diff --git "a/trainer_state.json" "b/trainer_state.json" new file mode 100644--- /dev/null +++ "b/trainer_state.json" @@ -0,0 +1,7031 @@ +{ + "best_metric": null, + "best_model_checkpoint": null, + "epoch": 0.999896960329727, + "eval_steps": 200, + "global_step": 4852, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.00020607934054611026, + "grad_norm": 0.22813845522641535, + "learning_rate": 4.1152263374485604e-07, + "loss": 0.3029, + "step": 1 + }, + { + "epoch": 0.0010303967027305513, + "grad_norm": 0.369767682912768, + "learning_rate": 2.05761316872428e-06, + "loss": 0.531, + "step": 5 + }, + { + "epoch": 0.0020607934054611026, + "grad_norm": 0.516859665385139, + "learning_rate": 4.11522633744856e-06, + "loss": 0.5792, + "step": 10 + }, + { + "epoch": 0.0030911901081916537, + "grad_norm": 0.39330735500575664, + "learning_rate": 6.172839506172839e-06, + "loss": 0.4978, + "step": 15 + }, + { + "epoch": 0.004121586810922205, + "grad_norm": 0.5867330845716273, + "learning_rate": 8.23045267489712e-06, + "loss": 0.5701, + "step": 20 + }, + { + "epoch": 0.005151983513652756, + "grad_norm": 0.5652480516594349, + "learning_rate": 1.02880658436214e-05, + "loss": 0.6254, + "step": 25 + }, + { + "epoch": 0.0061823802163833074, + "grad_norm": 0.44581621851474207, + "learning_rate": 1.2345679012345678e-05, + "loss": 0.4771, + "step": 30 + }, + { + "epoch": 0.0072127769191138585, + "grad_norm": 0.460575062400366, + "learning_rate": 1.440329218106996e-05, + "loss": 0.5632, + "step": 35 + }, + { + "epoch": 0.00824317362184441, + "grad_norm": 0.30751978801541957, + "learning_rate": 1.646090534979424e-05, + "loss": 0.4236, + "step": 40 + }, + { + "epoch": 0.00927357032457496, + "grad_norm": 0.38005292756389475, + "learning_rate": 1.8518518518518518e-05, + "loss": 0.4796, + "step": 45 + }, + { + "epoch": 0.010303967027305513, + "grad_norm": 0.7516556669584462, + "learning_rate": 2.05761316872428e-05, + "loss": 0.499, + "step": 50 + }, + { + "epoch": 0.011334363730036065, + "grad_norm": 0.43984231867613194, + "learning_rate": 2.2633744855967078e-05, + "loss": 0.3982, + "step": 55 + }, + { + "epoch": 0.012364760432766615, + "grad_norm": 0.48668013365999796, + "learning_rate": 2.4691358024691357e-05, + "loss": 0.4797, + "step": 60 + }, + { + "epoch": 0.013395157135497167, + "grad_norm": 0.414961574786946, + "learning_rate": 2.6748971193415638e-05, + "loss": 0.4253, + "step": 65 + }, + { + "epoch": 0.014425553838227717, + "grad_norm": 0.400815338908771, + "learning_rate": 2.880658436213992e-05, + "loss": 0.4563, + "step": 70 + }, + { + "epoch": 0.015455950540958269, + "grad_norm": 0.6713253921630139, + "learning_rate": 3.08641975308642e-05, + "loss": 0.4456, + "step": 75 + }, + { + "epoch": 0.01648634724368882, + "grad_norm": 0.46429013698478766, + "learning_rate": 3.292181069958848e-05, + "loss": 0.3943, + "step": 80 + }, + { + "epoch": 0.01751674394641937, + "grad_norm": 0.46368279851910454, + "learning_rate": 3.497942386831276e-05, + "loss": 0.4472, + "step": 85 + }, + { + "epoch": 0.01854714064914992, + "grad_norm": 0.6084053230691859, + "learning_rate": 3.7037037037037037e-05, + "loss": 0.4168, + "step": 90 + }, + { + "epoch": 0.019577537351880475, + "grad_norm": 0.3924187177282214, + "learning_rate": 3.909465020576132e-05, + "loss": 0.4242, + "step": 95 + }, + { + "epoch": 0.020607934054611025, + "grad_norm": 0.5391485143871717, + "learning_rate": 4.11522633744856e-05, + "loss": 0.405, + "step": 100 + }, + { + "epoch": 0.021638330757341576, + "grad_norm": 0.5391287793903055, + "learning_rate": 4.3209876543209875e-05, + "loss": 0.3786, + "step": 105 + }, + { + "epoch": 0.02266872746007213, + "grad_norm": 0.5860930775079664, + "learning_rate": 4.5267489711934157e-05, + "loss": 0.4353, + "step": 110 + }, + { + "epoch": 0.02369912416280268, + "grad_norm": 0.5142605508034142, + "learning_rate": 4.732510288065844e-05, + "loss": 0.3807, + "step": 115 + }, + { + "epoch": 0.02472952086553323, + "grad_norm": 0.5212600058334357, + "learning_rate": 4.938271604938271e-05, + "loss": 0.404, + "step": 120 + }, + { + "epoch": 0.02575991756826378, + "grad_norm": 0.5518790775686553, + "learning_rate": 5.1440329218106995e-05, + "loss": 0.3854, + "step": 125 + }, + { + "epoch": 0.026790314270994334, + "grad_norm": 0.5275493325802079, + "learning_rate": 5.3497942386831277e-05, + "loss": 0.3954, + "step": 130 + }, + { + "epoch": 0.027820710973724884, + "grad_norm": 0.5128585051120935, + "learning_rate": 5.555555555555556e-05, + "loss": 0.3937, + "step": 135 + }, + { + "epoch": 0.028851107676455434, + "grad_norm": 0.5148606074591149, + "learning_rate": 5.761316872427984e-05, + "loss": 0.3846, + "step": 140 + }, + { + "epoch": 0.029881504379185988, + "grad_norm": 0.5488455929859509, + "learning_rate": 5.9670781893004115e-05, + "loss": 0.4055, + "step": 145 + }, + { + "epoch": 0.030911901081916538, + "grad_norm": 0.6098923361895402, + "learning_rate": 6.17283950617284e-05, + "loss": 0.3934, + "step": 150 + }, + { + "epoch": 0.03194229778464709, + "grad_norm": 0.4754468226292264, + "learning_rate": 6.378600823045267e-05, + "loss": 0.3727, + "step": 155 + }, + { + "epoch": 0.03297269448737764, + "grad_norm": 0.4791430171004267, + "learning_rate": 6.584362139917696e-05, + "loss": 0.3787, + "step": 160 + }, + { + "epoch": 0.03400309119010819, + "grad_norm": 0.349013143936483, + "learning_rate": 6.790123456790123e-05, + "loss": 0.3746, + "step": 165 + }, + { + "epoch": 0.03503348789283874, + "grad_norm": 0.4634759570911148, + "learning_rate": 6.995884773662552e-05, + "loss": 0.3554, + "step": 170 + }, + { + "epoch": 0.03606388459556929, + "grad_norm": 0.6112517018171014, + "learning_rate": 7.20164609053498e-05, + "loss": 0.4185, + "step": 175 + }, + { + "epoch": 0.03709428129829984, + "grad_norm": 0.5549371605797818, + "learning_rate": 7.407407407407407e-05, + "loss": 0.3424, + "step": 180 + }, + { + "epoch": 0.0381246780010304, + "grad_norm": 0.4115597342413292, + "learning_rate": 7.613168724279836e-05, + "loss": 0.3756, + "step": 185 + }, + { + "epoch": 0.03915507470376095, + "grad_norm": 0.42107468069152587, + "learning_rate": 7.818930041152264e-05, + "loss": 0.3499, + "step": 190 + }, + { + "epoch": 0.0401854714064915, + "grad_norm": 0.4487911468673543, + "learning_rate": 8.024691358024692e-05, + "loss": 0.3414, + "step": 195 + }, + { + "epoch": 0.04121586810922205, + "grad_norm": 0.5916231604404705, + "learning_rate": 8.23045267489712e-05, + "loss": 0.3535, + "step": 200 + }, + { + "epoch": 0.04121586810922205, + "eval_loss": 0.35861507058143616, + "eval_runtime": 2884.7816, + "eval_samples_per_second": 2.773, + "eval_steps_per_second": 0.347, + "step": 200 + }, + { + "epoch": 0.0422462648119526, + "grad_norm": 0.48915095397808533, + "learning_rate": 8.436213991769549e-05, + "loss": 0.3254, + "step": 205 + }, + { + "epoch": 0.04327666151468315, + "grad_norm": 0.5120767478539743, + "learning_rate": 8.641975308641975e-05, + "loss": 0.4097, + "step": 210 + }, + { + "epoch": 0.0443070582174137, + "grad_norm": 0.4212019374481453, + "learning_rate": 8.847736625514404e-05, + "loss": 0.3389, + "step": 215 + }, + { + "epoch": 0.04533745492014426, + "grad_norm": 0.4396834951099101, + "learning_rate": 9.053497942386831e-05, + "loss": 0.3942, + "step": 220 + }, + { + "epoch": 0.04636785162287481, + "grad_norm": 0.6548047535409067, + "learning_rate": 9.25925925925926e-05, + "loss": 0.3656, + "step": 225 + }, + { + "epoch": 0.04739824832560536, + "grad_norm": 0.49422979322946825, + "learning_rate": 9.465020576131688e-05, + "loss": 0.3472, + "step": 230 + }, + { + "epoch": 0.04842864502833591, + "grad_norm": 0.45951373244390564, + "learning_rate": 9.670781893004116e-05, + "loss": 0.389, + "step": 235 + }, + { + "epoch": 0.04945904173106646, + "grad_norm": 0.3835541723420252, + "learning_rate": 9.876543209876543e-05, + "loss": 0.3268, + "step": 240 + }, + { + "epoch": 0.05048943843379701, + "grad_norm": 0.43616306650374015, + "learning_rate": 0.00010082304526748971, + "loss": 0.3813, + "step": 245 + }, + { + "epoch": 0.05151983513652756, + "grad_norm": 0.5081680558376234, + "learning_rate": 0.00010288065843621399, + "loss": 0.3842, + "step": 250 + }, + { + "epoch": 0.05255023183925812, + "grad_norm": 0.4927357690560021, + "learning_rate": 0.00010493827160493828, + "loss": 0.3078, + "step": 255 + }, + { + "epoch": 0.05358062854198867, + "grad_norm": 0.40808058841192324, + "learning_rate": 0.00010699588477366255, + "loss": 0.3535, + "step": 260 + }, + { + "epoch": 0.05461102524471922, + "grad_norm": 0.36331664339606895, + "learning_rate": 0.00010905349794238684, + "loss": 0.2998, + "step": 265 + }, + { + "epoch": 0.05564142194744977, + "grad_norm": 0.33960813975720344, + "learning_rate": 0.00011111111111111112, + "loss": 0.3557, + "step": 270 + }, + { + "epoch": 0.05667181865018032, + "grad_norm": 0.45056548565314214, + "learning_rate": 0.0001131687242798354, + "loss": 0.3631, + "step": 275 + }, + { + "epoch": 0.05770221535291087, + "grad_norm": 0.3659158092281187, + "learning_rate": 0.00011522633744855968, + "loss": 0.314, + "step": 280 + }, + { + "epoch": 0.05873261205564142, + "grad_norm": 0.4438288771100659, + "learning_rate": 0.00011728395061728397, + "loss": 0.36, + "step": 285 + }, + { + "epoch": 0.059763008758371976, + "grad_norm": 0.418631875979075, + "learning_rate": 0.00011934156378600823, + "loss": 0.3419, + "step": 290 + }, + { + "epoch": 0.060793405461102526, + "grad_norm": 0.4396365282210275, + "learning_rate": 0.0001213991769547325, + "loss": 0.3928, + "step": 295 + }, + { + "epoch": 0.061823802163833076, + "grad_norm": 0.4402925268314175, + "learning_rate": 0.0001234567901234568, + "loss": 0.4072, + "step": 300 + }, + { + "epoch": 0.06285419886656363, + "grad_norm": 0.4563210413645905, + "learning_rate": 0.00012551440329218108, + "loss": 0.3218, + "step": 305 + }, + { + "epoch": 0.06388459556929418, + "grad_norm": 0.4951662000563643, + "learning_rate": 0.00012757201646090534, + "loss": 0.3638, + "step": 310 + }, + { + "epoch": 0.06491499227202473, + "grad_norm": 0.3406684829409407, + "learning_rate": 0.00012962962962962963, + "loss": 0.3247, + "step": 315 + }, + { + "epoch": 0.06594538897475528, + "grad_norm": 0.3849603318857793, + "learning_rate": 0.00013168724279835392, + "loss": 0.3775, + "step": 320 + }, + { + "epoch": 0.06697578567748583, + "grad_norm": 0.47213282241970134, + "learning_rate": 0.0001337448559670782, + "loss": 0.4242, + "step": 325 + }, + { + "epoch": 0.06800618238021638, + "grad_norm": 0.4081937117941576, + "learning_rate": 0.00013580246913580247, + "loss": 0.3297, + "step": 330 + }, + { + "epoch": 0.06903657908294693, + "grad_norm": 0.4290389394884079, + "learning_rate": 0.00013786008230452676, + "loss": 0.3824, + "step": 335 + }, + { + "epoch": 0.07006697578567748, + "grad_norm": 0.40035955885155716, + "learning_rate": 0.00013991769547325105, + "loss": 0.3567, + "step": 340 + }, + { + "epoch": 0.07109737248840804, + "grad_norm": 0.35369148138173667, + "learning_rate": 0.00014197530864197534, + "loss": 0.3851, + "step": 345 + }, + { + "epoch": 0.07212776919113859, + "grad_norm": 0.3933488263087153, + "learning_rate": 0.0001440329218106996, + "loss": 0.3941, + "step": 350 + }, + { + "epoch": 0.07315816589386914, + "grad_norm": 0.3403535421027664, + "learning_rate": 0.00014609053497942386, + "loss": 0.3279, + "step": 355 + }, + { + "epoch": 0.07418856259659969, + "grad_norm": 0.3901955358278729, + "learning_rate": 0.00014814814814814815, + "loss": 0.3892, + "step": 360 + }, + { + "epoch": 0.07521895929933024, + "grad_norm": 0.3942505006346225, + "learning_rate": 0.00015020576131687243, + "loss": 0.3409, + "step": 365 + }, + { + "epoch": 0.0762493560020608, + "grad_norm": 0.36161132887938113, + "learning_rate": 0.00015226337448559672, + "loss": 0.3653, + "step": 370 + }, + { + "epoch": 0.07727975270479134, + "grad_norm": 0.35951863605570794, + "learning_rate": 0.00015432098765432098, + "loss": 0.369, + "step": 375 + }, + { + "epoch": 0.0783101494075219, + "grad_norm": 0.3508546244263811, + "learning_rate": 0.00015637860082304527, + "loss": 0.3609, + "step": 380 + }, + { + "epoch": 0.07934054611025244, + "grad_norm": 0.4027043746098036, + "learning_rate": 0.00015843621399176956, + "loss": 0.3736, + "step": 385 + }, + { + "epoch": 0.080370942812983, + "grad_norm": 0.32030210240674106, + "learning_rate": 0.00016049382716049385, + "loss": 0.3543, + "step": 390 + }, + { + "epoch": 0.08140133951571354, + "grad_norm": 0.3037761803574099, + "learning_rate": 0.0001625514403292181, + "loss": 0.3657, + "step": 395 + }, + { + "epoch": 0.0824317362184441, + "grad_norm": 0.47893634163074095, + "learning_rate": 0.0001646090534979424, + "loss": 0.4117, + "step": 400 + }, + { + "epoch": 0.0824317362184441, + "eval_loss": 0.3370859622955322, + "eval_runtime": 2880.7606, + "eval_samples_per_second": 2.777, + "eval_steps_per_second": 0.347, + "step": 400 + }, + { + "epoch": 0.08346213292117466, + "grad_norm": 0.3512494491395804, + "learning_rate": 0.0001666666666666667, + "loss": 0.3436, + "step": 405 + }, + { + "epoch": 0.0844925296239052, + "grad_norm": 0.35871617137745193, + "learning_rate": 0.00016872427983539098, + "loss": 0.3617, + "step": 410 + }, + { + "epoch": 0.08552292632663576, + "grad_norm": 0.3508752237247138, + "learning_rate": 0.00017078189300411524, + "loss": 0.3472, + "step": 415 + }, + { + "epoch": 0.0865533230293663, + "grad_norm": 0.40385959435891516, + "learning_rate": 0.0001728395061728395, + "loss": 0.3353, + "step": 420 + }, + { + "epoch": 0.08758371973209686, + "grad_norm": 0.346542199921827, + "learning_rate": 0.0001748971193415638, + "loss": 0.3251, + "step": 425 + }, + { + "epoch": 0.0886141164348274, + "grad_norm": 0.32327444973794744, + "learning_rate": 0.00017695473251028808, + "loss": 0.2782, + "step": 430 + }, + { + "epoch": 0.08964451313755796, + "grad_norm": 0.4605590792675189, + "learning_rate": 0.00017901234567901234, + "loss": 0.3569, + "step": 435 + }, + { + "epoch": 0.09067490984028852, + "grad_norm": 0.33078249165297896, + "learning_rate": 0.00018106995884773663, + "loss": 0.336, + "step": 440 + }, + { + "epoch": 0.09170530654301906, + "grad_norm": 0.3755269642559541, + "learning_rate": 0.00018312757201646091, + "loss": 0.3523, + "step": 445 + }, + { + "epoch": 0.09273570324574962, + "grad_norm": 0.3872427530097071, + "learning_rate": 0.0001851851851851852, + "loss": 0.3798, + "step": 450 + }, + { + "epoch": 0.09376609994848016, + "grad_norm": 0.3360831675941825, + "learning_rate": 0.00018724279835390946, + "loss": 0.294, + "step": 455 + }, + { + "epoch": 0.09479649665121072, + "grad_norm": 0.34721503385661145, + "learning_rate": 0.00018930041152263375, + "loss": 0.3474, + "step": 460 + }, + { + "epoch": 0.09582689335394126, + "grad_norm": 0.4035919572242035, + "learning_rate": 0.00019135802469135804, + "loss": 0.3334, + "step": 465 + }, + { + "epoch": 0.09685729005667182, + "grad_norm": 0.3634206383026372, + "learning_rate": 0.00019341563786008233, + "loss": 0.3529, + "step": 470 + }, + { + "epoch": 0.09788768675940238, + "grad_norm": 0.47867112342634655, + "learning_rate": 0.0001954732510288066, + "loss": 0.3766, + "step": 475 + }, + { + "epoch": 0.09891808346213292, + "grad_norm": 0.2683190757613072, + "learning_rate": 0.00019753086419753085, + "loss": 0.2898, + "step": 480 + }, + { + "epoch": 0.09994848016486348, + "grad_norm": 0.35359467324165766, + "learning_rate": 0.00019958847736625514, + "loss": 0.343, + "step": 485 + }, + { + "epoch": 0.10097887686759402, + "grad_norm": 0.3024165989651168, + "learning_rate": 0.00019999958578867407, + "loss": 0.2837, + "step": 490 + }, + { + "epoch": 0.10200927357032458, + "grad_norm": 0.3060378938897583, + "learning_rate": 0.00019999790306104336, + "loss": 0.3267, + "step": 495 + }, + { + "epoch": 0.10303967027305512, + "grad_norm": 0.3849754123838811, + "learning_rate": 0.0001999949259506647, + "loss": 0.367, + "step": 500 + }, + { + "epoch": 0.10407006697578568, + "grad_norm": 0.357883050197563, + "learning_rate": 0.00019999065449607402, + "loss": 0.2918, + "step": 505 + }, + { + "epoch": 0.10510046367851623, + "grad_norm": 0.41425939893167707, + "learning_rate": 0.00019998508875256158, + "loss": 0.3707, + "step": 510 + }, + { + "epoch": 0.10613086038124678, + "grad_norm": 0.30851216885867083, + "learning_rate": 0.0001999782287921708, + "loss": 0.3154, + "step": 515 + }, + { + "epoch": 0.10716125708397733, + "grad_norm": 0.3349002143749146, + "learning_rate": 0.00019997007470369773, + "loss": 0.3502, + "step": 520 + }, + { + "epoch": 0.10819165378670788, + "grad_norm": 0.37146332082387845, + "learning_rate": 0.0001999606265926897, + "loss": 0.3824, + "step": 525 + }, + { + "epoch": 0.10922205048943844, + "grad_norm": 0.30215937352060623, + "learning_rate": 0.000199949884581444, + "loss": 0.3011, + "step": 530 + }, + { + "epoch": 0.11025244719216898, + "grad_norm": 0.33400076650466676, + "learning_rate": 0.00019993784880900623, + "loss": 0.3178, + "step": 535 + }, + { + "epoch": 0.11128284389489954, + "grad_norm": 0.5100263753382147, + "learning_rate": 0.0001999245194311687, + "loss": 0.3327, + "step": 540 + }, + { + "epoch": 0.11231324059763009, + "grad_norm": 0.35486673055341267, + "learning_rate": 0.00019990989662046818, + "loss": 0.3397, + "step": 545 + }, + { + "epoch": 0.11334363730036064, + "grad_norm": 0.4302123439623329, + "learning_rate": 0.0001998939805661837, + "loss": 0.3359, + "step": 550 + }, + { + "epoch": 0.1143740340030912, + "grad_norm": 0.3424158782854524, + "learning_rate": 0.00019987677147433432, + "loss": 0.3034, + "step": 555 + }, + { + "epoch": 0.11540443070582174, + "grad_norm": 0.3322620075239061, + "learning_rate": 0.0001998582695676762, + "loss": 0.3561, + "step": 560 + }, + { + "epoch": 0.1164348274085523, + "grad_norm": 0.34133172554856017, + "learning_rate": 0.00019983847508569987, + "loss": 0.3092, + "step": 565 + }, + { + "epoch": 0.11746522411128284, + "grad_norm": 0.3114364442803864, + "learning_rate": 0.00019981738828462703, + "loss": 0.3485, + "step": 570 + }, + { + "epoch": 0.1184956208140134, + "grad_norm": 0.4465425096855801, + "learning_rate": 0.00019979500943740735, + "loss": 0.3786, + "step": 575 + }, + { + "epoch": 0.11952601751674395, + "grad_norm": 0.31716364711744127, + "learning_rate": 0.00019977133883371478, + "loss": 0.2759, + "step": 580 + }, + { + "epoch": 0.1205564142194745, + "grad_norm": 0.39768793428389815, + "learning_rate": 0.00019974637677994404, + "loss": 0.3397, + "step": 585 + }, + { + "epoch": 0.12158681092220505, + "grad_norm": 0.36329417504514366, + "learning_rate": 0.00019972012359920638, + "loss": 0.3363, + "step": 590 + }, + { + "epoch": 0.1226172076249356, + "grad_norm": 0.34276825012450907, + "learning_rate": 0.0001996925796313256, + "loss": 0.362, + "step": 595 + }, + { + "epoch": 0.12364760432766615, + "grad_norm": 0.3378267807010673, + "learning_rate": 0.00019966374523283347, + "loss": 0.3577, + "step": 600 + }, + { + "epoch": 0.12364760432766615, + "eval_loss": 0.3276773989200592, + "eval_runtime": 