diff --git "a/final/trainer_state.json" "b/final/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/final/trainer_state.json"
@@ -0,0 +1,4185 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.23788528598340614,
+  "eval_steps": 500,
+  "global_step": 5935,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0004008176680428073,
+      "grad_norm": 0.8771801590919495,
+      "learning_rate": 2.910112359550562e-06,
+      "loss": 0.5885,
+      "step": 10
+    },
+    {
+      "epoch": 0.0008016353360856146,
+      "grad_norm": 0.7135661840438843,
+      "learning_rate": 3.921348314606742e-06,
+      "loss": 0.5855,
+      "step": 20
+    },
+    {
+      "epoch": 0.001202453004128422,
+      "grad_norm": 0.684427797794342,
+      "learning_rate": 4.932584269662922e-06,
+      "loss": 0.6184,
+      "step": 30
+    },
+    {
+      "epoch": 0.0016032706721712292,
+      "grad_norm": 0.6721655130386353,
+      "learning_rate": 5.943820224719102e-06,
+      "loss": 0.6311,
+      "step": 40
+    },
+    {
+      "epoch": 0.0020040883402140364,
+      "grad_norm": 0.6820604801177979,
+      "learning_rate": 6.955056179775282e-06,
+      "loss": 0.6092,
+      "step": 50
+    },
+    {
+      "epoch": 0.002404906008256844,
+      "grad_norm": 0.6614678502082825,
+      "learning_rate": 7.966292134831462e-06,
+      "loss": 0.6381,
+      "step": 60
+    },
+    {
+      "epoch": 0.0028057236762996512,
+      "grad_norm": 0.7028018236160278,
+      "learning_rate": 8.977528089887641e-06,
+      "loss": 0.5996,
+      "step": 70
+    },
+    {
+      "epoch": 0.0032065413443424584,
+      "grad_norm": 0.618502140045166,
+      "learning_rate": 9.988764044943822e-06,
+      "loss": 0.5959,
+      "step": 80
+    },
+    {
+      "epoch": 0.003607359012385266,
+      "grad_norm": 1.102985143661499,
+      "learning_rate": 1.1000000000000001e-05,
+      "loss": 0.5787,
+      "step": 90
+    },
+    {
+      "epoch": 0.004008176680428073,
+      "grad_norm": 0.9070648550987244,
+      "learning_rate": 1.2011235955056182e-05,
+      "loss": 0.6572,
+      "step": 100
+    },
+    {
+      "epoch": 0.004408994348470881,
+      "grad_norm": 0.8499509692192078,
+      "learning_rate": 1.3022471910112362e-05,
+      "loss": 0.5635,
+      "step": 110
+    },
+    {
+      "epoch": 0.004809812016513688,
+      "grad_norm": 0.8052905797958374,
+      "learning_rate": 1.403370786516854e-05,
+      "loss": 0.637,
+      "step": 120
+    },
+    {
+      "epoch": 0.005210629684556495,
+      "grad_norm": 0.5784770250320435,
+      "learning_rate": 1.504494382022472e-05,
+      "loss": 0.621,
+      "step": 130
+    },
+    {
+      "epoch": 0.0056114473525993025,
+      "grad_norm": 0.7975672483444214,
+      "learning_rate": 1.60561797752809e-05,
+      "loss": 0.6094,
+      "step": 140
+    },
+    {
+      "epoch": 0.00601226502064211,
+      "grad_norm": 0.7009092569351196,
+      "learning_rate": 1.706741573033708e-05,
+      "loss": 0.5613,
+      "step": 150
+    },
+    {
+      "epoch": 0.006413082688684917,
+      "grad_norm": 0.6548627018928528,
+      "learning_rate": 1.8078651685393256e-05,
+      "loss": 0.6051,
+      "step": 160
+    },
+    {
+      "epoch": 0.006813900356727725,
+      "grad_norm": 0.6045688390731812,
+      "learning_rate": 1.908988764044944e-05,
+      "loss": 0.563,
+      "step": 170
+    },
+    {
+      "epoch": 0.007214718024770532,
+      "grad_norm": 0.6471413373947144,
+      "learning_rate": 2e-05,
+      "loss": 0.6005,
+      "step": 180
+    },
+    {
+      "epoch": 0.007615535692813339,
+      "grad_norm": 0.629910945892334,
+      "learning_rate": 2e-05,
+      "loss": 0.6198,
+      "step": 190
+    },
+    {
+      "epoch": 0.008016353360856146,
+      "grad_norm": 0.927505373954773,
+      "learning_rate": 2e-05,
+      "loss": 0.582,
+      "step": 200
+    },
+    {
+      "epoch": 0.008417171028898954,
+      "grad_norm": 0.5309529304504395,
+      "learning_rate": 2e-05,
+      "loss": 0.5922,
+      "step": 210
+    },
+    {
+      "epoch": 0.008817988696941762,
+      "grad_norm": 0.5308235287666321,
+      "learning_rate": 2e-05,
+      "loss": 0.62,
+      "step": 220
+    },
+    {
+      "epoch": 0.009218806364984568,
+      "grad_norm": 0.8441914319992065,
+      "learning_rate": 2e-05,
+      "loss": 0.5574,
+      "step": 230
+    },
+    {
+      "epoch": 0.009619624033027376,
+      "grad_norm": 0.6029269099235535,
+      "learning_rate": 2e-05,
+      "loss": 0.5807,
+      "step": 240
+    },
+    {
+      "epoch": 0.010020441701070183,
+      "grad_norm": 0.5120773911476135,
+      "learning_rate": 2e-05,
+      "loss": 0.6125,
+      "step": 250
+    },
+    {
+      "epoch": 0.01042125936911299,
+      "grad_norm": 0.5556898713111877,
+      "learning_rate": 2e-05,
+      "loss": 0.608,
+      "step": 260
+    },
+    {
+      "epoch": 0.010822077037155799,
+      "grad_norm": 0.550104558467865,
+      "learning_rate": 2e-05,
+      "loss": 0.603,
+      "step": 270
+    },
+    {
+      "epoch": 0.011222894705198605,
+      "grad_norm": 0.46029090881347656,
+      "learning_rate": 2e-05,
+      "loss": 0.6234,
+      "step": 280
+    },
+    {
+      "epoch": 0.011623712373241413,
+      "grad_norm": 0.4434625208377838,
+      "learning_rate": 2e-05,
+      "loss": 0.5865,
+      "step": 290
+    },
+    {
+      "epoch": 0.01202453004128422,
+      "grad_norm": 0.49364206194877625,
+      "learning_rate": 2e-05,
+      "loss": 0.5991,
+      "step": 300
+    },
+    {
+      "epoch": 0.012425347709327027,
+      "grad_norm": 0.41587820649147034,
+      "learning_rate": 2e-05,
+      "loss": 0.5954,
+      "step": 310
+    },
+    {
+      "epoch": 0.012826165377369834,
+      "grad_norm": 0.4016185700893402,
+      "learning_rate": 2e-05,
+      "loss": 0.5823,
+      "step": 320
+    },
+    {
+      "epoch": 0.013226983045412642,
+      "grad_norm": 0.42542213201522827,
+      "learning_rate": 2e-05,
+      "loss": 0.629,
+      "step": 330
+    },
+    {
+      "epoch": 0.01362780071345545,
+      "grad_norm": 0.5087456703186035,
+      "learning_rate": 2e-05,
+      "loss": 0.6184,
+      "step": 340
