1.0869565217391305e-05, + "loss": 0.4812, + "step": 70 + }, + { + "epoch": 0.06986492780624126, + "grad_norm": 0.5470354496253793, + "learning_rate": 1.1645962732919255e-05, + "loss": 0.4836, + "step": 75 + }, + { + "epoch": 0.07452258965999069, + "grad_norm": 0.5113258877886584, + "learning_rate": 1.2422360248447205e-05, + "loss": 0.4952, + "step": 80 + }, + { + "epoch": 0.0791802515137401, + "grad_norm": 0.5057618561684237, + "learning_rate": 1.3198757763975155e-05, + "loss": 0.4745, + "step": 85 + }, + { + "epoch": 0.08383791336748952, + "grad_norm": 0.5001007612786222, + "learning_rate": 1.3975155279503105e-05, + "loss": 0.4816, + "step": 90 + }, + { + "epoch": 0.08849557522123894, + "grad_norm": 0.5243509105553624, + "learning_rate": 1.4751552795031057e-05, + "loss": 0.4866, + "step": 95 + }, + { + "epoch": 0.09315323707498836, + "grad_norm": 0.5439557087344651, + "learning_rate": 1.5527950310559007e-05, + "loss": 0.4777, + "step": 100 + }, + { + "epoch": 0.09781089892873777, + "grad_norm": 0.7178104364022548, + "learning_rate": 1.630434782608696e-05, + "loss": 0.4807, + "step": 105 + }, + { + "epoch": 0.10246856078248719, + "grad_norm": 0.546495949795031, + "learning_rate": 1.7080745341614907e-05, + "loss": 0.4702, + "step": 110 + }, + { + "epoch": 0.10712622263623661, + "grad_norm": 0.5702506336081349, + "learning_rate": 1.785714285714286e-05, + "loss": 0.4705, + "step": 115 + }, + { + "epoch": 0.11178388448998602, + "grad_norm": 0.6325662707265639, + "learning_rate": 1.8633540372670807e-05, + "loss": 0.482, + "step": 120 + }, + { + "epoch": 0.11644154634373545, + "grad_norm": 0.5823217978669636, + "learning_rate": 1.940993788819876e-05, + "loss": 0.4627, + "step": 125 + }, + { + "epoch": 0.12109920819748486, + "grad_norm": 0.5479236260881903, + "learning_rate": 2.0186335403726707e-05, + "loss": 0.4659, + "step": 130 + }, + { + "epoch": 0.1257568700512343, + "grad_norm": 0.5486635640187691, + "learning_rate": 2.096273291925466e-05, + "loss": 0.4731, + "step": 135 + }, + { + "epoch": 0.1304145319049837, + "grad_norm": 0.6809487670952536, + "learning_rate": 2.173913043478261e-05, + "loss": 0.4671, + "step": 140 + }, + { + "epoch": 0.1350721937587331, + "grad_norm": 0.8855677856193335, + "learning_rate": 2.2515527950310562e-05, + "loss": 0.4774, + "step": 145 + }, + { + "epoch": 0.13972985561248252, + "grad_norm": 0.7587714736025803, + "learning_rate": 2.329192546583851e-05, + "loss": 0.4727, + "step": 150 + }, + { + "epoch": 0.14438751746623196, + "grad_norm": 0.6005017329879899, + "learning_rate": 2.4068322981366462e-05, + "loss": 0.4553, + "step": 155 + }, + { + "epoch": 0.14904517931998137, + "grad_norm": 0.7164655541521237, + "learning_rate": 2.484472049689441e-05, + "loss": 0.4624, + "step": 160 + }, + { + "epoch": 0.1537028411737308, + "grad_norm": 0.8225497978940743, + "learning_rate": 2.5621118012422362e-05, + "loss": 0.4637, + "step": 165 + }, + { + "epoch": 0.1583605030274802, + "grad_norm": 0.7574940659782605, + "learning_rate": 2.639751552795031e-05, + "loss": 0.4622, + "step": 170 + }, + { + "epoch": 0.1630181648812296, + "grad_norm": 0.6043591001115144, + "learning_rate": 2.7173913043478262e-05, + "loss": 0.4736, + "step": 175 + }, + { + "epoch": 0.16767582673497905, + "grad_norm": 0.6123842566150755, + "learning_rate": 2.795031055900621e-05, + "loss": 0.4688, + "step": 180 + }, + { + "epoch": 0.17233348858872846, + "grad_norm": 0.6573572807933171, + "learning_rate": 2.8726708074534165e-05, + "loss": 0.4674, + "step": 185 + }, + { + "epoch": 0.17699115044247787, + "grad_norm": 0.6569054098256357, + "learning_rate": 2.9503105590062114e-05, + "loss": 0.4614, + "step": 190 + }, + { + "epoch": 0.18164881229622729, + "grad_norm": 0.5669164627540662, + "learning_rate": 3.0279503105590062e-05, + "loss": 0.4673, + "step": 195 + }, + { + "epoch": 0.18630647414997673, + "grad_norm": 0.4741303438237046, + "learning_rate": 3.1055900621118014e-05, + "loss": 0.4421, + "step": 200 + }, + { + "epoch": 0.19096413600372614, + "grad_norm": 0.8316092944810866, + "learning_rate": 3.183229813664597e-05, + "loss": 0.4635, + "step": 205 + }, + { + "epoch": 0.19562179785747555, + "grad_norm": 0.5588719774621962, + "learning_rate": 3.260869565217392e-05, + "loss": 0.4534, + "step": 210 + }, + { + "epoch": 0.20027945971122496, + "grad_norm": 0.5722614945495902, + "learning_rate": 3.3385093167701865e-05, + "loss": 0.4649, + "step": 215 + }, + { + "epoch": 0.20493712156497437, + "grad_norm": 0.47183393664254064, + "learning_rate": 3.4161490683229814e-05, + "loss": 0.461, + "step": 220 + }, + { + "epoch": 0.2095947834187238, + "grad_norm": 0.4713075366570546, + "learning_rate": 3.493788819875777e-05, + "loss": 0.4641, + "step": 225 + }, + { + "epoch": 0.21425244527247322, + "grad_norm": 0.465225134633574, + "learning_rate": 3.571428571428572e-05, + "loss": 0.4624, + "step": 230 + }, + { + "epoch": 0.21891010712622264, + "grad_norm": 0.5103756288362324, + "learning_rate": 3.6490683229813665e-05, + "loss": 0.458, + "step": 235 + }, + { + "epoch": 0.22356776897997205, + "grad_norm": 0.4948340330469265, + "learning_rate": 3.7267080745341614e-05, + "loss": 0.457, + "step": 240 + }, + { + "epoch": 0.22822543083372146, + "grad_norm": 0.5735125415902487, + "learning_rate": 3.804347826086957e-05, + "loss": 0.4489, + "step": 245 + }, + { + "epoch": 0.2328830926874709, + "grad_norm": 0.6082116394835448, + "learning_rate": 3.881987577639752e-05, + "loss": 0.4471, + "step": 250 + }, + { + "epoch": 0.2375407545412203, + "grad_norm": 0.544434229947204, + "learning_rate": 3.9596273291925465e-05, + "loss": 0.4577, + "step": 255 + }, + { + "epoch": 0.24219841639496972, + "grad_norm": 0.48166421763412753, + "learning_rate": 4.0372670807453414e-05, + "loss": 0.4621, + "step": 260 + }, + { + "epoch": 0.24685607824871914, + "grad_norm": 0.6714168210349829, + "learning_rate": 4.114906832298137e-05, + "loss": 0.4569, + "step": 265 + }, + { + "epoch": 0.2515137401024686, + "grad_norm": 0.4407546736840361, + "learning_rate": 4.192546583850932e-05, + "loss": 0.4468, + "step": 270 + }, + { + "epoch": 0.25617140195621796, + "grad_norm": 0.4764022562959415, + "learning_rate": 4.270186335403727e-05, + "loss": 0.4627, + "step": 275 + }, + { + "epoch": 0.2608290638099674, + "grad_norm": 0.698057577633327, + "learning_rate": 4.347826086956522e-05, + "loss": 0.4516, + "step": 280 + }, + { + "epoch": 0.26548672566371684, + "grad_norm": 0.5447358555717683, + "learning_rate": 4.425465838509317e-05, + "loss": 0.4512, + "step": 285 + }, + { + "epoch": 0.2701443875174662, + "grad_norm": 0.7771738299993547, + "learning_rate": 4.5031055900621124e-05, + "loss": 0.4513, + "step": 290 + }, + { + "epoch": 0.27480204937121566, + "grad_norm": 0.6245714550577889, + "learning_rate": 4.580745341614907e-05, + "loss": 0.4451, + "step": 295 + }, + { + "epoch": 0.27945971122496505, + "grad_norm": 0.43790665434878945, + "learning_rate": 4.658385093167702e-05, + "loss": 0.4567, + "step": 300 + }, + { + "epoch": 0.2841173730787145, + "grad_norm": 0.5618750399986439, + "learning_rate": 4.736024844720497e-05, + "loss": 0.4623, + "step": 305 + }, + { + "epoch": 0.2887750349324639, + "grad_norm": 0.5231305670185117, + "learning_rate": 4.8136645962732924e-05, + "loss": 0.4578, + "step": 310 + }, + { + "epoch": 0.2934326967862133, + "grad_norm": 0.5361127188910412, + "learning_rate": 4.891304347826087e-05, + "loss": 0.4489, + "step": 315 + }, + { + "epoch": 0.29809035863996275, + "grad_norm": 0.4732170257669621, + "learning_rate": 4.968944099378882e-05, + "loss": 0.4563, + "step": 320 + }, + { + "epoch": 0.30274802049371213, + "grad_norm": 0.4501399431086543, + "learning_rate": 4.994822229892993e-05, + "loss": 0.4592, + "step": 325 + }, + { + "epoch": 0.3074056823474616, + "grad_norm": 0.6834758906392285, + "learning_rate": 4.986192613047981e-05, + "loss": 0.4575, + "step": 330 + }, + { + "epoch": 0.312063344201211, + "grad_norm": 0.5362283122366314, + "learning_rate": 4.977562996202969e-05, + "loss": 0.4427, + "step": 335 + }, + { + "epoch": 0.3167210060549604, + "grad_norm": 0.5938350871835776, + "learning_rate": 4.968933379357957e-05, + "loss": 0.4559, + "step": 340 + }, + { + "epoch": 0.32137866790870984, + "grad_norm": 0.6179062918046965, + "learning_rate": 4.9603037625129445e-05, + "loss": 0.4452, + "step": 345 + }, + { + "epoch": 0.3260363297624592, + "grad_norm": 0.503244246267909, + "learning_rate": 4.951674145667933e-05, + "loss": 0.4478, + "step": 350 + }, + { + "epoch": 0.33069399161620866, + "grad_norm": 0.4864627905979029, + "learning_rate": 4.94304452882292e-05, + "loss": 0.4452, + "step": 355 + }, + { + "epoch": 0.3353516534699581, + "grad_norm": 0.42350634326603037, + "learning_rate": 4.934414911977908e-05, + "loss": 0.4536, + "step": 360 + }, + { + "epoch": 0.3400093153237075, + "grad_norm": 0.415079269533796, + "learning_rate": 4.9257852951328965e-05, + "loss": 0.4535, + "step": 365 + }, + { + "epoch": 0.3446669771774569, + "grad_norm": 0.4443162625377456, + "learning_rate": 4.917155678287884e-05, + "loss": 0.4632, + "step": 370 + }, + { + "epoch": 0.3493246390312063, + "grad_norm": 0.4272948200404277, + "learning_rate": 4.908526061442872e-05, + "loss": 0.4573, + "step": 375 + }, + { + "epoch": 0.35398230088495575, + "grad_norm": 0.44159526857296133, + "learning_rate": 4.89989644459786e-05, + "loss": 0.4439, + "step": 380 + }, + { + "epoch": 0.3586399627387052, + "grad_norm": 0.45615653427248887, + "learning_rate": 4.891266827752848e-05, + "loss": 0.4391, + "step": 385 + }, + { + "epoch": 0.36329762459245457, + "grad_norm": 0.4876335217281381, + "learning_rate": 4.882637210907836e-05, + "loss": 0.4442, + "step": 390 + }, + { + "epoch": 0.367955286446204, + "grad_norm": 0.4058309541849574, + "learning_rate": 4.874007594062824e-05, + "loss": 0.439, + "step": 395 + }, + { + "epoch": 0.37261294829995345, + "grad_norm": 481.39439612564684, + "learning_rate": 4.865377977217811e-05, + "loss": 0.541, + "step": 400 + }, + { + "epoch": 0.37727061015370283, + "grad_norm": 0.47735090439252476, + "learning_rate": 4.8567483603728e-05, + "loss": 0.4431, + "step": 405 + }, + { + "epoch": 0.3819282720074523, + "grad_norm": 0.491303661903526, + "learning_rate": 4.8481187435277875e-05, + "loss": 0.4474, + "step": 410 + }, + { + "epoch": 0.38658593386120166, + "grad_norm": 0.41603407985945356, + "learning_rate": 4.839489126682776e-05, + "loss": 0.4458, + "step": 415 + }, + { + "epoch": 0.3912435957149511, + "grad_norm": 0.45507287709983313, + "learning_rate": 4.830859509837763e-05, + "loss": 0.4362, + "step": 420 + }, + { + "epoch": 0.39590125756870054, + "grad_norm": 0.46531195249792473, + "learning_rate": 4.822229892992751e-05, + "loss": 0.4517, + "step": 425 + }, + { + "epoch": 0.4005589194224499, + "grad_norm": 0.4902178812723198, + "learning_rate": 4.8136002761477395e-05, + "loss": 0.4446, + "step": 430 + }, + { + "epoch": 0.40521658127619936, + "grad_norm": 0.4847211890046387, + "learning_rate": 4.804970659302727e-05, + "loss": 0.448, + "step": 435 + }, + { + "epoch": 0.40987424312994875, + "grad_norm": 0.4394393821073969, + "learning_rate": 4.796341042457715e-05, + "loss": 0.4561, + "step": 440 + }, + { + "epoch": 0.4145319049836982, + "grad_norm": 0.43517418417308185, + "learning_rate": 4.787711425612703e-05, + "loss": 0.4355, + "step": 445 + }, + { + "epoch": 0.4191895668374476, + "grad_norm": 0.4547774758547587, + "learning_rate": 4.779081808767691e-05, + "loss": 0.4398, + "step": 450 + }, + { + "epoch": 0.423847228691197, + "grad_norm": 0.6065747366678503, + "learning_rate": 4.770452191922679e-05, + "loss": 0.4574, + "step": 455 + }, + { + "epoch": 0.42850489054494645, + "grad_norm": 0.38522024825247947, + "learning_rate": 4.761822575077667e-05, + "loss": 0.4382, + "step": 460 + }, + { + "epoch": 0.43316255239869583, + "grad_norm": 0.3876206766802, + "learning_rate": 4.753192958232654e-05, + "loss": 0.4473, + "step": 465 + }, + { + "epoch": 