2883.345, + "eval_samples_per_second": 2.775, + "eval_steps_per_second": 0.347, + "step": 600 + }, + { + "epoch": 0.12467800103039671, + "grad_norm": 0.3741701704271306, + "learning_rate": 0.00019963362077696537, + "loss": 0.2976, + "step": 605 + }, + { + "epoch": 0.12570839773312725, + "grad_norm": 0.3477249258246963, + "learning_rate": 0.00019960220665365518, + "loss": 0.3612, + "step": 610 + }, + { + "epoch": 0.1267387944358578, + "grad_norm": 0.3387684957358674, + "learning_rate": 0.0001995695032695305, + "loss": 0.3462, + "step": 615 + }, + { + "epoch": 0.12776919113858837, + "grad_norm": 0.3239206580729764, + "learning_rate": 0.0001995355110479071, + "loss": 0.3366, + "step": 620 + }, + { + "epoch": 0.1287995878413189, + "grad_norm": 0.405004728947829, + "learning_rate": 0.00019950023042878366, + "loss": 0.3746, + "step": 625 + }, + { + "epoch": 0.12982998454404945, + "grad_norm": 0.3198722424159481, + "learning_rate": 0.00019946366186883604, + "loss": 0.274, + "step": 630 + }, + { + "epoch": 0.13086038124678, + "grad_norm": 0.35143449361980517, + "learning_rate": 0.00019942580584141127, + "loss": 0.3399, + "step": 635 + }, + { + "epoch": 0.13189077794951057, + "grad_norm": 0.35776677388978995, + "learning_rate": 0.0001993866628365215, + "loss": 0.3188, + "step": 640 + }, + { + "epoch": 0.13292117465224113, + "grad_norm": 0.28985796662067814, + "learning_rate": 0.00019934623336083772, + "loss": 0.325, + "step": 645 + }, + { + "epoch": 0.13395157135497165, + "grad_norm": 0.4298738885622017, + "learning_rate": 0.00019930451793768298, + "loss": 0.3716, + "step": 650 + }, + { + "epoch": 0.1349819680577022, + "grad_norm": 0.4249611424637099, + "learning_rate": 0.00019926151710702588, + "loss": 0.316, + "step": 655 + }, + { + "epoch": 0.13601236476043277, + "grad_norm": 0.3264504540380892, + "learning_rate": 0.00019921723142547347, + "loss": 0.3485, + "step": 660 + }, + { + "epoch": 0.13704276146316333, + "grad_norm": 0.35251441037598413, + "learning_rate": 0.00019917166146626392, + "loss": 0.3001, + "step": 665 + }, + { + "epoch": 0.13807315816589386, + "grad_norm": 0.35147072953110736, + "learning_rate": 0.0001991248078192593, + "loss": 0.3482, + "step": 670 + }, + { + "epoch": 0.1391035548686244, + "grad_norm": 0.3954004821084264, + "learning_rate": 0.00019907667109093794, + "loss": 0.383, + "step": 675 + }, + { + "epoch": 0.14013395157135497, + "grad_norm": 0.3767557406947751, + "learning_rate": 0.00019902725190438627, + "loss": 0.3111, + "step": 680 + }, + { + "epoch": 0.14116434827408553, + "grad_norm": 0.37536811330325576, + "learning_rate": 0.00019897655089929126, + "loss": 0.3581, + "step": 685 + }, + { + "epoch": 0.14219474497681608, + "grad_norm": 0.3390394380307238, + "learning_rate": 0.00019892456873193165, + "loss": 0.3034, + "step": 690 + }, + { + "epoch": 0.1432251416795466, + "grad_norm": 0.3536586170228539, + "learning_rate": 0.00019887130607516978, + "loss": 0.3709, + "step": 695 + }, + { + "epoch": 0.14425553838227717, + "grad_norm": 0.4333572658960968, + "learning_rate": 0.00019881676361844275, + "loss": 0.365, + "step": 700 + }, + { + "epoch": 0.14528593508500773, + "grad_norm": 0.3450036993380974, + "learning_rate": 0.0001987609420677535, + "loss": 0.2859, + "step": 705 + }, + { + "epoch": 0.14631633178773829, + "grad_norm": 0.37669321445499737, + "learning_rate": 0.00019870384214566174, + "loss": 0.3351, + "step": 710 + }, + { + "epoch": 0.14734672849046884, + "grad_norm": 0.30441804003835093, + "learning_rate": 0.00019864546459127448, + "loss": 0.3173, + "step": 715 + }, + { + "epoch": 0.14837712519319937, + "grad_norm": 0.3562294792918495, + "learning_rate": 0.0001985858101602366, + "loss": 0.3527, + "step": 720 + }, + { + "epoch": 0.14940752189592993, + "grad_norm": 0.4228017846853497, + "learning_rate": 0.0001985248796247209, + "loss": 0.3503, + "step": 725 + }, + { + "epoch": 0.15043791859866049, + "grad_norm": 0.3669422341454514, + "learning_rate": 0.00019846267377341827, + "loss": 0.3211, + "step": 730 + }, + { + "epoch": 0.15146831530139104, + "grad_norm": 0.8193549662084493, + "learning_rate": 0.00019839919341152742, + "loss": 0.3723, + "step": 735 + }, + { + "epoch": 0.1524987120041216, + "grad_norm": 0.3192375696482019, + "learning_rate": 0.00019833443936074442, + "loss": 0.3307, + "step": 740 + }, + { + "epoch": 0.15352910870685213, + "grad_norm": 0.37787227997284556, + "learning_rate": 0.00019826841245925212, + "loss": 0.3543, + "step": 745 + }, + { + "epoch": 0.1545595054095827, + "grad_norm": 0.3977868146433894, + "learning_rate": 0.00019820111356170923, + "loss": 0.3901, + "step": 750 + }, + { + "epoch": 0.15558990211231324, + "grad_norm": 0.31941675552699234, + "learning_rate": 0.00019813254353923937, + "loss": 0.3238, + "step": 755 + }, + { + "epoch": 0.1566202988150438, + "grad_norm": 0.3308867953958594, + "learning_rate": 0.00019806270327941971, + "loss": 0.335, + "step": 760 + }, + { + "epoch": 0.15765069551777433, + "grad_norm": 0.35793702629959095, + "learning_rate": 0.00019799159368626945, + "loss": 0.2967, + "step": 765 + }, + { + "epoch": 0.1586810922205049, + "grad_norm": 0.3990749337961218, + "learning_rate": 0.00019791921568023822, + "loss": 0.3647, + "step": 770 + }, + { + "epoch": 0.15971148892323545, + "grad_norm": 0.37712646550547224, + "learning_rate": 0.00019784557019819404, + "loss": 0.3639, + "step": 775 + }, + { + "epoch": 0.160741885625966, + "grad_norm": 0.3039841465338024, + "learning_rate": 0.00019777065819341137, + "loss": 0.2798, + "step": 780 + }, + { + "epoch": 0.16177228232869656, + "grad_norm": 0.33026286060468363, + "learning_rate": 0.00019769448063555856, + "loss": 0.3811, + "step": 785 + }, + { + "epoch": 0.1628026790314271, + "grad_norm": 0.41703134958746274, + "learning_rate": 0.00019761703851068553, + "loss": 0.3082, + "step": 790 + }, + { + "epoch": 0.16383307573415765, + "grad_norm": 0.3508751011833616, + "learning_rate": 0.0001975383328212107, + "loss": 0.3379, + "step": 795 + }, + { + "epoch": 0.1648634724368882, + "grad_norm": 0.4074398780982895, + "learning_rate": 0.00019745836458590836, + "loss": 0.3594, + "step": 800 + }, + { + "epoch": 0.1648634724368882, + "eval_loss": 0.31942659616470337, + "eval_runtime": 2882.3337, + "eval_samples_per_second": 2.776, + "eval_steps_per_second": 0.347, + "step": 800 + }, + { + "epoch": 0.16589386913961876, + "grad_norm": 0.368070769838129, + "learning_rate": 0.0001973771348398953, + "loss": 0.2961, + "step": 805 + }, + { + "epoch": 0.16692426584234932, + "grad_norm": 0.38698709364654266, + "learning_rate": 0.0001972946446346173, + "loss": 0.3399, + "step": 810 + }, + { + "epoch": 0.16795466254507985, + "grad_norm": 0.36943660915830057, + "learning_rate": 0.00019721089503783577, + "loss": 0.3625, + "step": 815 + }, + { + "epoch": 0.1689850592478104, + "grad_norm": 0.2988194350307528, + "learning_rate": 0.00019712588713361378, + "loss": 0.3206, + "step": 820 + }, + { + "epoch": 0.17001545595054096, + "grad_norm": 0.4353851924786793, + "learning_rate": 0.00019703962202230203, + "loss": 0.3588, + "step": 825 + }, + { + "epoch": 0.17104585265327152, + "grad_norm": 0.31052731774203746, + "learning_rate": 0.00019695210082052472, + "loss": 0.3029, + "step": 830 + }, + { + "epoch": 0.17207624935600205, + "grad_norm": 0.34433049046241526, + "learning_rate": 0.00019686332466116487, + "loss": 0.3371, + "step": 835 + }, + { + "epoch": 0.1731066460587326, + "grad_norm": 0.30745452191681344, + "learning_rate": 0.0001967732946933499, + "loss": 0.2944, + "step": 840 + }, + { + "epoch": 0.17413704276146316, + "grad_norm": 0.31184697151248314, + "learning_rate": 0.00019668201208243658, + "loss": 0.3055, + "step": 845 + }, + { + "epoch": 0.17516743946419372, + "grad_norm": 0.30600883974803933, + "learning_rate": 0.0001965894780099961, + "loss": 0.3842, + "step": 850 + }, + { + "epoch": 0.17619783616692428, + "grad_norm": 0.2887356407325159, + "learning_rate": 0.00019649569367379867, + "loss": 0.3003, + "step": 855 + }, + { + "epoch": 0.1772282328696548, + "grad_norm": 0.3432548851322827, + "learning_rate": 0.00019640066028779794, + "loss": 0.3479, + "step": 860 + }, + { + "epoch": 0.17825862957238536, + "grad_norm": 0.3174948728433821, + "learning_rate": 0.00019630437908211548, + "loss": 0.3126, + "step": 865 + }, + { + "epoch": 0.17928902627511592, + "grad_norm": 0.3097638932288133, + "learning_rate": 0.00019620685130302478, + "loss": 0.3312, + "step": 870 + }, + { + "epoch": 0.18031942297784648, + "grad_norm": 0.42841659366456675, + "learning_rate": 0.00019610807821293503, + "loss": 0.3557, + "step": 875 + }, + { + "epoch": 0.18134981968057703, + "grad_norm": 0.32435713533284716, + "learning_rate": 0.00019600806109037485, + "loss": 0.3339, + "step": 880 + }, + { + "epoch": 0.18238021638330756, + "grad_norm": 0.353414630324643, + "learning_rate": 0.00019590680122997582, + "loss": 0.3283, + "step": 885 + }, + { + "epoch": 0.18341061308603812, + "grad_norm": 0.316231400356792, + "learning_rate": 0.00019580429994245555, + "loss": 0.3368, + "step": 890 + }, + { + "epoch": 0.18444100978876868, + "grad_norm": 0.3585149987799701, + "learning_rate": 0.0001957005585546009, + "loss": 0.3356, + "step": 895 + }, + { + "epoch": 0.18547140649149924, + "grad_norm": 0.36281772675718904, + "learning_rate": 0.00019559557840925055, + "loss": 0.3593, + "step": 900 + }, + { + "epoch": 0.18650180319422976, + "grad_norm": 0.33084410964043454, + "learning_rate": 0.00019548936086527798, + "loss": 0.2723, + "step": 905 + }, + { + "epoch": 0.18753219989696032, + "grad_norm": 0.30849057849040473, + "learning_rate": 0.00019538190729757356, + "loss": 0.3333, + "step": 910 + }, + { + "epoch": 0.18856259659969088, + "grad_norm": 0.3332033528248514, + "learning_rate": 0.00019527321909702688, + "loss": 0.3154, + "step": 915 + }, + { + "epoch": 0.18959299330242144, + "grad_norm": 0.39247405346568126, + "learning_rate": 0.00019516329767050878, + "loss": 0.3034, + "step": 920 + }, + { + "epoch": 0.190623390005152, + "grad_norm": 0.38738599137518354, + "learning_rate": 0.00019505214444085308, + "loss": 0.3755, + "step": 925 + }, + { + "epoch": 0.19165378670788252, + "grad_norm": 0.3353639819496788, + "learning_rate": 0.00019493976084683813, + "loss": 0.3224, + "step": 930 + }, + { + "epoch": 0.19268418341061308, + "grad_norm": 0.3475835339204704, + "learning_rate": 0.00019482614834316836, + "loss": 0.3368, + "step": 935 + }, + { + "epoch": 0.19371458011334364, + "grad_norm": 0.3383417577483814, + "learning_rate": 0.00019471130840045518, + "loss": 0.3114, + "step": 940 + }, + { + "epoch": 0.1947449768160742, + "grad_norm": 0.4095387754149621, + "learning_rate": 0.00019459524250519826, + "loss": 0.3466, + "step": 945 + }, + { + "epoch": 0.19577537351880475, + "grad_norm": 0.3425734449753018, + "learning_rate": 0.00019447795215976594, + "loss": 0.3426, + "step": 950 + }, + { + "epoch": 0.19680577022153528, + "grad_norm": 0.36205423513820917, + "learning_rate": 0.0001943594388823761, + "loss": 0.2825, + "step": 955 + }, + { + "epoch": 0.19783616692426584, + "grad_norm": 0.3722404265826129, + "learning_rate": 0.00019423970420707627, + "loss": 0.3471, + "step": 960 + }, + { + "epoch": 0.1988665636269964, + "grad_norm": 0.40847615253258024, + "learning_rate": 0.00019411874968372402, + "loss": 0.3151, + "step": 965 + }, + { + "epoch": 0.19989696032972695, + "grad_norm": 0.3379508791065496, + "learning_rate": 0.00019399657687796658, + "loss": 0.3414, + "step": 970 + }, + { + "epoch": 0.2009273570324575, + "grad_norm": 0.34821921677428874, + "learning_rate": 0.00019387318737122092, + "loss": 0.3034, + "step": 975 + }, + { + "epoch": 0.20195775373518804, + "grad_norm": 0.4453864463102829, + "learning_rate": 0.0001937485827606529, + "loss": 0.3158, + "step": 980 + }, + { + "epoch": 0.2029881504379186, + "grad_norm": 0.32502909512279515, + "learning_rate": 0.00019362276465915702, + "loss": 0.3426, + "step": 985 + }, + { + "epoch": 0.20401854714064915, + "grad_norm": 0.33635157429335094, + "learning_rate": 0.0001934957346953352, + "loss": 0.3088, + "step": 990 + }, + { + "epoch": 0.2050489438433797, + "grad_norm": 0.32532872733870094, + "learning_rate": 0.00019336749451347586, + "loss": 0.3311, + "step": 995 + }, + { + "epoch": 0.20607934054611024, + "grad_norm": 0.4664188938536663, + "learning_rate": 0.0001932380457735326, + "loss": 0.3603, + "step": 1000 + }, + { + "epoch": 0.20607934054611024, + "eval_loss": 0.30960944294929504, + "eval_runtime": 2881.9251, + "eval_samples_per_second": 2.776, + "eval_steps_per_second": 0.347, + "step": 1000 + }, + { + "epoch": 0.2071097372488408, + "grad_norm": 0.30204444974859856, + "learning_rate": 0.00019310739015110267, + "loss": 0.3048, + "step": 1005 + }, + { + "epoch": 0.20814013395157135, + "grad_norm": 0.366246259849553, + "learning_rate": 0.00019297552933740547, + "loss": 0.3329, + "step": 1010 + }, + { + "epoch": 0.2091705306543019, + "grad_norm": 0.42600306630277895, + "learning_rate": 0.0001928424650392603, + "loss": 0.3223, + "step": 1015 + }, + { + "epoch": 0.21020092735703247, + "grad_norm": 0.4447553605441752, + "learning_rate": 0.00019270819897906468, + "loss": 0.3232, + "step": 1020 + }, + { + "epoch": 0.211231324059763, + "grad_norm": 0.3949912941897582, + "learning_rate": 0.00019257273289477174, + "loss": 0.3383, + "step": 1025 + }, + { + "epoch": 0.21226172076249356, + "grad_norm": 0.34413838854516726, + "learning_rate": 0.00019243606853986786, + "loss": 0.2956, + "step": 1030 + }, + { + "epoch": 0.2132921174652241, + "grad_norm": 0.32739075098948817, + "learning_rate": 0.00019229820768335, + "loss": 0.3441, + "step": 1035 + }, + { + "epoch": 0.21432251416795467, + "grad_norm": 0.31062731875466804, + "learning_rate": 0.00019215915210970267, + "loss": 0.2817, + "step": 1040 + }, + { + "epoch": 0.21535291087068523, + "grad_norm": 0.3603886132870308, + "learning_rate": 0.00019201890361887506, + "loss": 0.3199, + "step": 1045 + }, + { + "epoch": 0.21638330757341576, + "grad_norm": 0.4203386465797232, + "learning_rate": 0.0001918774640262574, + "loss": 0.3599, + "step": 1050 + }, + { + "epoch": 0.2174137042761463, + "grad_norm": 0.37186453766073907, + "learning_rate": 0.00019173483516265788, + "loss": 0.3261, + "step": 1055 + }, + { + "epoch": 0.21844410097887687, + "grad_norm": 0.3357291647533494, + "learning_rate": 0.00019159101887427854, + "loss": 0.3101, + "step": 1060 + }, + { + "epoch": 0.21947449768160743, + "grad_norm": 0.2839969390914523, + "learning_rate": 0.00019144601702269162, + "loss": 0.3065, + "step": 1065 + }, + { + "epoch": 0.22050489438433796, + "grad_norm": 0.37955068309626705, + "learning_rate": 0.00019129983148481552, + "loss": 0.3205, + "step": 1070 + }, + { + "epoch": 0.22153529108706851, + "grad_norm": 0.3864113540369065, + "learning_rate": 0.0001911524641528902, + "loss": 0.3304, + "step": 1075 + }, + { + "epoch": 0.22256568778979907, + "grad_norm": 0.30427047902449084, + "learning_rate": 0.00019100391693445306, + "loss": 0.2778, + "step": 1080 + }, + { + "epoch": 0.22359608449252963, + "grad_norm": 0.3141521581653154, + "learning_rate": 0.00019085419175231394, + "loss": 0.3165, + "step": 1085 + }, + { + "epoch": 0.22462648119526019, + "grad_norm": 0.3274727504782773, + "learning_rate": 0.00019070329054453046, + "loss": 0.2788, + "step": 1090 + }, + { + "epoch": 0.22565687789799072, + "grad_norm": 0.307921166920738, + "learning_rate": 0.00019055121526438272, + "loss": 0.3032, + "step": 1095 + }, + { + "epoch": 0.22668727460072127, + "grad_norm": 0.5011206814587393, + "learning_rate": 0.00019039796788034822, + "loss": 0.3707, + "step": 1100 + }, + { + "epoch": 0.22771767130345183, + "grad_norm": 0.3031868388904063, + "learning_rate": 0.00019024355037607622, + "loss": 0.3075, + "step": 1105 + }, + { + "epoch": 0.2287480680061824, + "grad_norm": 0.4079301863341415, + "learning_rate": 0.00019008796475036222, + "loss": 0.3533, + "step": 1110 + }, + { + "epoch": 0.22977846470891294, + "grad_norm": 0.36502776550467453, + "learning_rate": 0.00018993121301712193, + "loss": 0.3018, + "step": 1115 + }, + { + "epoch": 0.23080886141164347, + "grad_norm": 0.36939878212398075, + "learning_rate": 0.00018977329720536529, + "loss": 0.3577, + "step": 1120 + }, + { + "epoch": 0.23183925811437403, + "grad_norm": 0.46361278352012, + "learning_rate": 0.00018961421935917016, + "loss": 0.3344, + "step": 1125 + }, + { + "epoch": 0.2328696548171046, + "grad_norm": 0.3264834812438753, + "learning_rate": 0.00018945398153765597, + "loss": 0.2821, + "step": 1130 + }, + { + "epoch": 0.23390005151983514, + "grad_norm": 0.3590984056606702, + "learning_rate": 0.00018929258581495685, + "loss": 0.3186, + "step": 1135 + }, + { + "epoch": 0.23493044822256567, + "grad_norm": 0.314422298134173, + "learning_rate": 0.00018913003428019506, + "loss": 0.2805, + "step": 1140 + }, + { + "epoch": 0.23596084492529623, + "grad_norm": 0.40817489184028294, + "learning_rate": 0.00018896632903745374, + "loss": 0.3382, + "step": 1145 + }, + { + "epoch": 0.2369912416280268, + "grad_norm": 0.4550813349162488, + "learning_rate": 0.00018880147220574976, + "loss": 0.3704, + "step": 1150 + }, + { + "epoch": 0.23802163833075735, + "grad_norm": 0.36999200007127014, + "learning_rate": 0.00018863546591900622, + "loss": 0.2956, + "step": 1155 + }, + { + "epoch": 0.2390520350334879, + "grad_norm": 0.3907765146123421, + "learning_rate": 0.00018846831232602492, + "loss": 0.3306, + "step": 1160 + }, + { + "epoch": 0.24008243173621843, + "grad_norm": 0.2290757131338595, + "learning_rate": 0.00018830001359045845, + "loss": 0.2949, + "step": 1165 + }, + { + "epoch": 0.241112828438949, + "grad_norm": 0.3209905523588552, + "learning_rate": 0.00018813057189078243, + "loss": 0.321, + "step": 1170 + }, + { + "epoch": 0.24214322514167955, + "grad_norm": 0.3773596332992971, + "learning_rate": 0.00018795998942026685, + "loss": 0.3522, + "step": 1175 + }, + { + "epoch": 0.2431736218444101, + "grad_norm": 0.34209228943160486, + "learning_rate": 0.00018778826838694812, + "loss": 0.2808, + "step": 1180 + }, + { + "epoch": 0.24420401854714066, + "grad_norm": 0.41181016625387107, + "learning_rate": 0.0001876154110136003, + "loss": 0.329, + "step": 1185 + }, + { + "epoch": 0.2452344152498712, + "grad_norm": 0.3066529005132519, + "learning_rate": 0.0001874414195377063, + "loss": 0.2656, + "step": 1190 + }, + { + "epoch": 0.24626481195260175, + "grad_norm": 0.3794689895293103, + "learning_rate": 0.000187266296211429, + "loss": 0.3052, + "step": 1195 + }, + { + "epoch": 0.2472952086553323, + "grad_norm": 0.4365846410577374, + "learning_rate": 0.0001870900433015821, + "loss": 0.3633, + "step": 1200 + }, + { + "epoch": 0.2472952086553323, + "eval_loss": 0.3063368499279022, + "eval_runtime": 2882.6283, + "eval_samples_per_second": 2.775, + "eval_steps_per_second": 0.347, + "step": 1200 + }, + { + "epoch": 0.24832560535806286, + "grad_norm": 0.38508135542075006, + "learning_rate": 0.00018691266308960066, + "loss": 0.3035, + "step": 1205 + }, + { + "epoch": 0.24935600206079342, + "grad_norm": 