+    },
+    {
+      "epoch": 0.014028618381498256,
+      "grad_norm": 0.46378353238105774,
+      "learning_rate": 2e-05,
+      "loss": 0.5894,
+      "step": 350
+    },
+    {
+      "epoch": 0.014429436049541064,
+      "grad_norm": 0.4463004171848297,
+      "learning_rate": 2e-05,
+      "loss": 0.6111,
+      "step": 360
+    },
+    {
+      "epoch": 0.01483025371758387,
+      "grad_norm": 0.5027539730072021,
+      "learning_rate": 2e-05,
+      "loss": 0.5779,
+      "step": 370
+    },
+    {
+      "epoch": 0.015231071385626679,
+      "grad_norm": 0.4367392361164093,
+      "learning_rate": 2e-05,
+      "loss": 0.6042,
+      "step": 380
+    },
+    {
+      "epoch": 0.015631889053669485,
+      "grad_norm": 0.4877663254737854,
+      "learning_rate": 2e-05,
+      "loss": 0.5847,
+      "step": 390
+    },
+    {
+      "epoch": 0.01603270672171229,
+      "grad_norm": 0.615467369556427,
+      "learning_rate": 2e-05,
+      "loss": 0.6312,
+      "step": 400
+    },
+    {
+      "epoch": 0.0164335243897551,
+      "grad_norm": 0.44901445508003235,
+      "learning_rate": 2e-05,
+      "loss": 0.5789,
+      "step": 410
+    },
+    {
+      "epoch": 0.016834342057797907,
+      "grad_norm": 0.5024731755256653,
+      "learning_rate": 2e-05,
+      "loss": 0.5949,
+      "step": 420
+    },
+    {
+      "epoch": 0.017235159725840714,
+      "grad_norm": 0.47739124298095703,
+      "learning_rate": 2e-05,
+      "loss": 0.6386,
+      "step": 430
+    },
+    {
+      "epoch": 0.017635977393883524,
+      "grad_norm": 0.3923758864402771,
+      "learning_rate": 2e-05,
+      "loss": 0.6097,
+      "step": 440
+    },
+    {
+      "epoch": 0.01803679506192633,
+      "grad_norm": 0.38089510798454285,
+      "learning_rate": 2e-05,
+      "loss": 0.5916,
+      "step": 450
+    },
+    {
+      "epoch": 0.018437612729969136,
+      "grad_norm": 0.37581101059913635,
+      "learning_rate": 2e-05,
+      "loss": 0.6205,
+      "step": 460
+    },
+    {
+      "epoch": 0.018838430398011946,
+      "grad_norm": 0.4052000343799591,
+      "learning_rate": 2e-05,
+      "loss": 0.6101,
+      "step": 470
+    },
+    {
+      "epoch": 0.019239248066054752,
+      "grad_norm": 0.4212661385536194,
+      "learning_rate": 2e-05,
+      "loss": 0.6051,
+      "step": 480
+    },
+    {
+      "epoch": 0.01964006573409756,
+      "grad_norm": 0.3834267556667328,
+      "learning_rate": 2e-05,
+      "loss": 0.5524,
+      "step": 490
+    },
+    {
+      "epoch": 0.020040883402140365,
+      "grad_norm": 0.38489559292793274,
+      "learning_rate": 2e-05,
+      "loss": 0.595,
+      "step": 500
+    },
+    {
+      "epoch": 0.020441701070183175,
+      "grad_norm": 0.42016276717185974,
+      "learning_rate": 2e-05,
+      "loss": 0.5996,
+      "step": 510
+    },
+    {
+      "epoch": 0.02084251873822598,
+      "grad_norm": 0.38279569149017334,
+      "learning_rate": 2e-05,
+      "loss": 0.586,
+      "step": 520
+    },
+    {
+      "epoch": 0.021243336406268788,
+      "grad_norm": 1.364205241203308,
+      "learning_rate": 2e-05,
+      "loss": 0.5536,
+      "step": 530
+    },
+    {
+      "epoch": 0.021644154074311597,
+      "grad_norm": 0.4498082995414734,
+      "learning_rate": 2e-05,
+      "loss": 0.5895,
+      "step": 540
+    },
+    {
+      "epoch": 0.022044971742354404,
+      "grad_norm": 0.38928699493408203,
+      "learning_rate": 2e-05,
+      "loss": 0.573,
+      "step": 550
+    },
+    {
+      "epoch": 0.02244578941039721,
+      "grad_norm": 0.39118388295173645,
+      "learning_rate": 2e-05,
+      "loss": 0.6096,
+      "step": 560
+    },
+    {
+      "epoch": 0.022846607078440016,
+      "grad_norm": 0.32673704624176025,
+      "learning_rate": 2e-05,
+      "loss": 0.5765,
+      "step": 570
+    },
+    {
+      "epoch": 0.023247424746482826,
+      "grad_norm": 0.36165913939476013,
+      "learning_rate": 2e-05,
+      "loss": 0.6114,
+      "step": 580
+    },
+    {
+      "epoch": 0.023648242414525632,
+      "grad_norm": 0.3407866954803467,
+      "learning_rate": 2e-05,
+      "loss": 0.5656,
+      "step": 590
+    },
+    {
+      "epoch": 0.02404906008256844,
+      "grad_norm": 0.3884231150150299,
+      "learning_rate": 2e-05,
+      "loss": 0.6137,
+      "step": 600
+    },
+    {
+      "epoch": 0.02444987775061125,
+      "grad_norm": 0.359744668006897,
+      "learning_rate": 2e-05,
+      "loss": 0.5952,
+      "step": 610
+    },
+    {
+      "epoch": 0.024850695418654055,
+      "grad_norm": 0.44357308745384216,
+      "learning_rate": 2e-05,
+      "loss": 0.5999,
+      "step": 620
+    },
+    {
+      "epoch": 0.02525151308669686,
+      "grad_norm": 0.3912670910358429,
+      "learning_rate": 2e-05,
+      "loss": 0.6191,
+      "step": 630
+    },
+    {
+      "epoch": 0.025652330754739668,
+      "grad_norm": 0.34180188179016113,
+      "learning_rate": 2e-05,
+      "loss": 0.597,
+      "step": 640
+    },
+    {
+      "epoch": 0.026053148422782477,
+      "grad_norm": 0.4104002118110657,
+      "learning_rate": 2e-05,
+      "loss": 0.5491,
+      "step": 650
+    },
+    {
+      "epoch": 0.026453966090825284,
+      "grad_norm": 0.3853461742401123,
+      "learning_rate": 2e-05,
+      "loss": 0.5516,
+      "step": 660
+    },
+    {
+      "epoch": 0.02685478375886809,
+      "grad_norm": 0.43015527725219727,
+      "learning_rate": 2e-05,
+      "loss": 0.6129,
+      "step": 670
+    },
+    {
+      "epoch": 0.0272556014269109,
+      "grad_norm": 0.3709610402584076,
+      "learning_rate": 2e-05,
+      "loss": 0.6228,
+      "step": 680
+    },
+    {
+      "epoch": 0.027656419094953706,
+      "grad_norm": 0.42639583349227905,
+      "learning_rate": 2e-05,
+      "loss": 0.6351,
+      "step": 690
+    },
+    {
+      "epoch": 0.028057236762996512,
+      "grad_norm": 0.402164489030838,
+      "learning_rate": 2e-05,
+      "loss": 0.6049,
+      "step": 700
+    },
+    {
+      "epoch": 0.02845805443103932,
+      "grad_norm": 0.6255981922149658,
+      "learning_rate": 2e-05,
+      "loss": 0.6139,
+      "step": 710
+    },
+    {
+      "epoch": 0.02885887209908213,