0.43782021425244527, + "grad_norm": 0.49840542772899743, + "learning_rate": 4.744563341387643e-05, + "loss": 0.44, + "step": 470 + }, + { + "epoch": 0.4424778761061947, + "grad_norm": 0.4023787142304999, + "learning_rate": 4.7359337245426306e-05, + "loss": 0.4479, + "step": 475 + }, + { + "epoch": 0.4471355379599441, + "grad_norm": 0.3285212211122666, + "learning_rate": 4.7273041076976184e-05, + "loss": 0.434, + "step": 480 + }, + { + "epoch": 0.45179319981369354, + "grad_norm": 0.3961706976207895, + "learning_rate": 4.718674490852606e-05, + "loss": 0.436, + "step": 485 + }, + { + "epoch": 0.4564508616674429, + "grad_norm": 0.396382884763074, + "learning_rate": 4.710044874007594e-05, + "loss": 0.4389, + "step": 490 + }, + { + "epoch": 0.46110852352119236, + "grad_norm": 0.3583524612839967, + "learning_rate": 4.7014152571625826e-05, + "loss": 0.4322, + "step": 495 + }, + { + "epoch": 0.4657661853749418, + "grad_norm": 0.3988323726838178, + "learning_rate": 4.6927856403175704e-05, + "loss": 0.4384, + "step": 500 + }, + { + "epoch": 0.4704238472286912, + "grad_norm": 0.34259386355774835, + "learning_rate": 4.684156023472558e-05, + "loss": 0.4331, + "step": 505 + }, + { + "epoch": 0.4750815090824406, + "grad_norm": 0.36356797696670407, + "learning_rate": 4.675526406627546e-05, + "loss": 0.4289, + "step": 510 + }, + { + "epoch": 0.47973917093619, + "grad_norm": 0.3901286598881084, + "learning_rate": 4.666896789782534e-05, + "loss": 0.4396, + "step": 515 + }, + { + "epoch": 0.48439683278993945, + "grad_norm": 0.370989030940083, + "learning_rate": 4.658267172937522e-05, + "loss": 0.4372, + "step": 520 + }, + { + "epoch": 0.4890544946436889, + "grad_norm": 0.39965598904455285, + "learning_rate": 4.64963755609251e-05, + "loss": 0.4455, + "step": 525 + }, + { + "epoch": 0.49371215649743827, + "grad_norm": 0.46668846506387435, + "learning_rate": 4.641007939247497e-05, + "loss": 0.4326, + "step": 530 + }, + { + "epoch": 0.4983698183511877, + "grad_norm": 0.582132009650607, + "learning_rate": 4.632378322402486e-05, + "loss": 0.4397, + "step": 535 + }, + { + "epoch": 0.5030274802049371, + "grad_norm": 0.4121012916753587, + "learning_rate": 4.6237487055574736e-05, + "loss": 0.4417, + "step": 540 + }, + { + "epoch": 0.5076851420586865, + "grad_norm": 0.3576279653758138, + "learning_rate": 4.6151190887124615e-05, + "loss": 0.4313, + "step": 545 + }, + { + "epoch": 0.5123428039124359, + "grad_norm": 0.3572720255717242, + "learning_rate": 4.606489471867449e-05, + "loss": 0.4457, + "step": 550 + }, + { + "epoch": 0.5170004657661854, + "grad_norm": 0.547766271076225, + "learning_rate": 4.597859855022437e-05, + "loss": 0.4345, + "step": 555 + }, + { + "epoch": 0.5216581276199348, + "grad_norm": 0.40390301623067965, + "learning_rate": 4.589230238177425e-05, + "loss": 0.4391, + "step": 560 + }, + { + "epoch": 0.5263157894736842, + "grad_norm": 0.360272003554287, + "learning_rate": 4.5806006213324134e-05, + "loss": 0.4378, + "step": 565 + }, + { + "epoch": 0.5309734513274337, + "grad_norm": 0.43764565431712693, + "learning_rate": 4.5719710044874006e-05, + "loss": 0.437, + "step": 570 + }, + { + "epoch": 0.5356311131811831, + "grad_norm": 0.35655227981632326, + "learning_rate": 4.563341387642389e-05, + "loss": 0.4338, + "step": 575 + }, + { + "epoch": 0.5402887750349324, + "grad_norm": 0.40521948022517407, + "learning_rate": 4.554711770797377e-05, + "loss": 0.4373, + "step": 580 + }, + { + "epoch": 0.5449464368886818, + "grad_norm": 0.3883906171159338, + "learning_rate": 4.546082153952365e-05, + "loss": 0.4338, + "step": 585 + }, + { + "epoch": 0.5496040987424313, + "grad_norm": 0.35569113255509077, + "learning_rate": 4.5374525371073526e-05, + "loss": 0.4477, + "step": 590 + }, + { + "epoch": 0.5542617605961807, + "grad_norm": 0.4258008397028241, + "learning_rate": 4.5288229202623404e-05, + "loss": 0.4381, + "step": 595 + }, + { + "epoch": 0.5589194224499301, + "grad_norm": 0.413449691817557, + "learning_rate": 4.520193303417328e-05, + "loss": 0.4434, + "step": 600 + }, + { + "epoch": 0.5635770843036796, + "grad_norm": 0.3430385595750802, + "learning_rate": 4.511563686572317e-05, + "loss": 0.4187, + "step": 605 + }, + { + "epoch": 0.568234746157429, + "grad_norm": 0.4042699670473815, + "learning_rate": 4.5029340697273045e-05, + "loss": 0.4363, + "step": 610 + }, + { + "epoch": 0.5728924080111784, + "grad_norm": 0.43480499151771707, + "learning_rate": 4.4943044528822923e-05, + "loss": 0.4286, + "step": 615 + }, + { + "epoch": 0.5775500698649279, + "grad_norm": 0.4652132056153088, + "learning_rate": 4.48567483603728e-05, + "loss": 0.4315, + "step": 620 + }, + { + "epoch": 0.5822077317186772, + "grad_norm": 0.36729682033831373, + "learning_rate": 4.477045219192268e-05, + "loss": 0.441, + "step": 625 + }, + { + "epoch": 0.5868653935724266, + "grad_norm": 0.350174779447523, + "learning_rate": 4.4684156023472565e-05, + "loss": 0.4283, + "step": 630 + }, + { + "epoch": 0.5915230554261761, + "grad_norm": 0.3988360790981172, + "learning_rate": 4.4597859855022436e-05, + "loss": 0.4277, + "step": 635 + }, + { + "epoch": 0.5961807172799255, + "grad_norm": 0.364409160363076, + "learning_rate": 4.4511563686572315e-05, + "loss": 0.4323, + "step": 640 + }, + { + "epoch": 0.6008383791336749, + "grad_norm": 0.4049958520953601, + "learning_rate": 4.44252675181222e-05, + "loss": 0.4319, + "step": 645 + }, + { + "epoch": 0.6054960409874243, + "grad_norm": 0.401092404832482, + "learning_rate": 4.433897134967208e-05, + "loss": 0.4298, + "step": 650 + }, + { + "epoch": 0.6101537028411738, + "grad_norm": 0.34234109065820656, + "learning_rate": 4.4252675181221956e-05, + "loss": 0.4357, + "step": 655 + }, + { + "epoch": 0.6148113646949231, + "grad_norm": 0.31842990925114173, + "learning_rate": 4.4166379012771834e-05, + "loss": 0.436, + "step": 660 + }, + { + "epoch": 0.6194690265486725, + "grad_norm": 0.32769017770224784, + "learning_rate": 4.408008284432171e-05, + "loss": 0.4308, + "step": 665 + }, + { + "epoch": 0.624126688402422, + "grad_norm": 0.34233152696132674, + "learning_rate": 4.39937866758716e-05, + "loss": 0.4332, + "step": 670 + }, + { + "epoch": 0.6287843502561714, + "grad_norm": 0.35536872530734553, + "learning_rate": 4.3907490507421476e-05, + "loss": 0.4258, + "step": 675 + }, + { + "epoch": 0.6334420121099208, + "grad_norm": 0.34026249002596654, + "learning_rate": 4.382119433897135e-05, + "loss": 0.4295, + "step": 680 + }, + { + "epoch": 0.6380996739636703, + "grad_norm": 0.41190356822364965, + "learning_rate": 4.373489817052123e-05, + "loss": 0.4406, + "step": 685 + }, + { + "epoch": 0.6427573358174197, + "grad_norm": 0.49421175909812803, + "learning_rate": 4.364860200207111e-05, + "loss": 0.4322, + "step": 690 + }, + { + "epoch": 0.6474149976711691, + "grad_norm": 0.3603660811499903, + "learning_rate": 4.356230583362099e-05, + "loss": 0.4302, + "step": 695 + }, + { + "epoch": 0.6520726595249184, + "grad_norm": 0.3125363535924705, + "learning_rate": 4.347600966517087e-05, + "loss": 0.4275, + "step": 700 + }, + { + "epoch": 0.6567303213786679, + "grad_norm": 0.30041905324736656, + "learning_rate": 4.3389713496720745e-05, + "loss": 0.4249, + "step": 705 + }, + { + "epoch": 0.6613879832324173, + "grad_norm": 0.345152816790548, + "learning_rate": 4.330341732827063e-05, + "loss": 0.4362, + "step": 710 + }, + { + "epoch": 0.6660456450861667, + "grad_norm": 0.37585421147119386, + "learning_rate": 4.321712115982051e-05, + "loss": 0.4306, + "step": 715 + }, + { + "epoch": 0.6707033069399162, + "grad_norm": 0.3555850152506123, + "learning_rate": 4.3130824991370387e-05, + "loss": 0.4225, + "step": 720 + }, + { + "epoch": 0.6753609687936656, + "grad_norm": 0.37642144873402433, + "learning_rate": 4.3044528822920265e-05, + "loss": 0.427, + "step": 725 + }, + { + "epoch": 0.680018630647415, + "grad_norm": 0.39601172436978427, + "learning_rate": 4.295823265447014e-05, + "loss": 0.4334, + "step": 730 + }, + { + "epoch": 0.6846762925011645, + "grad_norm": 0.30751773394560866, + "learning_rate": 4.287193648602002e-05, + "loss": 0.4221, + "step": 735 + }, + { + "epoch": 0.6893339543549138, + "grad_norm": 0.2987838209757166, + "learning_rate": 4.27856403175699e-05, + "loss": 0.4264, + "step": 740 + }, + { + "epoch": 0.6939916162086632, + "grad_norm": 0.33988559830495146, + "learning_rate": 4.269934414911978e-05, + "loss": 0.4233, + "step": 745 + }, + { + "epoch": 0.6986492780624126, + "grad_norm": 0.4343934463688201, + "learning_rate": 4.261304798066966e-05, + "loss": 0.4294, + "step": 750 + }, + { + "epoch": 0.7033069399161621, + "grad_norm": 0.37360553714178396, + "learning_rate": 4.252675181221954e-05, + "loss": 0.4205, + "step": 755 + }, + { + "epoch": 0.7079646017699115, + "grad_norm": 0.3518865734884153, + "learning_rate": 4.244045564376942e-05, + "loss": 0.4234, + "step": 760 + }, + { + "epoch": 0.7126222636236609, + "grad_norm": 0.3831682013029351, + "learning_rate": 4.23541594753193e-05, + "loss": 0.4306, + "step": 765 + }, + { + "epoch": 0.7172799254774104, + "grad_norm": 0.41718671802772334, + "learning_rate": 4.2267863306869176e-05, + "loss": 0.4302, + "step": 770 + }, + { + "epoch": 0.7219375873311598, + "grad_norm": 0.4835433327884266, + "learning_rate": 4.2181567138419054e-05, + "loss": 0.4193, + "step": 775 + }, + { + "epoch": 0.7265952491849091, + "grad_norm": 0.38728558689060666, + "learning_rate": 4.209527096996894e-05, + "loss": 0.4348, + "step": 780 + }, + { + "epoch": 0.7312529110386586, + "grad_norm": 0.34227700976916714, + "learning_rate": 4.200897480151881e-05, + "loss": 0.4282, + "step": 785 + }, + { + "epoch": 0.735910572892408, + "grad_norm": 0.36104964606844725, + "learning_rate": 4.1922678633068695e-05, + "loss": 0.4207, + "step": 790 + }, + { + "epoch": 0.7405682347461574, + "grad_norm": 0.3693699727676046, + "learning_rate": 4.1836382464618573e-05, + "loss": 0.4271, + "step": 795 + }, + { + "epoch": 0.7452258965999069, + "grad_norm": 0.3252684155467639, + "learning_rate": 4.175008629616845e-05, + "loss": 0.4234, + "step": 800 + }, + { + "epoch": 0.7498835584536563, + "grad_norm": 0.3135302717292993, + "learning_rate": 4.166379012771833e-05, + "loss": 0.4382, + "step": 805 + }, + { + "epoch": 0.7545412203074057, + "grad_norm": 0.4218553599465769, + "learning_rate": 4.157749395926821e-05, + "loss": 0.4261, + "step": 810 + }, + { + "epoch": 0.759198882161155, + "grad_norm": 0.3411302170331546, + "learning_rate": 4.1491197790818086e-05, + "loss": 0.4276, + "step": 815 + }, + { + "epoch": 0.7638565440149045, + "grad_norm": 0.3255638030179571, + "learning_rate": 4.140490162236797e-05, + "loss": 0.4272, + "step": 820 + }, + { + "epoch": 0.7685142058686539, + "grad_norm": 0.33969362630966377, + "learning_rate": 4.131860545391785e-05, + "loss": 0.4155, + "step": 825 + }, + { + "epoch": 0.7731718677224033, + "grad_norm": 0.3026933967426526, + "learning_rate": 4.123230928546773e-05, + "loss": 0.4227, + "step": 830 + }, + { + "epoch": 0.7778295295761528, + "grad_norm": 0.3772435054159253, + "learning_rate": 4.1146013117017606e-05, + "loss": 0.4302, + "step": 835 + }, + { + "epoch": 0.7824871914299022, + "grad_norm": 0.30135072993267004, + "learning_rate": 4.1059716948567484e-05, + "loss": 0.4158, + "step": 840 + }, + { + "epoch": 0.7871448532836516, + "grad_norm": 0.33346125457132675, + "learning_rate": 4.097342078011737e-05, + "loss": 0.4212, + "step": 845 + }, + { + "epoch": 0.7918025151374011, + "grad_norm": 0.282888962852797, + "learning_rate": 4.088712461166724e-05, + "loss": 0.4233, + "step": 850 + }, + { + "epoch": 0.7964601769911505, + "grad_norm": 0.28794732707700255, + "learning_rate": 4.080082844321712e-05, + "loss": 0.4153, + "step": 855 + }, + { + "epoch": 0.8011178388448998, + "grad_norm": 0.31081358602252523, + "learning_rate": 4.0714532274767004e-05, + "loss": 0.4178, + "step": 860 + }, + { + "epoch": 0.8057755006986492, + "grad_norm": 0.3191261560402839, + "learning_rate": 