0.3726705928374773, + "learning_rate": 0.00018673415787151166, + "loss": 0.3209, + "step": 1210 + }, + { + "epoch": 0.250386398763524, + "grad_norm": 0.2907676540641749, + "learning_rate": 0.00018655452995790435, + "loss": 0.2398, + "step": 1215 + }, + { + "epoch": 0.2514167954662545, + "grad_norm": 0.4457090162372571, + "learning_rate": 0.00018637378167390018, + "loss": 0.2955, + "step": 1220 + }, + { + "epoch": 0.25244719216898504, + "grad_norm": 0.4064739346307168, + "learning_rate": 0.0001861919153591228, + "loss": 0.3579, + "step": 1225 + }, + { + "epoch": 0.2534775888717156, + "grad_norm": 0.3439355157513527, + "learning_rate": 0.00018600893336766786, + "loss": 0.2909, + "step": 1230 + }, + { + "epoch": 0.25450798557444615, + "grad_norm": 0.30462368971059195, + "learning_rate": 0.00018582483806807228, + "loss": 0.3128, + "step": 1235 + }, + { + "epoch": 0.25553838227717673, + "grad_norm": 0.29050064091501104, + "learning_rate": 0.0001856396318432838, + "loss": 0.2849, + "step": 1240 + }, + { + "epoch": 0.25656877897990726, + "grad_norm": 0.30700131979933004, + "learning_rate": 0.0001854533170906302, + "loss": 0.3037, + "step": 1245 + }, + { + "epoch": 0.2575991756826378, + "grad_norm": 0.38518853571022627, + "learning_rate": 0.00018526589622178802, + "loss": 0.3175, + "step": 1250 + }, + { + "epoch": 0.2586295723853684, + "grad_norm": 0.38834107822360947, + "learning_rate": 0.00018507737166275154, + "loss": 0.2856, + "step": 1255 + }, + { + "epoch": 0.2596599690880989, + "grad_norm": 0.3984475799653097, + "learning_rate": 0.00018488774585380125, + "loss": 0.3323, + "step": 1260 + }, + { + "epoch": 0.2606903657908295, + "grad_norm": 0.2895823528396569, + "learning_rate": 0.00018469702124947245, + "loss": 0.2937, + "step": 1265 + }, + { + "epoch": 0.26172076249356, + "grad_norm": 0.2966226853911782, + "learning_rate": 0.00018450520031852325, + "loss": 0.3053, + "step": 1270 + }, + { + "epoch": 0.26275115919629055, + "grad_norm": 0.4177177396553655, + "learning_rate": 0.0001843122855439027, + "loss": 0.3241, + "step": 1275 + }, + { + "epoch": 0.26378155589902114, + "grad_norm": 0.3437444560725597, + "learning_rate": 0.00018411827942271884, + "loss": 0.2482, + "step": 1280 + }, + { + "epoch": 0.26481195260175167, + "grad_norm": 0.28500402984085127, + "learning_rate": 0.000183923184466206, + "loss": 0.3476, + "step": 1285 + }, + { + "epoch": 0.26584234930448225, + "grad_norm": 0.332018609909056, + "learning_rate": 0.0001837270031996926, + "loss": 0.2863, + "step": 1290 + }, + { + "epoch": 0.2668727460072128, + "grad_norm": 0.5825427423902888, + "learning_rate": 0.00018352973816256838, + "loss": 0.3228, + "step": 1295 + }, + { + "epoch": 0.2679031427099433, + "grad_norm": 0.35407010241176123, + "learning_rate": 0.0001833313919082515, + "loss": 0.3699, + "step": 1300 + }, + { + "epoch": 0.2689335394126739, + "grad_norm": 0.34044644075483865, + "learning_rate": 0.0001831319670041555, + "loss": 0.2764, + "step": 1305 + }, + { + "epoch": 0.2699639361154044, + "grad_norm": 0.35384626006434766, + "learning_rate": 0.00018293146603165603, + "loss": 0.3338, + "step": 1310 + }, + { + "epoch": 0.270994332818135, + "grad_norm": 0.3524599113148914, + "learning_rate": 0.00018272989158605752, + "loss": 0.2699, + "step": 1315 + }, + { + "epoch": 0.27202472952086554, + "grad_norm": 0.3581600001494241, + "learning_rate": 0.00018252724627655954, + "loss": 0.2907, + "step": 1320 + }, + { + "epoch": 0.27305512622359607, + "grad_norm": 0.3715546739847767, + "learning_rate": 0.00018232353272622302, + "loss": 0.3091, + "step": 1325 + }, + { + "epoch": 0.27408552292632665, + "grad_norm": 0.3125684903994619, + "learning_rate": 0.00018211875357193632, + "loss": 0.2815, + "step": 1330 + }, + { + "epoch": 0.2751159196290572, + "grad_norm": 0.31294105869026073, + "learning_rate": 0.00018191291146438105, + "loss": 0.3052, + "step": 1335 + }, + { + "epoch": 0.2761463163317877, + "grad_norm": 0.3175018199701703, + "learning_rate": 0.0001817060090679978, + "loss": 0.3345, + "step": 1340 + }, + { + "epoch": 0.2771767130345183, + "grad_norm": 0.2888029163944005, + "learning_rate": 0.00018149804906095163, + "loss": 0.2956, + "step": 1345 + }, + { + "epoch": 0.2782071097372488, + "grad_norm": 0.37092342339586304, + "learning_rate": 0.00018128903413509756, + "loss": 0.33, + "step": 1350 + }, + { + "epoch": 0.2792375064399794, + "grad_norm": 0.37102895224368077, + "learning_rate": 0.0001810789669959453, + "loss": 0.2904, + "step": 1355 + }, + { + "epoch": 0.28026790314270994, + "grad_norm": 0.2968345721462541, + "learning_rate": 0.0001808678503626248, + "loss": 0.3297, + "step": 1360 + }, + { + "epoch": 0.28129829984544047, + "grad_norm": 0.33998169922164373, + "learning_rate": 0.00018065568696785058, + "loss": 0.2835, + "step": 1365 + }, + { + "epoch": 0.28232869654817105, + "grad_norm": 0.3725382909503799, + "learning_rate": 0.00018044247955788662, + "loss": 0.3328, + "step": 1370 + }, + { + "epoch": 0.2833590932509016, + "grad_norm": 0.44691364364670483, + "learning_rate": 0.00018022823089251073, + "loss": 0.3633, + "step": 1375 + }, + { + "epoch": 0.28438948995363217, + "grad_norm": 0.36853190246642853, + "learning_rate": 0.00018001294374497882, + "loss": 0.2866, + "step": 1380 + }, + { + "epoch": 0.2854198866563627, + "grad_norm": 0.3266190569238548, + "learning_rate": 0.00017979662090198906, + "loss": 0.3011, + "step": 1385 + }, + { + "epoch": 0.2864502833590932, + "grad_norm": 0.3732859483546346, + "learning_rate": 0.00017957926516364565, + "loss": 0.3232, + "step": 1390 + }, + { + "epoch": 0.2874806800618238, + "grad_norm": 0.32682252149256485, + "learning_rate": 0.00017936087934342283, + "loss": 0.3277, + "step": 1395 + }, + { + "epoch": 0.28851107676455434, + "grad_norm": 0.37717993812571365, + "learning_rate": 0.00017914146626812823, + "loss": 0.3078, + "step": 1400 + }, + { + "epoch": 0.28851107676455434, + "eval_loss": 0.2999653220176697, + "eval_runtime": 2880.5087, + "eval_samples_per_second": 2.777, + "eval_steps_per_second": 0.347, + "step": 1400 + }, + { + "epoch": 0.2895414734672849, + "grad_norm": 0.2987256318589488, + "learning_rate": 0.0001789210287778664, + "loss": 0.2856, + "step": 1405 + }, + { + "epoch": 0.29057187017001546, + "grad_norm": 0.3199265296581454, + "learning_rate": 0.00017869956972600202, + "loss": 0.3176, + "step": 1410 + }, + { + "epoch": 0.291602266872746, + "grad_norm": 0.3362846877462833, + "learning_rate": 0.00017847709197912296, + "loss": 0.2992, + "step": 1415 + }, + { + "epoch": 0.29263266357547657, + "grad_norm": 0.3185294578549888, + "learning_rate": 0.0001782535984170032, + "loss": 0.2937, + "step": 1420 + }, + { + "epoch": 0.2936630602782071, + "grad_norm": 0.3243854055339378, + "learning_rate": 0.00017802909193256547, + "loss": 0.3524, + "step": 1425 + }, + { + "epoch": 0.2946934569809377, + "grad_norm": 0.33250072639136574, + "learning_rate": 0.00017780357543184397, + "loss": 0.2744, + "step": 1430 + }, + { + "epoch": 0.2957238536836682, + "grad_norm": 0.34028811291554906, + "learning_rate": 0.00017757705183394653, + "loss": 0.3326, + "step": 1435 + }, + { + "epoch": 0.29675425038639874, + "grad_norm": 0.2753584287911616, + "learning_rate": 0.00017734952407101706, + "loss": 0.2782, + "step": 1440 + }, + { + "epoch": 0.29778464708912933, + "grad_norm": 0.4202404483488457, + "learning_rate": 0.0001771209950881974, + "loss": 0.3234, + "step": 1445 + }, + { + "epoch": 0.29881504379185986, + "grad_norm": 0.4350364742880956, + "learning_rate": 0.00017689146784358927, + "loss": 0.3117, + "step": 1450 + }, + { + "epoch": 0.29984544049459044, + "grad_norm": 0.31309620861911946, + "learning_rate": 0.0001766609453082161, + "loss": 0.2912, + "step": 1455 + }, + { + "epoch": 0.30087583719732097, + "grad_norm": 0.32354138707483654, + "learning_rate": 0.00017642943046598436, + "loss": 0.3373, + "step": 1460 + }, + { + "epoch": 0.3019062339000515, + "grad_norm": 0.3515037763945781, + "learning_rate": 0.00017619692631364506, + "loss": 0.3193, + "step": 1465 + }, + { + "epoch": 0.3029366306027821, + "grad_norm": 0.3483804999160483, + "learning_rate": 0.00017596343586075497, + "loss": 0.329, + "step": 1470 + }, + { + "epoch": 0.3039670273055126, + "grad_norm": 0.39993094570722865, + "learning_rate": 0.00017572896212963754, + "loss": 0.3514, + "step": 1475 + }, + { + "epoch": 0.3049974240082432, + "grad_norm": 0.36781639256435866, + "learning_rate": 0.000175493508155344, + "loss": 0.3263, + "step": 1480 + }, + { + "epoch": 0.30602782071097373, + "grad_norm": 0.29161988646871145, + "learning_rate": 0.00017525707698561385, + "loss": 0.3461, + "step": 1485 + }, + { + "epoch": 0.30705821741370426, + "grad_norm": 0.3154471225790784, + "learning_rate": 0.00017501967168083557, + "loss": 0.2986, + "step": 1490 + }, + { + "epoch": 0.30808861411643484, + "grad_norm": 0.34492695895689485, + "learning_rate": 0.00017478129531400688, + "loss": 0.3079, + "step": 1495 + }, + { + "epoch": 0.3091190108191654, + "grad_norm": 0.4725349308354448, + "learning_rate": 0.00017454195097069505, + "loss": 0.3115, + "step": 1500 + }, + { + "epoch": 0.3101494075218959, + "grad_norm": 0.3436076430515735, + "learning_rate": 0.00017430164174899696, + "loss": 0.2844, + "step": 1505 + }, + { + "epoch": 0.3111798042246265, + "grad_norm": 0.3199620943087237, + "learning_rate": 0.0001740603707594989, + "loss": 0.3363, + "step": 1510 + }, + { + "epoch": 0.312210200927357, + "grad_norm": 0.3294543675032788, + "learning_rate": 0.00017381814112523648, + "loss": 0.2831, + "step": 1515 + }, + { + "epoch": 0.3132405976300876, + "grad_norm": 0.4442315813048831, + "learning_rate": 0.000173574955981654, + "loss": 0.3239, + "step": 1520 + }, + { + "epoch": 0.31427099433281813, + "grad_norm": 0.3315063178638844, + "learning_rate": 0.00017333081847656397, + "loss": 0.3282, + "step": 1525 + }, + { + "epoch": 0.31530139103554866, + "grad_norm": 0.38910409480102465, + "learning_rate": 0.00017308573177010652, + "loss": 0.3091, + "step": 1530 + }, + { + "epoch": 0.31633178773827925, + "grad_norm": 0.3610118589070733, + "learning_rate": 0.00017283969903470815, + "loss": 0.3047, + "step": 1535 + }, + { + "epoch": 0.3173621844410098, + "grad_norm": 0.3342784924557649, + "learning_rate": 0.0001725927234550409, + "loss": 0.2829, + "step": 1540 + }, + { + "epoch": 0.31839258114374036, + "grad_norm": 0.3065893071381057, + "learning_rate": 0.00017234480822798113, + "loss": 0.327, + "step": 1545 + }, + { + "epoch": 0.3194229778464709, + "grad_norm": 0.3678298005788044, + "learning_rate": 0.00017209595656256807, + "loss": 0.3401, + "step": 1550 + }, + { + "epoch": 0.3204533745492014, + "grad_norm": 0.333367656104773, + "learning_rate": 0.0001718461716799623, + "loss": 0.2857, + "step": 1555 + }, + { + "epoch": 0.321483771251932, + "grad_norm": 0.37925023182652434, + "learning_rate": 0.000171595456813404, + "loss": 0.3491, + "step": 1560 + }, + { + "epoch": 0.32251416795466253, + "grad_norm": 0.3765681871737522, + "learning_rate": 0.00017134381520817127, + "loss": 0.3001, + "step": 1565 + }, + { + "epoch": 0.3235445646573931, + "grad_norm": 0.29672578991036613, + "learning_rate": 0.00017109125012153783, + "loss": 0.2959, + "step": 1570 + }, + { + "epoch": 0.32457496136012365, + "grad_norm": 0.42011792115497876, + "learning_rate": 0.00017083776482273126, + "loss": 0.3532, + "step": 1575 + }, + { + "epoch": 0.3256053580628542, + "grad_norm": 0.4222336181716507, + "learning_rate": 0.00017058336259289026, + "loss": 0.247, + "step": 1580 + }, + { + "epoch": 0.32663575476558476, + "grad_norm": 0.31097703592676695, + "learning_rate": 0.0001703280467250225, + "loss": 0.3212, + "step": 1585 + }, + { + "epoch": 0.3276661514683153, + "grad_norm": 0.2963584098165209, + "learning_rate": 0.0001700718205239618, + "loss": 0.285, + "step": 1590 + }, + { + "epoch": 0.3286965481710459, + "grad_norm": 0.27473277659713885, + "learning_rate": 0.0001698146873063255, + "loss": 0.2852, + "step": 1595 + }, + { + "epoch": 0.3297269448737764, + "grad_norm": 0.41875809452677054, + "learning_rate": 0.00016955665040047134, + "loss": 0.3274, + "step": 1600 + }, + { + "epoch": 0.3297269448737764, + "eval_loss": 0.2948421239852905, + "eval_runtime": 2881.5407, + "eval_samples_per_second": 2.776, + "eval_steps_per_second": 0.347, + "step": 1600 + }, + { + "epoch": 0.33075734157650694, + "grad_norm": 0.3202391256850226, + "learning_rate": 0.00016929771314645454, + "loss": 0.2402, + "step": 1605 + }, + { + "epoch": 0.3317877382792375, + "grad_norm": 0.3612689826913437, + "learning_rate": 0.00016903787889598458, + "loss": 0.3088, + "step": 1610 + }, + { + "epoch": 0.33281813498196805, + "grad_norm": 0.31051336019913206, + "learning_rate": 0.00016877715101238172, + "loss": 0.3202, + "step": 1615 + }, + { + "epoch": 0.33384853168469864, + "grad_norm": 0.3462332312673768, + "learning_rate": 0.00016851553287053342, + "loss": 0.3489, + "step": 1620 + }, + { + "epoch": 0.33487892838742916, + "grad_norm": 0.4197430450739659, + "learning_rate": 0.00016825302785685077, + "loss": 0.3453, + "step": 1625 + }, + { + "epoch": 0.3359093250901597, + "grad_norm": 0.355766620140618, + "learning_rate": 0.00016798963936922467, + "loss": 0.3032, + "step": 1630 + }, + { + "epoch": 0.3369397217928903, + "grad_norm": 0.452678294800493, + "learning_rate": 0.00016772537081698175, + "loss": 0.3333, + "step": 1635 + }, + { + "epoch": 0.3379701184956208, + "grad_norm": 0.38423566249718305, + "learning_rate": 0.00016746022562084026, + "loss": 0.2674, + "step": 1640 + }, + { + "epoch": 0.33900051519835134, + "grad_norm": 0.2811598903049604, + "learning_rate": 0.0001671942072128659, + "loss": 0.3195, + "step": 1645 + }, + { + "epoch": 0.3400309119010819, + "grad_norm": 0.3585975448009611, + "learning_rate": 0.00016692731903642725, + "loss": 0.3445, + "step": 1650 + }, + { + "epoch": 0.34106130860381245, + "grad_norm": 0.3724482026933407, + "learning_rate": 0.0001666595645461512, + "loss": 0.3199, + "step": 1655 + }, + { + "epoch": 0.34209170530654304, + "grad_norm": 0.3867823350912187, + "learning_rate": 0.0001663909472078784, + "loss": 0.3194, + "step": 1660 + }, + { + "epoch": 0.34312210200927357, + "grad_norm": 0.33254169042512904, + "learning_rate": 0.0001661214704986182, + "loss": 0.3144, + "step": 1665 + }, + { + "epoch": 0.3441524987120041, + "grad_norm": 0.3478869105292329, + "learning_rate": 0.00016585113790650388, + "loss": 0.3242, + "step": 1670 + }, + { + "epoch": 0.3451828954147347, + "grad_norm": 0.34064698003921173, + "learning_rate": 0.00016557995293074715, + "loss": 0.3102, + "step": 1675 + }, + { + "epoch": 0.3462132921174652, + "grad_norm": 0.4527188823274613, + "learning_rate": 0.00016530791908159323, + "loss": 0.2957, + "step": 1680 + }, + { + "epoch": 0.3472436888201958, + "grad_norm": 0.3309454383786622, + "learning_rate": 0.0001650350398802751, + "loss": 0.3244, + "step": 1685 + }, + { + "epoch": 0.3482740855229263, + "grad_norm": 0.3589818975912585, + "learning_rate": 0.0001647613188589682, + "loss": 0.2871, + "step": 1690 + }, + { + "epoch": 0.34930448222565685, + "grad_norm": 0.369801877686259, + "learning_rate": 0.00016448675956074444, + "loss": 0.2991, + "step": 1695 + }, + { + "epoch": 0.35033487892838744, + "grad_norm": 0.3664277823673914, + "learning_rate": 0.0001642113655395266, + "loss": 0.3374, + "step": 1700 + }, + { + "epoch": 0.35136527563111797, + "grad_norm": 0.30826498189575957, + "learning_rate": 0.00016393514036004204, + "loss": 0.2722, + "step": 1705 + }, + { + "epoch": 0.35239567233384855, + "grad_norm": 0.2991931556332237, + "learning_rate": 0.0001636580875977769, + "loss": 0.317, + "step": 1710 + }, + { + "epoch": 0.3534260690365791, + "grad_norm": 0.3253832462040809, + "learning_rate": 0.0001633802108389295, + "loss": 0.3, + "step": 1715 + }, + { + "epoch": 0.3544564657393096, + "grad_norm": 0.37843130847255274, + "learning_rate": 0.00016310151368036408, + "loss": 0.3036, + "step": 1720 + }, + { + "epoch": 0.3554868624420402, + "grad_norm": 0.3399737695505088, + "learning_rate": 0.00016282199972956425, + "loss": 0.2905, + "step": 1725 + }, + { + "epoch": 0.3565172591447707, + "grad_norm": 0.32815030354752983, + "learning_rate": 0.00016254167260458622, + "loss": 0.3056, + "step": 1730 + }, + { + "epoch": 0.3575476558475013, + "grad_norm": 0.3322854407295662, + "learning_rate": 0.000162260535934012, + "loss": 0.2974, + "step": 1735 + }, + { + "epoch": 0.35857805255023184, + "grad_norm": 0.3141453045464179, + "learning_rate": 0.00016197859335690247, + "loss": 0.289, + "step": 1740 + }, + { + "epoch": 0.35960844925296237, + "grad_norm": 0.32100160310926246, + "learning_rate": 0.0001616958485227503, + "loss": 0.3173, + "step": 1745 + }, + { + "epoch": 0.36063884595569295, + "grad_norm": 0.39886105757863777, + "learning_rate": 0.0001614123050914325, + "loss": 0.3074, + "step": 1750 + }, + { + "epoch": 0.3616692426584235, + "grad_norm": 0.3604472382794737, + "learning_rate": 0.00016112796673316336, + "loss": 0.2689, + "step": 1755 + }, + { + "epoch": 0.36269963936115407, + "grad_norm": 0.2784618832581671, + "learning_rate": 0.00016084283712844666, + "loss": 0.3148, + "step": 1760 + }, + { + "epoch": 0.3637300360638846, + "grad_norm": 0.3021780644976563, + "learning_rate": 0.00016055691996802823, + "loss": 0.2802, + "step": 1765 + }, + { + "epoch": 0.36476043276661513, + "grad_norm": 0.34664455247499926, + "learning_rate": 0.00016027021895284808, + "loss": 0.3214, + "step": 1770 + }, + { + "epoch": 0.3657908294693457, + "grad_norm": 0.3169725333496023, + "learning_rate": 0.0001599827377939925, + "loss": 0.2969, + "step": 1775 + }, + { + "epoch": 0.36682122617207624, + "grad_norm": 0.3171890414737335, + "learning_rate": 0.00015969448021264606, + "loss": 0.2762, + "step": 1780 + }, + { + "epoch": 0.3678516228748068, + "grad_norm": 0.32861761572712384, + "learning_rate": 0.00015940544994004334, + "loss": 0.3453, + "step": 1785 + }, + { + "epoch": 0.36888201957753736, + "grad_norm": 0.31257042261002754, + "learning_rate": 0.00015911565071742088, + "loss": 0.3054, + "step": 1790 + }, + { + "epoch": 0.3699124162802679, + "grad_norm": 0.3241031150716069, + "learning_rate": 0.00015882508629596836, + "loss": 0.3076, + "step": 1795 + }, + { + "epoch": 0.37094281298299847, + "grad_norm": 0.39403630552828167, + "learning_rate": 0.00015853376043678053, + "loss": 0.3474, + "step": 1800 + }, + { + "epoch": 0.37094281298299847, + "eval_loss": 0.2924995422363281, + "eval_runtime": 2880.2214, + "eval_samples_per_second": 2.778, + "eval_steps_per_second": 0.347, + "step": 1800 + }, + { + "epoch": 0.371973209685729, + "grad_norm": 0.2832222312603273, + "learning_rate": 0.00015824167691080802, + "loss": 0.2464, + "step": 1805 + }, + { + "epoch": 0.37300360638845953, + "grad_norm": 0.34854474448900724, + "learning_rate": 0.00015794883949880894, + "loss": 0.3153, + "step": 1810 + }, + { + "epoch": 0.3740340030911901, + "grad_norm": 0.411995708678398, + "learning_rate": 0.00015765525199129966, + "loss": 0.3075, + "step": 1815 + }, + { + "epoch": 