+      "grad_norm": 0.3667871952056885,
+      "learning_rate": 2e-05,
+      "loss": 0.5691,
+      "step": 720
+    },
+    {
+      "epoch": 0.029259689767124935,
+      "grad_norm": 0.36819183826446533,
+      "learning_rate": 2e-05,
+      "loss": 0.6111,
+      "step": 730
+    },
+    {
+      "epoch": 0.02966050743516774,
+      "grad_norm": 0.40006792545318604,
+      "learning_rate": 2e-05,
+      "loss": 0.613,
+      "step": 740
+    },
+    {
+      "epoch": 0.03006132510321055,
+      "grad_norm": 0.397798091173172,
+      "learning_rate": 2e-05,
+      "loss": 0.5872,
+      "step": 750
+    },
+    {
+      "epoch": 0.030462142771253357,
+      "grad_norm": 0.3581616282463074,
+      "learning_rate": 2e-05,
+      "loss": 0.5831,
+      "step": 760
+    },
+    {
+      "epoch": 0.030862960439296164,
+      "grad_norm": 0.3883630931377411,
+      "learning_rate": 2e-05,
+      "loss": 0.6182,
+      "step": 770
+    },
+    {
+      "epoch": 0.03126377810733897,
+      "grad_norm": 0.4035130739212036,
+      "learning_rate": 2e-05,
+      "loss": 0.58,
+      "step": 780
+    },
+    {
+      "epoch": 0.031664595775381776,
+      "grad_norm": 0.4688993990421295,
+      "learning_rate": 2e-05,
+      "loss": 0.5971,
+      "step": 790
+    },
+    {
+      "epoch": 0.03206541344342458,
+      "grad_norm": 0.37623828649520874,
+      "learning_rate": 2e-05,
+      "loss": 0.5682,
+      "step": 800
+    },
+    {
+      "epoch": 0.032466231111467396,
+      "grad_norm": 0.38642242550849915,
+      "learning_rate": 2e-05,
+      "loss": 0.6227,
+      "step": 810
+    },
+    {
+      "epoch": 0.0328670487795102,
+      "grad_norm": 0.3684792220592499,
+      "learning_rate": 2e-05,
+      "loss": 0.5819,
+      "step": 820
+    },
+    {
+      "epoch": 0.03326786644755301,
+      "grad_norm": 0.4786039888858795,
+      "learning_rate": 2e-05,
+      "loss": 0.5936,
+      "step": 830
+    },
+    {
+      "epoch": 0.033668684115595815,
+      "grad_norm": 0.3609310984611511,
+      "learning_rate": 2e-05,
+      "loss": 0.5941,
+      "step": 840
+    },
+    {
+      "epoch": 0.03406950178363862,
+      "grad_norm": 0.3404565751552582,
+      "learning_rate": 2e-05,
+      "loss": 0.6299,
+      "step": 850
+    },
+    {
+      "epoch": 0.03447031945168143,
+      "grad_norm": 0.3853689134120941,
+      "learning_rate": 2e-05,
+      "loss": 0.604,
+      "step": 860
+    },
+    {
+      "epoch": 0.03487113711972424,
+      "grad_norm": 0.3917834162712097,
+      "learning_rate": 2e-05,
+      "loss": 0.6053,
+      "step": 870
+    },
+    {
+      "epoch": 0.03527195478776705,
+      "grad_norm": 0.36823081970214844,
+      "learning_rate": 2e-05,
+      "loss": 0.6171,
+      "step": 880
+    },
+    {
+      "epoch": 0.035672772455809854,
+      "grad_norm": 0.3762242794036865,
+      "learning_rate": 2e-05,
+      "loss": 0.64,
+      "step": 890
+    },
+    {
+      "epoch": 0.03607359012385266,
+      "grad_norm": 0.8369109034538269,
+      "learning_rate": 2e-05,
+      "loss": 0.6405,
+      "step": 900
+    },
+    {
+      "epoch": 0.036474407791895466,
+      "grad_norm": 0.4162234365940094,
+      "learning_rate": 2e-05,
+      "loss": 0.6021,
+      "step": 910
+    },
+    {
+      "epoch": 0.03687522545993827,
+      "grad_norm": 0.4115590453147888,
+      "learning_rate": 2e-05,
+      "loss": 0.6031,
+      "step": 920
+    },
+    {
+      "epoch": 0.03727604312798108,
+      "grad_norm": 0.3745420277118683,
+      "learning_rate": 2e-05,
+      "loss": 0.5785,
+      "step": 930
+    },
+    {
+      "epoch": 0.03767686079602389,
+      "grad_norm": 0.38241758942604065,
+      "learning_rate": 2e-05,
+      "loss": 0.6068,
+      "step": 940
+    },
+    {
+      "epoch": 0.0380776784640667,
+      "grad_norm": 0.37572914361953735,
+      "learning_rate": 2e-05,
+      "loss": 0.6126,
+      "step": 950
+    },
+    {
+      "epoch": 0.038478496132109505,
+      "grad_norm": 0.3598463535308838,
+      "learning_rate": 2e-05,
+      "loss": 0.5647,
+      "step": 960
+    },
+    {
+      "epoch": 0.03887931380015231,
+      "grad_norm": 0.3645531237125397,
+      "learning_rate": 2e-05,
+      "loss": 0.5874,
+      "step": 970
+    },
+    {
+      "epoch": 0.03928013146819512,
+      "grad_norm": 0.37088748812675476,
+      "learning_rate": 2e-05,
+      "loss": 0.5966,
+      "step": 980
+    },
+    {
+      "epoch": 0.039680949136237924,
+      "grad_norm": 0.3901059031486511,
+      "learning_rate": 2e-05,
+      "loss": 0.6214,
+      "step": 990
+    },
+    {
+      "epoch": 0.04008176680428073,
+      "grad_norm": 0.3765749931335449,
+      "learning_rate": 2e-05,
+      "loss": 0.5771,
+      "step": 1000
+    },
+    {
+      "epoch": 0.04048258447232354,
+      "grad_norm": 0.3934212028980255,
+      "learning_rate": 2e-05,
+      "loss": 0.5661,
+      "step": 1010
+    },
+    {
+      "epoch": 0.04088340214036635,
+      "grad_norm": 0.3081901967525482,
+      "learning_rate": 2e-05,
+      "loss": 0.5772,
+      "step": 1020
+    },
+    {
+      "epoch": 0.041284219808409156,
+      "grad_norm": 0.5176882743835449,
+      "learning_rate": 2e-05,
+      "loss": 0.5925,
+      "step": 1030
+    },
+    {
+      "epoch": 0.04168503747645196,
+      "grad_norm": 0.3315117657184601,
+      "learning_rate": 2e-05,
+      "loss": 0.588,
+      "step": 1040
+    },
+    {
+      "epoch": 0.04208585514449477,
+      "grad_norm": 0.36918768286705017,
+      "learning_rate": 2e-05,
+      "loss": 0.6049,
+      "step": 1050
+    },
+    {
+      "epoch": 0.042486672812537575,
+      "grad_norm": 0.42960917949676514,
+      "learning_rate": 2e-05,
+      "loss": 0.6315,
+      "step": 1060
+    },
+    {
+      "epoch": 0.04288749048058038,
+      "grad_norm": 0.38764506578445435,
+      "learning_rate": 2e-05,
+      "loss": 0.5556,
+      "step": 1070
+    },
+    {
+      "epoch": 0.043288308148623195,
+      "grad_norm": 0.38045603036880493,
+      "learning_rate": 2e-05,
+      "loss": 0.578,
+      "step": 1080
+    },
+    {
+      "epoch": 0.043689125816666,
+      "grad_norm": 0.40927988290786743,