4.062823610631688e-05, + "loss": 0.4205, + "step": 865 + }, + { + "epoch": 0.8104331625523987, + "grad_norm": 0.3142363512759355, + "learning_rate": 4.054193993786676e-05, + "loss": 0.4154, + "step": 870 + }, + { + "epoch": 0.8150908244061481, + "grad_norm": 0.3364363550776446, + "learning_rate": 4.045564376941664e-05, + "loss": 0.4246, + "step": 875 + }, + { + "epoch": 0.8197484862598975, + "grad_norm": 0.2747088202515003, + "learning_rate": 4.036934760096652e-05, + "loss": 0.4094, + "step": 880 + }, + { + "epoch": 0.824406148113647, + "grad_norm": 0.301772676161776, + "learning_rate": 4.02830514325164e-05, + "loss": 0.4241, + "step": 885 + }, + { + "epoch": 0.8290638099673964, + "grad_norm": 0.3016593662082275, + "learning_rate": 4.019675526406628e-05, + "loss": 0.4227, + "step": 890 + }, + { + "epoch": 0.8337214718211458, + "grad_norm": 0.33138408886331777, + "learning_rate": 4.011045909561615e-05, + "loss": 0.4232, + "step": 895 + }, + { + "epoch": 0.8383791336748952, + "grad_norm": 0.2786490697206707, + "learning_rate": 4.0024162927166037e-05, + "loss": 0.4182, + "step": 900 + }, + { + "epoch": 0.8430367955286446, + "grad_norm": 0.29649339366141875, + "learning_rate": 3.9937866758715915e-05, + "loss": 0.4211, + "step": 905 + }, + { + "epoch": 0.847694457382394, + "grad_norm": 0.3222105000197682, + "learning_rate": 3.98515705902658e-05, + "loss": 0.427, + "step": 910 + }, + { + "epoch": 0.8523521192361434, + "grad_norm": 0.3341349853429822, + "learning_rate": 3.976527442181567e-05, + "loss": 0.4189, + "step": 915 + }, + { + "epoch": 0.8570097810898929, + "grad_norm": 0.34063605933590474, + "learning_rate": 3.967897825336555e-05, + "loss": 0.4263, + "step": 920 + }, + { + "epoch": 0.8616674429436423, + "grad_norm": 0.30926572421159065, + "learning_rate": 3.9592682084915434e-05, + "loss": 0.4208, + "step": 925 + }, + { + "epoch": 0.8663251047973917, + "grad_norm": 0.30756608599962054, + "learning_rate": 3.950638591646531e-05, + "loss": 0.4219, + "step": 930 + }, + { + "epoch": 0.8709827666511412, + "grad_norm": 0.2998808322519085, + "learning_rate": 3.942008974801519e-05, + "loss": 0.4115, + "step": 935 + }, + { + "epoch": 0.8756404285048905, + "grad_norm": 0.3388893716296557, + "learning_rate": 3.933379357956507e-05, + "loss": 0.4116, + "step": 940 + }, + { + "epoch": 0.8802980903586399, + "grad_norm": 0.3130985744320382, + "learning_rate": 3.924749741111495e-05, + "loss": 0.4152, + "step": 945 + }, + { + "epoch": 0.8849557522123894, + "grad_norm": 0.30426689978117843, + "learning_rate": 3.916120124266483e-05, + "loss": 0.4048, + "step": 950 + }, + { + "epoch": 0.8896134140661388, + "grad_norm": 0.2886330489061831, + "learning_rate": 3.9074905074214704e-05, + "loss": 0.4173, + "step": 955 + }, + { + "epoch": 0.8942710759198882, + "grad_norm": 1.289716934090849, + "learning_rate": 3.898860890576458e-05, + "loss": 0.426, + "step": 960 + }, + { + "epoch": 0.8989287377736377, + "grad_norm": 0.2921968866113809, + "learning_rate": 3.890231273731447e-05, + "loss": 0.4179, + "step": 965 + }, + { + "epoch": 0.9035863996273871, + "grad_norm": 0.3069214291008881, + "learning_rate": 3.8816016568864345e-05, + "loss": 0.4211, + "step": 970 + }, + { + "epoch": 0.9082440614811365, + "grad_norm": 0.30084230027853326, + "learning_rate": 3.8729720400414224e-05, + "loss": 0.4159, + "step": 975 + }, + { + "epoch": 0.9129017233348858, + "grad_norm": 0.30750699298398976, + "learning_rate": 3.86434242319641e-05, + "loss": 0.419, + "step": 980 + }, + { + "epoch": 0.9175593851886353, + "grad_norm": 0.3004553953211842, + "learning_rate": 3.855712806351398e-05, + "loss": 0.4235, + "step": 985 + }, + { + "epoch": 0.9222170470423847, + "grad_norm": 0.2875475095748544, + "learning_rate": 3.8470831895063865e-05, + "loss": 0.3969, + "step": 990 + }, + { + "epoch": 0.9268747088961341, + "grad_norm": 0.30257179456095296, + "learning_rate": 3.838453572661374e-05, + "loss": 0.4122, + "step": 995 + }, + { + "epoch": 0.9315323707498836, + "grad_norm": 0.3325193460041206, + "learning_rate": 3.8298239558163615e-05, + "loss": 0.4166, + "step": 1000 + }, + { + "epoch": 0.936190032603633, + "grad_norm": 0.2877923670180117, + "learning_rate": 3.82119433897135e-05, + "loss": 0.4238, + "step": 1005 + }, + { + "epoch": 0.9408476944573824, + "grad_norm": 0.3098748526987193, + "learning_rate": 3.812564722126338e-05, + "loss": 0.4142, + "step": 1010 + }, + { + "epoch": 0.9455053563111319, + "grad_norm": 0.27274244149432386, + "learning_rate": 3.8039351052813256e-05, + "loss": 0.4157, + "step": 1015 + }, + { + "epoch": 0.9501630181648812, + "grad_norm": 0.2853193572022348, + "learning_rate": 3.7953054884363134e-05, + "loss": 0.4235, + "step": 1020 + }, + { + "epoch": 0.9548206800186306, + "grad_norm": 0.379133486073973, + "learning_rate": 3.786675871591301e-05, + "loss": 0.4157, + "step": 1025 + }, + { + "epoch": 0.95947834187238, + "grad_norm": 0.32560077911033863, + "learning_rate": 3.77804625474629e-05, + "loss": 0.4078, + "step": 1030 + }, + { + "epoch": 0.9641360037261295, + "grad_norm": 0.2715212772469474, + "learning_rate": 3.7694166379012776e-05, + "loss": 0.4108, + "step": 1035 + }, + { + "epoch": 0.9687936655798789, + "grad_norm": 0.3202605689576077, + "learning_rate": 3.7607870210562654e-05, + "loss": 0.4102, + "step": 1040 + }, + { + "epoch": 0.9734513274336283, + "grad_norm": 0.2922435947862865, + "learning_rate": 3.752157404211253e-05, + "loss": 0.4111, + "step": 1045 + }, + { + "epoch": 0.9781089892873778, + "grad_norm": 0.2957561031633552, + "learning_rate": 3.743527787366241e-05, + "loss": 0.4113, + "step": 1050 + }, + { + "epoch": 0.9827666511411272, + "grad_norm": 0.31048108883800707, + "learning_rate": 3.734898170521229e-05, + "loss": 0.4122, + "step": 1055 + }, + { + "epoch": 0.9874243129948765, + "grad_norm": 0.2953875423271191, + "learning_rate": 3.7262685536762174e-05, + "loss": 0.4152, + "step": 1060 + }, + { + "epoch": 0.992081974848626, + "grad_norm": 0.286137496896298, + "learning_rate": 3.7176389368312045e-05, + "loss": 0.4121, + "step": 1065 + }, + { + "epoch": 0.9967396367023754, + "grad_norm": 0.2831385238087909, + "learning_rate": 3.709009319986193e-05, + "loss": 0.4043, + "step": 1070 + }, + { + "epoch": 1.00093153237075, + "grad_norm": 0.4684349512233376, + "learning_rate": 3.700379703141181e-05, + "loss": 0.414, + "step": 1075 + }, + { + "epoch": 1.0055891942244992, + "grad_norm": 0.3538765149793847, + "learning_rate": 3.6917500862961687e-05, + "loss": 0.3541, + "step": 1080 + }, + { + "epoch": 1.0102468560782487, + "grad_norm": 0.30560129560141963, + "learning_rate": 3.6831204694511565e-05, + "loss": 0.3423, + "step": 1085 + }, + { + "epoch": 1.0149045179319982, + "grad_norm": 0.30243535095385876, + "learning_rate": 3.674490852606144e-05, + "loss": 0.3553, + "step": 1090 + }, + { + "epoch": 1.0195621797857475, + "grad_norm": 0.3507483425624034, + "learning_rate": 3.665861235761132e-05, + "loss": 0.3504, + "step": 1095 + }, + { + "epoch": 1.024219841639497, + "grad_norm": 0.32221792171674635, + "learning_rate": 3.6572316189161206e-05, + "loss": 0.3541, + "step": 1100 + }, + { + "epoch": 1.0288775034932465, + "grad_norm": 0.37215297022295046, + "learning_rate": 3.6486020020711085e-05, + "loss": 0.3534, + "step": 1105 + }, + { + "epoch": 1.0335351653469957, + "grad_norm": 1.1943932059811095, + "learning_rate": 3.639972385226096e-05, + "loss": 0.3546, + "step": 1110 + }, + { + "epoch": 1.0381928272007452, + "grad_norm": 0.3516112606363948, + "learning_rate": 3.631342768381084e-05, + "loss": 0.3469, + "step": 1115 + }, + { + "epoch": 1.0428504890544947, + "grad_norm": 0.28850962245298706, + "learning_rate": 3.622713151536072e-05, + "loss": 0.3538, + "step": 1120 + }, + { + "epoch": 1.047508150908244, + "grad_norm": 0.3314624580634308, + "learning_rate": 3.6140835346910604e-05, + "loss": 0.359, + "step": 1125 + }, + { + "epoch": 1.0521658127619935, + "grad_norm": 0.3790414890626171, + "learning_rate": 3.6054539178460476e-05, + "loss": 0.3443, + "step": 1130 + }, + { + "epoch": 1.056823474615743, + "grad_norm": 0.30670131819474084, + "learning_rate": 3.5968243010010354e-05, + "loss": 0.353, + "step": 1135 + }, + { + "epoch": 1.0614811364694923, + "grad_norm": 0.35114327907437415, + "learning_rate": 3.588194684156024e-05, + "loss": 0.3526, + "step": 1140 + }, + { + "epoch": 1.0661387983232418, + "grad_norm": 0.32134248505751767, + "learning_rate": 3.579565067311012e-05, + "loss": 0.3531, + "step": 1145 + }, + { + "epoch": 1.0707964601769913, + "grad_norm": 0.31651076269082523, + "learning_rate": 3.5709354504659995e-05, + "loss": 0.3491, + "step": 1150 + }, + { + "epoch": 1.0754541220307405, + "grad_norm": 0.2976154864236664, + "learning_rate": 3.5623058336209874e-05, + "loss": 0.3522, + "step": 1155 + }, + { + "epoch": 1.08011178388449, + "grad_norm": 0.32449478555851013, + "learning_rate": 3.553676216775975e-05, + "loss": 0.3595, + "step": 1160 + }, + { + "epoch": 1.0847694457382393, + "grad_norm": 0.27878607188094323, + "learning_rate": 3.545046599930964e-05, + "loss": 0.3583, + "step": 1165 + }, + { + "epoch": 1.0894271075919888, + "grad_norm": 0.31988584789330404, + "learning_rate": 3.536416983085951e-05, + "loss": 0.3556, + "step": 1170 + }, + { + "epoch": 1.0940847694457383, + "grad_norm": 0.3032847585581242, + "learning_rate": 3.5277873662409386e-05, + "loss": 0.3555, + "step": 1175 + }, + { + "epoch": 1.0987424312994876, + "grad_norm": 0.309027578439761, + "learning_rate": 3.519157749395927e-05, + "loss": 0.3492, + "step": 1180 + }, + { + "epoch": 1.103400093153237, + "grad_norm": 0.27391324423231966, + "learning_rate": 3.510528132550915e-05, + "loss": 0.3502, + "step": 1185 + }, + { + "epoch": 1.1080577550069866, + "grad_norm": 0.3061764334729438, + "learning_rate": 3.501898515705903e-05, + "loss": 0.3335, + "step": 1190 + }, + { + "epoch": 1.1127154168607358, + "grad_norm": 0.2875710397797018, + "learning_rate": 3.4932688988608906e-05, + "loss": 0.3417, + "step": 1195 + }, + { + "epoch": 1.1173730787144853, + "grad_norm": 0.2624004234850815, + "learning_rate": 3.4846392820158784e-05, + "loss": 0.3511, + "step": 1200 + }, + { + "epoch": 1.1220307405682348, + "grad_norm": 0.3248110653316974, + "learning_rate": 3.476009665170867e-05, + "loss": 0.3583, + "step": 1205 + }, + { + "epoch": 1.126688402421984, + "grad_norm": 0.3633571821910072, + "learning_rate": 3.467380048325855e-05, + "loss": 0.3533, + "step": 1210 + }, + { + "epoch": 1.1313460642757336, + "grad_norm": 0.33993904005897413, + "learning_rate": 3.458750431480842e-05, + "loss": 0.3488, + "step": 1215 + }, + { + "epoch": 1.136003726129483, + "grad_norm": 0.2870516215844625, + "learning_rate": 3.4501208146358304e-05, + "loss": 0.3613, + "step": 1220 + }, + { + "epoch": 1.1406613879832324, + "grad_norm": 0.3032689766119555, + "learning_rate": 3.441491197790818e-05, + "loss": 0.3495, + "step": 1225 + }, + { + "epoch": 1.1453190498369819, + "grad_norm": 0.28433837156321073, + "learning_rate": 3.432861580945806e-05, + "loss": 0.3505, + "step": 1230 + }, + { + "epoch": 1.1499767116907313, + "grad_norm": 0.291464118593231, + "learning_rate": 3.424231964100794e-05, + "loss": 0.3514, + "step": 1235 + }, + { + "epoch": 1.1546343735444806, + "grad_norm": 0.2885273892747404, + "learning_rate": 3.415602347255782e-05, + "loss": 0.3561, + "step": 1240 + }, + { + "epoch": 1.1592920353982301, + "grad_norm": 0.31723881770265705, + "learning_rate": 3.40697273041077e-05, + "loss": 0.3584, + "step": 1245 + }, + { + "epoch": 1.1639496972519794, + "grad_norm": 2.0745129939580536, + "learning_rate": 3.398343113565758e-05, + "loss": 0.3663, + "step": 1250 + }, + { + "epoch": 1.1686073591057289, + "grad_norm": 0.5108539849873057, + "learning_rate": 3.389713496720746e-05, + "loss": 0.3561, + "step": 1255 + }, + { + "epoch": 1.1732650209594784, + "grad_norm": 0.30039700859419805, + "learning_rate": 3.381083879875734e-05, + "loss": 0.3642, + "step": 1260 + }, + { + "epoch": 1.1779226828132279, + "grad_norm": 0.35686121830383705, + "learning_rate": 3.3724542630307215e-05, + "loss": 0.3563, + "step": 1265 + }, + { + "epoch": 1.1825803446669771, + "grad_norm": 0.3187570633882832, + "learning_rate": 3.363824646185709e-05, + "loss": 0.3435, + "step": 1270 + }, + { + "epoch": 1.1872380065207266, + "grad_norm": 0.332204557041565, + "learning_rate": 3.355195029340698e-05, + "loss": 0.3563, + "step": 1275 + }, + { + "epoch": 1.191895668374476, + "grad_norm": 0.3257790715751293, + "learning_rate": 3.346565412495685e-05, + "loss": 0.3572, + "step": 1280 + }, + { + "epoch": 1.1965533302282254, + "grad_norm": 0.3022135881303542, + "learning_rate": 3.3379357956506735e-05, + "loss": 0.3604, + "step": 1285 + }, + { + "epoch": 1.201210992081975, + "grad_norm": 0.3177687317250914, + "learning_rate": 3.329306178805661e-05, + "loss": 0.3532, + "step": 1290 + }, + { + "epoch": 1.2058686539357242, + "grad_norm": 1.2396857071159404, + "learning_rate": 3.320676561960649e-05, + "loss": 0.3574, + "step": 1295 + }, + { + "epoch": 1.2105263157894737, + "grad_norm": 0.3664373829434122, + "learning_rate": 3.312046945115637e-05, + "loss": 0.3533, + "step": 1300 + }, + { + "epoch": 1.2151839776432232, + "grad_norm": 0.3023963308504667, + "learning_rate": 3.303417328270625e-05, + "loss": 0.3613, + "step": 1305 + }, + { + "epoch": 1.2198416394969724, + "grad_norm": 0.3082958509214965, + "learning_rate": 3.2947877114256126e-05, + "loss": 0.3687, + "step": 1310 + }, + { + "epoch": 1.224499301350722, + "grad_norm": 0.36918987030345646, + "learning_rate": 3.286158094580601e-05, + "loss": 0.4267, + "step": 1315 + }, + { + "epoch": 1.2291569632044714, + "grad_norm": 0.35602913397714164, + "learning_rate": 3.277528477735589e-05, + "loss": 0.3634, + "step": 1320 + }, + { + "epoch": 1.2338146250582207, + "grad_norm": 0.3105028968314904, + "learning_rate": 3.268898860890577e-05, + "loss": 0.355, + "step": 1325 + }, + { + "epoch": 1.2384722869119702, + "grad_norm": 0.3141898633738905, + "learning_rate": 3.2602692440455645e-05, + "loss": 0.3599, + "step": 1330 + }, + { + "epoch": 1.2431299487657197, + "grad_norm": 0.34247754511180467, + "learning_rate": 3.2516396272005524e-05, + "loss": 0.3547, + "step": 1335 + }, + { + "epoch": 1.247787610619469, + "grad_norm": 0.3434502610535775, + "learning_rate": 3.243010010355541e-05, + "loss": 0.3547, + "step": 1340 + }, + { + "epoch": 1.2524452724732185, + "grad_norm": 0.3083559532522247, + "learning_rate": 3.234380393510528e-05, + "loss": 0.3438, + "step": 1345 + }, + { + "epoch": 1.257102934326968, + "grad_norm": 8.089934674724914, + "learning_rate": 3.225750776665516e-05, + "loss": 0.3566, + "step": 1350 + }, + { + "epoch": 1.2617605961807172, + "grad_norm": 23.609984098862725, + "learning_rate": 3.217121159820504e-05, + "loss": 0.3954, + "step": 1355 + }, + { + "epoch": 1.2664182580344667, + "grad_norm": 0.6589262851727549, + "learning_rate": 3.208491542975492e-05, + "loss": 0.4094, + "step": 1360 + }, + { + "epoch": 1.271075919888216, + "grad_norm": 0.37969701440271525, + "learning_rate": 3.19986192613048e-05, + "loss": 0.3618, + "step": 1365 + }, + { + "epoch": 1.2757335817419655, + "grad_norm": 0.35419265438638503, + "learning_rate": 3.191232309285468e-05, + "loss": 0.357, + "step": 1370 + }, + { + "epoch": 1.280391243595715, + "grad_norm": 0.3113731460367887, + "learning_rate": 3.1826026924404556e-05, + "loss": 0.3547, + "step": 1375 + }, + { + "epoch": 1.2850489054494645, + "grad_norm": 0.3711646260264564, + "learning_rate": 3.173973075595444e-05, + "loss": 0.3557, + "step": 1380 + }, + { + "epoch": 1.2897065673032138, + "grad_norm": 0.28159626489540107, + "learning_rate": 3.165343458750431e-05, + "loss": 0.3523, + "step": 1385 + }, + { + "epoch": 1.2943642291569633, + "grad_norm": 0.4368347910396357, + "learning_rate": 3.156713841905419e-05, + "loss": 0.3612, + "step": 1390 + }, + { + "epoch": 1.2990218910107125, + "grad_norm": 0.292048370076934, + "learning_rate": 3.1480842250604076e-05, + "loss": 0.3551, + "step": 1395 + }, + { + "epoch": 1.303679552864462, + "grad_norm": 0.28783970311720675, + "learning_rate": 3.1394546082153954e-05, + "loss": 0.3566, + "step": 1400 + }, + { + "epoch": 1.3083372147182115, + "grad_norm": 0.39691810792427434, + "learning_rate": 3.130824991370383e-05, + "loss": 0.3499, + "step": 1405 + }, + { + "epoch": 1.312994876571961, + "grad_norm": 5.1812004529766, + "learning_rate": 3.122195374525371e-05, + "loss": 0.3509, + "step": 1410 + }, + { + "epoch": 1.3176525384257103, + "grad_norm": 3.5459716211561947, + "learning_rate": 3.113565757680359e-05, + "loss": 0.3695, + "step": 1415 + }, + { + "epoch": 1.3223102002794598, + "grad_norm": 0.4018389678575513, + "learning_rate": 3.1049361408353474e-05, + "loss": 0.358, + "step": 1420 + }, + { + "epoch": 1.326967862133209, + "grad_norm": 0.37316511070692127, + "learning_rate": 3.096306523990335e-05, + "loss": 0.3503, + "step": 1425 + }, + { + "epoch": 1.3316255239869585, + "grad_norm": 0.4253066965910869, + "learning_rate": 3.0876769071453223e-05, + "loss": 0.3519, + "step": 1430 + }, + { + "epoch": 1.336283185840708, + "grad_norm": 0.41012648718576017, + "learning_rate": 3.079047290300311e-05, + "loss": 0.3577, + "step": 1435 + }, + { + "epoch": 1.3409408476944573, + "grad_norm": 0.4660255215590989, + "learning_rate": 3.070417673455299e-05, + "loss": 0.3547, + "step": 1440 + }, + { + "epoch": 1.3455985095482068, + "grad_norm": 0.9606859956101274, + "learning_rate": 3.061788056610287e-05, + "loss": 0.3603, + "step": 1445 + }, + { + "epoch": 1.350256171401956, + "grad_norm": 0.2729737086954493, + "learning_rate": 3.053158439765274e-05, + "loss": 0.3536, + "step": 1450 + }, + { + "epoch": 1.3549138332557056, + "grad_norm": 0.3185706265133299, + "learning_rate": 3.0445288229202625e-05, + "loss": 0.363, + "step": 1455 + }, + { + "epoch": 1.359571495109455, + "grad_norm": 0.29368292075603664, + "learning_rate": 3.0358992060752506e-05, + "loss": 0.346, + "step": 1460 + }, + { + "epoch": 1.3642291569632046, + "grad_norm": 0.6188898818836498, + "learning_rate": 3.027269589230238e-05, + "loss": 0.3522, + "step": 1465 + }, + { + "epoch": 1.3688868188169538, + "grad_norm": 0.29275072649894884, + "learning_rate": 3.018639972385226e-05, + "loss": 0.3561, + "step": 1470 + }, + { + "epoch": 1.3735444806707033, + "grad_norm": 0.3123873178331315, + "learning_rate": 3.010010355540214e-05, + "loss": 0.3494, + "step": 1475 + }, + { + "epoch": 1.3782021425244526, + "grad_norm": 0.3339526911219289, + "learning_rate": 3.001380738695202e-05, + "loss": 0.3518, + "step": 1480 + }, + { + "epoch": 1.382859804378202, + "grad_norm": 0.41947552863552484, + "learning_rate": 2.99275112185019e-05, + "loss": 0.3576, + "step": 1485 + }, + { + "epoch": 1.3875174662319516, + "grad_norm": 0.290834349215534, + "learning_rate": 2.984121505005178e-05, + "loss": 0.358, + "step": 1490 + }, + { + "epoch": 1.392175128085701, + "grad_norm": 0.3257789570431207, + "learning_rate": 2.9754918881601657e-05, + "loss": 0.3589, + "step": 1495 + }, + { + "epoch": 1.3968327899394504, + "grad_norm": 0.2697638582059203, + "learning_rate": 2.966862271315154e-05, + "loss": 0.3468, + "step": 1500 + }, + { + "epoch": 1.4014904517931999, + "grad_norm": 0.34596498712608775, + "learning_rate": 2.9582326544701417e-05, + "loss": 0.3516, + "step": 1505 + }, + { + "epoch": 1.4061481136469491, + "grad_norm": 0.3424362255444567, + "learning_rate": 2.9496030376251292e-05, + "loss": 0.3505, + "step": 1510 + }, + { + "epoch": 1.4108057755006986, + "grad_norm": 0.28515100314876, + "learning_rate": 2.9409734207801177e-05, + "loss": 0.3591, + "step": 1515 + }, + { + "epoch": 1.4154634373544481, + "grad_norm": 0.27947785201095854, + "learning_rate": 2.9323438039351052e-05, + "loss": 0.3522, + "step": 1520 + }, + { + "epoch": 1.4201210992081974, + "grad_norm": 0.5037464671470212, + "learning_rate": 2.9237141870900937e-05, + "loss": 0.352, + "step": 1525 + }, + { + "epoch": 1.424778761061947, + "grad_norm": 0.3427793038690582, + "learning_rate": 2.9150845702450812e-05, + "loss": 0.3435, + "step": 1530 + }, + { + "epoch": 1.4294364229156964, + "grad_norm": 0.2882568116228797, + "learning_rate": 2.906454953400069e-05, + "loss": 0.3584, + "step": 1535 + }, + { + "epoch": 1.4340940847694457, + "grad_norm": 0.3135134737473443, + "learning_rate": 2.897825336555057e-05, + "loss": 0.3499, + "step": 1540 + }, + { + "epoch": 1.4387517466231952, + "grad_norm": 0.30701305078292784, + "learning_rate": 2.889195719710045e-05, + "loss": 0.3498, + "step": 1545 + }, + { + "epoch": 1.4434094084769447, + "grad_norm": 0.9739810284082362, + "learning_rate": 2.8805661028650328e-05, + "loss": 0.3461, + "step": 1550 + }, + { + "epoch": 1.448067070330694, + "grad_norm": 0.30165717081554694, + "learning_rate": 2.871936486020021e-05, + "loss": 0.342, + "step": 1555 + }, + { + "epoch": 1.4527247321844434, + "grad_norm": 0.2613673867436863, + "learning_rate": 2.8633068691750088e-05, + "loss": 0.3489, + "step": 1560 + }, + { + "epoch": 1.4573823940381927, + "grad_norm": 0.32713429396994886, + "learning_rate": 2.854677252329997e-05, + "loss": 0.357, + "step": 1565 + }, + { + "epoch": 1.4620400558919422, + "grad_norm": 3.226986295868854, + "learning_rate": 2.8460476354849848e-05, + "loss": 0.3683, + "step": 1570 + }, + { + "epoch": 1.4666977177456917, + "grad_norm": 0.26293280030526456, + "learning_rate": 2.8374180186399723e-05, + "loss": 0.3511, + "step": 1575 + }, + { + "epoch": 1.4713553795994412, + "grad_norm": 0.28666579972329054, + "learning_rate": 2.8287884017949608e-05, + "loss": 0.344, + "step": 1580 + }, + { + "epoch": 1.4760130414531905, + "grad_norm": 0.27704759464972445, + "learning_rate": 2.8201587849499482e-05, + "loss": 0.3524, + "step": 1585 + }, + { + "epoch": 1.48067070330694, + "grad_norm": 0.28594874213631005, + "learning_rate": 2.811529168104936e-05, + "loss": 0.3512, + "step": 1590 + }, + { + "epoch": 1.4853283651606892, + "grad_norm": 0.27066447710153146, + "learning_rate": 2.8028995512599242e-05, + "loss": 0.3493, + "step": 1595 + }, + { + "epoch": 1.4899860270144387, + "grad_norm": 0.2600899746581506, + "learning_rate": 2.794269934414912e-05, + "loss": 0.3546, + "step": 1600 + }, + { + "epoch": 1.4946436888681882, + "grad_norm": 0.2500046923889755, + "learning_rate": 2.7856403175699002e-05, + "loss": 0.3465, + "step": 1605 + }, + { + "epoch": 1.4993013507219377, + "grad_norm": 0.2567828872835996, + "learning_rate": 2.777010700724888e-05, + "loss": 0.352, + "step": 1610 + }, + { + "epoch": 1.503959012575687, + "grad_norm": 0.2779666234603821, + "learning_rate": 2.768381083879876e-05, + "loss": 0.3507, + "step": 1615 + }, + { + "epoch": 1.5086166744294365, + "grad_norm": 0.27186485130837373, + "learning_rate": 2.759751467034864e-05, + "loss": 0.3516, + "step": 1620 + }, + { + "epoch": 1.5132743362831858, + "grad_norm": 0.24645955549365214, + "learning_rate": 2.751121850189852e-05, + "loss": 0.3546, + "step": 1625 + }, + { + "epoch": 1.5179319981369352, + "grad_norm": 0.2542323054350234, + "learning_rate": 2.7424922333448393e-05, + "loss": 0.3515, + "step": 1630 + }, + { + "epoch": 1.5225896599906847, + "grad_norm": 0.25933000013528096, + "learning_rate": 2.7338626164998278e-05, + "loss": 0.3616, + "step": 1635 + }, + { + "epoch": 1.5272473218444342, + "grad_norm": 0.2578174399758813, + "learning_rate": 2.7252329996548153e-05, + "loss": 0.3555, + "step": 1640 + }, + { + "epoch": 1.5319049836981835, + "grad_norm": 0.27068988078684453, + "learning_rate": 2.7166033828098038e-05, + "loss": 0.359, + "step": 1645 + }, + { + "epoch": 1.5365626455519328, + "grad_norm": 0.2677801002022169, + "learning_rate": 2.7079737659647913e-05, + "loss": 0.3437, + "step": 1650 + }, + { + "epoch": 1.5412203074056823, + "grad_norm": 0.2660552731038306, + "learning_rate": 2.699344149119779e-05, + "loss": 0.3424, + "step": 1655 + }, + { + "epoch": 