0.37506439979392064, + "grad_norm": 0.35807431649917565, + "learning_rate": 0.000157360918188506, + "loss": 0.3233, + "step": 1820 + }, + { + "epoch": 0.37609479649665123, + "grad_norm": 0.39599363094279105, + "learning_rate": 0.0001570658419003137, + "loss": 0.3298, + "step": 1825 + }, + { + "epoch": 0.37712519319938176, + "grad_norm": 0.38707229731214027, + "learning_rate": 0.00015677002694621948, + "loss": 0.269, + "step": 1830 + }, + { + "epoch": 0.3781555899021123, + "grad_norm": 0.32749605968678447, + "learning_rate": 0.00015647347715528137, + "loss": 0.3334, + "step": 1835 + }, + { + "epoch": 0.3791859866048429, + "grad_norm": 0.35761830471268774, + "learning_rate": 0.00015617619636606924, + "loss": 0.2933, + "step": 1840 + }, + { + "epoch": 0.3802163833075734, + "grad_norm": 0.35967453920717335, + "learning_rate": 0.00015587818842661494, + "loss": 0.2949, + "step": 1845 + }, + { + "epoch": 0.381246780010304, + "grad_norm": 0.3143527248467094, + "learning_rate": 0.00015557945719436278, + "loss": 0.329, + "step": 1850 + }, + { + "epoch": 0.3822771767130345, + "grad_norm": 0.372991583134114, + "learning_rate": 0.00015528000653611935, + "loss": 0.3037, + "step": 1855 + }, + { + "epoch": 0.38330757341576505, + "grad_norm": 0.35004314943597253, + "learning_rate": 0.0001549798403280036, + "loss": 0.3377, + "step": 1860 + }, + { + "epoch": 0.38433797011849563, + "grad_norm": 0.3231196426185917, + "learning_rate": 0.0001546789624553966, + "loss": 0.3208, + "step": 1865 + }, + { + "epoch": 0.38536836682122616, + "grad_norm": 0.35488369318279184, + "learning_rate": 0.00015437737681289128, + "loss": 0.296, + "step": 1870 + }, + { + "epoch": 0.38639876352395675, + "grad_norm": 0.3718463756429486, + "learning_rate": 0.00015407508730424206, + "loss": 0.3374, + "step": 1875 + }, + { + "epoch": 0.3874291602266873, + "grad_norm": 0.2802376403000415, + "learning_rate": 0.00015377209784231424, + "loss": 0.2588, + "step": 1880 + }, + { + "epoch": 0.3884595569294178, + "grad_norm": 0.3799105689825581, + "learning_rate": 0.00015346841234903337, + "loss": 0.3026, + "step": 1885 + }, + { + "epoch": 0.3894899536321484, + "grad_norm": 0.3296377663164442, + "learning_rate": 0.0001531640347553345, + "loss": 0.2751, + "step": 1890 + }, + { + "epoch": 0.3905203503348789, + "grad_norm": 0.36279328661672916, + "learning_rate": 0.00015285896900111133, + "loss": 0.3242, + "step": 1895 + }, + { + "epoch": 0.3915507470376095, + "grad_norm": 0.3552709404621765, + "learning_rate": 0.00015255321903516516, + "loss": 0.3264, + "step": 1900 + }, + { + "epoch": 0.39258114374034003, + "grad_norm": 0.41641063282756574, + "learning_rate": 0.00015224678881515377, + "loss": 0.2637, + "step": 1905 + }, + { + "epoch": 0.39361154044307056, + "grad_norm": 0.3878387710201773, + "learning_rate": 0.00015193968230754024, + "loss": 0.3148, + "step": 1910 + }, + { + "epoch": 0.39464193714580115, + "grad_norm": 0.3421084341160711, + "learning_rate": 0.00015163190348754162, + "loss": 0.2826, + "step": 1915 + }, + { + "epoch": 0.3956723338485317, + "grad_norm": 0.2898537738332495, + "learning_rate": 0.00015132345633907734, + "loss": 0.3069, + "step": 1920 + }, + { + "epoch": 0.39670273055126226, + "grad_norm": 0.3594731425839707, + "learning_rate": 0.0001510143448547178, + "loss": 0.347, + "step": 1925 + }, + { + "epoch": 0.3977331272539928, + "grad_norm": 0.33029549881450443, + "learning_rate": 0.00015070457303563268, + "loss": 0.3231, + "step": 1930 + }, + { + "epoch": 0.3987635239567233, + "grad_norm": 0.40244939641132027, + "learning_rate": 0.000150394144891539, + "loss": 0.3372, + "step": 1935 + }, + { + "epoch": 0.3997939206594539, + "grad_norm": 0.35291733189988017, + "learning_rate": 0.00015008306444064942, + "loss": 0.2737, + "step": 1940 + }, + { + "epoch": 0.40082431736218443, + "grad_norm": 0.32337967389685746, + "learning_rate": 0.00014977133570961997, + "loss": 0.2988, + "step": 1945 + }, + { + "epoch": 0.401854714064915, + "grad_norm": 0.36198496665746605, + "learning_rate": 0.00014945896273349827, + "loss": 0.3199, + "step": 1950 + }, + { + "epoch": 0.40288511076764555, + "grad_norm": 0.2982085727716395, + "learning_rate": 0.00014914594955567099, + "loss": 0.2734, + "step": 1955 + }, + { + "epoch": 0.4039155074703761, + "grad_norm": 0.35550939579352364, + "learning_rate": 0.00014883230022781163, + "loss": 0.3455, + "step": 1960 + }, + { + "epoch": 0.40494590417310666, + "grad_norm": 0.3483385027728902, + "learning_rate": 0.00014851801880982814, + "loss": 0.2716, + "step": 1965 + }, + { + "epoch": 0.4059763008758372, + "grad_norm": 0.348108964846049, + "learning_rate": 0.00014820310936981026, + "loss": 0.2873, + "step": 1970 + }, + { + "epoch": 0.4070066975785677, + "grad_norm": 0.3953405122632469, + "learning_rate": 0.000147887575983977, + "loss": 0.3316, + "step": 1975 + }, + { + "epoch": 0.4080370942812983, + "grad_norm": 0.27428554427495483, + "learning_rate": 0.00014757142273662358, + "loss": 0.2857, + "step": 1980 + }, + { + "epoch": 0.40906749098402884, + "grad_norm": 0.3584297752000547, + "learning_rate": 0.00014725465372006905, + "loss": 0.3355, + "step": 1985 + }, + { + "epoch": 0.4100978876867594, + "grad_norm": 0.2972999446567122, + "learning_rate": 0.0001469372730346028, + "loss": 0.2656, + "step": 1990 + }, + { + "epoch": 0.41112828438948995, + "grad_norm": 0.3627876088336718, + "learning_rate": 0.00014661928478843186, + "loss": 0.3345, + "step": 1995 + }, + { + "epoch": 0.4121586810922205, + "grad_norm": 0.3495779600410771, + "learning_rate": 0.00014630069309762753, + "loss": 0.3401, + "step": 2000 + }, + { + "epoch": 0.4121586810922205, + "eval_loss": 0.2875462770462036, + "eval_runtime": 2879.4184, + "eval_samples_per_second": 2.778, + "eval_steps_per_second": 0.347, + "step": 2000 + }, + { + "epoch": 0.41318907779495107, + "grad_norm": 0.3391439590455538, + "learning_rate": 0.00014598150208607212, + "loss": 0.2705, + "step": 2005 + }, + { + "epoch": 0.4142194744976816, + "grad_norm": 0.29356233596458986, + "learning_rate": 0.00014566171588540572, + "loss": 0.3306, + "step": 2010 + }, + { + "epoch": 0.4152498712004122, + "grad_norm": 0.33751989432866675, + "learning_rate": 0.00014534133863497252, + "loss": 0.2923, + "step": 2015 + }, + { + "epoch": 0.4162802679031427, + "grad_norm": 0.32503045452079943, + "learning_rate": 0.00014502037448176734, + "loss": 0.2681, + "step": 2020 + }, + { + "epoch": 0.41731066460587324, + "grad_norm": 0.4256762276514426, + "learning_rate": 0.00014469882758038193, + "loss": 0.3142, + "step": 2025 + }, + { + "epoch": 0.4183410613086038, + "grad_norm": 0.31736929564499033, + "learning_rate": 0.00014437670209295112, + "loss": 0.281, + "step": 2030 + }, + { + "epoch": 0.41937145801133435, + "grad_norm": 0.289199714492364, + "learning_rate": 0.0001440540021890992, + "loss": 0.3127, + "step": 2035 + }, + { + "epoch": 0.42040185471406494, + "grad_norm": 0.32970220123556293, + "learning_rate": 0.00014373073204588556, + "loss": 0.2875, + "step": 2040 + }, + { + "epoch": 0.42143225141679547, + "grad_norm": 0.32861612982909, + "learning_rate": 0.0001434068958477509, + "loss": 0.3219, + "step": 2045 + }, + { + "epoch": 0.422462648119526, + "grad_norm": 0.4097507208513791, + "learning_rate": 0.00014308249778646306, + "loss": 0.3185, + "step": 2050 + }, + { + "epoch": 0.4234930448222566, + "grad_norm": 0.3696930062252253, + "learning_rate": 0.0001427575420610626, + "loss": 0.2635, + "step": 2055 + }, + { + "epoch": 0.4245234415249871, + "grad_norm": 0.3440933088510053, + "learning_rate": 0.00014243203287780856, + "loss": 0.322, + "step": 2060 + }, + { + "epoch": 0.4255538382277177, + "grad_norm": 0.3639003692887089, + "learning_rate": 0.00014210597445012398, + "loss": 0.2911, + "step": 2065 + }, + { + "epoch": 0.4265842349304482, + "grad_norm": 0.3552632151370132, + "learning_rate": 0.0001417793709985415, + "loss": 0.3233, + "step": 2070 + }, + { + "epoch": 0.42761463163317875, + "grad_norm": 0.385489568046544, + "learning_rate": 0.0001414522267506484, + "loss": 0.348, + "step": 2075 + }, + { + "epoch": 0.42864502833590934, + "grad_norm": 0.34307097509378426, + "learning_rate": 0.0001411245459410322, + "loss": 0.2649, + "step": 2080 + }, + { + "epoch": 0.42967542503863987, + "grad_norm": 0.3018369278756531, + "learning_rate": 0.00014079633281122573, + "loss": 0.3082, + "step": 2085 + }, + { + "epoch": 0.43070582174137045, + "grad_norm": 0.286861468385612, + "learning_rate": 0.00014046759160965224, + "loss": 0.2935, + "step": 2090 + }, + { + "epoch": 0.431736218444101, + "grad_norm": 0.3506128021819822, + "learning_rate": 0.0001401383265915703, + "loss": 0.303, + "step": 2095 + }, + { + "epoch": 0.4327666151468315, + "grad_norm": 0.40503992585942733, + "learning_rate": 0.00013980854201901886, + "loss": 0.3254, + "step": 2100 + }, + { + "epoch": 0.4337970118495621, + "grad_norm": 0.3697571032927392, + "learning_rate": 0.00013947824216076207, + "loss": 0.2567, + "step": 2105 + }, + { + "epoch": 0.4348274085522926, + "grad_norm": 0.399061699010693, + "learning_rate": 0.00013914743129223405, + "loss": 0.3413, + "step": 2110 + }, + { + "epoch": 0.43585780525502316, + "grad_norm": 0.32157629254065767, + "learning_rate": 0.00013881611369548325, + "loss": 0.2802, + "step": 2115 + }, + { + "epoch": 0.43688820195775374, + "grad_norm": 0.3567204984816549, + "learning_rate": 0.00013848429365911753, + "loss": 0.3155, + "step": 2120 + }, + { + "epoch": 0.43791859866048427, + "grad_norm": 0.30654912781148985, + "learning_rate": 0.00013815197547824824, + "loss": 0.3, + "step": 2125 + }, + { + "epoch": 0.43894899536321486, + "grad_norm": 0.3866305793168129, + "learning_rate": 0.00013781916345443474, + "loss": 0.285, + "step": 2130 + }, + { + "epoch": 0.4399793920659454, + "grad_norm": 0.38963039998932314, + "learning_rate": 0.00013748586189562878, + "loss": 0.3271, + "step": 2135 + }, + { + "epoch": 0.4410097887686759, + "grad_norm": 0.2944680903796273, + "learning_rate": 0.00013715207511611876, + "loss": 0.2796, + "step": 2140 + }, + { + "epoch": 0.4420401854714065, + "grad_norm": 0.3287140509685993, + "learning_rate": 0.0001368178074364737, + "loss": 0.2794, + "step": 2145 + }, + { + "epoch": 0.44307058217413703, + "grad_norm": 0.3497708699847718, + "learning_rate": 0.00013648306318348762, + "loss": 0.348, + "step": 2150 + }, + { + "epoch": 0.4441009788768676, + "grad_norm": 0.3577124630797181, + "learning_rate": 0.0001361478466901231, + "loss": 0.2638, + "step": 2155 + }, + { + "epoch": 0.44513137557959814, + "grad_norm": 0.3160258522774726, + "learning_rate": 0.0001358121622954557, + "loss": 0.305, + "step": 2160 + }, + { + "epoch": 0.4461617722823287, + "grad_norm": 0.3245915685090122, + "learning_rate": 0.00013547601434461733, + "loss": 0.2794, + "step": 2165 + }, + { + "epoch": 0.44719216898505926, + "grad_norm": 0.3081359389684238, + "learning_rate": 0.0001351394071887404, + "loss": 0.3119, + "step": 2170 + }, + { + "epoch": 0.4482225656877898, + "grad_norm": 0.3562399653740551, + "learning_rate": 0.0001348023451849012, + "loss": 0.3133, + "step": 2175 + }, + { + "epoch": 0.44925296239052037, + "grad_norm": 0.33659092073928976, + "learning_rate": 0.00013446483269606362, + "loss": 0.2584, + "step": 2180 + }, + { + "epoch": 0.4502833590932509, + "grad_norm": 0.3486043316585054, + "learning_rate": 0.00013412687409102277, + "loss": 0.3073, + "step": 2185 + }, + { + "epoch": 0.45131375579598143, + "grad_norm": 0.3184771967280283, + "learning_rate": 0.00013378847374434814, + "loss": 0.2868, + "step": 2190 + }, + { + "epoch": 0.452344152498712, + "grad_norm": 0.32065577051849475, + "learning_rate": 0.0001334496360363274, + "loss": 0.2804, + "step": 2195 + }, + { + "epoch": 0.45337454920144254, + "grad_norm": 0.3597968288617642, + "learning_rate": 0.00013311036535290925, + "loss": 0.3124, + "step": 2200 + }, + { + "epoch": 0.45337454920144254, + "eval_loss": 0.2839260995388031, + "eval_runtime": 2877.4926, + "eval_samples_per_second": 2.78, + "eval_steps_per_second": 0.348, + "step": 2200 + }, + { + "epoch": 0.45440494590417313, + "grad_norm": 0.3642061316908434, + "learning_rate": 0.00013277066608564713, + "loss": 0.2602, + "step": 2205 + }, + { + "epoch": 0.45543534260690366, + "grad_norm": 0.4070129402915311, + "learning_rate": 0.0001324305426316418, + "loss": 0.3135, + "step": 2210 + }, + { + "epoch": 0.4564657393096342, + "grad_norm": 0.28705492007355193, + "learning_rate": 0.000132089999393485, + "loss": 0.2955, + "step": 2215 + }, + { + "epoch": 0.4574961360123648, + "grad_norm": 0.4192774207761234, + "learning_rate": 0.000131749040779202, + "loss": 0.3107, + "step": 2220 + }, + { + "epoch": 0.4585265327150953, + "grad_norm": 0.35888440532939486, + "learning_rate": 0.0001314076712021949, + "loss": 0.3141, + "step": 2225 + }, + { + "epoch": 0.4595569294178259, + "grad_norm": 0.36636355983534064, + "learning_rate": 0.0001310658950811852, + "loss": 0.2645, + "step": 2230 + }, + { + "epoch": 0.4605873261205564, + "grad_norm": 0.2950156999856839, + "learning_rate": 0.00013072371684015688, + "loss": 0.2863, + "step": 2235 + }, + { + "epoch": 0.46161772282328695, + "grad_norm": 0.3006839955052212, + "learning_rate": 0.00013038114090829892, + "loss": 0.2907, + "step": 2240 + }, + { + "epoch": 0.46264811952601753, + "grad_norm": 0.4458786907297178, + "learning_rate": 0.00013003817171994807, + "loss": 0.3088, + "step": 2245 + }, + { + "epoch": 0.46367851622874806, + "grad_norm": 0.45350359238505367, + "learning_rate": 0.00012969481371453135, + "loss": 0.3292, + "step": 2250 + }, + { + "epoch": 0.46470891293147865, + "grad_norm": 0.32874718511199297, + "learning_rate": 0.00012935107133650885, + "loss": 0.26, + "step": 2255 + }, + { + "epoch": 0.4657393096342092, + "grad_norm": 0.3640981247486453, + "learning_rate": 0.00012900694903531586, + "loss": 0.3098, + "step": 2260 + }, + { + "epoch": 0.4667697063369397, + "grad_norm": 0.3753366193823528, + "learning_rate": 0.0001286624512653055, + "loss": 0.2864, + "step": 2265 + }, + { + "epoch": 0.4678001030396703, + "grad_norm": 0.3932136104409213, + "learning_rate": 0.00012831758248569097, + "loss": 0.3311, + "step": 2270 + }, + { + "epoch": 0.4688304997424008, + "grad_norm": 0.40583272633582196, + "learning_rate": 0.00012797234716048784, + "loss": 0.2953, + "step": 2275 + }, + { + "epoch": 0.46986089644513135, + "grad_norm": 0.3932055476008604, + "learning_rate": 0.00012762674975845637, + "loss": 0.2474, + "step": 2280 + }, + { + "epoch": 0.47089129314786193, + "grad_norm": 0.30115467677227103, + "learning_rate": 0.00012728079475304345, + "loss": 0.3102, + "step": 2285 + }, + { + "epoch": 0.47192168985059246, + "grad_norm": 0.3161465840153421, + "learning_rate": 0.0001269344866223249, + "loss": 0.27, + "step": 2290 + }, + { + "epoch": 0.47295208655332305, + "grad_norm": 0.2984326355380522, + "learning_rate": 0.00012658782984894743, + "loss": 0.2889, + "step": 2295 + }, + { + "epoch": 0.4739824832560536, + "grad_norm": 0.42593663924882696, + "learning_rate": 0.00012624082892007064, + "loss": 0.3334, + "step": 2300 + }, + { + "epoch": 0.4750128799587841, + "grad_norm": 0.33769395333885094, + "learning_rate": 0.00012589348832730882, + "loss": 0.2998, + "step": 2305 + }, + { + "epoch": 0.4760432766615147, + "grad_norm": 0.303923325646938, + "learning_rate": 0.00012554581256667296, + "loss": 0.3035, + "step": 2310 + }, + { + "epoch": 0.4770736733642452, + "grad_norm": 0.2858575579612513, + "learning_rate": 0.00012519780613851254, + "loss": 0.2485, + "step": 2315 + }, + { + "epoch": 0.4781040700669758, + "grad_norm": 0.35956368876613565, + "learning_rate": 0.00012484947354745714, + "loss": 0.3149, + "step": 2320 + }, + { + "epoch": 0.47913446676970634, + "grad_norm": 0.37980409546574345, + "learning_rate": 0.0001245008193023583, + "loss": 0.3311, + "step": 2325 + }, + { + "epoch": 0.48016486347243686, + "grad_norm": 0.33696103332976024, + "learning_rate": 0.00012415184791623101, + "loss": 0.2588, + "step": 2330 + }, + { + "epoch": 0.48119526017516745, + "grad_norm": 0.29566476069771774, + "learning_rate": 0.00012380256390619548, + "loss": 0.3085, + "step": 2335 + }, + { + "epoch": 0.482225656877898, + "grad_norm": 0.33290532311027576, + "learning_rate": 0.00012345297179341844, + "loss": 0.2796, + "step": 2340 + }, + { + "epoch": 0.48325605358062856, + "grad_norm": 0.3250440560893066, + "learning_rate": 0.00012310307610305477, + "loss": 0.3314, + "step": 2345 + }, + { + "epoch": 0.4842864502833591, + "grad_norm": 0.4026792239398559, + "learning_rate": 0.00012275288136418889, + "loss": 0.3271, + "step": 2350 + }, + { + "epoch": 0.4853168469860896, + "grad_norm": 0.34006109198287215, + "learning_rate": 0.0001224023921097762, + "loss": 0.2546, + "step": 2355 + }, + { + "epoch": 0.4863472436888202, + "grad_norm": 0.34339764904284936, + "learning_rate": 0.0001220516128765842, + "loss": 0.2988, + "step": 2360 + }, + { + "epoch": 0.48737764039155074, + "grad_norm": 0.33688366804796616, + "learning_rate": 0.00012170054820513401, + "loss": 0.2532, + "step": 2365 + }, + { + "epoch": 0.4884080370942813, + "grad_norm": 0.38985512420501045, + "learning_rate": 0.00012134920263964147, + "loss": 0.3374, + "step": 2370 + }, + { + "epoch": 0.48943843379701185, + "grad_norm": 0.35409668644772796, + "learning_rate": 0.00012099758072795832, + "loss": 0.2997, + "step": 2375 + }, + { + "epoch": 0.4904688304997424, + "grad_norm": 0.30022085659766934, + "learning_rate": 0.00012064568702151335, + "loss": 0.2606, + "step": 2380 + }, + { + "epoch": 0.49149922720247297, + "grad_norm": 0.3461846293405654, + "learning_rate": 0.0001202935260752535, + "loss": 0.3363, + "step": 2385 + }, + { + "epoch": 0.4925296239052035, + "grad_norm": 0.28834357398525434, + "learning_rate": 0.00011994110244758496, + "loss": 0.2773, + "step": 2390 + }, + { + "epoch": 0.4935600206079341, + "grad_norm": 0.3332635903203983, + "learning_rate": 0.00011958842070031395, + "loss": 0.3057, + "step": 2395 + }, + { + "epoch": 0.4945904173106646, + "grad_norm": 0.35364367272743974, + "learning_rate": 0.0001192354853985879, + "loss": 0.3095, + "step": 2400 + }, + { + "epoch": 0.4945904173106646, + "eval_loss": 0.28022027015686035, + "eval_runtime": 2878.1786, + "eval_samples_per_second": 2.78, + "eval_steps_per_second": 0.347, + "step": 2400 + }, + { + "epoch": 0.49562081401339514, + "grad_norm": 0.3454343108599935, + "learning_rate": 0.00011888230111083627, + "loss": 0.2574, + "step": 2405 + }, + { + "epoch": 0.4966512107161257, + "grad_norm": 0.4247051943255858, + "learning_rate": 0.00011852887240871145, + "loss": 0.3067, + "step": 2410 + }, + { + "epoch": 0.49768160741885625, + "grad_norm": 0.36774522098501977, + "learning_rate": 0.00011817520386702947, + "loss": 0.3384, + "step": 2415 + }, + { + "epoch": 0.49871200412158684, + "grad_norm": 0.3272725276077021, + "learning_rate": 0.00011782130006371092, + "loss": 0.3136, + "step": 2420 + }, + { + "epoch": 0.49974240082431737, + "grad_norm": 0.46257832400717397, + "learning_rate": 0.00011746716557972167, + "loss": 