+      "learning_rate": 2e-05,
+      "loss": 0.5735,
+      "step": 1090
+    },
+    {
+      "epoch": 0.04408994348470881,
+      "grad_norm": 0.37116703391075134,
+      "learning_rate": 2e-05,
+      "loss": 0.5898,
+      "step": 1100
+    },
+    {
+      "epoch": 0.044490761152751614,
+      "grad_norm": 0.4863120913505554,
+      "learning_rate": 2e-05,
+      "loss": 0.6207,
+      "step": 1110
+    },
+    {
+      "epoch": 0.04489157882079442,
+      "grad_norm": 0.4080751836299896,
+      "learning_rate": 2e-05,
+      "loss": 0.6031,
+      "step": 1120
+    },
+    {
+      "epoch": 0.045292396488837226,
+      "grad_norm": 0.38050946593284607,
+      "learning_rate": 2e-05,
+      "loss": 0.5693,
+      "step": 1130
+    },
+    {
+      "epoch": 0.04569321415688003,
+      "grad_norm": 0.4002411365509033,
+      "learning_rate": 2e-05,
+      "loss": 0.5546,
+      "step": 1140
+    },
+    {
+      "epoch": 0.046094031824922846,
+      "grad_norm": 0.337520569562912,
+      "learning_rate": 2e-05,
+      "loss": 0.5649,
+      "step": 1150
+    },
+    {
+      "epoch": 0.04649484949296565,
+      "grad_norm": 0.4152458608150482,
+      "learning_rate": 2e-05,
+      "loss": 0.5985,
+      "step": 1160
+    },
+    {
+      "epoch": 0.04689566716100846,
+      "grad_norm": 0.3427383601665497,
+      "learning_rate": 2e-05,
+      "loss": 0.5974,
+      "step": 1170
+    },
+    {
+      "epoch": 0.047296484829051265,
+      "grad_norm": 0.48308777809143066,
+      "learning_rate": 2e-05,
+      "loss": 0.5676,
+      "step": 1180
+    },
+    {
+      "epoch": 0.04769730249709407,
+      "grad_norm": 0.3868969976902008,
+      "learning_rate": 2e-05,
+      "loss": 0.5791,
+      "step": 1190
+    },
+    {
+      "epoch": 0.04809812016513688,
+      "grad_norm": 0.4252076745033264,
+      "learning_rate": 2e-05,
+      "loss": 0.5949,
+      "step": 1200
+    },
+    {
+      "epoch": 0.048498937833179684,
+      "grad_norm": 0.3403521478176117,
+      "learning_rate": 2e-05,
+      "loss": 0.5721,
+      "step": 1210
+    },
+    {
+      "epoch": 0.0488997555012225,
+      "grad_norm": 0.3935258090496063,
+      "learning_rate": 2e-05,
+      "loss": 0.6194,
+      "step": 1220
+    },
+    {
+      "epoch": 0.049300573169265303,
+      "grad_norm": 0.39958739280700684,
+      "learning_rate": 2e-05,
+      "loss": 0.5838,
+      "step": 1230
+    },
+    {
+      "epoch": 0.04970139083730811,
+      "grad_norm": 0.3761332035064697,
+      "learning_rate": 2e-05,
+      "loss": 0.5746,
+      "step": 1240
+    },
+    {
+      "epoch": 0.050102208505350916,
+      "grad_norm": 0.3709917962551117,
+      "learning_rate": 2e-05,
+      "loss": 0.5925,
+      "step": 1250
+    },
+    {
+      "epoch": 0.05050302617339372,
+      "grad_norm": 0.4449377954006195,
+      "learning_rate": 2e-05,
+      "loss": 0.5975,
+      "step": 1260
+    },
+    {
+      "epoch": 0.05090384384143653,
+      "grad_norm": 0.4874597489833832,
+      "learning_rate": 2e-05,
+      "loss": 0.6075,
+      "step": 1270
+    },
+    {
+      "epoch": 0.051304661509479335,
+      "grad_norm": 0.5181390047073364,
+      "learning_rate": 2e-05,
+      "loss": 0.5873,
+      "step": 1280
+    },
+    {
+      "epoch": 0.05170547917752215,
+      "grad_norm": 0.38032063841819763,
+      "learning_rate": 2e-05,
+      "loss": 0.5734,
+      "step": 1290
+    },
+    {
+      "epoch": 0.052106296845564955,
+      "grad_norm": 0.3566950857639313,
+      "learning_rate": 2e-05,
+      "loss": 0.5876,
+      "step": 1300
+    },
+    {
+      "epoch": 0.05250711451360776,
+      "grad_norm": 0.3316211998462677,
+      "learning_rate": 2e-05,
+      "loss": 0.5729,
+      "step": 1310
+    },
+    {
+      "epoch": 0.05290793218165057,
+      "grad_norm": 0.3991880416870117,
+      "learning_rate": 2e-05,
+      "loss": 0.5903,
+      "step": 1320
+    },
+    {
+      "epoch": 0.053308749849693374,
+      "grad_norm": 0.4019741714000702,
+      "learning_rate": 2e-05,
+      "loss": 0.5768,
+      "step": 1330
+    },
+    {
+      "epoch": 0.05370956751773618,
+      "grad_norm": 0.417432963848114,
+      "learning_rate": 2e-05,
+      "loss": 0.6152,
+      "step": 1340
+    },
+    {
+      "epoch": 0.054110385185778986,
+      "grad_norm": 0.3510989248752594,
+      "learning_rate": 2e-05,
+      "loss": 0.5694,
+      "step": 1350
+    },
+    {
+      "epoch": 0.0545112028538218,
+      "grad_norm": 0.4420473575592041,
+      "learning_rate": 2e-05,
+      "loss": 0.5697,
+      "step": 1360
+    },
+    {
+      "epoch": 0.054912020521864606,
+      "grad_norm": 0.4099363684654236,
+      "learning_rate": 2e-05,
+      "loss": 0.5891,
+      "step": 1370
+    },
+    {
+      "epoch": 0.05531283818990741,
+      "grad_norm": 0.44445106387138367,
+      "learning_rate": 2e-05,
+      "loss": 0.5602,
+      "step": 1380
+    },
+    {
+      "epoch": 0.05571365585795022,
+      "grad_norm": 0.35209330916404724,
+      "learning_rate": 2e-05,
+      "loss": 0.6442,
+      "step": 1390
+    },
+    {
+      "epoch": 0.056114473525993025,
+      "grad_norm": 0.3788399398326874,
+      "learning_rate": 2e-05,
+      "loss": 0.6103,
+      "step": 1400
+    },
+    {
+      "epoch": 0.05651529119403583,
+      "grad_norm": 0.3592466711997986,
+      "learning_rate": 2e-05,
+      "loss": 0.5739,
+      "step": 1410
+    },
+    {
+      "epoch": 0.05691610886207864,
+      "grad_norm": 0.3659421503543854,
+      "learning_rate": 2e-05,
+      "loss": 0.5754,
+      "step": 1420
+    },
+    {
+      "epoch": 0.05731692653012145,
+      "grad_norm": 0.33240798115730286,
+      "learning_rate": 2e-05,
+      "loss": 0.5657,
+      "step": 1430
+    },
+    {
+      "epoch": 0.05771774419816426,
+      "grad_norm": 0.3574551045894623,
+      "learning_rate": 2e-05,
+      "loss": 0.599,
+      "step": 1440
+    },
+    {
+      "epoch": 0.058118561866207064,
+      "grad_norm": 0.3870832920074463,
+      "learning_rate": 2e-05,
+      "loss": 0.586,
+      "step": 1450
+    },
+    {
+      "epoch": 0.05851937953424987,
+      "grad_norm": 0.3373236656188965,
+      "learning_rate": 2e-05,