1.5458779692594318, + "grad_norm": 0.28523039735998723, + "learning_rate": 2.6907145322747673e-05, + "loss": 0.3418, + "step": 1660 + }, + { + "epoch": 1.5505356311131813, + "grad_norm": 0.28731844231489023, + "learning_rate": 2.682084915429755e-05, + "loss": 0.3543, + "step": 1665 + }, + { + "epoch": 1.5551932929669308, + "grad_norm": 1.4775974232314364, + "learning_rate": 2.673455298584743e-05, + "loss": 0.3518, + "step": 1670 + }, + { + "epoch": 1.55985095482068, + "grad_norm": 0.24943372088294952, + "learning_rate": 2.664825681739731e-05, + "loss": 0.348, + "step": 1675 + }, + { + "epoch": 1.5645086166744293, + "grad_norm": 0.27406670344929485, + "learning_rate": 2.6561960648947186e-05, + "loss": 0.3548, + "step": 1680 + }, + { + "epoch": 1.5691662785281788, + "grad_norm": 0.26608558873313726, + "learning_rate": 2.647566448049707e-05, + "loss": 0.3444, + "step": 1685 + }, + { + "epoch": 1.5738239403819283, + "grad_norm": 0.27726086768562713, + "learning_rate": 2.6389368312046945e-05, + "loss": 0.3522, + "step": 1690 + }, + { + "epoch": 1.5784816022356778, + "grad_norm": 0.24678701765325478, + "learning_rate": 2.6303072143596824e-05, + "loss": 0.3487, + "step": 1695 + }, + { + "epoch": 1.583139264089427, + "grad_norm": 0.2712396806672742, + "learning_rate": 2.6216775975146705e-05, + "loss": 0.3519, + "step": 1700 + }, + { + "epoch": 1.5877969259431766, + "grad_norm": 0.2517013095851643, + "learning_rate": 2.6130479806696584e-05, + "loss": 0.3463, + "step": 1705 + }, + { + "epoch": 1.5924545877969258, + "grad_norm": 0.250371729590157, + "learning_rate": 2.6044183638246462e-05, + "loss": 0.3538, + "step": 1710 + }, + { + "epoch": 1.5971122496506753, + "grad_norm": 0.275342753890394, + "learning_rate": 2.5957887469796343e-05, + "loss": 0.3555, + "step": 1715 + }, + { + "epoch": 1.6017699115044248, + "grad_norm": 0.24313053813658697, + "learning_rate": 2.587159130134622e-05, + "loss": 0.342, + "step": 1720 + }, + { + "epoch": 1.6064275733581743, + "grad_norm": 0.2722072414317194, + "learning_rate": 2.5785295132896096e-05, + "loss": 0.3488, + "step": 1725 + }, + { + "epoch": 1.6110852352119236, + "grad_norm": 0.2987110807378229, + "learning_rate": 2.569899896444598e-05, + "loss": 0.3599, + "step": 1730 + }, + { + "epoch": 1.6157428970656729, + "grad_norm": 0.2774231229703875, + "learning_rate": 2.5612702795995856e-05, + "loss": 0.3587, + "step": 1735 + }, + { + "epoch": 1.6204005589194224, + "grad_norm": 0.27635376671669754, + "learning_rate": 2.552640662754574e-05, + "loss": 0.3402, + "step": 1740 + }, + { + "epoch": 1.6250582207731719, + "grad_norm": 0.25478389391029976, + "learning_rate": 2.5440110459095616e-05, + "loss": 0.3687, + "step": 1745 + }, + { + "epoch": 1.6297158826269214, + "grad_norm": 0.2781430915699508, + "learning_rate": 2.5353814290645494e-05, + "loss": 0.3483, + "step": 1750 + }, + { + "epoch": 1.6343735444806708, + "grad_norm": 0.2586169106331023, + "learning_rate": 2.5267518122195376e-05, + "loss": 0.356, + "step": 1755 + }, + { + "epoch": 1.6390312063344201, + "grad_norm": 0.25373603422413227, + "learning_rate": 2.5181221953745254e-05, + "loss": 0.3433, + "step": 1760 + }, + { + "epoch": 1.6436888681881694, + "grad_norm": 0.25761180042874776, + "learning_rate": 2.5094925785295132e-05, + "loss": 0.3454, + "step": 1765 + }, + { + "epoch": 1.648346530041919, + "grad_norm": 0.2832378466169248, + "learning_rate": 2.5008629616845014e-05, + "loss": 0.3419, + "step": 1770 + }, + { + "epoch": 1.6530041918956684, + "grad_norm": 0.24635731465935254, + "learning_rate": 2.4922333448394892e-05, + "loss": 0.3561, + "step": 1775 + }, + { + "epoch": 1.6576618537494179, + "grad_norm": 0.9005827663380318, + "learning_rate": 2.483603727994477e-05, + "loss": 0.3441, + "step": 1780 + }, + { + "epoch": 1.6623195156031674, + "grad_norm": 0.3437481829556501, + "learning_rate": 2.4749741111494652e-05, + "loss": 0.3573, + "step": 1785 + }, + { + "epoch": 1.6669771774569166, + "grad_norm": 0.36819542687532286, + "learning_rate": 2.466344494304453e-05, + "loss": 0.3461, + "step": 1790 + }, + { + "epoch": 1.671634839310666, + "grad_norm": 0.3068015778936282, + "learning_rate": 2.457714877459441e-05, + "loss": 0.3444, + "step": 1795 + }, + { + "epoch": 1.6762925011644154, + "grad_norm": 0.2618150436269955, + "learning_rate": 2.4490852606144287e-05, + "loss": 0.3426, + "step": 1800 + }, + { + "epoch": 1.680950163018165, + "grad_norm": 0.25451524024562244, + "learning_rate": 2.440455643769417e-05, + "loss": 0.3511, + "step": 1805 + }, + { + "epoch": 1.6856078248719144, + "grad_norm": 0.2695440900949316, + "learning_rate": 2.4318260269244047e-05, + "loss": 0.3477, + "step": 1810 + }, + { + "epoch": 1.6902654867256637, + "grad_norm": 0.2674915183495278, + "learning_rate": 2.4231964100793925e-05, + "loss": 0.3485, + "step": 1815 + }, + { + "epoch": 1.6949231485794132, + "grad_norm": 0.31579487413450547, + "learning_rate": 2.4145667932343803e-05, + "loss": 0.3375, + "step": 1820 + }, + { + "epoch": 1.6995808104331624, + "grad_norm": 0.2731513416518806, + "learning_rate": 2.4059371763893685e-05, + "loss": 0.3404, + "step": 1825 + }, + { + "epoch": 1.704238472286912, + "grad_norm": 0.26736069371041915, + "learning_rate": 2.3973075595443563e-05, + "loss": 0.3432, + "step": 1830 + }, + { + "epoch": 1.7088961341406614, + "grad_norm": 0.2871495858381633, + "learning_rate": 2.388677942699344e-05, + "loss": 0.3508, + "step": 1835 + }, + { + "epoch": 1.713553795994411, + "grad_norm": 0.24116315103304178, + "learning_rate": 2.3800483258543323e-05, + "loss": 0.3446, + "step": 1840 + }, + { + "epoch": 1.7182114578481602, + "grad_norm": 0.25078460812146336, + "learning_rate": 2.37141870900932e-05, + "loss": 0.349, + "step": 1845 + }, + { + "epoch": 1.7228691197019095, + "grad_norm": 0.29558613517452564, + "learning_rate": 2.3627890921643083e-05, + "loss": 0.357, + "step": 1850 + }, + { + "epoch": 1.727526781555659, + "grad_norm": 0.25402857290536573, + "learning_rate": 2.3541594753192957e-05, + "loss": 0.341, + "step": 1855 + }, + { + "epoch": 1.7321844434094085, + "grad_norm": 0.26708730846147416, + "learning_rate": 2.345529858474284e-05, + "loss": 0.3625, + "step": 1860 + }, + { + "epoch": 1.736842105263158, + "grad_norm": 0.25810926942570817, + "learning_rate": 2.3369002416292717e-05, + "loss": 0.3424, + "step": 1865 + }, + { + "epoch": 1.7414997671169075, + "grad_norm": 0.25777408289568604, + "learning_rate": 2.32827062478426e-05, + "loss": 0.3469, + "step": 1870 + }, + { + "epoch": 1.7461574289706567, + "grad_norm": 0.28033996112574766, + "learning_rate": 2.3196410079392474e-05, + "loss": 0.3604, + "step": 1875 + }, + { + "epoch": 1.750815090824406, + "grad_norm": 0.23366951536520036, + "learning_rate": 2.3110113910942355e-05, + "loss": 0.3525, + "step": 1880 + }, + { + "epoch": 1.7554727526781555, + "grad_norm": 0.25296859342059685, + "learning_rate": 2.3023817742492234e-05, + "loss": 0.3531, + "step": 1885 + }, + { + "epoch": 1.760130414531905, + "grad_norm": 0.24983643896715807, + "learning_rate": 2.2937521574042115e-05, + "loss": 0.3512, + "step": 1890 + }, + { + "epoch": 1.7647880763856545, + "grad_norm": 0.2716754177639391, + "learning_rate": 2.285122540559199e-05, + "loss": 0.3589, + "step": 1895 + }, + { + "epoch": 1.7694457382394038, + "grad_norm": 0.24135095593523273, + "learning_rate": 2.276492923714187e-05, + "loss": 0.3492, + "step": 1900 + }, + { + "epoch": 1.7741034000931533, + "grad_norm": 0.26280121470301687, + "learning_rate": 2.267863306869175e-05, + "loss": 0.35, + "step": 1905 + }, + { + "epoch": 1.7787610619469025, + "grad_norm": 0.24986188415439609, + "learning_rate": 2.259233690024163e-05, + "loss": 0.341, + "step": 1910 + }, + { + "epoch": 1.783418723800652, + "grad_norm": 0.24809669885312474, + "learning_rate": 2.250604073179151e-05, + "loss": 0.3383, + "step": 1915 + }, + { + "epoch": 1.7880763856544015, + "grad_norm": 1.4295723229351893, + "learning_rate": 2.2419744563341388e-05, + "loss": 0.3429, + "step": 1920 + }, + { + "epoch": 1.792734047508151, + "grad_norm": 0.29048553056880866, + "learning_rate": 2.233344839489127e-05, + "loss": 0.3522, + "step": 1925 + }, + { + "epoch": 1.7973917093619003, + "grad_norm": 0.2738000474300528, + "learning_rate": 2.2247152226441148e-05, + "loss": 0.3406, + "step": 1930 + }, + { + "epoch": 1.8020493712156498, + "grad_norm": 0.26755797749035115, + "learning_rate": 2.2160856057991026e-05, + "loss": 0.339, + "step": 1935 + }, + { + "epoch": 1.806707033069399, + "grad_norm": 0.24051646148161943, + "learning_rate": 2.2074559889540904e-05, + "loss": 0.3415, + "step": 1940 + }, + { + "epoch": 1.8113646949231486, + "grad_norm": 0.4480396154759198, + "learning_rate": 2.1988263721090786e-05, + "loss": 0.3504, + "step": 1945 + }, + { + "epoch": 1.816022356776898, + "grad_norm": 0.2585475391328862, + "learning_rate": 2.1901967552640664e-05, + "loss": 0.3423, + "step": 1950 + }, + { + "epoch": 1.8206800186306475, + "grad_norm": 0.23815718303661385, + "learning_rate": 2.1815671384190542e-05, + "loss": 0.3447, + "step": 1955 + }, + { + "epoch": 1.8253376804843968, + "grad_norm": 0.308300042750051, + "learning_rate": 2.172937521574042e-05, + "loss": 0.347, + "step": 1960 + }, + { + "epoch": 1.829995342338146, + "grad_norm": 0.25553311828654945, + "learning_rate": 2.1643079047290302e-05, + "loss": 0.3487, + "step": 1965 + }, + { + "epoch": 1.8346530041918956, + "grad_norm": 0.3588389772067993, + "learning_rate": 2.155678287884018e-05, + "loss": 0.3434, + "step": 1970 + }, + { + "epoch": 1.839310666045645, + "grad_norm": 0.26552906632603784, + "learning_rate": 2.147048671039006e-05, + "loss": 0.3503, + "step": 1975 + }, + { + "epoch": 1.8439683278993946, + "grad_norm": 0.25976308131809595, + "learning_rate": 2.1384190541939937e-05, + "loss": 0.3418, + "step": 1980 + }, + { + "epoch": 1.848625989753144, + "grad_norm": 0.2608984454672737, + "learning_rate": 2.129789437348982e-05, + "loss": 0.3539, + "step": 1985 + }, + { + "epoch": 1.8532836516068933, + "grad_norm": 0.6841187702059441, + "learning_rate": 2.1211598205039697e-05, + "loss": 0.3506, + "step": 1990 + }, + { + "epoch": 1.8579413134606426, + "grad_norm": 0.29300984860851426, + "learning_rate": 2.1125302036589575e-05, + "loss": 0.3378, + "step": 1995 + }, + { + "epoch": 1.8625989753143921, + "grad_norm": 0.25853621490849804, + "learning_rate": 2.1039005868139457e-05, + "loss": 0.3471, + "step": 2000 + }, + { + "epoch": 1.8672566371681416, + "grad_norm": 0.271087074061142, + "learning_rate": 2.0952709699689335e-05, + "loss": 0.3526, + "step": 2005 + }, + { + "epoch": 1.871914299021891, + "grad_norm": 0.2601235728752933, + "learning_rate": 2.0866413531239216e-05, + "loss": 0.3417, + "step": 2010 + }, + { + "epoch": 1.8765719608756404, + "grad_norm": 0.2588273475069737, + "learning_rate": 2.078011736278909e-05, + "loss": 0.3526, + "step": 2015 + }, + { + "epoch": 1.8812296227293899, + "grad_norm": 0.4189872439881, + "learning_rate": 2.0693821194338973e-05, + "loss": 0.3411, + "step": 2020 + }, + { + "epoch": 1.8858872845831391, + "grad_norm": 0.26516086453032495, + "learning_rate": 2.060752502588885e-05, + "loss": 0.3452, + "step": 2025 + }, + { + "epoch": 1.8905449464368886, + "grad_norm": 0.25754850179921396, + "learning_rate": 2.0521228857438733e-05, + "loss": 0.3428, + "step": 2030 + }, + { + "epoch": 1.8952026082906381, + "grad_norm": 0.2254389135020473, + "learning_rate": 2.0434932688988608e-05, + "loss": 0.3407, + "step": 2035 + }, + { + "epoch": 1.8998602701443876, + "grad_norm": 1.2877944565555752, + "learning_rate": 2.034863652053849e-05, + "loss": 0.3553, + "step": 2040 + }, + { + "epoch": 1.904517931998137, + "grad_norm": 0.23303512944550436, + "learning_rate": 2.0262340352088367e-05, + "loss": 0.3512, + "step": 2045 + }, + { + "epoch": 1.9091755938518864, + "grad_norm": 0.39910536743505765, + "learning_rate": 2.017604418363825e-05, + "loss": 0.3472, + "step": 2050 + }, + { + "epoch": 