0.3025, + "step": 2425 + }, + { + "epoch": 0.500772797527048, + "grad_norm": 0.28584536205539957, + "learning_rate": 0.00011711280499901347, + "loss": 0.2736, + "step": 2430 + }, + { + "epoch": 0.5018031942297785, + "grad_norm": 0.36350270232137516, + "learning_rate": 0.00011675822290846474, + "loss": 0.3114, + "step": 2435 + }, + { + "epoch": 0.502833590932509, + "grad_norm": 0.3515382888246873, + "learning_rate": 0.00011640342389782114, + "loss": 0.2756, + "step": 2440 + }, + { + "epoch": 0.5038639876352395, + "grad_norm": 0.333990361886686, + "learning_rate": 0.00011604841255963616, + "loss": 0.2982, + "step": 2445 + }, + { + "epoch": 0.5048943843379701, + "grad_norm": 0.3774121482190172, + "learning_rate": 0.00011569319348921168, + "loss": 0.2888, + "step": 2450 + }, + { + "epoch": 0.5059247810407007, + "grad_norm": 0.3449765743536067, + "learning_rate": 0.00011533777128453844, + "loss": 0.239, + "step": 2455 + }, + { + "epoch": 0.5069551777434312, + "grad_norm": 0.39508888175442797, + "learning_rate": 0.00011498215054623664, + "loss": 0.3099, + "step": 2460 + }, + { + "epoch": 0.5079855744461618, + "grad_norm": 0.29997399648896544, + "learning_rate": 0.00011462633587749629, + "loss": 0.2553, + "step": 2465 + }, + { + "epoch": 0.5090159711488923, + "grad_norm": 0.3946262619553709, + "learning_rate": 0.00011427033188401768, + "loss": 0.3003, + "step": 2470 + }, + { + "epoch": 0.5100463678516228, + "grad_norm": 0.3979213104058746, + "learning_rate": 0.00011391414317395167, + "loss": 0.3272, + "step": 2475 + }, + { + "epoch": 0.5110767645543535, + "grad_norm": 0.36162636563510103, + "learning_rate": 0.00011355777435784016, + "loss": 0.2696, + "step": 2480 + }, + { + "epoch": 0.512107161257084, + "grad_norm": 0.3436202202392731, + "learning_rate": 0.0001132012300485564, + "loss": 0.3231, + "step": 2485 + }, + { + "epoch": 0.5131375579598145, + "grad_norm": 0.358384231565426, + "learning_rate": 0.00011284451486124514, + "loss": 0.2691, + "step": 2490 + }, + { + "epoch": 0.5141679546625451, + "grad_norm": 0.32204877501229845, + "learning_rate": 0.00011248763341326307, + "loss": 0.3002, + "step": 2495 + }, + { + "epoch": 0.5151983513652756, + "grad_norm": 0.30701193700017093, + "learning_rate": 0.00011213059032411897, + "loss": 0.3097, + "step": 2500 + }, + { + "epoch": 0.5162287480680062, + "grad_norm": 0.3394950970385267, + "learning_rate": 0.00011177339021541387, + "loss": 0.2587, + "step": 2505 + }, + { + "epoch": 0.5172591447707368, + "grad_norm": 0.3366588908099883, + "learning_rate": 0.00011141603771078133, + "loss": 0.3194, + "step": 2510 + }, + { + "epoch": 0.5182895414734673, + "grad_norm": 0.3981768141047026, + "learning_rate": 0.00011105853743582751, + "loss": 0.2968, + "step": 2515 + }, + { + "epoch": 0.5193199381761978, + "grad_norm": 0.32297584447574446, + "learning_rate": 0.00011070089401807129, + "loss": 0.2836, + "step": 2520 + }, + { + "epoch": 0.5203503348789283, + "grad_norm": 0.357520694115771, + "learning_rate": 0.0001103431120868845, + "loss": 0.2945, + "step": 2525 + }, + { + "epoch": 0.521380731581659, + "grad_norm": 0.3150716632390592, + "learning_rate": 0.00010998519627343182, + "loss": 0.2665, + "step": 2530 + }, + { + "epoch": 0.5224111282843895, + "grad_norm": 0.3979335237201818, + "learning_rate": 0.00010962715121061095, + "loss": 0.3219, + "step": 2535 + }, + { + "epoch": 0.52344152498712, + "grad_norm": 0.2755501921687075, + "learning_rate": 0.00010926898153299259, + "loss": 0.2851, + "step": 2540 + }, + { + "epoch": 0.5244719216898506, + "grad_norm": 0.3265322073550003, + "learning_rate": 0.00010891069187676051, + "loss": 0.2908, + "step": 2545 + }, + { + "epoch": 0.5255023183925811, + "grad_norm": 0.2997864769906755, + "learning_rate": 0.00010855228687965138, + "loss": 0.2829, + "step": 2550 + }, + { + "epoch": 0.5265327150953117, + "grad_norm": 0.2857206537360907, + "learning_rate": 0.000108193771180895, + "loss": 0.2539, + "step": 2555 + }, + { + "epoch": 0.5275631117980423, + "grad_norm": 0.36091935273374165, + "learning_rate": 0.00010783514942115398, + "loss": 0.3289, + "step": 2560 + }, + { + "epoch": 0.5285935085007728, + "grad_norm": 0.3436525132226202, + "learning_rate": 0.0001074764262424639, + "loss": 0.2741, + "step": 2565 + }, + { + "epoch": 0.5296239052035033, + "grad_norm": 0.27339341095603925, + "learning_rate": 0.00010711760628817304, + "loss": 0.2826, + "step": 2570 + }, + { + "epoch": 0.5306543019062339, + "grad_norm": 0.40588239990731534, + "learning_rate": 0.0001067586942028824, + "loss": 0.3024, + "step": 2575 + }, + { + "epoch": 0.5316846986089645, + "grad_norm": 0.4102653825925909, + "learning_rate": 0.00010639969463238553, + "loss": 0.2714, + "step": 2580 + }, + { + "epoch": 0.532715095311695, + "grad_norm": 0.3099370106047446, + "learning_rate": 0.00010604061222360828, + "loss": 0.2808, + "step": 2585 + }, + { + "epoch": 0.5337454920144256, + "grad_norm": 0.2822749846247971, + "learning_rate": 0.00010568145162454896, + "loss": 0.2801, + "step": 2590 + }, + { + "epoch": 0.5347758887171561, + "grad_norm": 0.3490990673692889, + "learning_rate": 0.00010532221748421787, + "loss": 0.2945, + "step": 2595 + }, + { + "epoch": 0.5358062854198866, + "grad_norm": 0.4053913540554717, + "learning_rate": 0.00010496291445257725, + "loss": 0.3532, + "step": 2600 + }, + { + "epoch": 0.5358062854198866, + "eval_loss": 0.27754950523376465, + "eval_runtime": 2877.7009, + "eval_samples_per_second": 2.78, + "eval_steps_per_second": 0.347, + "step": 2600 + }, + { + "epoch": 0.5368366821226173, + "grad_norm": 0.33039659690717843, + "learning_rate": 0.00010460354718048109, + "loss": 0.2886, + "step": 2605 + }, + { + "epoch": 0.5378670788253478, + "grad_norm": 0.3749436861438668, + "learning_rate": 0.00010424412031961484, + "loss": 0.3254, + "step": 2610 + }, + { + "epoch": 0.5388974755280783, + "grad_norm": 0.21141370926954733, + "learning_rate": 0.0001038846385224354, + "loss": 0.2522, + "step": 2615 + }, + { + "epoch": 0.5399278722308088, + "grad_norm": 0.3834437318572923, + "learning_rate": 0.00010352510644211074, + "loss": 0.2977, + "step": 2620 + }, + { + "epoch": 0.5409582689335394, + "grad_norm": 0.3725001960205959, + "learning_rate": 0.0001031655287324596, + "loss": 0.3162, + "step": 2625 + }, + { + "epoch": 0.54198866563627, + "grad_norm": 0.3203518352250769, + "learning_rate": 0.00010280591004789144, + "loss": 0.2699, + "step": 2630 + }, + { + "epoch": 0.5430190623390005, + "grad_norm": 0.29941898288284835, + "learning_rate": 0.00010244625504334609, + "loss": 0.2931, + "step": 2635 + }, + { + "epoch": 0.5440494590417311, + "grad_norm": 0.30304378110760377, + "learning_rate": 0.00010208656837423351, + "loss": 0.2689, + "step": 2640 + }, + { + "epoch": 0.5450798557444616, + "grad_norm": 0.3348107931428455, + "learning_rate": 0.00010172685469637351, + "loss": 0.2764, + "step": 2645 + }, + { + "epoch": 0.5461102524471921, + "grad_norm": 0.42244879169591937, + "learning_rate": 0.00010136711866593551, + "loss": 0.3216, + "step": 2650 + }, + { + "epoch": 0.5471406491499228, + "grad_norm": 0.3226252753934826, + "learning_rate": 0.00010100736493937828, + "loss": 0.29, + "step": 2655 + }, + { + "epoch": 0.5481710458526533, + "grad_norm": 0.4102152257267717, + "learning_rate": 0.00010064759817338965, + "loss": 0.3126, + "step": 2660 + }, + { + "epoch": 0.5492014425553838, + "grad_norm": 0.35520661791811914, + "learning_rate": 0.00010028782302482617, + "loss": 0.2796, + "step": 2665 + }, + { + "epoch": 0.5502318392581144, + "grad_norm": 0.34575679827777267, + "learning_rate": 9.992804415065305e-05, + "loss": 0.305, + "step": 2670 + }, + { + "epoch": 0.5512622359608449, + "grad_norm": 0.39375752670785547, + "learning_rate": 9.956826620788352e-05, + "loss": 0.3136, + "step": 2675 + }, + { + "epoch": 0.5522926326635754, + "grad_norm": 0.3612100553953111, + "learning_rate": 9.920849385351889e-05, + "loss": 0.2581, + "step": 2680 + }, + { + "epoch": 0.5533230293663061, + "grad_norm": 0.2712636740543298, + "learning_rate": 9.884873174448811e-05, + "loss": 0.32, + "step": 2685 + }, + { + "epoch": 0.5543534260690366, + "grad_norm": 0.3709056434567233, + "learning_rate": 9.848898453758752e-05, + "loss": 0.2832, + "step": 2690 + }, + { + "epoch": 0.5553838227717671, + "grad_norm": 0.31865604636217015, + "learning_rate": 9.812925688942054e-05, + "loss": 0.2877, + "step": 2695 + }, + { + "epoch": 0.5564142194744977, + "grad_norm": 0.38997754979522803, + "learning_rate": 9.776955345633739e-05, + "loss": 0.3212, + "step": 2700 + }, + { + "epoch": 0.5574446161772282, + "grad_norm": 0.33716767476645676, + "learning_rate": 9.740987889437492e-05, + "loss": 0.2427, + "step": 2705 + }, + { + "epoch": 0.5584750128799588, + "grad_norm": 0.33728797221096607, + "learning_rate": 9.705023785919623e-05, + "loss": 0.2934, + "step": 2710 + }, + { + "epoch": 0.5595054095826894, + "grad_norm": 0.29970002424929326, + "learning_rate": 9.669063500603049e-05, + "loss": 0.2824, + "step": 2715 + }, + { + "epoch": 0.5605358062854199, + "grad_norm": 0.4423023226513398, + "learning_rate": 9.633107498961251e-05, + "loss": 0.2951, + "step": 2720 + }, + { + "epoch": 0.5615662029881504, + "grad_norm": 0.3832845615289885, + "learning_rate": 9.597156246412277e-05, + "loss": 0.3217, + "step": 2725 + }, + { + "epoch": 0.5625965996908809, + "grad_norm": 0.3243338286963795, + "learning_rate": 9.561210208312694e-05, + "loss": 0.2701, + "step": 2730 + }, + { + "epoch": 0.5636269963936116, + "grad_norm": 0.3962929861495999, + "learning_rate": 9.525269849951578e-05, + "loss": 0.2894, + "step": 2735 + }, + { + "epoch": 0.5646573930963421, + "grad_norm": 0.2617163116841798, + "learning_rate": 9.489335636544476e-05, + "loss": 0.2122, + "step": 2740 + }, + { + "epoch": 0.5656877897990726, + "grad_norm": 0.3646251465358233, + "learning_rate": 9.453408033227406e-05, + "loss": 0.3145, + "step": 2745 + }, + { + "epoch": 0.5667181865018032, + "grad_norm": 0.3319778563330451, + "learning_rate": 9.417487505050816e-05, + "loss": 0.2919, + "step": 2750 + }, + { + "epoch": 0.5677485832045337, + "grad_norm": 0.372171978724275, + "learning_rate": 9.38157451697358e-05, + "loss": 0.2757, + "step": 2755 + }, + { + "epoch": 0.5687789799072643, + "grad_norm": 0.3752921419845626, + "learning_rate": 9.345669533856961e-05, + "loss": 0.3013, + "step": 2760 + }, + { + "epoch": 0.5698093766099949, + "grad_norm": 0.3043674671804571, + "learning_rate": 9.309773020458616e-05, + "loss": 0.2543, + "step": 2765 + }, + { + "epoch": 0.5708397733127254, + "grad_norm": 0.2550633494163026, + "learning_rate": 9.273885441426562e-05, + "loss": 0.2497, + "step": 2770 + }, + { + "epoch": 0.5718701700154559, + "grad_norm": 0.36955800478440715, + "learning_rate": 9.238007261293176e-05, + "loss": 0.3131, + "step": 2775 + }, + { + "epoch": 0.5729005667181865, + "grad_norm": 0.35332539238207206, + "learning_rate": 9.202138944469168e-05, + "loss": 0.2658, + "step": 2780 + }, + { + "epoch": 0.5739309634209171, + "grad_norm": 0.3264554515234995, + "learning_rate": 9.16628095523758e-05, + "loss": 0.2897, + "step": 2785 + }, + { + "epoch": 0.5749613601236476, + "grad_norm": 0.29610170158655286, + "learning_rate": 9.130433757747772e-05, + "loss": 0.283, + "step": 2790 + }, + { + "epoch": 0.5759917568263782, + "grad_norm": 0.33278027754490686, + "learning_rate": 9.094597816009409e-05, + "loss": 0.2842, + "step": 2795 + }, + { + "epoch": 0.5770221535291087, + "grad_norm": 0.4528453579289396, + "learning_rate": 9.058773593886469e-05, + "loss": 0.301, + "step": 2800 + }, + { + "epoch": 0.5770221535291087, + "eval_loss": 0.27565160393714905, + "eval_runtime": 2878.5613, + "eval_samples_per_second": 2.779, + "eval_steps_per_second": 0.347, + "step": 2800 + }, + { + "epoch": 0.5780525502318392, + "grad_norm": 0.35089860682025525, + "learning_rate": 9.022961555091226e-05, + "loss": 0.2567, + "step": 2805 + }, + { + "epoch": 0.5790829469345699, + "grad_norm": 0.33263682642433734, + "learning_rate": 8.987162163178256e-05, + "loss": 0.2926, + "step": 2810 + }, + { + "epoch": 0.5801133436373004, + "grad_norm": 0.3451318549909309, + "learning_rate": 8.951375881538421e-05, + "loss": 0.2747, + "step": 2815 + }, + { + "epoch": 0.5811437403400309, + "grad_norm": 0.32240079639742575, + "learning_rate": 8.915603173392895e-05, + "loss": 0.2879, + "step": 2820 + }, + { + "epoch": 0.5821741370427614, + "grad_norm": 0.399328888829942, + "learning_rate": 8.87984450178715e-05, + "loss": 0.2948, + "step": 2825 + }, + { + "epoch": 0.583204533745492, + "grad_norm": 0.36055298325517393, + "learning_rate": 8.84410032958497e-05, + "loss": 0.2633, + "step": 2830 + }, + { + "epoch": 0.5842349304482226, + "grad_norm": 0.3920621364225242, + "learning_rate": 8.808371119462452e-05, + "loss": 0.3164, + "step": 2835 + }, + { + "epoch": 0.5852653271509531, + "grad_norm": 0.34432478887516577, + "learning_rate": 8.772657333902027e-05, + "loss": 0.2683, + "step": 2840 + }, + { + "epoch": 0.5862957238536837, + "grad_norm": 0.3158164048351637, + "learning_rate": 8.736959435186466e-05, + "loss": 0.3049, + "step": 2845 + }, + { + "epoch": 0.5873261205564142, + "grad_norm": 0.3999813171101764, + "learning_rate": 8.701277885392909e-05, + "loss": 0.325, + "step": 2850 + }, + { + "epoch": 0.5883565172591447, + "grad_norm": 0.32600872728866054, + "learning_rate": 8.665613146386854e-05, + "loss": 0.2686, + "step": 2855 + }, + { + "epoch": 0.5893869139618754, + "grad_norm": 0.352195036886543, + "learning_rate": 8.629965679816217e-05, + "loss": 0.3215, + "step": 2860 + }, + { + "epoch": 0.5904173106646059, + "grad_norm": 0.31676293306405257, + "learning_rate": 8.594335947105328e-05, + "loss": 0.2762, + "step": 2865 + }, + { + "epoch": 0.5914477073673364, + "grad_norm": 0.3596596959340484, + "learning_rate": 8.558724409448974e-05, + "loss": 0.2708, + "step": 2870 + }, + { + "epoch": 0.592478104070067, + "grad_norm": 0.36548567729401993, + "learning_rate": 8.523131527806423e-05, + "loss": 0.3089, + "step": 2875 + }, + { + "epoch": 0.5935085007727975, + "grad_norm": 0.3315130832888099, + "learning_rate": 8.48755776289545e-05, + "loss": 0.2527, + "step": 2880 + }, + { + "epoch": 0.5945388974755281, + "grad_norm": 0.4241057519904373, + "learning_rate": 8.452003575186394e-05, + "loss": 0.3049, + "step": 2885 + }, + { + "epoch": 0.5955692941782587, + "grad_norm": 0.30694798340386065, + "learning_rate": 8.416469424896167e-05, + "loss": 0.2523, + "step": 2890 + }, + { + "epoch": 0.5965996908809892, + "grad_norm": 0.28092729723258986, + "learning_rate": 8.380955771982332e-05, + "loss": 0.3147, + "step": 2895 + }, + { + "epoch": 0.5976300875837197, + "grad_norm": 0.3701838757603822, + "learning_rate": 8.345463076137125e-05, + "loss": 0.2936, + "step": 2900 + }, + { + "epoch": 0.5986604842864502, + "grad_norm": 0.33596931371013156, + "learning_rate": 8.309991796781511e-05, + "loss": 0.2313, + "step": 2905 + }, + { + "epoch": 0.5996908809891809, + "grad_norm": 0.358626386553519, + "learning_rate": 8.274542393059235e-05, + "loss": 0.2965, + "step": 2910 + }, + { + "epoch": 0.6007212776919114, + "grad_norm": 0.34139489578529375, + "learning_rate": 8.239115323830889e-05, + "loss": 0.2526, + "step": 2915 + }, + { + "epoch": 0.6017516743946419, + "grad_norm": 0.3258606452817875, + "learning_rate": 8.203711047667958e-05, + "loss": 0.2881, + "step": 2920 + }, + { + "epoch": 0.6027820710973725, + "grad_norm": 0.3847954972920331, + "learning_rate": 8.1683300228469e-05, + "loss": 0.337, + "step": 2925 + }, + { + "epoch": 0.603812467800103, + "grad_norm": 0.29131611380441297, + "learning_rate": 8.132972707343192e-05, + "loss": 0.2275, + "step": 2930 + }, + { + "epoch": 0.6048428645028336, + "grad_norm": 0.27037063359019814, + "learning_rate": 8.097639558825427e-05, + "loss": 0.2976, + "step": 2935 + }, + { + "epoch": 0.6058732612055642, + "grad_norm": 0.32696985705968035, + "learning_rate": 8.062331034649376e-05, + "loss": 0.2878, + "step": 2940 + }, + { + "epoch": 0.6069036579082947, + "grad_norm": 0.44146891732351323, + "learning_rate": 8.027047591852069e-05, + "loss": 0.2888, + "step": 2945 + }, + { + "epoch": 0.6079340546110252, + "grad_norm": 0.39221179639528275, + "learning_rate": 7.991789687145873e-05, + "loss": 0.3014, + "step": 2950 + }, + { + "epoch": 0.6089644513137558, + "grad_norm": 0.3510472981653684, + "learning_rate": 7.956557776912596e-05, + "loss": 0.2546, + "step": 2955 + }, + { + "epoch": 0.6099948480164864, + "grad_norm": 0.38058040305787494, + "learning_rate": 7.921352317197574e-05, + "loss": 0.2949, + "step": 2960 + }, + { + "epoch": 0.6110252447192169, + "grad_norm": 0.35902085682972984, + "learning_rate": 7.886173763703757e-05, + "loss": 0.2827, + "step": 2965 + }, + { + "epoch": 0.6120556414219475, + "grad_norm": 0.3927327124338447, + "learning_rate": 7.851022571785819e-05, + "loss": 0.3116, + "step": 2970 + }, + { + "epoch": 0.613086038124678, + "grad_norm": 0.3935039943632793, + "learning_rate": 7.815899196444267e-05, + "loss": 0.3131, + "step": 2975 + }, + { + "epoch": 0.6141164348274085, + "grad_norm": 0.3561243208088519, + "learning_rate": 7.780804092319547e-05, + "loss": 0.2535, + "step": 2980 + }, + { + "epoch": 0.615146831530139, + "grad_norm": 0.3479054336232948, + "learning_rate": 7.745737713686152e-05, + "loss": 0.2948, + "step": 2985 + }, + { + "epoch": 0.6161772282328697, + "grad_norm": 0.3898086480384796, + "learning_rate": 7.710700514446762e-05, + "loss": 0.305, + "step": 2990 + }, + { + "epoch": 0.6172076249356002, + "grad_norm": 0.2782916809425304, + "learning_rate": 7.675692948126345e-05, + "loss": 0.3028, + "step": 2995 + }, + { + "epoch": 0.6182380216383307, + "grad_norm": 0.3827931144256995, + "learning_rate": 7.640715467866307e-05, + "loss": 0.3204, + "step": 3000 + }, + { + "epoch": 0.6182380216383307, + "eval_loss": 0.2711670994758606, + "eval_runtime": 2879.0056, + "eval_samples_per_second": 2.779, + "eval_steps_per_second": 0.347, + "step": 3000 + }, + { + "epoch": 0.6192684183410613, + "grad_norm": 0.3311400596855034, + "learning_rate": 7.605768526418605e-05, + "loss": 0.2534, + "step": 3005 + }, + { + "epoch": 0.6202988150437918, + "grad_norm": 0.4024348674942288, + "learning_rate": 7.57085257613991e-05, + "loss": 0.2999, + "step": 3010 + }, + { + "epoch": 0.6213292117465224, + "grad_norm": 0.30615727991401664, + "learning_rate": 7.535968068985737e-05, + "loss": 0.277, + "step": 3015 + }, + { + "epoch": 0.622359608449253, + "grad_norm": 0.3434621463110342, + "learning_rate": 7.501115456504595e-05, + "loss": 0.2832, + "step": 3020 + }, + { + "epoch": 0.6233900051519835, + "grad_norm": 0.34574651517676536, + "learning_rate": 7.466295189832148e-05, + "loss": 0.2949, + "step": 3025 + }, + { + "epoch": 0.624420401854714, + "grad_norm": 0.3352676719646584, + "learning_rate": 7.431507719685371e-05, + "loss": 0.2449, + "step": 3030 + }, + { + "epoch": 0.6254507985574446, + "grad_norm": 0.4336028485216119, + "learning_rate": 7.396753496356718e-05, + "loss": 0.3216, + "step": 3035 + }, + { + "epoch": 0.6264811952601752, + "grad_norm": 