+      "loss": 0.6046,
+      "step": 1460
+    },
+    {
+      "epoch": 0.058920197202292676,
+      "grad_norm": 0.3652302026748657,
+      "learning_rate": 2e-05,
+      "loss": 0.5895,
+      "step": 1470
+    },
+    {
+      "epoch": 0.05932101487033548,
+      "grad_norm": 0.3693019449710846,
+      "learning_rate": 2e-05,
+      "loss": 0.6038,
+      "step": 1480
+    },
+    {
+      "epoch": 0.05972183253837829,
+      "grad_norm": 0.4179181158542633,
+      "learning_rate": 2e-05,
+      "loss": 0.5379,
+      "step": 1490
+    },
+    {
+      "epoch": 0.0601226502064211,
+      "grad_norm": 0.3929535746574402,
+      "learning_rate": 2e-05,
+      "loss": 0.5814,
+      "step": 1500
+    },
+    {
+      "epoch": 0.06052346787446391,
+      "grad_norm": 0.34511131048202515,
+      "learning_rate": 2e-05,
+      "loss": 0.5608,
+      "step": 1510
+    },
+    {
+      "epoch": 0.060924285542506715,
+      "grad_norm": 0.3240802586078644,
+      "learning_rate": 2e-05,
+      "loss": 0.6026,
+      "step": 1520
+    },
+    {
+      "epoch": 0.06132510321054952,
+      "grad_norm": 0.3732409179210663,
+      "learning_rate": 2e-05,
+      "loss": 0.5922,
+      "step": 1530
+    },
+    {
+      "epoch": 0.06172592087859233,
+      "grad_norm": 0.3403308391571045,
+      "learning_rate": 2e-05,
+      "loss": 0.5784,
+      "step": 1540
+    },
+    {
+      "epoch": 0.062126738546635134,
+      "grad_norm": 0.43892017006874084,
+      "learning_rate": 2e-05,
+      "loss": 0.5733,
+      "step": 1550
+    },
+    {
+      "epoch": 0.06252755621467794,
+      "grad_norm": 0.32453539967536926,
+      "learning_rate": 2e-05,
+      "loss": 0.6164,
+      "step": 1560
+    },
+    {
+      "epoch": 0.06292837388272075,
+      "grad_norm": 0.4057670533657074,
+      "learning_rate": 2e-05,
+      "loss": 0.5654,
+      "step": 1570
+    },
+    {
+      "epoch": 0.06332919155076355,
+      "grad_norm": 0.401593953371048,
+      "learning_rate": 2e-05,
+      "loss": 0.5708,
+      "step": 1580
+    },
+    {
+      "epoch": 0.06373000921880637,
+      "grad_norm": 0.9239656329154968,
+      "learning_rate": 2e-05,
+      "loss": 0.6037,
+      "step": 1590
+    },
+    {
+      "epoch": 0.06413082688684917,
+      "grad_norm": 0.43078750371932983,
+      "learning_rate": 2e-05,
+      "loss": 0.5834,
+      "step": 1600
+    },
+    {
+      "epoch": 0.06453164455489198,
+      "grad_norm": 0.3466830253601074,
+      "learning_rate": 2e-05,
+      "loss": 0.5737,
+      "step": 1610
+    },
+    {
+      "epoch": 0.06493246222293479,
+      "grad_norm": 0.3531036078929901,
+      "learning_rate": 2e-05,
+      "loss": 0.5849,
+      "step": 1620
+    },
+    {
+      "epoch": 0.06533327989097759,
+      "grad_norm": 0.5453006029129028,
+      "learning_rate": 2e-05,
+      "loss": 0.5575,
+      "step": 1630
+    },
+    {
+      "epoch": 0.0657340975590204,
+      "grad_norm": 0.3632678687572479,
+      "learning_rate": 2e-05,
+      "loss": 0.61,
+      "step": 1640
+    },
+    {
+      "epoch": 0.0661349152270632,
+      "grad_norm": 0.410371869802475,
+      "learning_rate": 2e-05,
+      "loss": 0.5943,
+      "step": 1650
+    },
+    {
+      "epoch": 0.06653573289510602,
+      "grad_norm": 0.3581090569496155,
+      "learning_rate": 2e-05,
+      "loss": 0.583,
+      "step": 1660
+    },
+    {
+      "epoch": 0.06693655056314882,
+      "grad_norm": 0.3194616436958313,
+      "learning_rate": 2e-05,
+      "loss": 0.5872,
+      "step": 1670
+    },
+    {
+      "epoch": 0.06733736823119163,
+      "grad_norm": 0.3554430603981018,
+      "learning_rate": 2e-05,
+      "loss": 0.6039,
+      "step": 1680
+    },
+    {
+      "epoch": 0.06773818589923444,
+      "grad_norm": 0.3838396966457367,
+      "learning_rate": 2e-05,
+      "loss": 0.6044,
+      "step": 1690
+    },
+    {
+      "epoch": 0.06813900356727724,
+      "grad_norm": 0.3924729824066162,
+      "learning_rate": 2e-05,
+      "loss": 0.6208,
+      "step": 1700
+    },
+    {
+      "epoch": 0.06853982123532006,
+      "grad_norm": 0.3942250907421112,
+      "learning_rate": 2e-05,
+      "loss": 0.5827,
+      "step": 1710
+    },
+    {
+      "epoch": 0.06894063890336286,
+      "grad_norm": 0.3890213072299957,
+      "learning_rate": 2e-05,
+      "loss": 0.5413,
+      "step": 1720
+    },
+    {
+      "epoch": 0.06934145657140567,
+      "grad_norm": 0.3526347577571869,
+      "learning_rate": 2e-05,
+      "loss": 0.5835,
+      "step": 1730
+    },
+    {
+      "epoch": 0.06974227423944848,
+      "grad_norm": 0.3879394829273224,
+      "learning_rate": 2e-05,
+      "loss": 0.5952,
+      "step": 1740
+    },
+    {
+      "epoch": 0.07014309190749128,
+      "grad_norm": 0.33095234632492065,
+      "learning_rate": 2e-05,
+      "loss": 0.5891,
+      "step": 1750
+    },
+    {
+      "epoch": 0.0705439095755341,
+      "grad_norm": 0.38446444272994995,
+      "learning_rate": 2e-05,
+      "loss": 0.5971,
+      "step": 1760
+    },
+    {
+      "epoch": 0.0709447272435769,
+      "grad_norm": 0.4128134250640869,
+      "learning_rate": 2e-05,
+      "loss": 0.5613,
+      "step": 1770
+    },
+    {
+      "epoch": 0.07134554491161971,
+      "grad_norm": 0.33474040031433105,
+      "learning_rate": 2e-05,
+      "loss": 0.5702,
+      "step": 1780
+    },
+    {
+      "epoch": 0.0717463625796625,
+      "grad_norm": 0.36933767795562744,
+      "learning_rate": 2e-05,
+      "loss": 0.5887,
+      "step": 1790
+    },
+    {
+      "epoch": 0.07214718024770532,
+      "grad_norm": 0.40623337030410767,
+      "learning_rate": 2e-05,
+      "loss": 0.5752,
+      "step": 1800
+    },
+    {
+      "epoch": 0.07254799791574813,
+      "grad_norm": 0.34252020716667175,
+      "learning_rate": 2e-05,
+      "loss": 0.5703,
+      "step": 1810
+    },
+    {
+      "epoch": 0.07294881558379093,
+      "grad_norm": 0.3257846236228943,
+      "learning_rate": 2e-05,
+      "loss": 0.5845,
+      "step": 1820
+    },
+    {
+      "epoch": 0.07334963325183375,
+      "grad_norm": 0.3836072087287903,
+      "learning_rate": 2e-05,
+      "loss": 0.5863,
+      "step": 1830