1.9138332557056357, + "grad_norm": 0.23961981952299882, + "learning_rate": 2.0089748015188127e-05, + "loss": 0.3465, + "step": 2055 + }, + { + "epoch": 1.9184909175593852, + "grad_norm": 0.2790757827986385, + "learning_rate": 2.0003451846738005e-05, + "loss": 0.3416, + "step": 2060 + }, + { + "epoch": 1.9231485794131347, + "grad_norm": 0.26541183998192874, + "learning_rate": 1.9917155678287887e-05, + "loss": 0.3426, + "step": 2065 + }, + { + "epoch": 1.9278062412668842, + "grad_norm": 0.26020009713009595, + "learning_rate": 1.9830859509837765e-05, + "loss": 0.3454, + "step": 2070 + }, + { + "epoch": 1.9324639031206334, + "grad_norm": 0.26844426587023945, + "learning_rate": 1.9744563341387643e-05, + "loss": 0.3453, + "step": 2075 + }, + { + "epoch": 1.9371215649743827, + "grad_norm": 0.2526468533029043, + "learning_rate": 1.9658267172937522e-05, + "loss": 0.339, + "step": 2080 + }, + { + "epoch": 1.9417792268281322, + "grad_norm": 0.25099498909407186, + "learning_rate": 1.9571971004487403e-05, + "loss": 0.3492, + "step": 2085 + }, + { + "epoch": 1.9464368886818817, + "grad_norm": 0.22893003769931286, + "learning_rate": 1.948567483603728e-05, + "loss": 0.3337, + "step": 2090 + }, + { + "epoch": 1.9510945505356312, + "grad_norm": 0.2499469298292734, + "learning_rate": 1.939937866758716e-05, + "loss": 0.3414, + "step": 2095 + }, + { + "epoch": 1.9557522123893807, + "grad_norm": 3.01516996652908, + "learning_rate": 1.9313082499137038e-05, + "loss": 0.3425, + "step": 2100 + }, + { + "epoch": 1.96040987424313, + "grad_norm": 0.235620525897121, + "learning_rate": 1.922678633068692e-05, + "loss": 0.3507, + "step": 2105 + }, + { + "epoch": 1.9650675360968792, + "grad_norm": 0.2271304553516188, + "learning_rate": 1.9140490162236798e-05, + "loss": 0.336, + "step": 2110 + }, + { + "epoch": 1.9697251979506287, + "grad_norm": 0.26095880436114127, + "learning_rate": 1.9054193993786676e-05, + "loss": 0.3429, + "step": 2115 + }, + { + "epoch": 1.9743828598043782, + "grad_norm": 0.2481845530711676, + "learning_rate": 1.8967897825336554e-05, + "loss": 0.346, + "step": 2120 + }, + { + "epoch": 1.9790405216581277, + "grad_norm": 0.2646587212417876, + "learning_rate": 1.8881601656886436e-05, + "loss": 0.3417, + "step": 2125 + }, + { + "epoch": 1.983698183511877, + "grad_norm": 0.23802879335165028, + "learning_rate": 1.8795305488436314e-05, + "loss": 0.3436, + "step": 2130 + }, + { + "epoch": 1.9883558453656265, + "grad_norm": 0.24832645092634265, + "learning_rate": 1.8709009319986192e-05, + "loss": 0.3487, + "step": 2135 + }, + { + "epoch": 1.9930135072193758, + "grad_norm": 0.2849361054588884, + "learning_rate": 1.8622713151536074e-05, + "loss": 0.347, + "step": 2140 + }, + { + "epoch": 1.9976711690731253, + "grad_norm": 0.24796434427364134, + "learning_rate": 1.8536416983085952e-05, + "loss": 0.3487, + "step": 2145 + }, + { + "epoch": 2.0018630647415, + "grad_norm": 0.4670077104840539, + "learning_rate": 1.8450120814635834e-05, + "loss": 0.3226, + "step": 2150 + }, + { + "epoch": 2.0065207265952494, + "grad_norm": 0.45452033829534516, + "learning_rate": 1.836382464618571e-05, + "loss": 0.2786, + "step": 2155 + }, + { + "epoch": 2.0111783884489984, + "grad_norm": 0.2794876251591008, + "learning_rate": 1.827752847773559e-05, + "loss": 0.2715, + "step": 2160 + }, + { + "epoch": 2.015836050302748, + "grad_norm": 0.2801100784062534, + "learning_rate": 1.819123230928547e-05, + "loss": 0.272, + "step": 2165 + }, + { + "epoch": 2.0204937121564974, + "grad_norm": 0.2817318982378413, + "learning_rate": 1.810493614083535e-05, + "loss": 0.2638, + "step": 2170 + }, + { + "epoch": 2.025151374010247, + "grad_norm": 0.274443503547923, + "learning_rate": 1.8018639972385225e-05, + "loss": 0.2749, + "step": 2175 + }, + { + "epoch": 2.0298090358639964, + "grad_norm": 0.2693048090720915, + "learning_rate": 1.7932343803935107e-05, + "loss": 0.2662, + "step": 2180 + }, + { + "epoch": 2.034466697717746, + "grad_norm": 0.29616883885176765, + "learning_rate": 1.7846047635484985e-05, + "loss": 0.2698, + "step": 2185 + }, + { + "epoch": 2.039124359571495, + "grad_norm": 0.2861132373580431, + "learning_rate": 1.7759751467034866e-05, + "loss": 0.2747, + "step": 2190 + }, + { + "epoch": 2.0437820214252445, + "grad_norm": 0.2673065715270604, + "learning_rate": 1.767345529858474e-05, + "loss": 0.2707, + "step": 2195 + }, + { + "epoch": 2.048439683278994, + "grad_norm": 0.2984804602495079, + "learning_rate": 1.7587159130134623e-05, + "loss": 0.2679, + "step": 2200 + }, + { + "epoch": 2.0530973451327434, + "grad_norm": 0.2651492582204811, + "learning_rate": 1.75008629616845e-05, + "loss": 0.2722, + "step": 2205 + }, + { + "epoch": 2.057755006986493, + "grad_norm": 0.2535902737834232, + "learning_rate": 1.7414566793234383e-05, + "loss": 0.2616, + "step": 2210 + }, + { + "epoch": 2.062412668840242, + "grad_norm": 0.38724908413622083, + "learning_rate": 1.732827062478426e-05, + "loss": 0.2706, + "step": 2215 + }, + { + "epoch": 2.0670703306939915, + "grad_norm": 0.29418307464847004, + "learning_rate": 1.724197445633414e-05, + "loss": 0.2686, + "step": 2220 + }, + { + "epoch": 2.071727992547741, + "grad_norm": 0.2685619174819881, + "learning_rate": 1.715567828788402e-05, + "loss": 0.2668, + "step": 2225 + }, + { + "epoch": 2.0763856544014905, + "grad_norm": 0.258294334458138, + "learning_rate": 1.70693821194339e-05, + "loss": 0.2696, + "step": 2230 + }, + { + "epoch": 2.08104331625524, + "grad_norm": 0.2654842147579374, + "learning_rate": 1.6983085950983777e-05, + "loss": 0.2638, + "step": 2235 + }, + { + "epoch": 2.0857009781089895, + "grad_norm": 0.29454304913296336, + "learning_rate": 1.6896789782533655e-05, + "loss": 0.2744, + "step": 2240 + }, + { + "epoch": 2.0903586399627385, + "grad_norm": 0.2853537828947393, + "learning_rate": 1.6810493614083537e-05, + "loss": 0.2704, + "step": 2245 + }, + { + "epoch": 2.095016301816488, + "grad_norm": 0.287794357838766, + "learning_rate": 1.6724197445633415e-05, + "loss": 0.2649, + "step": 2250 + }, + { + "epoch": 2.0996739636702375, + "grad_norm": 0.2689915249768445, + "learning_rate": 1.6637901277183294e-05, + "loss": 0.2755, + "step": 2255 + }, + { + "epoch": 2.104331625523987, + "grad_norm": 0.2763831642932531, + "learning_rate": 1.6551605108733172e-05, + "loss": 0.2668, + "step": 2260 + }, + { + "epoch": 2.1089892873777365, + "grad_norm": 0.27304407290976945, + "learning_rate": 1.6465308940283053e-05, + "loss": 0.2745, + "step": 2265 + }, + { + "epoch": 2.113646949231486, + "grad_norm": 0.2647848732867209, + "learning_rate": 1.637901277183293e-05, + "loss": 0.2682, + "step": 2270 + }, + { + "epoch": 2.118304611085235, + "grad_norm": 0.5195120921816434, + "learning_rate": 1.629271660338281e-05, + "loss": 0.2694, + "step": 2275 + }, + { + "epoch": 2.1229622729389845, + "grad_norm": 0.2525362731340643, + "learning_rate": 1.620642043493269e-05, + "loss": 0.27, + "step": 2280 + }, + { + "epoch": 2.127619934792734, + "grad_norm": 0.271141237976331, + "learning_rate": 1.612012426648257e-05, + "loss": 0.2736, + "step": 2285 + }, + { + "epoch": 2.1322775966464835, + "grad_norm": 0.29015424881388735, + "learning_rate": 1.6033828098032448e-05, + "loss": 0.2721, + "step": 2290 + }, + { + "epoch": 2.136935258500233, + "grad_norm": 0.25251985345507644, + "learning_rate": 1.5947531929582326e-05, + "loss": 0.2746, + "step": 2295 + }, + { + "epoch": 2.1415929203539825, + "grad_norm": 0.264331389896045, + "learning_rate": 1.5861235761132208e-05, + "loss": 0.2756, + "step": 2300 + }, + { + "epoch": 2.1462505822077316, + "grad_norm": 0.2818467534538804, + "learning_rate": 1.5774939592682086e-05, + "loss": 0.2736, + "step": 2305 + }, + { + "epoch": 2.150908244061481, + "grad_norm": 0.24602014197101432, + "learning_rate": 1.5688643424231964e-05, + "loss": 0.2615, + "step": 2310 + }, + { + "epoch": 2.1555659059152306, + "grad_norm": 0.26167388498791416, + "learning_rate": 1.5602347255781842e-05, + "loss": 0.2674, + "step": 2315 + }, + { + "epoch": 2.16022356776898, + "grad_norm": 0.2517441075418996, + "learning_rate": 1.5516051087331724e-05, + "loss": 0.2658, + "step": 2320 + }, + { + "epoch": 2.1648812296227296, + "grad_norm": 0.2753289696569991, + "learning_rate": 1.5429754918881602e-05, + "loss": 0.2706, + "step": 2325 + }, + { + "epoch": 2.1695388914764786, + "grad_norm": 0.24519322947304775, + "learning_rate": 1.534345875043148e-05, + "loss": 0.2711, + "step": 2330 + }, + { + "epoch": 2.174196553330228, + "grad_norm": 0.2709491329166551, + "learning_rate": 1.525716258198136e-05, + "loss": 0.2653, + "step": 2335 + }, + { + "epoch": 2.1788542151839776, + "grad_norm": 0.313781192883122, + "learning_rate": 1.517086641353124e-05, + "loss": 0.2686, + "step": 2340 + }, + { + "epoch": 2.183511877037727, + "grad_norm": 0.2652081894883003, + "learning_rate": 1.508457024508112e-05, + "loss": 0.2766, + "step": 2345 + }, + { + "epoch": 2.1881695388914766, + "grad_norm": 0.3001191925728167, + "learning_rate": 1.4998274076630997e-05, + "loss": 0.2677, + "step": 2350 + }, + { + "epoch": 2.192827200745226, + "grad_norm": 0.2451904092276869, + "learning_rate": 1.4911977908180877e-05, + "loss": 0.2672, + "step": 2355 + }, + { + "epoch": 2.197484862598975, + "grad_norm": 0.6591340342436288, + "learning_rate": 1.4825681739730757e-05, + "loss": 0.2762, + "step": 2360 + }, + { + "epoch": 2.2021425244527246, + "grad_norm": 0.300791091079865, + "learning_rate": 1.4739385571280637e-05, + "loss": 0.2734, + "step": 2365 + }, + { + "epoch": 2.206800186306474, + "grad_norm": 0.2653884810898289, + "learning_rate": 1.4653089402830513e-05, + "loss": 0.2727, + "step": 2370 + }, + { + "epoch": 2.2114578481602236, + "grad_norm": 0.2764326256981041, + "learning_rate": 1.4566793234380393e-05, + "loss": 0.2719, + "step": 2375 + }, + { + "epoch": 2.216115510013973, + "grad_norm": 0.2837879018636987, + "learning_rate": 1.4480497065930273e-05, + "loss": 0.2688, + "step": 2380 + }, + { + "epoch": 2.2207731718677226, + "grad_norm": 0.2548119239798833, + "learning_rate": 1.4394200897480153e-05, + "loss": 0.2814, + "step": 2385 + }, + { + "epoch": 2.2254308337214717, + "grad_norm": 0.2336755783098942, + "learning_rate": 1.4307904729030031e-05, + "loss": 0.2753, + "step": 2390 + }, + { + "epoch": 2.230088495575221, + "grad_norm": 0.2656480140545371, + "learning_rate": 1.4221608560579911e-05, + "loss": 0.273, + "step": 2395 + }, + { + "epoch": 2.2347461574289706, + "grad_norm": 0.2569155777441175, + "learning_rate": 1.4135312392129791e-05, + "loss": 0.2612, + "step": 2400 + }, + { + "epoch": 2.23940381928272, + "grad_norm": 0.2492998571163356, + "learning_rate": 1.404901622367967e-05, + "loss": 0.27, + "step": 2405 + }, + { + "epoch": 2.2440614811364696, + "grad_norm": 0.25590130320854276, + "learning_rate": 1.3962720055229547e-05, + "loss": 0.2713, + "step": 2410 + }, + { + "epoch": 2.248719142990219, + "grad_norm": 0.2539137547467478, + "learning_rate": 1.3876423886779427e-05, + "loss": 0.2692, + "step": 2415 + }, + { + "epoch": 2.253376804843968, + "grad_norm": 0.25764579758807804, + "learning_rate": 1.3790127718329307e-05, + "loss": 0.261, + "step": 2420 + }, + { + "epoch": 2.2580344666977177, + "grad_norm": 0.25551995299495456, + "learning_rate": 1.3703831549879187e-05, + "loss": 0.2724, + "step": 2425 + }, + { + "epoch": 2.262692128551467, + "grad_norm": 0.24266270441527438, + "learning_rate": 1.3617535381429064e-05, + "loss": 0.2642, + "step": 2430 + }, + { + "epoch": 2.2673497904052167, + "grad_norm": 0.2588361279242763, + "learning_rate": 1.3531239212978944e-05, + "loss": 0.2718, + "step": 2435 + }, + { + "epoch": 2.272007452258966, + "grad_norm": 0.28429578662647526, + "learning_rate": 1.3444943044528824e-05, + "loss": 0.2705, + "step": 2440 + }, + { + "epoch": 2.276665114112715, + "grad_norm": 0.24525404341789867, + "learning_rate": 1.3358646876078703e-05, + "loss": 0.2693, + "step": 2445 + }, + { + "epoch": 2.2813227759664647, + "grad_norm": 