0.33779434998514446, + "learning_rate": 7.362032969708297e-05, + "loss": 0.2632, + "step": 3040 + }, + { + "epoch": 0.6275115919629057, + "grad_norm": 0.3337649698653029, + "learning_rate": 7.327346589166035e-05, + "loss": 0.2749, + "step": 3045 + }, + { + "epoch": 0.6285419886656363, + "grad_norm": 0.4278703603131763, + "learning_rate": 7.292694803713871e-05, + "loss": 0.3101, + "step": 3050 + }, + { + "epoch": 0.6295723853683668, + "grad_norm": 0.33098852192128414, + "learning_rate": 7.258078061887947e-05, + "loss": 0.2577, + "step": 3055 + }, + { + "epoch": 0.6306027820710973, + "grad_norm": 0.4010491539631291, + "learning_rate": 7.223496811770796e-05, + "loss": 0.3134, + "step": 3060 + }, + { + "epoch": 0.631633178773828, + "grad_norm": 0.46500732529362854, + "learning_rate": 7.188951500985533e-05, + "loss": 0.2926, + "step": 3065 + }, + { + "epoch": 0.6326635754765585, + "grad_norm": 0.3282290070792828, + "learning_rate": 7.154442576690083e-05, + "loss": 0.3185, + "step": 3070 + }, + { + "epoch": 0.633693972179289, + "grad_norm": 0.43208806306822944, + "learning_rate": 7.119970485571375e-05, + "loss": 0.3238, + "step": 3075 + }, + { + "epoch": 0.6347243688820196, + "grad_norm": 0.31953297640440387, + "learning_rate": 7.08553567383956e-05, + "loss": 0.2961, + "step": 3080 + }, + { + "epoch": 0.6357547655847501, + "grad_norm": 0.37608984484932495, + "learning_rate": 7.051138587222255e-05, + "loss": 0.3062, + "step": 3085 + }, + { + "epoch": 0.6367851622874807, + "grad_norm": 0.30155315001331834, + "learning_rate": 7.016779670958746e-05, + "loss": 0.2951, + "step": 3090 + }, + { + "epoch": 0.6378155589902113, + "grad_norm": 0.3072620029346922, + "learning_rate": 6.982459369794247e-05, + "loss": 0.2783, + "step": 3095 + }, + { + "epoch": 0.6388459556929418, + "grad_norm": 0.30677479360829335, + "learning_rate": 6.948178127974127e-05, + "loss": 0.275, + "step": 3100 + }, + { + "epoch": 0.6398763523956723, + "grad_norm": 0.31058594621709634, + "learning_rate": 6.913936389238174e-05, + "loss": 0.2751, + "step": 3105 + }, + { + "epoch": 0.6409067490984028, + "grad_norm": 0.38892915494581226, + "learning_rate": 6.879734596814839e-05, + "loss": 0.3148, + "step": 3110 + }, + { + "epoch": 0.6419371458011335, + "grad_norm": 0.3166973015889613, + "learning_rate": 6.845573193415508e-05, + "loss": 0.308, + "step": 3115 + }, + { + "epoch": 0.642967542503864, + "grad_norm": 0.3572971206745672, + "learning_rate": 6.811452621228766e-05, + "loss": 0.3069, + "step": 3120 + }, + { + "epoch": 0.6439979392065945, + "grad_norm": 0.3735142202617577, + "learning_rate": 6.777373321914671e-05, + "loss": 0.3258, + "step": 3125 + }, + { + "epoch": 0.6450283359093251, + "grad_norm": 0.34484584719959194, + "learning_rate": 6.743335736599045e-05, + "loss": 0.254, + "step": 3130 + }, + { + "epoch": 0.6460587326120556, + "grad_norm": 0.40642185135429687, + "learning_rate": 6.709340305867762e-05, + "loss": 0.3387, + "step": 3135 + }, + { + "epoch": 0.6470891293147862, + "grad_norm": 0.34408235252291053, + "learning_rate": 6.675387469761033e-05, + "loss": 0.2852, + "step": 3140 + }, + { + "epoch": 0.6481195260175168, + "grad_norm": 0.2760204462481522, + "learning_rate": 6.641477667767738e-05, + "loss": 0.2979, + "step": 3145 + }, + { + "epoch": 0.6491499227202473, + "grad_norm": 0.3521257726089709, + "learning_rate": 6.607611338819697e-05, + "loss": 0.2972, + "step": 3150 + }, + { + "epoch": 0.6501803194229778, + "grad_norm": 0.3570108107553612, + "learning_rate": 6.573788921286028e-05, + "loss": 0.2556, + "step": 3155 + }, + { + "epoch": 0.6512107161257084, + "grad_norm": 0.29080835096475477, + "learning_rate": 6.540010852967447e-05, + "loss": 0.2909, + "step": 3160 + }, + { + "epoch": 0.652241112828439, + "grad_norm": 0.35414069594841313, + "learning_rate": 6.506277571090613e-05, + "loss": 0.2618, + "step": 3165 + }, + { + "epoch": 0.6532715095311695, + "grad_norm": 0.34570363781855507, + "learning_rate": 6.47258951230246e-05, + "loss": 0.2927, + "step": 3170 + }, + { + "epoch": 0.6543019062339, + "grad_norm": 0.3516727468458287, + "learning_rate": 6.438947112664555e-05, + "loss": 0.2738, + "step": 3175 + }, + { + "epoch": 0.6553323029366306, + "grad_norm": 0.2504400994918305, + "learning_rate": 6.405350807647444e-05, + "loss": 0.2446, + "step": 3180 + }, + { + "epoch": 0.6563626996393611, + "grad_norm": 0.3041195746043791, + "learning_rate": 6.371801032125026e-05, + "loss": 0.282, + "step": 3185 + }, + { + "epoch": 0.6573930963420918, + "grad_norm": 0.3395467056509759, + "learning_rate": 6.338298220368912e-05, + "loss": 0.2616, + "step": 3190 + }, + { + "epoch": 0.6584234930448223, + "grad_norm": 0.36750181241259344, + "learning_rate": 6.304842806042812e-05, + "loss": 0.2649, + "step": 3195 + }, + { + "epoch": 0.6594538897475528, + "grad_norm": 0.41382005017445883, + "learning_rate": 6.271435222196916e-05, + "loss": 0.3158, + "step": 3200 + }, + { + "epoch": 0.6594538897475528, + "eval_loss": 0.26873406767845154, + "eval_runtime": 2877.9738, + "eval_samples_per_second": 2.78, + "eval_steps_per_second": 0.347, + "step": 3200 + }, + { + "epoch": 0.6604842864502833, + "grad_norm": 0.3166096758409699, + "learning_rate": 6.238075901262293e-05, + "loss": 0.2522, + "step": 3205 + }, + { + "epoch": 0.6615146831530139, + "grad_norm": 0.3330010541870663, + "learning_rate": 6.204765275045298e-05, + "loss": 0.2961, + "step": 3210 + }, + { + "epoch": 0.6625450798557445, + "grad_norm": 0.26670607401675234, + "learning_rate": 6.171503774721966e-05, + "loss": 0.2548, + "step": 3215 + }, + { + "epoch": 0.663575476558475, + "grad_norm": 0.3302326211327972, + "learning_rate": 6.13829183083245e-05, + "loss": 0.314, + "step": 3220 + }, + { + "epoch": 0.6646058732612056, + "grad_norm": 0.3353496397076012, + "learning_rate": 6.105129873275435e-05, + "loss": 0.3014, + "step": 3225 + }, + { + "epoch": 0.6656362699639361, + "grad_norm": 0.31520904347140727, + "learning_rate": 6.072018331302577e-05, + "loss": 0.2333, + "step": 3230 + }, + { + "epoch": 0.6666666666666666, + "grad_norm": 0.35782950763156535, + "learning_rate": 6.038957633512957e-05, + "loss": 0.2881, + "step": 3235 + }, + { + "epoch": 0.6676970633693973, + "grad_norm": 0.2891661798517709, + "learning_rate": 6.005948207847516e-05, + "loss": 0.2499, + "step": 3240 + }, + { + "epoch": 0.6687274600721278, + "grad_norm": 0.29473346542335793, + "learning_rate": 5.97299048158352e-05, + "loss": 0.2756, + "step": 3245 + }, + { + "epoch": 0.6697578567748583, + "grad_norm": 0.4272344871197883, + "learning_rate": 5.940084881329042e-05, + "loss": 0.3062, + "step": 3250 + }, + { + "epoch": 0.6707882534775889, + "grad_norm": 0.4206884593213127, + "learning_rate": 5.907231833017424e-05, + "loss": 0.2517, + "step": 3255 + }, + { + "epoch": 0.6718186501803194, + "grad_norm": 0.3441386348303789, + "learning_rate": 5.8744317619017755e-05, + "loss": 0.3093, + "step": 3260 + }, + { + "epoch": 0.67284904688305, + "grad_norm": 0.33030990891048667, + "learning_rate": 5.841685092549456e-05, + "loss": 0.285, + "step": 3265 + }, + { + "epoch": 0.6738794435857806, + "grad_norm": 0.3414165248105595, + "learning_rate": 5.8089922488365975e-05, + "loss": 0.2906, + "step": 3270 + }, + { + "epoch": 0.6749098402885111, + "grad_norm": 0.36693139617736636, + "learning_rate": 5.776353653942602e-05, + "loss": 0.3077, + "step": 3275 + }, + { + "epoch": 0.6759402369912416, + "grad_norm": 0.35952213271475214, + "learning_rate": 5.743769730344666e-05, + "loss": 0.2588, + "step": 3280 + }, + { + "epoch": 0.6769706336939721, + "grad_norm": 0.3817072974116425, + "learning_rate": 5.7112408998123256e-05, + "loss": 0.3116, + "step": 3285 + }, + { + "epoch": 0.6780010303967027, + "grad_norm": 0.35226830622590205, + "learning_rate": 5.678767583401974e-05, + "loss": 0.2631, + "step": 3290 + }, + { + "epoch": 0.6790314270994333, + "grad_norm": 0.30682229922613086, + "learning_rate": 5.646350201451438e-05, + "loss": 0.3023, + "step": 3295 + }, + { + "epoch": 0.6800618238021638, + "grad_norm": 0.3752337210236874, + "learning_rate": 5.613989173574512e-05, + "loss": 0.2816, + "step": 3300 + }, + { + "epoch": 0.6810922205048944, + "grad_norm": 0.3785279140180591, + "learning_rate": 5.5816849186555386e-05, + "loss": 0.2239, + "step": 3305 + }, + { + "epoch": 0.6821226172076249, + "grad_norm": 0.32589901787515324, + "learning_rate": 5.549437854843995e-05, + "loss": 0.2436, + "step": 3310 + }, + { + "epoch": 0.6831530139103554, + "grad_norm": 0.3268785279060166, + "learning_rate": 5.517248399549063e-05, + "loss": 0.2332, + "step": 3315 + }, + { + "epoch": 0.6841834106130861, + "grad_norm": 0.3341851818333058, + "learning_rate": 5.48511696943423e-05, + "loss": 0.293, + "step": 3320 + }, + { + "epoch": 0.6852138073158166, + "grad_norm": 0.3438105755717694, + "learning_rate": 5.4530439804119096e-05, + "loss": 0.3301, + "step": 3325 + }, + { + "epoch": 0.6862442040185471, + "grad_norm": 0.3356276432272127, + "learning_rate": 5.4210298476380484e-05, + "loss": 0.2566, + "step": 3330 + }, + { + "epoch": 0.6872746007212777, + "grad_norm": 0.27575134631458276, + "learning_rate": 5.38907498550674e-05, + "loss": 0.29, + "step": 3335 + }, + { + "epoch": 0.6883049974240082, + "grad_norm": 0.6488176265940205, + "learning_rate": 5.357179807644887e-05, + "loss": 0.264, + "step": 3340 + }, + { + "epoch": 0.6893353941267388, + "grad_norm": 0.3468629435100644, + "learning_rate": 5.3253447269068245e-05, + "loss": 0.3106, + "step": 3345 + }, + { + "epoch": 0.6903657908294694, + "grad_norm": 0.3959804063346456, + "learning_rate": 5.293570155368981e-05, + "loss": 0.2888, + "step": 3350 + }, + { + "epoch": 0.6913961875321999, + "grad_norm": 0.29830853292625725, + "learning_rate": 5.261856504324563e-05, + "loss": 0.2553, + "step": 3355 + }, + { + "epoch": 0.6924265842349304, + "grad_norm": 0.32422270047266666, + "learning_rate": 5.230204184278195e-05, + "loss": 0.2863, + "step": 3360 + }, + { + "epoch": 0.693456980937661, + "grad_norm": 0.3472569191866048, + "learning_rate": 5.198613604940649e-05, + "loss": 0.2957, + "step": 3365 + }, + { + "epoch": 0.6944873776403916, + "grad_norm": 0.3756763363166075, + "learning_rate": 5.1670851752235025e-05, + "loss": 0.305, + "step": 3370 + }, + { + "epoch": 0.6955177743431221, + "grad_norm": 0.34873701828232423, + "learning_rate": 5.135619303233867e-05, + "loss": 0.3007, + "step": 3375 + }, + { + "epoch": 0.6965481710458526, + "grad_norm": 0.3521257255678766, + "learning_rate": 5.104216396269109e-05, + "loss": 0.2589, + "step": 3380 + }, + { + "epoch": 0.6975785677485832, + "grad_norm": 0.36867262641478793, + "learning_rate": 5.072876860811553e-05, + "loss": 0.3187, + "step": 3385 + }, + { + "epoch": 0.6986089644513137, + "grad_norm": 0.3582623295688211, + "learning_rate": 5.0416011025232546e-05, + "loss": 0.2496, + "step": 3390 + }, + { + "epoch": 0.6996393611540443, + "grad_norm": 0.2938091378788691, + "learning_rate": 5.010389526240719e-05, + "loss": 0.267, + "step": 3395 + }, + { + "epoch": 0.7006697578567749, + "grad_norm": 0.44117444865964234, + "learning_rate": 4.979242535969672e-05, + "loss": 0.3032, + "step": 3400 + }, + { + "epoch": 0.7006697578567749, + "eval_loss": 0.2666991353034973, + "eval_runtime": 2879.7665, + "eval_samples_per_second": 2.778, + "eval_steps_per_second": 0.347, + "step": 3400 + }, + { + "epoch": 0.7017001545595054, + "grad_norm": 0.32564728902576306, + "learning_rate": 4.9481605348798435e-05, + "loss": 0.279, + "step": 3405 + }, + { + "epoch": 0.7027305512622359, + "grad_norm": 0.3341012868943753, + "learning_rate": 4.917143925299728e-05, + "loss": 0.2746, + "step": 3410 + }, + { + "epoch": 0.7037609479649665, + "grad_norm": 0.34149796395178733, + "learning_rate": 4.886193108711378e-05, + "loss": 0.2561, + "step": 3415 + }, + { + "epoch": 0.7047913446676971, + "grad_norm": 0.32243235346772253, + "learning_rate": 4.8553084857452426e-05, + "loss": 0.2944, + "step": 3420 + }, + { + "epoch": 0.7058217413704276, + "grad_norm": 0.44891237315770505, + "learning_rate": 4.824490456174926e-05, + "loss": 0.3099, + "step": 3425 + }, + { + "epoch": 0.7068521380731582, + "grad_norm": 0.3713432210872996, + "learning_rate": 4.7937394189120485e-05, + "loss": 0.279, + "step": 3430 + }, + { + "epoch": 0.7078825347758887, + "grad_norm": 0.32387159462298654, + "learning_rate": 4.763055772001086e-05, + "loss": 0.2954, + "step": 3435 + }, + { + "epoch": 0.7089129314786192, + "grad_norm": 0.3655049214248582, + "learning_rate": 4.732439912614195e-05, + "loss": 0.2599, + "step": 3440 + }, + { + "epoch": 0.7099433281813499, + "grad_norm": 0.3585347195553639, + "learning_rate": 4.7018922370460835e-05, + "loss": 0.2819, + "step": 3445 + }, + { + "epoch": 0.7109737248840804, + "grad_norm": 0.39037453777103204, + "learning_rate": 4.671413140708893e-05, + "loss": 0.2956, + "step": 3450 + }, + { + "epoch": 0.7120041215868109, + "grad_norm": 0.44234657360712804, + "learning_rate": 4.6410030181270546e-05, + "loss": 0.2585, + "step": 3455 + }, + { + "epoch": 0.7130345182895415, + "grad_norm": 0.4181782981247106, + "learning_rate": 4.610662262932209e-05, + "loss": 0.3028, + "step": 3460 + }, + { + "epoch": 0.714064914992272, + "grad_norm": 0.36884003440220187, + "learning_rate": 4.5803912678580906e-05, + "loss": 0.2893, + "step": 3465 + }, + { + "epoch": 0.7150953116950026, + "grad_norm": 0.3437139757005691, + "learning_rate": 4.5501904247354474e-05, + "loss": 0.2672, + "step": 3470 + }, + { + "epoch": 0.7161257083977332, + "grad_norm": 0.3692077816092001, + "learning_rate": 4.520060124486989e-05, + "loss": 0.313, + "step": 3475 + }, + { + "epoch": 0.7171561051004637, + "grad_norm": 0.3688505909089892, + "learning_rate": 4.4900007571222946e-05, + "loss": 0.2338, + "step": 3480 + }, + { + "epoch": 0.7181865018031942, + "grad_norm": 0.33900982896467213, + "learning_rate": 4.460012711732795e-05, + "loss": 0.2938, + "step": 3485 + }, + { + "epoch": 0.7192168985059247, + "grad_norm": 0.37856168681933267, + "learning_rate": 4.430096376486713e-05, + "loss": 0.2671, + "step": 3490 + }, + { + "epoch": 0.7202472952086554, + "grad_norm": 0.2878154556878028, + "learning_rate": 4.4002521386240466e-05, + "loss": 0.2982, + "step": 3495 + }, + { + "epoch": 0.7212776919113859, + "grad_norm": 0.3454781004229804, + "learning_rate": 4.3704803844515705e-05, + "loss": 0.3049, + "step": 3500 + }, + { + "epoch": 0.7223080886141164, + "grad_norm": 0.3950485263991763, + "learning_rate": 4.3407814993378095e-05, + "loss": 0.2416, + "step": 3505 + }, + { + "epoch": 0.723338485316847, + "grad_norm": 0.3441530397526022, + "learning_rate": 4.311155867708071e-05, + "loss": 0.296, + "step": 3510 + }, + { + "epoch": 0.7243688820195775, + "grad_norm": 0.3142921040902395, + "learning_rate": 4.2816038730394656e-05, + "loss": 0.2544, + "step": 3515 + }, + { + "epoch": 0.7253992787223081, + "grad_norm": 0.34653726659247286, + "learning_rate": 4.252125897855932e-05, + "loss": 0.2707, + "step": 3520 + }, + { + "epoch": 0.7264296754250387, + "grad_norm": 0.4116989451226589, + "learning_rate": 4.222722323723294e-05, + "loss": 0.3414, + "step": 3525 + }, + { + "epoch": 0.7274600721277692, + "grad_norm": 0.32774342903134085, + "learning_rate": 4.1933935312443286e-05, + "loss": 0.2357, + "step": 3530 + }, + { + "epoch": 0.7284904688304997, + "grad_norm": 0.40918991230864377, + "learning_rate": 4.164139900053824e-05, + "loss": 0.3185, + "step": 3535 + }, + { + "epoch": 0.7295208655332303, + "grad_norm": 0.3188798828984705, + "learning_rate": 4.134961808813672e-05, + "loss": 0.255, + "step": 3540 + }, + { + "epoch": 0.7305512622359609, + "grad_norm": 0.3630307170455085, + "learning_rate": 4.1058596352079805e-05, + "loss": 0.2985, + "step": 3545 + }, + { + "epoch": 0.7315816589386914, + "grad_norm": 0.3547013162494809, + "learning_rate": 4.076833755938153e-05, + "loss": 0.3032, + "step": 3550 + }, + { + "epoch": 0.732612055641422, + "grad_norm": 0.3387183224894239, + "learning_rate": 4.0478845467180506e-05, + "loss": 0.2515, + "step": 3555 + }, + { + "epoch": 0.7336424523441525, + "grad_norm": 0.36466954036357224, + "learning_rate": 4.0190123822690965e-05, + "loss": 0.2937, + "step": 3560 + }, + { + "epoch": 0.734672849046883, + "grad_norm": 0.3143229990949987, + "learning_rate": 3.990217636315441e-05, + "loss": 0.2423, + "step": 3565 + }, + { + "epoch": 0.7357032457496137, + "grad_norm": 0.335420380404653, + "learning_rate": 3.961500681579129e-05, + "loss": 0.2678, + "step": 3570 + }, + { + "epoch": 0.7367336424523442, + "grad_norm": 0.37119641635914863, + "learning_rate": 3.9328618897752566e-05, + "loss": 0.2926, + "step": 3575 + }, + { + "epoch": 0.7377640391550747, + "grad_norm": 0.36479696667762745, + "learning_rate": 3.904301631607186e-05, + "loss": 0.2714, + "step": 3580 + }, + { + "epoch": 0.7387944358578052, + "grad_norm": 0.3767501561850481, + "learning_rate": 3.875820276761717e-05, + "loss": 0.3114, + "step": 3585 + }, + { + "epoch": 0.7398248325605358, + "grad_norm": 0.31019709873450946, + "learning_rate": 3.847418193904324e-05, + "loss": 0.2397, + "step": 3590 + }, + { + "epoch": 0.7408552292632664, + "grad_norm": 0.34360053586472283, + "learning_rate": 3.8190957506743806e-05, + "loss": 0.2985, + "step": 3595 + }, + { + "epoch": 0.7418856259659969, + "grad_norm": 0.3495720979590242, + "learning_rate": 3.790853313680389e-05, + "loss": 0.2851, + "step": 3600 + }, + { + "epoch": 0.7418856259659969, + "eval_loss": 0.2644506096839905, + "eval_runtime": 2882.0514, + "eval_samples_per_second": 2.776, + "eval_steps_per_second": 0.347, + "step": 3600 + }, + { + "epoch": 0.7429160226687275, + "grad_norm": 0.39945130493412156, + "learning_rate": 3.7626912484952495e-05, + "loss": 0.2578, + "step": 3605 + }, + { + "epoch": 0.743946419371458, + "grad_norm": 0.3148594700458464, + "learning_rate": 3.734609919651523e-05, + "loss": 0.2948, + "step": 3610 + }, + { + "epoch": 0.7449768160741885, + "grad_norm": 0.3215798364969972, + "learning_rate": 3.706609690636703e-05, + "loss": 0.2627, + "step": 3615 + }, + { + "epoch": 0.7460072127769191, + "grad_norm": 0.3548798973562654, + "learning_rate": 3.6786909238885215e-05, + "loss": 0.3064, + "step": 3620 + }, + { + "epoch": 0.7470376094796497, + "grad_norm": 0.31115745685492907, + "learning_rate": 3.650853980790262e-05, + "loss": 0.2837, + "step": 3625 + }, + { + "epoch": 0.7480680061823802, + "grad_norm": 0.27553549207406003, + "learning_rate": 3.6230992216660664e-05, + "loss": 0.255, + "step": 3630 + }, + { + "epoch": 0.7490984028851108, + "grad_norm": 0.3096210760605592, + "learning_rate": 3.595427005776281e-05, + "loss": 0.282, + "step": 3635 + }, + { + "epoch": 0.7501287995878413, + "grad_norm": 0.3605700004322446, + "learning_rate": 3.5678376913128075e-05, + "loss": 0.2682, + "step": 3640 + }, + { + "epoch": 0.7511591962905718, + "grad_norm": 0.32678097806644013, + "learning_rate": 3.540331635394458e-05, + "loss": 0.2891, + "step": 3645 + }, + { + "epoch": 0.7521895929933025, + "grad_norm": 0.37435989227694727, + "learning_rate": 3.512909194062347e-05, + "loss": 0.3123, + "step": 