+    },
+    {
+      "epoch": 0.07375045091987655,
+      "grad_norm": 0.4434497058391571,
+      "learning_rate": 2e-05,
+      "loss": 0.5607,
+      "step": 1840
+    },
+    {
+      "epoch": 0.07415126858791936,
+      "grad_norm": 0.369035929441452,
+      "learning_rate": 2e-05,
+      "loss": 0.5759,
+      "step": 1850
+    },
+    {
+      "epoch": 0.07455208625596216,
+      "grad_norm": 0.4021732807159424,
+      "learning_rate": 2e-05,
+      "loss": 0.5941,
+      "step": 1860
+    },
+    {
+      "epoch": 0.07495290392400497,
+      "grad_norm": 0.343211829662323,
+      "learning_rate": 2e-05,
+      "loss": 0.6035,
+      "step": 1870
+    },
+    {
+      "epoch": 0.07535372159204778,
+      "grad_norm": 0.3816221356391907,
+      "learning_rate": 2e-05,
+      "loss": 0.5661,
+      "step": 1880
+    },
+    {
+      "epoch": 0.07575453926009058,
+      "grad_norm": 0.35929223895072937,
+      "learning_rate": 2e-05,
+      "loss": 0.5867,
+      "step": 1890
+    },
+    {
+      "epoch": 0.0761553569281334,
+      "grad_norm": 0.36963996291160583,
+      "learning_rate": 2e-05,
+      "loss": 0.5842,
+      "step": 1900
+    },
+    {
+      "epoch": 0.0765561745961762,
+      "grad_norm": 0.4415839612483978,
+      "learning_rate": 2e-05,
+      "loss": 0.6102,
+      "step": 1910
+    },
+    {
+      "epoch": 0.07695699226421901,
+      "grad_norm": 0.3494766354560852,
+      "learning_rate": 2e-05,
+      "loss": 0.5586,
+      "step": 1920
+    },
+    {
+      "epoch": 0.07735780993226181,
+      "grad_norm": 0.434241384267807,
+      "learning_rate": 2e-05,
+      "loss": 0.609,
+      "step": 1930
+    },
+    {
+      "epoch": 0.07775862760030462,
+      "grad_norm": 0.39981046319007874,
+      "learning_rate": 2e-05,
+      "loss": 0.5942,
+      "step": 1940
+    },
+    {
+      "epoch": 0.07815944526834744,
+      "grad_norm": 0.3525627553462982,
+      "learning_rate": 2e-05,
+      "loss": 0.5905,
+      "step": 1950
+    },
+    {
+      "epoch": 0.07856026293639023,
+      "grad_norm": 0.48807239532470703,
+      "learning_rate": 2e-05,
+      "loss": 0.5696,
+      "step": 1960
+    },
+    {
+      "epoch": 0.07896108060443305,
+      "grad_norm": 0.35398364067077637,
+      "learning_rate": 2e-05,
+      "loss": 0.5505,
+      "step": 1970
+    },
+    {
+      "epoch": 0.07936189827247585,
+      "grad_norm": 0.4212177097797394,
+      "learning_rate": 2e-05,
+      "loss": 0.582,
+      "step": 1980
+    },
+    {
+      "epoch": 0.07976271594051866,
+      "grad_norm": 0.33485740423202515,
+      "learning_rate": 2e-05,
+      "loss": 0.5564,
+      "step": 1990
+    },
+    {
+      "epoch": 0.08016353360856146,
+      "grad_norm": 0.39947226643562317,
+      "learning_rate": 2e-05,
+      "loss": 0.562,
+      "step": 2000
+    },
+    {
+      "epoch": 0.08056435127660427,
+      "grad_norm": 0.3385741412639618,
+      "learning_rate": 2e-05,
+      "loss": 0.6098,
+      "step": 2010
+    },
+    {
+      "epoch": 0.08096516894464709,
+      "grad_norm": 0.3568948209285736,
+      "learning_rate": 2e-05,
+      "loss": 0.5853,
+      "step": 2020
+    },
+    {
+      "epoch": 0.08136598661268989,
+      "grad_norm": 0.36856064200401306,
+      "learning_rate": 2e-05,
+      "loss": 0.5616,
+      "step": 2030
+    },
+    {
+      "epoch": 0.0817668042807327,
+      "grad_norm": 0.36307284235954285,
+      "learning_rate": 2e-05,
+      "loss": 0.581,
+      "step": 2040
+    },
+    {
+      "epoch": 0.0821676219487755,
+      "grad_norm": 0.3536205589771271,
+      "learning_rate": 2e-05,
+      "loss": 0.5635,
+      "step": 2050
+    },
+    {
+      "epoch": 0.08256843961681831,
+      "grad_norm": 0.43149474263191223,
+      "learning_rate": 2e-05,
+      "loss": 0.6025,
+      "step": 2060
+    },
+    {
+      "epoch": 0.08296925728486111,
+      "grad_norm": 0.3605656325817108,
+      "learning_rate": 2e-05,
+      "loss": 0.5634,
+      "step": 2070
+    },
+    {
+      "epoch": 0.08337007495290392,
+      "grad_norm": 0.393462598323822,
+      "learning_rate": 2e-05,
+      "loss": 0.596,
+      "step": 2080
+    },
+    {
+      "epoch": 0.08377089262094674,
+      "grad_norm": 0.318276047706604,
+      "learning_rate": 2e-05,
+      "loss": 0.5504,
+      "step": 2090
+    },
+    {
+      "epoch": 0.08417171028898954,
+      "grad_norm": 0.46484410762786865,
+      "learning_rate": 2e-05,
+      "loss": 0.5638,
+      "step": 2100
+    },
+    {
+      "epoch": 0.08457252795703235,
+      "grad_norm": 0.4350844919681549,
+      "learning_rate": 2e-05,
+      "loss": 0.5854,
+      "step": 2110
+    },
+    {
+      "epoch": 0.08497334562507515,
+      "grad_norm": 0.3690831661224365,
+      "learning_rate": 2e-05,
+      "loss": 0.5923,
+      "step": 2120
+    },
+    {
+      "epoch": 0.08537416329311796,
+      "grad_norm": 0.32239291071891785,
+      "learning_rate": 2e-05,
+      "loss": 0.557,
+      "step": 2130
+    },
+    {
+      "epoch": 0.08577498096116076,
+      "grad_norm": 0.37218692898750305,
+      "learning_rate": 2e-05,
+      "loss": 0.5719,
+      "step": 2140
+    },
+    {
+      "epoch": 0.08617579862920358,
+      "grad_norm": 0.3764131963253021,
+      "learning_rate": 2e-05,
+      "loss": 0.5597,
+      "step": 2150
+    },
+    {
+      "epoch": 0.08657661629724639,
+      "grad_norm": 0.3383973240852356,
+      "learning_rate": 2e-05,
+      "loss": 0.5722,
+      "step": 2160
+    },
+    {
+      "epoch": 0.08697743396528919,
+      "grad_norm": 0.4094143211841583,
+      "learning_rate": 2e-05,
+      "loss": 0.57,
+      "step": 2170
+    },
+    {
+      "epoch": 0.087378251633332,
+      "grad_norm": 0.406859427690506,
+      "learning_rate": 2e-05,
+      "loss": 0.5704,
+      "step": 2180
+    },
+    {
+      "epoch": 0.0877790693013748,
+      "grad_norm": 0.3579443395137787,
+      "learning_rate": 2e-05,
+      "loss": 0.5896,
+      "step": 2190
+    },
+    {
+      "epoch": 0.08817988696941761,
+      "grad_norm": 0.3546210527420044,
+      "learning_rate": 2e-05,
+      "loss": 0.559,
+      "step": 2200
+    },
+    {
+      "epoch": 0.08858070463746041,