0.24607287108948245, + "learning_rate": 1.327235070762858e-05, + "loss": 0.2666, + "step": 2450 + }, + { + "epoch": 2.285980437820214, + "grad_norm": 0.24282683752499631, + "learning_rate": 1.318605453917846e-05, + "loss": 0.2697, + "step": 2455 + }, + { + "epoch": 2.2906380996739637, + "grad_norm": 0.305835699533878, + "learning_rate": 1.309975837072834e-05, + "loss": 0.2663, + "step": 2460 + }, + { + "epoch": 2.295295761527713, + "grad_norm": 0.2669796220360565, + "learning_rate": 1.301346220227822e-05, + "loss": 0.2782, + "step": 2465 + }, + { + "epoch": 2.2999534233814627, + "grad_norm": 0.3195367557398448, + "learning_rate": 1.2927166033828098e-05, + "loss": 0.269, + "step": 2470 + }, + { + "epoch": 2.3046110852352117, + "grad_norm": 0.260616710535662, + "learning_rate": 1.2840869865377978e-05, + "loss": 0.2743, + "step": 2475 + }, + { + "epoch": 2.3092687470889612, + "grad_norm": 0.2802244774876612, + "learning_rate": 1.2754573696927858e-05, + "loss": 0.2663, + "step": 2480 + }, + { + "epoch": 2.3139264089427107, + "grad_norm": 0.24626536789973147, + "learning_rate": 1.2668277528477738e-05, + "loss": 0.2741, + "step": 2485 + }, + { + "epoch": 2.3185840707964602, + "grad_norm": 0.2525988837688525, + "learning_rate": 1.2581981360027614e-05, + "loss": 0.2763, + "step": 2490 + }, + { + "epoch": 2.3232417326502097, + "grad_norm": 0.2419543341208871, + "learning_rate": 1.2495685191577494e-05, + "loss": 0.2684, + "step": 2495 + }, + { + "epoch": 2.3278993945039588, + "grad_norm": 0.22636835901167812, + "learning_rate": 1.2409389023127374e-05, + "loss": 0.2623, + "step": 2500 + }, + { + "epoch": 2.3325570563577083, + "grad_norm": 0.24683053840387006, + "learning_rate": 1.2323092854677252e-05, + "loss": 0.2689, + "step": 2505 + }, + { + "epoch": 2.3372147182114578, + "grad_norm": 0.39925021955339624, + "learning_rate": 1.2236796686227132e-05, + "loss": 0.2716, + "step": 2510 + }, + { + "epoch": 2.3418723800652073, + "grad_norm": 0.2460428498113885, + "learning_rate": 1.215050051777701e-05, + "loss": 0.2698, + "step": 2515 + }, + { + "epoch": 2.3465300419189568, + "grad_norm": 0.24081391446730407, + "learning_rate": 1.206420434932689e-05, + "loss": 0.2701, + "step": 2520 + }, + { + "epoch": 2.3511877037727063, + "grad_norm": 0.25896045081013075, + "learning_rate": 1.1977908180876769e-05, + "loss": 0.2727, + "step": 2525 + }, + { + "epoch": 2.3558453656264557, + "grad_norm": 0.2855436451430716, + "learning_rate": 1.1891612012426649e-05, + "loss": 0.2706, + "step": 2530 + }, + { + "epoch": 2.360503027480205, + "grad_norm": 0.26964620631891395, + "learning_rate": 1.1805315843976528e-05, + "loss": 0.2645, + "step": 2535 + }, + { + "epoch": 2.3651606893339543, + "grad_norm": 0.2553273475265153, + "learning_rate": 1.1719019675526408e-05, + "loss": 0.2611, + "step": 2540 + }, + { + "epoch": 2.369818351187704, + "grad_norm": 0.24598386932397875, + "learning_rate": 1.1632723507076287e-05, + "loss": 0.272, + "step": 2545 + }, + { + "epoch": 2.3744760130414533, + "grad_norm": 0.24820744671461673, + "learning_rate": 1.1546427338626167e-05, + "loss": 0.2715, + "step": 2550 + }, + { + "epoch": 2.3791336748952028, + "grad_norm": 0.2546651811004806, + "learning_rate": 1.1460131170176045e-05, + "loss": 0.2675, + "step": 2555 + }, + { + "epoch": 2.383791336748952, + "grad_norm": 0.28012240083023987, + "learning_rate": 1.1373835001725925e-05, + "loss": 0.2686, + "step": 2560 + }, + { + "epoch": 2.3884489986027013, + "grad_norm": 0.25496187070989224, + "learning_rate": 1.1287538833275803e-05, + "loss": 0.2725, + "step": 2565 + }, + { + "epoch": 2.393106660456451, + "grad_norm": 0.2459565835067793, + "learning_rate": 1.1201242664825683e-05, + "loss": 0.2747, + "step": 2570 + }, + { + "epoch": 2.3977643223102003, + "grad_norm": 0.24992557924789235, + "learning_rate": 1.1114946496375561e-05, + "loss": 0.2702, + "step": 2575 + }, + { + "epoch": 2.40242198416395, + "grad_norm": 0.24548790397731274, + "learning_rate": 1.1028650327925441e-05, + "loss": 0.2694, + "step": 2580 + }, + { + "epoch": 2.4070796460176993, + "grad_norm": 0.2540166581404125, + "learning_rate": 1.094235415947532e-05, + "loss": 0.2673, + "step": 2585 + }, + { + "epoch": 2.4117373078714484, + "grad_norm": 0.24956814468845384, + "learning_rate": 1.0856057991025199e-05, + "loss": 0.26, + "step": 2590 + }, + { + "epoch": 2.416394969725198, + "grad_norm": 0.2544462078803143, + "learning_rate": 1.0769761822575077e-05, + "loss": 0.2757, + "step": 2595 + }, + { + "epoch": 2.4210526315789473, + "grad_norm": 0.25477644710778236, + "learning_rate": 1.0683465654124957e-05, + "loss": 0.2728, + "step": 2600 + }, + { + "epoch": 2.425710293432697, + "grad_norm": 0.2463193145902546, + "learning_rate": 1.0597169485674835e-05, + "loss": 0.2642, + "step": 2605 + }, + { + "epoch": 2.4303679552864463, + "grad_norm": 0.25119398756180855, + "learning_rate": 1.0510873317224715e-05, + "loss": 0.2623, + "step": 2610 + }, + { + "epoch": 2.4350256171401954, + "grad_norm": 0.2554200605154604, + "learning_rate": 1.0424577148774595e-05, + "loss": 0.2673, + "step": 2615 + }, + { + "epoch": 2.439683278993945, + "grad_norm": 0.24850654131551053, + "learning_rate": 1.0338280980324475e-05, + "loss": 0.2691, + "step": 2620 + }, + { + "epoch": 2.4443409408476944, + "grad_norm": 0.25529790540475594, + "learning_rate": 1.0251984811874353e-05, + "loss": 0.2623, + "step": 2625 + }, + { + "epoch": 2.448998602701444, + "grad_norm": 0.24935654678620325, + "learning_rate": 1.0165688643424233e-05, + "loss": 0.266, + "step": 2630 + }, + { + "epoch": 2.4536562645551934, + "grad_norm": 0.24948434199922845, + "learning_rate": 1.0079392474974112e-05, + "loss": 0.2697, + "step": 2635 + }, + { + "epoch": 2.458313926408943, + "grad_norm": 0.5728746729322969, + "learning_rate": 9.993096306523992e-06, + "loss": 0.2709, + "step": 2640 + }, + { + "epoch": 2.4629715882626924, + "grad_norm": 0.2604484130291829, + "learning_rate": 9.90680013807387e-06, + "loss": 0.2701, + "step": 2645 + }, + { + "epoch": 2.4676292501164414, + "grad_norm": 0.23543493626242007, + "learning_rate": 9.82050396962375e-06, + "loss": 0.2672, + "step": 2650 + }, + { + "epoch": 2.472286911970191, + "grad_norm": 0.24954954048333747, + "learning_rate": 9.734207801173628e-06, + "loss": 0.2681, + "step": 2655 + }, + { + "epoch": 2.4769445738239404, + "grad_norm": 0.24700156741936974, + "learning_rate": 9.647911632723508e-06, + "loss": 0.265, + "step": 2660 + }, + { + "epoch": 2.48160223567769, + "grad_norm": 0.24034208079197983, + "learning_rate": 9.561615464273386e-06, + "loss": 0.2662, + "step": 2665 + }, + { + "epoch": 2.4862598975314394, + "grad_norm": 0.24046526182978087, + "learning_rate": 9.475319295823266e-06, + "loss": 0.27, + "step": 2670 + }, + { + "epoch": 2.4909175593851884, + "grad_norm": 0.23736799959590765, + "learning_rate": 9.389023127373144e-06, + "loss": 0.2656, + "step": 2675 + }, + { + "epoch": 2.495575221238938, + "grad_norm": 0.2653182659378759, + "learning_rate": 9.302726958923024e-06, + "loss": 0.2635, + "step": 2680 + }, + { + "epoch": 2.5002328830926874, + "grad_norm": 0.24189834679652392, + "learning_rate": 9.216430790472904e-06, + "loss": 0.2688, + "step": 2685 + }, + { + "epoch": 2.504890544946437, + "grad_norm": 0.2535012437719162, + "learning_rate": 9.130134622022784e-06, + "loss": 0.2598, + "step": 2690 + }, + { + "epoch": 2.5095482068001864, + "grad_norm": 0.4021752590698154, + "learning_rate": 9.043838453572662e-06, + "loss": 0.2656, + "step": 2695 + }, + { + "epoch": 2.514205868653936, + "grad_norm": 0.2396407337253643, + "learning_rate": 8.957542285122542e-06, + "loss": 0.2609, + "step": 2700 + }, + { + "epoch": 2.5188635305076854, + "grad_norm": 0.2539861103581641, + "learning_rate": 8.87124611667242e-06, + "loss": 0.2688, + "step": 2705 + }, + { + "epoch": 2.5235211923614345, + "grad_norm": 0.2359234652862627, + "learning_rate": 8.7849499482223e-06, + "loss": 0.2653, + "step": 2710 + }, + { + "epoch": 2.528178854215184, + "grad_norm": 0.25756160385693805, + "learning_rate": 8.698653779772179e-06, + "loss": 0.2702, + "step": 2715 + }, + { + "epoch": 2.5328365160689335, + "grad_norm": 0.2407337745730763, + "learning_rate": 8.612357611322058e-06, + "loss": 0.2737, + "step": 2720 + }, + { + "epoch": 2.537494177922683, + "grad_norm": 0.250173406027027, + "learning_rate": 8.526061442871937e-06, + "loss": 0.2678, + "step": 2725 + }, + { + "epoch": 2.542151839776432, + "grad_norm": 0.2397210288165992, + "learning_rate": 8.439765274421817e-06, + "loss": 0.2763, + "step": 2730 + }, + { + "epoch": 2.5468095016301815, + "grad_norm": 0.2401796074674243, + "learning_rate": 8.353469105971695e-06, + "loss": 0.2645, + "step": 2735 + }, + { + "epoch": 2.551467163483931, + "grad_norm": 0.23868068898159894, + "learning_rate": 8.267172937521575e-06, + "loss": 0.2633, + "step": 2740 + }, + { + "epoch": 2.5561248253376805, + "grad_norm": 0.2547368143136742, + "learning_rate": 8.180876769071453e-06, + "loss": 0.265, + "step": 2745 + }, + { + "epoch": 2.56078248719143, + "grad_norm": 0.23073042137252345, + "learning_rate": 8.094580600621333e-06, + "loss": 0.2645, + "step": 2750 + }, + { + "epoch": 2.5654401490451795, + "grad_norm": 0.24607684349630346, + "learning_rate": 8.008284432171211e-06, + "loss": 0.2614, + "step": 2755 + }, + { + "epoch": 2.570097810898929, + "grad_norm": 0.24201219527867612, + "learning_rate": 7.921988263721091e-06, + "loss": 0.2686, + "step": 2760 + }, + { + "epoch": 2.574755472752678, + "grad_norm": 0.24091627691070966, + "learning_rate": 7.835692095270971e-06, + "loss": 0.2627, + "step": 2765 + }, + { + "epoch": 2.5794131346064275, + "grad_norm": 0.2474852577907246, + "learning_rate": 7.749395926820851e-06, + "loss": 0.266, + "step": 2770 + }, + { + "epoch": 2.584070796460177, + "grad_norm": 0.2508786185963276, + "learning_rate": 7.663099758370729e-06, + "loss": 0.2606, + "step": 2775 + }, + { + "epoch": 2.5887284583139265, + "grad_norm": 0.26765346867793416, + "learning_rate": 7.576803589920608e-06, + "loss": 0.2699, + "step": 2780 + }, + { + "epoch": 2.5933861201676756, + "grad_norm": 0.6100376969601905, + "learning_rate": 7.490507421470487e-06, + "loss": 0.2709, + "step": 2785 + }, + { + "epoch": 2.598043782021425, + "grad_norm": 0.24495451940584023, + "learning_rate": 7.4042112530203655e-06, + "loss": 0.2733, + "step": 2790 + }, + { + "epoch": 2.6027014438751745, + "grad_norm": 0.2791048492990286, + "learning_rate": 7.317915084570245e-06, + "loss": 0.2671, + "step": 2795 + }, + { + "epoch": 2.607359105728924, + "grad_norm": 0.24944148889122386, + "learning_rate": 7.231618916120124e-06, + "loss": 0.2682, + "step": 2800 + }, + { + "epoch": 2.6120167675826735, + "grad_norm": 0.250766764140112, + "learning_rate": 7.1453227476700035e-06, + "loss": 0.2666, + "step": 2805 + }, + { + "epoch": 2.616674429436423, + "grad_norm": 0.2781145664814796, + "learning_rate": 7.059026579219883e-06, + "loss": 0.2676, + "step": 2810 + }, + { + "epoch": 2.6213320912901725, + "grad_norm": 0.22829160732591977, + "learning_rate": 6.9727304107697625e-06, + "loss": 0.2598, + "step": 2815 + }, + { + "epoch": 2.625989753143922, + "grad_norm": 0.23886526932116794, + "learning_rate": 6.886434242319641e-06, + "loss": 0.2702, + "step": 2820 + }, + { + "epoch": 2.630647414997671, + "grad_norm": 0.2532267531205448, + "learning_rate": 6.800138073869521e-06, + "loss": 0.2683, + "step": 2825 + }, + { + "epoch": 2.6353050768514206, + "grad_norm": 0.2778041631063136, + "learning_rate": 6.713841905419399e-06, + "loss": 0.263, + "step": 2830 + }, + { + "epoch": 2.63996273870517, + "grad_norm": 0.23987705253673325, + "learning_rate": 6.627545736969279e-06, + "loss": 0.2722, + "step": 2835 + }, + { + "epoch": 2.6446204005589196, + "grad_norm": 0.2336359656913526, + "learning_rate": 6.541249568519157e-06, + "loss": 0.2762, + "step": 2840 + }, + { + "epoch": 2.6492780624126686, + "grad_norm": 0.2522255896799647, + "learning_rate": 