3650 + }, + { + "epoch": 0.753219989696033, + "grad_norm": 0.3132183633883141, + "learning_rate": 3.485570722275264e-05, + "loss": 0.2516, + "step": 3655 + }, + { + "epoch": 0.7542503863987635, + "grad_norm": 0.39780383939283487, + "learning_rate": 3.458316573905087e-05, + "loss": 0.2857, + "step": 3660 + }, + { + "epoch": 0.755280783101494, + "grad_norm": 0.34139632684317156, + "learning_rate": 3.4311471017322175e-05, + "loss": 0.2345, + "step": 3665 + }, + { + "epoch": 0.7563111798042246, + "grad_norm": 0.3438982360840127, + "learning_rate": 3.4040626574409815e-05, + "loss": 0.275, + "step": 3670 + }, + { + "epoch": 0.7573415765069552, + "grad_norm": 0.34410794574144105, + "learning_rate": 3.377063591615113e-05, + "loss": 0.3175, + "step": 3675 + }, + { + "epoch": 0.7583719732096857, + "grad_norm": 0.3966945722421853, + "learning_rate": 3.350150253733186e-05, + "loss": 0.245, + "step": 3680 + }, + { + "epoch": 0.7594023699124163, + "grad_norm": 0.34416471278077815, + "learning_rate": 3.3233229921641064e-05, + "loss": 0.2788, + "step": 3685 + }, + { + "epoch": 0.7604327666151468, + "grad_norm": 0.3070507519002996, + "learning_rate": 3.296582154162604e-05, + "loss": 0.2541, + "step": 3690 + }, + { + "epoch": 0.7614631633178773, + "grad_norm": 0.3156034424176268, + "learning_rate": 3.2699280858647327e-05, + "loss": 0.3, + "step": 3695 + }, + { + "epoch": 0.762493560020608, + "grad_norm": 0.34194205180770326, + "learning_rate": 3.2433611322833845e-05, + "loss": 0.3081, + "step": 3700 + }, + { + "epoch": 0.7635239567233385, + "grad_norm": 0.33451394695168063, + "learning_rate": 3.216881637303839e-05, + "loss": 0.2352, + "step": 3705 + }, + { + "epoch": 0.764554353426069, + "grad_norm": 0.30760730599169134, + "learning_rate": 3.190489943679297e-05, + "loss": 0.3016, + "step": 3710 + }, + { + "epoch": 0.7655847501287996, + "grad_norm": 0.33011461282751464, + "learning_rate": 3.164186393026445e-05, + "loss": 0.2455, + "step": 3715 + }, + { + "epoch": 0.7666151468315301, + "grad_norm": 0.3557839605304464, + "learning_rate": 3.137971325821054e-05, + "loss": 0.3085, + "step": 3720 + }, + { + "epoch": 0.7676455435342607, + "grad_norm": 0.4005436680671154, + "learning_rate": 3.111845081393542e-05, + "loss": 0.2881, + "step": 3725 + }, + { + "epoch": 0.7686759402369913, + "grad_norm": 0.33159697370088725, + "learning_rate": 3.0858079979245965e-05, + "loss": 0.2463, + "step": 3730 + }, + { + "epoch": 0.7697063369397218, + "grad_norm": 0.2682129393375418, + "learning_rate": 3.059860412440811e-05, + "loss": 0.2776, + "step": 3735 + }, + { + "epoch": 0.7707367336424523, + "grad_norm": 0.31234701048600605, + "learning_rate": 3.0340026608102902e-05, + "loss": 0.2284, + "step": 3740 + }, + { + "epoch": 0.7717671303451829, + "grad_norm": 0.3499928250572632, + "learning_rate": 3.008235077738334e-05, + "loss": 0.3077, + "step": 3745 + }, + { + "epoch": 0.7727975270479135, + "grad_norm": 0.385108342102361, + "learning_rate": 2.9825579967630846e-05, + "loss": 0.3223, + "step": 3750 + }, + { + "epoch": 0.773827923750644, + "grad_norm": 0.3716570069552568, + "learning_rate": 2.956971750251215e-05, + "loss": 0.2398, + "step": 3755 + }, + { + "epoch": 0.7748583204533745, + "grad_norm": 0.29486612653904987, + "learning_rate": 2.9314766693936356e-05, + "loss": 0.2659, + "step": 3760 + }, + { + "epoch": 0.7758887171561051, + "grad_norm": 0.3691838137762809, + "learning_rate": 2.906073084201191e-05, + "loss": 0.2553, + "step": 3765 + }, + { + "epoch": 0.7769191138588356, + "grad_norm": 0.3573299514624386, + "learning_rate": 2.8807613235004037e-05, + "loss": 0.2966, + "step": 3770 + }, + { + "epoch": 0.7779495105615662, + "grad_norm": 0.2972422322880715, + "learning_rate": 2.855541714929206e-05, + "loss": 0.2817, + "step": 3775 + }, + { + "epoch": 0.7789799072642968, + "grad_norm": 0.31746513053860076, + "learning_rate": 2.8304145849327036e-05, + "loss": 0.2492, + "step": 3780 + }, + { + "epoch": 0.7800103039670273, + "grad_norm": 0.33326098116022557, + "learning_rate": 2.8053802587589538e-05, + "loss": 0.2821, + "step": 3785 + }, + { + "epoch": 0.7810407006697578, + "grad_norm": 0.343516862256668, + "learning_rate": 2.7804390604547557e-05, + "loss": 0.253, + "step": 3790 + }, + { + "epoch": 0.7820710973724884, + "grad_norm": 0.29542386402077736, + "learning_rate": 2.7555913128614398e-05, + "loss": 0.2846, + "step": 3795 + }, + { + "epoch": 0.783101494075219, + "grad_norm": 0.40216308008328466, + "learning_rate": 2.7308373376107142e-05, + "loss": 0.2903, + "step": 3800 + }, + { + "epoch": 0.783101494075219, + "eval_loss": 0.2628938853740692, + "eval_runtime": 2894.5935, + "eval_samples_per_second": 2.764, + "eval_steps_per_second": 0.345, + "step": 3800 + }, + { + "epoch": 0.7841318907779495, + "grad_norm": 0.40257332364286125, + "learning_rate": 2.706177455120482e-05, + "loss": 0.2771, + "step": 3805 + }, + { + "epoch": 0.7851622874806801, + "grad_norm": 0.33675471922875544, + "learning_rate": 2.681611984590696e-05, + "loss": 0.3142, + "step": 3810 + }, + { + "epoch": 0.7861926841834106, + "grad_norm": 0.30015764012340485, + "learning_rate": 2.6571412439992437e-05, + "loss": 0.2565, + "step": 3815 + }, + { + "epoch": 0.7872230808861411, + "grad_norm": 0.2809235294807618, + "learning_rate": 2.6327655500978076e-05, + "loss": 0.272, + "step": 3820 + }, + { + "epoch": 0.7882534775888718, + "grad_norm": 0.3792796949891521, + "learning_rate": 2.608485218407779e-05, + "loss": 0.2691, + "step": 3825 + }, + { + "epoch": 0.7892838742916023, + "grad_norm": 0.3665809699953418, + "learning_rate": 2.5843005632161787e-05, + "loss": 0.2615, + "step": 3830 + }, + { + "epoch": 0.7903142709943328, + "grad_norm": 0.30924641336214526, + "learning_rate": 2.5602118975715683e-05, + "loss": 0.2668, + "step": 3835 + }, + { + "epoch": 0.7913446676970634, + "grad_norm": 0.3535640843972266, + "learning_rate": 2.5362195332800253e-05, + "loss": 0.251, + "step": 3840 + }, + { + "epoch": 0.7923750643997939, + "grad_norm": 0.31442694000184856, + "learning_rate": 2.5123237809010836e-05, + "loss": 0.277, + "step": 3845 + }, + { + "epoch": 0.7934054611025245, + "grad_norm": 0.473383912475728, + "learning_rate": 2.4885249497437223e-05, + "loss": 0.3069, + "step": 3850 + }, + { + "epoch": 0.794435857805255, + "grad_norm": 0.39185959855652625, + "learning_rate": 2.4648233478623705e-05, + "loss": 0.2589, + "step": 3855 + }, + { + "epoch": 0.7954662545079856, + "grad_norm": 0.3644503327681114, + "learning_rate": 2.4412192820529034e-05, + "loss": 0.2912, + "step": 3860 + }, + { + "epoch": 0.7964966512107161, + "grad_norm": 0.28558482434608623, + "learning_rate": 2.4177130578486885e-05, + "loss": 0.2455, + "step": 3865 + }, + { + "epoch": 0.7975270479134466, + "grad_norm": 0.3567026701188835, + "learning_rate": 2.3943049795166126e-05, + "loss": 0.2901, + "step": 3870 + }, + { + "epoch": 0.7985574446161773, + "grad_norm": 0.3398153100653687, + "learning_rate": 2.370995350053157e-05, + "loss": 0.2763, + "step": 3875 + }, + { + "epoch": 0.7995878413189078, + "grad_norm": 0.3234949943482735, + "learning_rate": 2.3477844711804708e-05, + "loss": 0.2644, + "step": 3880 + }, + { + "epoch": 0.8006182380216383, + "grad_norm": 0.33467823842360356, + "learning_rate": 2.3246726433424716e-05, + "loss": 0.2914, + "step": 3885 + }, + { + "epoch": 0.8016486347243689, + "grad_norm": 0.32056191813609336, + "learning_rate": 2.301660165700936e-05, + "loss": 0.2654, + "step": 3890 + }, + { + "epoch": 0.8026790314270994, + "grad_norm": 0.3034671419615242, + "learning_rate": 2.2787473361316592e-05, + "loss": 0.2756, + "step": 3895 + }, + { + "epoch": 0.80370942812983, + "grad_norm": 0.37786668109879534, + "learning_rate": 2.2559344512205705e-05, + "loss": 0.292, + "step": 3900 + }, + { + "epoch": 0.8047398248325606, + "grad_norm": 0.34403274315844035, + "learning_rate": 2.233221806259903e-05, + "loss": 0.2334, + "step": 3905 + }, + { + "epoch": 0.8057702215352911, + "grad_norm": 0.32304818372501226, + "learning_rate": 2.2106096952443888e-05, + "loss": 0.2962, + "step": 3910 + }, + { + "epoch": 0.8068006182380216, + "grad_norm": 0.28048590872312423, + "learning_rate": 2.188098410867424e-05, + "loss": 0.2304, + "step": 3915 + }, + { + "epoch": 0.8078310149407522, + "grad_norm": 0.40871189926827856, + "learning_rate": 2.165688244517299e-05, + "loss": 0.3092, + "step": 3920 + }, + { + "epoch": 0.8088614116434827, + "grad_norm": 0.37394901632285205, + "learning_rate": 2.143379486273428e-05, + "loss": 0.3139, + "step": 3925 + }, + { + "epoch": 0.8098918083462133, + "grad_norm": 0.3813857683853878, + "learning_rate": 2.1211724249025787e-05, + "loss": 0.2522, + "step": 3930 + }, + { + "epoch": 0.8109222050489439, + "grad_norm": 0.3634405006796172, + "learning_rate": 2.099067347855157e-05, + "loss": 0.2892, + "step": 3935 + }, + { + "epoch": 0.8119526017516744, + "grad_norm": 0.3055737527512765, + "learning_rate": 2.077064541261462e-05, + "loss": 0.2376, + "step": 3940 + }, + { + "epoch": 0.8129829984544049, + "grad_norm": 0.29484654000322413, + "learning_rate": 2.0551642899279975e-05, + "loss": 0.2752, + "step": 3945 + }, + { + "epoch": 0.8140133951571354, + "grad_norm": 0.32834988086926475, + "learning_rate": 2.0333668773337866e-05, + "loss": 0.2711, + "step": 3950 + }, + { + "epoch": 0.8150437918598661, + "grad_norm": 0.36111993999911474, + "learning_rate": 2.0116725856266926e-05, + "loss": 0.2563, + "step": 3955 + }, + { + "epoch": 0.8160741885625966, + "grad_norm": 0.34560793782552396, + "learning_rate": 1.9900816956197698e-05, + "loss": 0.2868, + "step": 3960 + }, + { + "epoch": 0.8171045852653271, + "grad_norm": 0.3575615068921621, + "learning_rate": 1.9685944867876373e-05, + "loss": 0.2475, + "step": 3965 + }, + { + "epoch": 0.8181349819680577, + "grad_norm": 0.40119120114689133, + "learning_rate": 1.9472112372628536e-05, + "loss": 0.3028, + "step": 3970 + }, + { + "epoch": 0.8191653786707882, + "grad_norm": 0.3996261960750051, + "learning_rate": 1.9259322238323095e-05, + "loss": 0.2891, + "step": 3975 + }, + { + "epoch": 0.8201957753735188, + "grad_norm": 0.3045180790248437, + "learning_rate": 1.9047577219336665e-05, + "loss": 0.2379, + "step": 3980 + }, + { + "epoch": 0.8212261720762494, + "grad_norm": 0.32648574145564824, + "learning_rate": 1.8836880056517658e-05, + "loss": 0.3136, + "step": 3985 + }, + { + "epoch": 0.8222565687789799, + "grad_norm": 0.3368467272156829, + "learning_rate": 1.862723347715103e-05, + "loss": 0.2565, + "step": 3990 + }, + { + "epoch": 0.8232869654817104, + "grad_norm": 0.3785392791415945, + "learning_rate": 1.841864019492282e-05, + "loss": 0.3129, + "step": 3995 + }, + { + "epoch": 0.824317362184441, + "grad_norm": 0.37942467936258845, + "learning_rate": 1.821110290988509e-05, + "loss": 0.2943, + "step": 4000 + }, + { + "epoch": 0.824317362184441, + "eval_loss": 0.2613193392753601, + "eval_runtime": 2912.2794, + "eval_samples_per_second": 2.747, + "eval_steps_per_second": 0.343, + "step": 4000 + }, + { + "epoch": 0.8253477588871716, + "grad_norm": 0.37476871355187913, + "learning_rate": 1.8004624308421026e-05, + "loss": 0.2454, + "step": 4005 + }, + { + "epoch": 0.8263781555899021, + "grad_norm": 0.432774306799512, + "learning_rate": 1.7799207063210044e-05, + "loss": 0.3059, + "step": 4010 + }, + { + "epoch": 0.8274085522926327, + "grad_norm": 0.2649022181395363, + "learning_rate": 1.759485383319326e-05, + "loss": 0.2358, + "step": 4015 + }, + { + "epoch": 0.8284389489953632, + "grad_norm": 0.4262757717672499, + "learning_rate": 1.7391567263539144e-05, + "loss": 0.3068, + "step": 4020 + }, + { + "epoch": 0.8294693456980937, + "grad_norm": 0.40389188939685916, + "learning_rate": 1.7189349985609115e-05, + "loss": 0.3148, + "step": 4025 + }, + { + "epoch": 0.8304997424008244, + "grad_norm": 0.34848990344269987, + "learning_rate": 1.6988204616923666e-05, + "loss": 0.2645, + "step": 4030 + }, + { + "epoch": 0.8315301391035549, + "grad_norm": 0.3081339412565511, + "learning_rate": 1.6788133761128312e-05, + "loss": 0.2901, + "step": 4035 + }, + { + "epoch": 0.8325605358062854, + "grad_norm": 0.31249793235144235, + "learning_rate": 1.658914000795999e-05, + "loss": 0.2607, + "step": 4040 + }, + { + "epoch": 0.833590932509016, + "grad_norm": 0.4513911289927334, + "learning_rate": 1.639122593321357e-05, + "loss": 0.3071, + "step": 4045 + }, + { + "epoch": 0.8346213292117465, + "grad_norm": 0.36902215017120704, + "learning_rate": 1.6194394098708377e-05, + "loss": 0.2841, + "step": 4050 + }, + { + "epoch": 0.8356517259144771, + "grad_norm": 0.34291388974227893, + "learning_rate": 1.59986470522551e-05, + "loss": 0.256, + "step": 4055 + }, + { + "epoch": 0.8366821226172076, + "grad_norm": 0.3484068120394701, + "learning_rate": 1.580398732762297e-05, + "loss": 0.2998, + "step": 4060 + }, + { + "epoch": 0.8377125193199382, + "grad_norm": 0.3076306390902032, + "learning_rate": 1.5610417444506664e-05, + "loss": 0.2772, + "step": 4065 + }, + { + "epoch": 0.8387429160226687, + "grad_norm": 0.3147872649091281, + "learning_rate": 1.541793990849387e-05, + "loss": 0.2549, + "step": 4070 + }, + { + "epoch": 0.8397733127253992, + "grad_norm": 0.39284501205730754, + "learning_rate": 1.522655721103291e-05, + "loss": 0.335, + "step": 4075 + }, + { + "epoch": 0.8408037094281299, + "grad_norm": 0.4446246586077643, + "learning_rate": 1.5036271829400294e-05, + "loss": 0.2516, + "step": 4080 + }, + { + "epoch": 0.8418341061308604, + "grad_norm": 0.355547901160868, + "learning_rate": 1.4847086226668872e-05, + "loss": 0.2764, + "step": 4085 + }, + { + "epoch": 0.8428645028335909, + "grad_norm": 0.3688730084692743, + "learning_rate": 1.4659002851675774e-05, + "loss": 0.2716, + "step": 4090 + }, + { + "epoch": 0.8438948995363215, + "grad_norm": 0.40395424185354845, + "learning_rate": 1.447202413899078e-05, + "loss": 0.3036, + "step": 4095 + }, + { + "epoch": 0.844925296239052, + "grad_norm": 0.3798137311999296, + "learning_rate": 1.4286152508884898e-05, + "loss": 0.3078, + "step": 4100 + }, + { + "epoch": 0.8459556929417826, + "grad_norm": 0.3136631297841343, + "learning_rate": 1.4101390367298861e-05, + "loss": 0.239, + "step": 4105 + }, + { + "epoch": 0.8469860896445132, + "grad_norm": 0.3133675747466587, + "learning_rate": 1.3917740105812094e-05, + "loss": 0.2685, + "step": 4110 + }, + { + "epoch": 0.8480164863472437, + "grad_norm": 0.3840975313756144, + "learning_rate": 1.3735204101611776e-05, + "loss": 0.2685, + "step": 4115 + }, + { + "epoch": 0.8490468830499742, + "grad_norm": 0.34167013481311503, + "learning_rate": 1.355378471746196e-05, + "loss": 0.2835, + "step": 4120 + }, + { + "epoch": 0.8500772797527048, + "grad_norm": 0.37678807026533556, + "learning_rate": 1.3373484301673145e-05, + "loss": 0.3225, + "step": 4125 + }, + { + "epoch": 0.8511076764554354, + "grad_norm": 0.36169730897373126, + "learning_rate": 1.3194305188071732e-05, + "loss": 0.2297, + "step": 4130 + }, + { + "epoch": 0.8521380731581659, + "grad_norm": 0.34627350397620255, + "learning_rate": 1.301624969596985e-05, + "loss": 0.308, + "step": 4135 + }, + { + "epoch": 0.8531684698608965, + "grad_norm": 0.3855153639798906, + "learning_rate": 1.2839320130135468e-05, + "loss": 0.2484, + "step": 4140 + }, + { + "epoch": 0.854198866563627, + "grad_norm": 0.2963869582281001, + "learning_rate": 1.266351878076234e-05, + "loss": 0.2646, + "step": 4145 + }, + { + "epoch": 0.8552292632663575, + "grad_norm": 0.3094943477185959, + "learning_rate": 1.2488847923440483e-05, + "loss": 0.284, + "step": 4150 + }, + { + "epoch": 0.8562596599690881, + "grad_norm": 0.3854839502521428, + "learning_rate": 1.2315309819126852e-05, + "loss": 0.2085, + "step": 4155 + }, + { + "epoch": 0.8572900566718187, + "grad_norm": 0.3087600519704163, + "learning_rate": 1.2142906714115787e-05, + "loss": 0.2962, + "step": 4160 + }, + { + "epoch": 0.8583204533745492, + "grad_norm": 0.3979580261424234, + "learning_rate": 1.197164084001009e-05, + "loss": 0.2516, + "step": 4165 + }, + { + "epoch": 0.8593508500772797, + "grad_norm": 0.28766440994452486, + "learning_rate": 1.1801514413692239e-05, + "loss": 0.2795, + "step": 4170 + }, + { + "epoch": 0.8603812467800103, + "grad_norm": 0.3167014720203987, + "learning_rate": 1.1632529637295475e-05, + "loss": 0.2857, + "step": 4175 + }, + { + "epoch": 0.8614116434827409, + "grad_norm": 0.33414263857523546, + "learning_rate": 1.1464688698175497e-05, + "loss": 0.2431, + "step": 4180 + }, + { + "epoch": 0.8624420401854714, + "grad_norm": 0.29115194585667353, + "learning_rate": 1.1297993768881998e-05, + "loss": 0.301, + "step": 4185 + }, + { + "epoch": 0.863472436888202, + "grad_norm": 0.34951793739233555, + "learning_rate": 1.113244700713063e-05, + "loss": 0.2559, + "step": 4190 + }, + { + "epoch": 0.8645028335909325, + "grad_norm": 0.3322485561015013, + "learning_rate": 1.0968050555775067e-05, + "loss": 0.3051, + "step": 4195 + }, + { + "epoch": 0.865533230293663, + "grad_norm": 0.3598063007969742, + "learning_rate": 1.0804806542779223e-05, + "loss": 0.2787, + "step": 4200 + }, + { + "epoch": 0.865533230293663, + "eval_loss": 0.2603052854537964, + "eval_runtime": 2913.7463, + "eval_samples_per_second": 2.746, + "eval_steps_per_second": 0.343, + "step": 4200 + }, + { + "epoch": 0.8665636269963937, + "grad_norm": 0.37342349900756117, + "learning_rate": 1.0642717081189735e-05, + "loss": 0.2767, + "step": 4205 + }, + { + "epoch": 0.8675940236991242, + "grad_norm": 0.3264967979693496, + "learning_rate": 1.0481784269108664e-05, + "loss": 0.2836, + "step": 4210 + }, + { + "epoch": 0.8686244204018547, + "grad_norm": 0.2837466654297017, + "learning_rate": 1.032201018966621e-05, + "loss": 0.2616, + "step": 4215 + }, + { + "epoch": 0.8696548171045853, + "grad_norm": 0.3880834543063498, + "learning_rate": 1.0163396910993883e-05, + "loss": 0.2641, + "step": 4220 + }, + { + "epoch": 0.8706852138073158, + "grad_norm": 0.40926771296668485, + "learning_rate": 1.0005946486197648e-05, + "loss": 0.2946, + "step": 4225 + }, + { + "epoch": 0.8717156105100463, + "grad_norm": 0.34072299231820374, + "learning_rate": 9.849660953331363e-06, + "loss": 0.2257, + "step": 4230 + }, + { + "epoch": 0.872746007212777, + "grad_norm": 0.3607472251947616, + "learning_rate": 9.694542335370437e-06, + "loss": 0.3061, + "step": 4235 + }, + { + "epoch": 0.8737764039155075, + "grad_norm": 0.2710285195917331, + "learning_rate": 9.540592640185597e-06, + "loss": 0.2725, + "step": 4240 + }, + { + "epoch": 0.874806800618238, + "grad_norm": 0.36432299015695907, + "learning_rate": 9.387813860516915e-06, + "loss": 0.2851, + "step": 4245 + }, + { + "epoch": 0.8758371973209685, + "grad_norm": 0.35703992526720524, + "learning_rate": 9.236207973948063e-06, + "loss": 0.2972, + "step": 4250 + }, + { + "epoch": 0.8768675940236991, + "grad_norm": 0.3173002625132457, + "learning_rate": 9.085776942880608e-06, + "loss": 0.2876, + "step": 4255 + }, + { + "epoch": 0.8778979907264297, + "grad_norm": 0.28131579729158257, + "learning_rate": 8.936522714508678e-06, + "loss": 0.2867, + "step": 4260 + }, + { + "epoch": 