+      "grad_norm": 0.3182368278503418,
+      "learning_rate": 2e-05,
+      "loss": 0.5713,
+      "step": 2210
+    },
+    {
+      "epoch": 0.08898152230550323,
+      "grad_norm": 0.3846684396266937,
+      "learning_rate": 2e-05,
+      "loss": 0.5724,
+      "step": 2220
+    },
+    {
+      "epoch": 0.08938233997354604,
+      "grad_norm": 0.37794479727745056,
+      "learning_rate": 2e-05,
+      "loss": 0.6047,
+      "step": 2230
+    },
+    {
+      "epoch": 0.08978315764158884,
+      "grad_norm": 0.31841838359832764,
+      "learning_rate": 2e-05,
+      "loss": 0.5544,
+      "step": 2240
+    },
+    {
+      "epoch": 0.09018397530963165,
+      "grad_norm": 0.37598782777786255,
+      "learning_rate": 2e-05,
+      "loss": 0.5687,
+      "step": 2250
+    },
+    {
+      "epoch": 0.09058479297767445,
+      "grad_norm": 1.1034108400344849,
+      "learning_rate": 2e-05,
+      "loss": 0.5758,
+      "step": 2260
+    },
+    {
+      "epoch": 0.09098561064571727,
+      "grad_norm": 0.34686070680618286,
+      "learning_rate": 2e-05,
+      "loss": 0.5782,
+      "step": 2270
+    },
+    {
+      "epoch": 0.09138642831376007,
+      "grad_norm": 0.34697771072387695,
+      "learning_rate": 2e-05,
+      "loss": 0.5602,
+      "step": 2280
+    },
+    {
+      "epoch": 0.09178724598180288,
+      "grad_norm": 0.370018869638443,
+      "learning_rate": 2e-05,
+      "loss": 0.5912,
+      "step": 2290
+    },
+    {
+      "epoch": 0.09218806364984569,
+      "grad_norm": 0.390052855014801,
+      "learning_rate": 2e-05,
+      "loss": 0.5788,
+      "step": 2300
+    },
+    {
+      "epoch": 0.09258888131788849,
+      "grad_norm": 0.3554167151451111,
+      "learning_rate": 2e-05,
+      "loss": 0.56,
+      "step": 2310
+    },
+    {
+      "epoch": 0.0929896989859313,
+      "grad_norm": 0.343339204788208,
+      "learning_rate": 2e-05,
+      "loss": 0.6,
+      "step": 2320
+    },
+    {
+      "epoch": 0.0933905166539741,
+      "grad_norm": 0.3061367869377136,
+      "learning_rate": 2e-05,
+      "loss": 0.604,
+      "step": 2330
+    },
+    {
+      "epoch": 0.09379133432201692,
+      "grad_norm": 0.3962571322917938,
+      "learning_rate": 2e-05,
+      "loss": 0.5975,
+      "step": 2340
+    },
+    {
+      "epoch": 0.09419215199005972,
+      "grad_norm": 0.3495456576347351,
+      "learning_rate": 2e-05,
+      "loss": 0.5676,
+      "step": 2350
+    },
+    {
+      "epoch": 0.09459296965810253,
+      "grad_norm": 0.964829683303833,
+      "learning_rate": 2e-05,
+      "loss": 0.5851,
+      "step": 2360
+    },
+    {
+      "epoch": 0.09499378732614534,
+      "grad_norm": 0.4385446012020111,
+      "learning_rate": 2e-05,
+      "loss": 0.5538,
+      "step": 2370
+    },
+    {
+      "epoch": 0.09539460499418814,
+      "grad_norm": 0.37227344512939453,
+      "learning_rate": 2e-05,
+      "loss": 0.5631,
+      "step": 2380
+    },
+    {
+      "epoch": 0.09579542266223096,
+      "grad_norm": 0.3347318470478058,
+      "learning_rate": 2e-05,
+      "loss": 0.5681,
+      "step": 2390
+    },
+    {
+      "epoch": 0.09619624033027376,
+      "grad_norm": 0.3704676032066345,
+      "learning_rate": 2e-05,
+      "loss": 0.5869,
+      "step": 2400
+    },
+    {
+      "epoch": 0.09659705799831657,
+      "grad_norm": 0.35813140869140625,
+      "learning_rate": 2e-05,
+      "loss": 0.5553,
+      "step": 2410
+    },
+    {
+      "epoch": 0.09699787566635937,
+      "grad_norm": 0.3712270259857178,
+      "learning_rate": 2e-05,
+      "loss": 0.5476,
+      "step": 2420
+    },
+    {
+      "epoch": 0.09739869333440218,
+      "grad_norm": 0.4175005257129669,
+      "learning_rate": 2e-05,
+      "loss": 0.6143,
+      "step": 2430
+    },
+    {
+      "epoch": 0.097799511002445,
+      "grad_norm": 0.36217042803764343,
+      "learning_rate": 2e-05,
+      "loss": 0.5752,
+      "step": 2440
+    },
+    {
+      "epoch": 0.0982003286704878,
+      "grad_norm": 0.3934085965156555,
+      "learning_rate": 2e-05,
+      "loss": 0.5822,
+      "step": 2450
+    },
+    {
+      "epoch": 0.09860114633853061,
+      "grad_norm": 0.3451460599899292,
+      "learning_rate": 2e-05,
+      "loss": 0.5783,
+      "step": 2460
+    },
+    {
+      "epoch": 0.0990019640065734,
+      "grad_norm": 0.35573357343673706,
+      "learning_rate": 2e-05,
+      "loss": 0.5586,
+      "step": 2470
+    },
+    {
+      "epoch": 0.09940278167461622,
+      "grad_norm": 0.34175917506217957,
+      "learning_rate": 2e-05,
+      "loss": 0.5564,
+      "step": 2480
+    },
+    {
+      "epoch": 0.09980359934265902,
+      "grad_norm": 0.3565382659435272,
+      "learning_rate": 2e-05,
+      "loss": 0.5524,
+      "step": 2490
+    },
+    {
+      "epoch": 0.10020441701070183,
+      "grad_norm": 0.4152110517024994,
+      "learning_rate": 2e-05,
+      "loss": 0.5831,
+      "step": 2500
+    },
+    {
+      "epoch": 0.10060523467874465,
+      "grad_norm": 0.3308890163898468,
+      "learning_rate": 2e-05,
+      "loss": 0.5485,
+      "step": 2510
+    },
+    {
+      "epoch": 0.10100605234678744,
+      "grad_norm": 0.45242831110954285,
+      "learning_rate": 2e-05,
+      "loss": 0.5838,
+      "step": 2520
+    },
+    {
+      "epoch": 0.10140687001483026,
+      "grad_norm": 0.37882164120674133,
+      "learning_rate": 2e-05,
+      "loss": 0.5674,
+      "step": 2530
+    },
+    {
+      "epoch": 0.10180768768287306,
+      "grad_norm": 0.37527206540107727,
+      "learning_rate": 2e-05,
+      "loss": 0.56,
+      "step": 2540
+    },
+    {
+      "epoch": 0.10220850535091587,
+      "grad_norm": 0.45134904980659485,
+      "learning_rate": 2e-05,
+      "loss": 0.5398,
+      "step": 2550
+    },
+    {
+      "epoch": 0.10260932301895867,
+      "grad_norm": 0.44729816913604736,
+      "learning_rate": 2e-05,
+      "loss": 0.5569,
+      "step": 2560
+    },
+    {
+      "epoch": 0.10301014068700148,
+      "grad_norm": 0.29179856181144714,
+      "learning_rate": 2e-05,
+      "loss": 0.5806,
+      "step": 2570
+    },
+    {
+      "epoch": 0.1034109583550443,
+      "grad_norm": 0.3566204905509949,