6.454953400069037e-06, + "loss": 0.2677, + "step": 2845 + }, + { + "epoch": 2.653935724266418, + "grad_norm": 0.2362771386522153, + "learning_rate": 6.368657231618916e-06, + "loss": 0.2702, + "step": 2850 + }, + { + "epoch": 2.6585933861201676, + "grad_norm": 0.2603907654875355, + "learning_rate": 6.282361063168796e-06, + "loss": 0.2606, + "step": 2855 + }, + { + "epoch": 2.663251047973917, + "grad_norm": 0.2384253786782358, + "learning_rate": 6.196064894718675e-06, + "loss": 0.2665, + "step": 2860 + }, + { + "epoch": 2.6679087098276666, + "grad_norm": 0.2417383086843217, + "learning_rate": 6.109768726268554e-06, + "loss": 0.2654, + "step": 2865 + }, + { + "epoch": 2.672566371681416, + "grad_norm": 0.23682724066001656, + "learning_rate": 6.023472557818433e-06, + "loss": 0.2651, + "step": 2870 + }, + { + "epoch": 2.6772240335351656, + "grad_norm": 0.28125485834884123, + "learning_rate": 5.937176389368312e-06, + "loss": 0.2676, + "step": 2875 + }, + { + "epoch": 2.6818816953889146, + "grad_norm": 0.24560709342381493, + "learning_rate": 5.850880220918191e-06, + "loss": 0.2656, + "step": 2880 + }, + { + "epoch": 2.686539357242664, + "grad_norm": 0.24142604350710792, + "learning_rate": 5.76458405246807e-06, + "loss": 0.2736, + "step": 2885 + }, + { + "epoch": 2.6911970190964136, + "grad_norm": 0.23021865789214346, + "learning_rate": 5.67828788401795e-06, + "loss": 0.2673, + "step": 2890 + }, + { + "epoch": 2.695854680950163, + "grad_norm": 0.24645998952166318, + "learning_rate": 5.591991715567829e-06, + "loss": 0.2621, + "step": 2895 + }, + { + "epoch": 2.700512342803912, + "grad_norm": 0.23281318258868106, + "learning_rate": 5.5056955471177085e-06, + "loss": 0.2645, + "step": 2900 + }, + { + "epoch": 2.7051700046576617, + "grad_norm": 0.2326586772152503, + "learning_rate": 5.4193993786675876e-06, + "loss": 0.2593, + "step": 2905 + }, + { + "epoch": 2.709827666511411, + "grad_norm": 0.23989592399985699, + "learning_rate": 5.333103210217467e-06, + "loss": 0.2625, + "step": 2910 + }, + { + "epoch": 2.7144853283651607, + "grad_norm": 0.24264635453844383, + "learning_rate": 5.246807041767346e-06, + "loss": 0.2685, + "step": 2915 + }, + { + "epoch": 2.71914299021891, + "grad_norm": 0.23505622614412772, + "learning_rate": 5.160510873317225e-06, + "loss": 0.2619, + "step": 2920 + }, + { + "epoch": 2.7238006520726596, + "grad_norm": 0.2954774090778297, + "learning_rate": 5.074214704867105e-06, + "loss": 0.2709, + "step": 2925 + }, + { + "epoch": 2.728458313926409, + "grad_norm": 0.23449375146214052, + "learning_rate": 4.987918536416984e-06, + "loss": 0.2686, + "step": 2930 + }, + { + "epoch": 2.7331159757801586, + "grad_norm": 0.22660197484627229, + "learning_rate": 4.901622367966863e-06, + "loss": 0.2639, + "step": 2935 + }, + { + "epoch": 2.7377736376339077, + "grad_norm": 0.25366849181683887, + "learning_rate": 4.815326199516742e-06, + "loss": 0.2652, + "step": 2940 + }, + { + "epoch": 2.742431299487657, + "grad_norm": 0.23773508357884918, + "learning_rate": 4.729030031066621e-06, + "loss": 0.272, + "step": 2945 + }, + { + "epoch": 2.7470889613414067, + "grad_norm": 0.23883706049912928, + "learning_rate": 4.6427338626165e-06, + "loss": 0.262, + "step": 2950 + }, + { + "epoch": 2.751746623195156, + "grad_norm": 0.245082821723228, + "learning_rate": 4.556437694166379e-06, + "loss": 0.266, + "step": 2955 + }, + { + "epoch": 2.7564042850489052, + "grad_norm": 0.24388812731912868, + "learning_rate": 4.470141525716258e-06, + "loss": 0.2665, + "step": 2960 + }, + { + "epoch": 2.7610619469026547, + "grad_norm": 0.23080388123509749, + "learning_rate": 4.383845357266138e-06, + "loss": 0.2687, + "step": 2965 + }, + { + "epoch": 2.765719608756404, + "grad_norm": 0.23304757623637526, + "learning_rate": 4.297549188816017e-06, + "loss": 0.2629, + "step": 2970 + }, + { + "epoch": 2.7703772706101537, + "grad_norm": 0.2629554897037301, + "learning_rate": 4.211253020365896e-06, + "loss": 0.2656, + "step": 2975 + }, + { + "epoch": 2.775034932463903, + "grad_norm": 0.24540971434329928, + "learning_rate": 4.124956851915775e-06, + "loss": 0.2629, + "step": 2980 + }, + { + "epoch": 2.7796925943176527, + "grad_norm": 0.30014532921383136, + "learning_rate": 4.0386606834656544e-06, + "loss": 0.2678, + "step": 2985 + }, + { + "epoch": 2.784350256171402, + "grad_norm": 0.2342133541506607, + "learning_rate": 3.9523645150155335e-06, + "loss": 0.2654, + "step": 2990 + }, + { + "epoch": 2.7890079180251512, + "grad_norm": 0.23959451694876363, + "learning_rate": 3.8660683465654126e-06, + "loss": 0.2652, + "step": 2995 + }, + { + "epoch": 2.7936655798789007, + "grad_norm": 0.24027322925970268, + "learning_rate": 3.779772178115292e-06, + "loss": 0.2683, + "step": 3000 + }, + { + "epoch": 2.7983232417326502, + "grad_norm": 0.22866196108603312, + "learning_rate": 3.693476009665171e-06, + "loss": 0.2638, + "step": 3005 + }, + { + "epoch": 2.8029809035863997, + "grad_norm": 0.23207917421270907, + "learning_rate": 3.6071798412150506e-06, + "loss": 0.2697, + "step": 3010 + }, + { + "epoch": 2.807638565440149, + "grad_norm": 0.24148817561468802, + "learning_rate": 3.5208836727649297e-06, + "loss": 0.2548, + "step": 3015 + }, + { + "epoch": 2.8122962272938983, + "grad_norm": 0.24826400013212727, + "learning_rate": 3.434587504314809e-06, + "loss": 0.2669, + "step": 3020 + }, + { + "epoch": 2.8169538891476478, + "grad_norm": 0.22758778242341457, + "learning_rate": 3.348291335864688e-06, + "loss": 0.2631, + "step": 3025 + }, + { + "epoch": 2.8216115510013973, + "grad_norm": 0.2278230958357378, + "learning_rate": 3.2619951674145674e-06, + "loss": 0.2617, + "step": 3030 + }, + { + "epoch": 2.8262692128551468, + "grad_norm": 0.2301525600023757, + "learning_rate": 3.1756989989644464e-06, + "loss": 0.2652, + "step": 3035 + }, + { + "epoch": 2.8309268747088963, + "grad_norm": 0.24711954628030958, + "learning_rate": 3.089402830514325e-06, + "loss": 0.2751, + "step": 3040 + }, + { + "epoch": 2.8355845365626458, + "grad_norm": 0.26581279477281244, + "learning_rate": 3.0031066620642046e-06, + "loss": 0.2609, + "step": 3045 + }, + { + "epoch": 2.840242198416395, + "grad_norm": 0.24890741319507606, + "learning_rate": 2.9168104936140837e-06, + "loss": 0.261, + "step": 3050 + }, + { + "epoch": 2.8448998602701443, + "grad_norm": 0.30572357087890906, + "learning_rate": 2.8305143251639627e-06, + "loss": 0.269, + "step": 3055 + }, + { + "epoch": 2.849557522123894, + "grad_norm": 0.3070212619911331, + "learning_rate": 2.7442181567138422e-06, + "loss": 0.2717, + "step": 3060 + }, + { + "epoch": 2.8542151839776433, + "grad_norm": 0.24123580523943625, + "learning_rate": 2.6579219882637213e-06, + "loss": 0.2681, + "step": 3065 + }, + { + "epoch": 2.858872845831393, + "grad_norm": 0.23087439262318699, + "learning_rate": 2.5716258198136004e-06, + "loss": 0.2582, + "step": 3070 + }, + { + "epoch": 2.863530507685142, + "grad_norm": 0.2391097559052989, + "learning_rate": 2.4853296513634795e-06, + "loss": 0.2631, + "step": 3075 + }, + { + "epoch": 2.8681881695388913, + "grad_norm": 1.6021977536357102, + "learning_rate": 2.399033482913359e-06, + "loss": 0.2619, + "step": 3080 + }, + { + "epoch": 2.872845831392641, + "grad_norm": 0.2257095993666073, + "learning_rate": 2.312737314463238e-06, + "loss": 0.2638, + "step": 3085 + }, + { + "epoch": 2.8775034932463903, + "grad_norm": 0.23447988430791128, + "learning_rate": 2.226441146013117e-06, + "loss": 0.2695, + "step": 3090 + }, + { + "epoch": 2.88216115510014, + "grad_norm": 0.23336542967563148, + "learning_rate": 2.140144977562996e-06, + "loss": 0.258, + "step": 3095 + }, + { + "epoch": 2.8868188169538893, + "grad_norm": 0.22960523248355422, + "learning_rate": 2.0538488091128757e-06, + "loss": 0.2701, + "step": 3100 + }, + { + "epoch": 2.891476478807639, + "grad_norm": 0.23768769948413135, + "learning_rate": 1.9675526406627547e-06, + "loss": 0.2651, + "step": 3105 + }, + { + "epoch": 2.896134140661388, + "grad_norm": 0.2264847690514202, + "learning_rate": 1.8812564722126338e-06, + "loss": 0.2592, + "step": 3110 + }, + { + "epoch": 2.9007918025151374, + "grad_norm": 0.22932272675979026, + "learning_rate": 1.794960303762513e-06, + "loss": 0.2565, + "step": 3115 + }, + { + "epoch": 2.905449464368887, + "grad_norm": 0.23637996003822423, + "learning_rate": 1.7086641353123924e-06, + "loss": 0.2659, + "step": 3120 + }, + { + "epoch": 2.9101071262226363, + "grad_norm": 3.6839450033328283, + "learning_rate": 1.6223679668622715e-06, + "loss": 0.2672, + "step": 3125 + }, + { + "epoch": 2.9147647880763854, + "grad_norm": 0.2227886774143204, + "learning_rate": 1.5360717984121505e-06, + "loss": 0.255, + "step": 3130 + }, + { + "epoch": 2.919422449930135, + "grad_norm": 0.2271396022815255, + "learning_rate": 1.4497756299620296e-06, + "loss": 0.263, + "step": 3135 + }, + { + "epoch": 2.9240801117838844, + "grad_norm": 0.22277258393762092, + "learning_rate": 1.363479461511909e-06, + "loss": 0.272, + "step": 3140 + }, + { + "epoch": 2.928737773637634, + "grad_norm": 0.23371618200567057, + "learning_rate": 1.277183293061788e-06, + "loss": 0.2733, + "step": 3145 + }, + { + "epoch": 2.9333954354913834, + "grad_norm": 0.2264313972112628, + "learning_rate": 1.1908871246116673e-06, + "loss": 0.2639, + "step": 3150 + }, + { + "epoch": 2.938053097345133, + "grad_norm": 0.4570714544778108, + "learning_rate": 1.1045909561615463e-06, + "loss": 0.2642, + "step": 3155 + }, + { + "epoch": 2.9427107591988824, + "grad_norm": 0.2366565080467962, + "learning_rate": 1.0182947877114256e-06, + "loss": 0.2639, + "step": 3160 + }, + { + "epoch": 2.9473684210526314, + "grad_norm": 0.2346391676286511, + "learning_rate": 9.319986192613048e-07, + "loss": 0.2659, + "step": 3165 + }, + { + "epoch": 2.952026082906381, + "grad_norm": 0.21986695916960491, + "learning_rate": 8.45702450811184e-07, + "loss": 0.2613, + "step": 3170 + }, + { + "epoch": 2.9566837447601304, + "grad_norm": 0.23068227390735385, + "learning_rate": 7.594062823610632e-07, + "loss": 0.2652, + "step": 3175 + }, + { + "epoch": 2.96134140661388, + "grad_norm": 0.2218014778660938, + "learning_rate": 6.731101139109423e-07, + "loss": 0.2607, + "step": 3180 + }, + { + "epoch": 2.9659990684676294, + "grad_norm": 0.23904524486615045, + "learning_rate": 5.868139454608215e-07, + "loss": 0.2647, + "step": 3185 + }, + { + "epoch": 2.9706567303213784, + "grad_norm": 0.22329952211908216, + "learning_rate": 5.005177770107007e-07, + "loss": 0.2562, + "step": 3190 + }, + { + "epoch": 2.975314392175128, + "grad_norm": 0.23326673144161586, + "learning_rate": 4.142216085605799e-07, + "loss": 0.2678, + "step": 3195 + }, + { + "epoch": 2.9799720540288774, + "grad_norm": 0.23348286530329374, + "learning_rate": 3.279254401104591e-07, + "loss": 0.2668, + "step": 3200 + }, + { + "epoch": 2.984629715882627, + "grad_norm": 0.23276521431022687, + "learning_rate": 2.416292716603383e-07, + "loss": 0.2684, + "step": 3205 + }, + { + "epoch": 2.9892873777363764, + "grad_norm": 0.2257062545640731, + "learning_rate": 1.5533310321021747e-07, + "loss": 0.2529, + "step": 3210 + }, + { + "epoch": 2.993945039590126, + "grad_norm": 0.23253102905210304, + "learning_rate": 6.903693476009665e-08, + "loss": 0.2549, + "step": 3215 + }, + { + "epoch": 2.9976711690731253, + "step": 3219, + "total_flos": 2.754977641940386e+18, + "train_loss": 0.3562816010013283, + "train_runtime": 43513.4977, + "train_samples_per_second": 1.184, + "train_steps_per_second": 0.074 + } + ], + "logging_steps": 5, + "max_steps": 3219, + "num_input_tokens_seen": 0, + "num_train_epochs": 3, + "save_steps": 500, + "stateful_callbacks": { + "TrainerControl": { + "args": { + "should_epoch_stop": false, + "should_evaluate": false, + "should_log": false, + "should_save": true, + "should_training_stop": true + }, + "attributes": {} + } + }, + "total_flos": 2.754977641940386e+18, + "train_batch_size": 1, + "trial_name": null, + "trial_params": null +}