0.8789283874291602, + "grad_norm": 0.3530032076872567, + "learning_rate": 8.788447220793806e-06, + "loss": 0.2798, + "step": 4265 + }, + { + "epoch": 0.8799587841318908, + "grad_norm": 0.347013694472638, + "learning_rate": 8.641552378439776e-06, + "loss": 0.2937, + "step": 4270 + }, + { + "epoch": 0.8809891808346213, + "grad_norm": 0.357377959486403, + "learning_rate": 8.495840088868024e-06, + "loss": 0.334, + "step": 4275 + }, + { + "epoch": 0.8820195775373518, + "grad_norm": 0.27316974502275887, + "learning_rate": 8.351312238192787e-06, + "loss": 0.2675, + "step": 4280 + }, + { + "epoch": 0.8830499742400825, + "grad_norm": 0.35924727199943307, + "learning_rate": 8.207970697196866e-06, + "loss": 0.2961, + "step": 4285 + }, + { + "epoch": 0.884080370942813, + "grad_norm": 0.3577190138371286, + "learning_rate": 8.065817321307367e-06, + "loss": 0.254, + "step": 4290 + }, + { + "epoch": 0.8851107676455435, + "grad_norm": 0.5576399803787185, + "learning_rate": 7.924853950571642e-06, + "loss": 0.2952, + "step": 4295 + }, + { + "epoch": 0.8861411643482741, + "grad_norm": 0.3763241455754686, + "learning_rate": 7.78508240963347e-06, + "loss": 0.2863, + "step": 4300 + }, + { + "epoch": 0.8871715610510046, + "grad_norm": 0.30880739002832336, + "learning_rate": 7.646504507709563e-06, + "loss": 0.2824, + "step": 4305 + }, + { + "epoch": 0.8882019577537352, + "grad_norm": 0.3137337962583597, + "learning_rate": 7.50912203856593e-06, + "loss": 0.2674, + "step": 4310 + }, + { + "epoch": 0.8892323544564658, + "grad_norm": 0.31308889569105514, + "learning_rate": 7.372936780494877e-06, + "loss": 0.2755, + "step": 4315 + }, + { + "epoch": 0.8902627511591963, + "grad_norm": 0.4028987419767302, + "learning_rate": 7.237950496291856e-06, + "loss": 0.2871, + "step": 4320 + }, + { + "epoch": 0.8912931478619268, + "grad_norm": 0.36938026316991573, + "learning_rate": 7.104164933232649e-06, + "loss": 0.2786, + "step": 4325 + }, + { + "epoch": 0.8923235445646573, + "grad_norm": 0.3208025100288799, + "learning_rate": 6.971581823050832e-06, + "loss": 0.2474, + "step": 4330 + }, + { + "epoch": 0.893353941267388, + "grad_norm": 0.3598206195922243, + "learning_rate": 6.840202881915325e-06, + "loss": 0.3023, + "step": 4335 + }, + { + "epoch": 0.8943843379701185, + "grad_norm": 0.29364044659572436, + "learning_rate": 6.710029810408092e-06, + "loss": 0.2397, + "step": 4340 + }, + { + "epoch": 0.895414734672849, + "grad_norm": 0.31914310843659316, + "learning_rate": 6.581064293502293e-06, + "loss": 0.2883, + "step": 4345 + }, + { + "epoch": 0.8964451313755796, + "grad_norm": 0.45022448364382284, + "learning_rate": 6.453308000540304e-06, + "loss": 0.2756, + "step": 4350 + }, + { + "epoch": 0.8974755280783101, + "grad_norm": 0.310267786867979, + "learning_rate": 6.326762585212209e-06, + "loss": 0.2415, + "step": 4355 + }, + { + "epoch": 0.8985059247810407, + "grad_norm": 0.3112505151505233, + "learning_rate": 6.20142968553441e-06, + "loss": 0.2821, + "step": 4360 + }, + { + "epoch": 0.8995363214837713, + "grad_norm": 0.3141837487350365, + "learning_rate": 6.077310923828328e-06, + "loss": 0.2725, + "step": 4365 + }, + { + "epoch": 0.9005667181865018, + "grad_norm": 0.27623461345523886, + "learning_rate": 5.954407906699511e-06, + "loss": 0.268, + "step": 4370 + }, + { + "epoch": 0.9015971148892323, + "grad_norm": 0.3891601937753882, + "learning_rate": 5.8327222250167735e-06, + "loss": 0.2778, + "step": 4375 + }, + { + "epoch": 0.9026275115919629, + "grad_norm": 0.3558581335957565, + "learning_rate": 5.71225545389158e-06, + "loss": 0.2241, + "step": 4380 + }, + { + "epoch": 0.9036579082946935, + "grad_norm": 0.3693584144347684, + "learning_rate": 5.59300915265778e-06, + "loss": 0.293, + "step": 4385 + }, + { + "epoch": 0.904688304997424, + "grad_norm": 0.412387091213602, + "learning_rate": 5.4749848648512624e-06, + "loss": 0.2638, + "step": 4390 + }, + { + "epoch": 0.9057187017001546, + "grad_norm": 0.4134072879227268, + "learning_rate": 5.358184118190068e-06, + "loss": 0.2996, + "step": 4395 + }, + { + "epoch": 0.9067490984028851, + "grad_norm": 0.33905018808481746, + "learning_rate": 5.242608424554651e-06, + "loss": 0.2558, + "step": 4400 + }, + { + "epoch": 0.9067490984028851, + "eval_loss": 0.25964951515197754, + "eval_runtime": 2914.0015, + "eval_samples_per_second": 2.745, + "eval_steps_per_second": 0.343, + "step": 4400 + }, + { + "epoch": 0.9077794951056156, + "grad_norm": 0.3147365134234666, + "learning_rate": 5.128259279968195e-06, + "loss": 0.2516, + "step": 4405 + }, + { + "epoch": 0.9088098918083463, + "grad_norm": 0.34810883229758166, + "learning_rate": 5.01513816457736e-06, + "loss": 0.2778, + "step": 4410 + }, + { + "epoch": 0.9098402885110768, + "grad_norm": 0.3558739397328132, + "learning_rate": 4.903246542633033e-06, + "loss": 0.2516, + "step": 4415 + }, + { + "epoch": 0.9108706852138073, + "grad_norm": 0.3434310486600019, + "learning_rate": 4.7925858624714215e-06, + "loss": 0.2658, + "step": 4420 + }, + { + "epoch": 0.9119010819165378, + "grad_norm": 0.4739996869148237, + "learning_rate": 4.683157556495343e-06, + "loss": 0.3177, + "step": 4425 + }, + { + "epoch": 0.9129314786192684, + "grad_norm": 0.3432195383447999, + "learning_rate": 4.574963041155622e-06, + "loss": 0.2456, + "step": 4430 + }, + { + "epoch": 0.913961875321999, + "grad_norm": 0.39865656812853095, + "learning_rate": 4.468003716932734e-06, + "loss": 0.3004, + "step": 4435 + }, + { + "epoch": 0.9149922720247295, + "grad_norm": 0.3074797243983616, + "learning_rate": 4.362280968318777e-06, + "loss": 0.2468, + "step": 4440 + }, + { + "epoch": 0.9160226687274601, + "grad_norm": 0.3821770273420256, + "learning_rate": 4.257796163799455e-06, + "loss": 0.2569, + "step": 4445 + }, + { + "epoch": 0.9170530654301906, + "grad_norm": 0.3846408351560208, + "learning_rate": 4.154550655836409e-06, + "loss": 0.2915, + "step": 4450 + }, + { + "epoch": 0.9180834621329211, + "grad_norm": 0.3328477896869499, + "learning_rate": 4.052545780849715e-06, + "loss": 0.2382, + "step": 4455 + }, + { + "epoch": 0.9191138588356518, + "grad_norm": 0.39094030418826553, + "learning_rate": 3.9517828592005475e-06, + "loss": 0.3148, + "step": 4460 + }, + { + "epoch": 0.9201442555383823, + "grad_norm": 0.32981008377515736, + "learning_rate": 3.852263195174155e-06, + "loss": 0.2671, + "step": 4465 + }, + { + "epoch": 0.9211746522411128, + "grad_norm": 0.31369535441554675, + "learning_rate": 3.7539880769628998e-06, + "loss": 0.2947, + "step": 4470 + }, + { + "epoch": 0.9222050489438434, + "grad_norm": 0.37611489272202936, + "learning_rate": 3.6569587766496216e-06, + "loss": 0.2808, + "step": 4475 + }, + { + "epoch": 0.9232354456465739, + "grad_norm": 0.26359972435462287, + "learning_rate": 3.561176550191203e-06, + "loss": 0.2394, + "step": 4480 + }, + { + "epoch": 0.9242658423493045, + "grad_norm": 0.3951324034345502, + "learning_rate": 3.46664263740224e-06, + "loss": 0.2848, + "step": 4485 + }, + { + "epoch": 0.9252962390520351, + "grad_norm": 0.3608819993192937, + "learning_rate": 3.3733582619390523e-06, + "loss": 0.2632, + "step": 4490 + }, + { + "epoch": 0.9263266357547656, + "grad_norm": 0.3555456945529403, + "learning_rate": 3.281324631283833e-06, + "loss": 0.279, + "step": 4495 + }, + { + "epoch": 0.9273570324574961, + "grad_norm": 0.39293143583676265, + "learning_rate": 3.1905429367289795e-06, + "loss": 0.31, + "step": 4500 + }, + { + "epoch": 0.9283874291602267, + "grad_norm": 0.3368920177589104, + "learning_rate": 3.101014353361753e-06, + "loss": 0.2295, + "step": 4505 + }, + { + "epoch": 0.9294178258629573, + "grad_norm": 0.3034071042496814, + "learning_rate": 3.012740040048978e-06, + "loss": 0.2922, + "step": 4510 + }, + { + "epoch": 0.9304482225656878, + "grad_norm": 0.30952410133286584, + "learning_rate": 2.9257211394220773e-06, + "loss": 0.2719, + "step": 4515 + }, + { + "epoch": 0.9314786192684184, + "grad_norm": 0.30563095374433163, + "learning_rate": 2.8399587778623505e-06, + "loss": 0.2688, + "step": 4520 + }, + { + "epoch": 0.9325090159711489, + "grad_norm": 0.3782690565607755, + "learning_rate": 2.755454065486263e-06, + "loss": 0.2842, + "step": 4525 + }, + { + "epoch": 0.9335394126738794, + "grad_norm": 0.37596385739941673, + "learning_rate": 2.672208096131157e-06, + "loss": 0.2412, + "step": 4530 + }, + { + "epoch": 0.93456980937661, + "grad_norm": 0.33654733598583986, + "learning_rate": 2.5902219473411204e-06, + "loss": 0.2903, + "step": 4535 + }, + { + "epoch": 0.9356002060793406, + "grad_norm": 0.3317164278402486, + "learning_rate": 2.509496680352963e-06, + "loss": 0.2443, + "step": 4540 + }, + { + "epoch": 0.9366306027820711, + "grad_norm": 0.30341596651692276, + "learning_rate": 2.430033340082516e-06, + "loss": 0.2532, + "step": 4545 + }, + { + "epoch": 0.9376609994848016, + "grad_norm": 0.38567400726184814, + "learning_rate": 2.3518329551111217e-06, + "loss": 0.3219, + "step": 4550 + }, + { + "epoch": 0.9386913961875322, + "grad_norm": 0.5114420556484347, + "learning_rate": 2.2748965376723e-06, + "loss": 0.2911, + "step": 4555 + }, + { + "epoch": 0.9397217928902627, + "grad_norm": 0.2935585297022278, + "learning_rate": 2.199225083638656e-06, + "loss": 0.2843, + "step": 4560 + }, + { + "epoch": 0.9407521895929933, + "grad_norm": 0.28858986310591855, + "learning_rate": 2.1248195725089624e-06, + "loss": 0.2327, + "step": 4565 + }, + { + "epoch": 0.9417825862957239, + "grad_norm": 0.2847670925427268, + "learning_rate": 2.0516809673955083e-06, + "loss": 0.2866, + "step": 4570 + }, + { + "epoch": 0.9428129829984544, + "grad_norm": 0.3816164009941783, + "learning_rate": 1.9798102150116573e-06, + "loss": 0.3146, + "step": 4575 + }, + { + "epoch": 0.9438433797011849, + "grad_norm": 0.34153941420407485, + "learning_rate": 1.909208245659522e-06, + "loss": 0.2655, + "step": 4580 + }, + { + "epoch": 0.9448737764039155, + "grad_norm": 0.3775194213326933, + "learning_rate": 1.8398759732179637e-06, + "loss": 0.2893, + "step": 4585 + }, + { + "epoch": 0.9459041731066461, + "grad_norm": 0.30788658820871284, + "learning_rate": 1.7718142951307914e-06, + "loss": 0.2611, + "step": 4590 + }, + { + "epoch": 0.9469345698093766, + "grad_norm": 0.3383435069564331, + "learning_rate": 1.705024092395091e-06, + "loss": 0.2504, + "step": 4595 + }, + { + "epoch": 0.9479649665121072, + "grad_norm": 0.4137029730454973, + "learning_rate": 1.6395062295498698e-06, + "loss": 0.3107, + "step": 4600 + }, + { + "epoch": 0.9479649665121072, + "eval_loss": 0.2593182921409607, + "eval_runtime": 2914.0057, + "eval_samples_per_second": 2.745, + "eval_steps_per_second": 0.343, + "step": 4600 + }, + { + "epoch": 0.9489953632148377, + "grad_norm": 0.2756456550382528, + "learning_rate": 1.5752615546647975e-06, + "loss": 0.2688, + "step": 4605 + }, + { + "epoch": 0.9500257599175682, + "grad_norm": 0.3558221128709261, + "learning_rate": 1.5122908993293273e-06, + "loss": 0.2962, + "step": 4610 + }, + { + "epoch": 0.9510561566202989, + "grad_norm": 0.30866148215688255, + "learning_rate": 1.4505950786418255e-06, + "loss": 0.2396, + "step": 4615 + }, + { + "epoch": 0.9520865533230294, + "grad_norm": 0.32718060116494296, + "learning_rate": 1.3901748911991253e-06, + "loss": 0.259, + "step": 4620 + }, + { + "epoch": 0.9531169500257599, + "grad_norm": 0.41916606085025016, + "learning_rate": 1.331031119086079e-06, + "loss": 0.2914, + "step": 4625 + }, + { + "epoch": 0.9541473467284904, + "grad_norm": 0.32801299914013593, + "learning_rate": 1.2731645278655445e-06, + "loss": 0.2627, + "step": 4630 + }, + { + "epoch": 0.955177743431221, + "grad_norm": 0.35340318213193234, + "learning_rate": 1.2165758665683924e-06, + "loss": 0.2832, + "step": 4635 + }, + { + "epoch": 0.9562081401339516, + "grad_norm": 0.3450871009199193, + "learning_rate": 1.1612658676838473e-06, + "loss": 0.2402, + "step": 4640 + }, + { + "epoch": 0.9572385368366821, + "grad_norm": 0.35663476220665696, + "learning_rate": 1.107235247150018e-06, + "loss": 0.2892, + "step": 4645 + }, + { + "epoch": 0.9582689335394127, + "grad_norm": 0.3491909974711667, + "learning_rate": 1.0544847043445938e-06, + "loss": 0.2838, + "step": 4650 + }, + { + "epoch": 0.9592993302421432, + "grad_norm": 0.47789307717475066, + "learning_rate": 1.0030149220758288e-06, + "loss": 0.2633, + "step": 4655 + }, + { + "epoch": 0.9603297269448737, + "grad_norm": 0.3220693161443208, + "learning_rate": 9.528265665736502e-07, + "loss": 0.2983, + "step": 4660 + }, + { + "epoch": 0.9613601236476044, + "grad_norm": 0.3355491215986634, + "learning_rate": 9.039202874811192e-07, + "loss": 0.2606, + "step": 4665 + }, + { + "epoch": 0.9623905203503349, + "grad_norm": 0.3331685126111554, + "learning_rate": 8.562967178459391e-07, + "loss": 0.2836, + "step": 4670 + }, + { + "epoch": 0.9634209170530654, + "grad_norm": 0.33838080949230487, + "learning_rate": 8.099564741123166e-07, + "loss": 0.2912, + "step": 4675 + }, + { + "epoch": 0.964451313755796, + "grad_norm": 0.3638971937799169, + "learning_rate": 7.649001561129354e-07, + "loss": 0.2357, + "step": 4680 + }, + { + "epoch": 0.9654817104585265, + "grad_norm": 0.30728028400390417, + "learning_rate": 7.211283470612395e-07, + "loss": 0.2763, + "step": 4685 + }, + { + "epoch": 0.9665121071612571, + "grad_norm": 0.42508518073646434, + "learning_rate": 6.786416135438512e-07, + "loss": 0.2632, + "step": 4690 + }, + { + "epoch": 0.9675425038639877, + "grad_norm": 0.319418982554995, + "learning_rate": 6.374405055132537e-07, + "loss": 0.2666, + "step": 4695 + }, + { + "epoch": 0.9685729005667182, + "grad_norm": 0.3449145369969141, + "learning_rate": 5.975255562806647e-07, + "loss": 0.2924, + "step": 4700 + }, + { + "epoch": 0.9696032972694487, + "grad_norm": 0.4202821282926822, + "learning_rate": 5.58897282509141e-07, + "loss": 0.2151, + "step": 4705 + }, + { + "epoch": 0.9706336939721792, + "grad_norm": 0.34298503928511215, + "learning_rate": 5.215561842068728e-07, + "loss": 0.2776, + "step": 4710 + }, + { + "epoch": 0.9716640906749099, + "grad_norm": 0.3226411709221547, + "learning_rate": 4.855027447207338e-07, + "loss": 0.2726, + "step": 4715 + }, + { + "epoch": 0.9726944873776404, + "grad_norm": 0.32438079260005576, + "learning_rate": 4.507374307299972e-07, + "loss": 0.2897, + "step": 4720 + }, + { + "epoch": 0.973724884080371, + "grad_norm": 0.29250744283879526, + "learning_rate": 4.172606922403399e-07, + "loss": 0.3075, + "step": 4725 + }, + { + "epoch": 0.9747552807831015, + "grad_norm": 0.3734192410658037, + "learning_rate": 3.8507296257798145e-07, + "loss": 0.2382, + "step": 4730 + }, + { + "epoch": 0.975785677485832, + "grad_norm": 0.32324581396516133, + "learning_rate": 3.541746583840655e-07, + "loss": 0.2776, + "step": 4735 + }, + { + "epoch": 0.9768160741885626, + "grad_norm": 0.33899080785526653, + "learning_rate": 3.24566179609298e-07, + "loss": 0.278, + "step": 4740 + }, + { + "epoch": 0.9778464708912932, + "grad_norm": 0.31722996073708115, + "learning_rate": 2.9624790950875113e-07, + "loss": 0.2854, + "step": 4745 + }, + { + "epoch": 0.9788768675940237, + "grad_norm": 0.3653754490411951, + "learning_rate": 2.692202146369338e-07, + "loss": 0.2903, + "step": 4750 + }, + { + "epoch": 0.9799072642967542, + "grad_norm": 0.3440659912321346, + "learning_rate": 2.434834448429957e-07, + "loss": 0.2624, + "step": 4755 + }, + { + "epoch": 0.9809376609994848, + "grad_norm": 0.3265654789285722, + "learning_rate": 2.1903793326621957e-07, + "loss": 0.2785, + "step": 4760 + }, + { + "epoch": 0.9819680577022154, + "grad_norm": 0.2763203967812842, + "learning_rate": 1.9588399633173605e-07, + "loss": 0.2387, + "step": 4765 + }, + { + "epoch": 0.9829984544049459, + "grad_norm": 0.32114982480782467, + "learning_rate": 1.740219337463822e-07, + "loss": 0.2601, + "step": 4770 + }, + { + "epoch": 0.9840288511076765, + "grad_norm": 0.34288896513991585, + "learning_rate": 1.534520284948715e-07, + "loss": 0.2752, + "step": 4775 + }, + { + "epoch": 0.985059247810407, + "grad_norm": 0.3760081860947557, + "learning_rate": 1.3417454683608554e-07, + "loss": 0.2583, + "step": 4780 + }, + { + "epoch": 0.9860896445131375, + "grad_norm": 0.362145568445466, + "learning_rate": 1.1618973829966572e-07, + "loss": 0.3256, + "step": 4785 + }, + { + "epoch": 0.9871200412158682, + "grad_norm": 0.29692027485612293, + "learning_rate": 9.949783568272697e-08, + "loss": 0.2302, + "step": 4790 + }, + { + "epoch": 0.9881504379185987, + "grad_norm": 0.35544835709990513, + "learning_rate": 8.409905504693782e-08, + "loss": 0.2742, + "step": 4795 + }, + { + "epoch": 0.9891808346213292, + "grad_norm": 0.34329714361053626, + "learning_rate": 6.999359571561171e-08, + "loss": 0.2894, + "step": 4800 + }, + { + "epoch": 0.9891808346213292, + "eval_loss": 0.25925347208976746, + "eval_runtime": 2913.0942, + "eval_samples_per_second": 2.746, + "eval_steps_per_second": 0.343, + "step": 4800 + }, + { + "epoch": 0.9902112313240597, + "grad_norm": 0.3542301832349851, + "learning_rate": 5.718164027121997e-08, + "loss": 0.2343, + "step": 4805 + }, + { + "epoch": 0.9912416280267903, + "grad_norm": 0.3086510062926568, + "learning_rate": 4.566335455299387e-08, + "loss": 0.2788, + "step": 4810 + }, + { + "epoch": 0.9922720247295209, + "grad_norm": 0.3507547113812826, + "learning_rate": 3.5438887654737355e-08, + "loss": 0.2747, + "step": 4815 + }, + { + "epoch": 0.9933024214322514, + "grad_norm": 0.3265356296027687, + "learning_rate": 2.6508371922984166e-08, + "loss": 0.2711, + "step": 4820 + }, + { + "epoch": 0.994332818134982, + "grad_norm": 0.3943623315986484, + "learning_rate": 1.887192295521034e-08, + "loss": 0.2788, + "step": 4825 + }, + { + "epoch": 0.9953632148377125, + "grad_norm": 0.3158484623921147, + "learning_rate": 1.252963959834652e-08, + "loss": 0.2249, + "step": 4830 + }, + { + "epoch": 0.996393611540443, + "grad_norm": 0.2997635546788507, + "learning_rate": 7.481603947556703e-09, + "loss": 0.2663, + "step": 4835 + }, + { + "epoch": 0.9974240082431737, + "grad_norm": 0.32227151020376166, + "learning_rate": 3.727881345105821e-09, + "loss": 0.2558, + "step": 4840 + }, + { + "epoch": 0.9984544049459042, + "grad_norm": 0.33469168529537796, + "learning_rate": 1.2685203795492762e-09, + "loss": 0.259, + "step": 4845 + }, + { + "epoch": 0.9994848016486347, + "grad_norm": 0.3630544783943566, + "learning_rate": 1.0355288510011107e-10, + "loss": 0.3056, + "step": 4850 + }, + { + "epoch": 0.999896960329727, + "step": 4852, + "total_flos": 2.4144679757152256e+16, + "train_loss": 0.3040192322367887, + "train_runtime": 137718.1009, + "train_samples_per_second": 0.564, + "train_steps_per_second": 0.035 + } + ], + "logging_steps": 5, + "max_steps": 4852, + "num_input_tokens_seen": 0, + "num_train_epochs": 1, + "save_steps": 200, + "stateful_callbacks": { + "TrainerControl": { + "args": { + "should_epoch_stop": false, + "should_evaluate": false, + "should_log": false, + "should_save": true, + "should_training_stop": true + }, + "attributes": {} + } + }, + "total_flos": 2.4144679757152256e+16, + "train_batch_size": 2, + "trial_name": null, + "trial_params": null +}