+      "learning_rate": 2e-05,
+      "loss": 0.5792,
+      "step": 2580
+    },
+    {
+      "epoch": 0.1038117760230871,
+      "grad_norm": 0.34388139843940735,
+      "learning_rate": 2e-05,
+      "loss": 0.5607,
+      "step": 2590
+    },
+    {
+      "epoch": 0.10421259369112991,
+      "grad_norm": 0.4333122670650482,
+      "learning_rate": 2e-05,
+      "loss": 0.5567,
+      "step": 2600
+    },
+    {
+      "epoch": 0.10461341135917271,
+      "grad_norm": 0.3270779252052307,
+      "learning_rate": 2e-05,
+      "loss": 0.572,
+      "step": 2610
+    },
+    {
+      "epoch": 0.10501422902721552,
+      "grad_norm": 0.3311755657196045,
+      "learning_rate": 2e-05,
+      "loss": 0.5705,
+      "step": 2620
+    },
+    {
+      "epoch": 0.10541504669525832,
+      "grad_norm": 0.355700820684433,
+      "learning_rate": 2e-05,
+      "loss": 0.5679,
+      "step": 2630
+    },
+    {
+      "epoch": 0.10581586436330113,
+      "grad_norm": 0.3758179247379303,
+      "learning_rate": 2e-05,
+      "loss": 0.5837,
+      "step": 2640
+    },
+    {
+      "epoch": 0.10621668203134395,
+      "grad_norm": 0.4235946536064148,
+      "learning_rate": 2e-05,
+      "loss": 0.5706,
+      "step": 2650
+    },
+    {
+      "epoch": 0.10661749969938675,
+      "grad_norm": 0.4534910023212433,
+      "learning_rate": 2e-05,
+      "loss": 0.545,
+      "step": 2660
+    },
+    {
+      "epoch": 0.10701831736742956,
+      "grad_norm": 0.3604424297809601,
+      "learning_rate": 2e-05,
+      "loss": 0.5662,
+      "step": 2670
+    },
+    {
+      "epoch": 0.10741913503547236,
+      "grad_norm": 0.32635748386383057,
+      "learning_rate": 2e-05,
+      "loss": 0.5581,
+      "step": 2680
+    },
+    {
+      "epoch": 0.10781995270351517,
+      "grad_norm": 0.3974153697490692,
+      "learning_rate": 2e-05,
+      "loss": 0.5658,
+      "step": 2690
+    },
+    {
+      "epoch": 0.10822077037155797,
+      "grad_norm": 0.37879955768585205,
+      "learning_rate": 2e-05,
+      "loss": 0.5591,
+      "step": 2700
+    },
+    {
+      "epoch": 0.10862158803960079,
+      "grad_norm": 0.39657601714134216,
+      "learning_rate": 2e-05,
+      "loss": 0.5888,
+      "step": 2710
+    },
+    {
+      "epoch": 0.1090224057076436,
+      "grad_norm": 0.38731491565704346,
+      "learning_rate": 2e-05,
+      "loss": 0.5804,
+      "step": 2720
+    },
+    {
+      "epoch": 0.1094232233756864,
+      "grad_norm": 0.5076712965965271,
+      "learning_rate": 2e-05,
+      "loss": 0.5961,
+      "step": 2730
+    },
+    {
+      "epoch": 0.10982404104372921,
+      "grad_norm": 0.3596253991127014,
+      "learning_rate": 2e-05,
+      "loss": 0.5465,
+      "step": 2740
+    },
+    {
+      "epoch": 0.11022485871177201,
+      "grad_norm": 0.6818852424621582,
+      "learning_rate": 2e-05,
+      "loss": 0.5708,
+      "step": 2750
+    },
+    {
+      "epoch": 0.11062567637981482,
+      "grad_norm": 0.3373611569404602,
+      "learning_rate": 2e-05,
+      "loss": 0.5681,
+      "step": 2760
+    },
+    {
+      "epoch": 0.11102649404785762,
+      "grad_norm": 0.41528621315956116,
+      "learning_rate": 2e-05,
+      "loss": 0.5874,
+      "step": 2770
+    },
+    {
+      "epoch": 0.11142731171590044,
+      "grad_norm": 0.3721466064453125,
+      "learning_rate": 2e-05,
+      "loss": 0.5433,
+      "step": 2780
+    },
+    {
+      "epoch": 0.11182812938394325,
+      "grad_norm": 0.3605429232120514,
+      "learning_rate": 2e-05,
+      "loss": 0.5582,
+      "step": 2790
+    },
+    {
+      "epoch": 0.11222894705198605,
+      "grad_norm": 0.4084005653858185,
+      "learning_rate": 2e-05,
+      "loss": 0.6014,
+      "step": 2800
+    },
+    {
+      "epoch": 0.11262976472002886,
+      "grad_norm": 0.35510143637657166,
+      "learning_rate": 2e-05,
+      "loss": 0.5312,
+      "step": 2810
+    },
+    {
+      "epoch": 0.11303058238807166,
+      "grad_norm": 0.4171420633792877,
+      "learning_rate": 2e-05,
+      "loss": 0.5833,
+      "step": 2820
+    },
+    {
+      "epoch": 0.11343140005611448,
+      "grad_norm": 0.38897544145584106,
+      "learning_rate": 2e-05,
+      "loss": 0.543,
+      "step": 2830
+    },
+    {
+      "epoch": 0.11383221772415728,
+      "grad_norm": 0.3851448595523834,
+      "learning_rate": 2e-05,
+      "loss": 0.5484,
+      "step": 2840
+    },
+    {
+      "epoch": 0.11423303539220009,
+      "grad_norm": 0.41326257586479187,
+      "learning_rate": 2e-05,
+      "loss": 0.5719,
+      "step": 2850
+    },
+    {
+      "epoch": 0.1146338530602429,
+      "grad_norm": 0.37410998344421387,
+      "learning_rate": 2e-05,
+      "loss": 0.5597,
+      "step": 2860
+    },
+    {
+      "epoch": 0.1150346707282857,
+      "grad_norm": 0.4749361574649811,
+      "learning_rate": 2e-05,
+      "loss": 0.583,
+      "step": 2870
+    },
+    {
+      "epoch": 0.11543548839632851,
+      "grad_norm": 0.4170619547367096,
+      "learning_rate": 2e-05,
+      "loss": 0.5713,
+      "step": 2880
+    },
+    {
+      "epoch": 0.11583630606437131,
+      "grad_norm": 0.41376549005508423,
+      "learning_rate": 2e-05,
+      "loss": 0.5582,
+      "step": 2890
+    },
+    {
+      "epoch": 0.11623712373241413,
+      "grad_norm": 0.37296295166015625,
+      "learning_rate": 2e-05,
+      "loss": 0.5574,
+      "step": 2900
+    },
+    {
+      "epoch": 0.11663794140045693,
+      "grad_norm": 0.3766053020954132,
+      "learning_rate": 2e-05,
+      "loss": 0.5911,
+      "step": 2910
+    },
+    {
+      "epoch": 0.11703875906849974,
+      "grad_norm": 0.40055564045906067,
+      "learning_rate": 2e-05,
+      "loss": 0.6162,
+      "step": 2920
+    },
+    {
+      "epoch": 0.11743957673654255,
+      "grad_norm": 0.34108811616897583,
+      "learning_rate": 2e-05,
+      "loss": 0.5413,
+      "step": 2930
+    },
+    {
+      "epoch": 0.11784039440458535,
+      "grad_norm": 0.3507615327835083,
+      "learning_rate": 2e-05,
+      "loss": 0.5779,
+      "step": 2940
+    },
+    {
+      "epoch": 0.11824121207262817,
+      "grad_norm": 0.39660125970840454,
+      "learning_rate": 2e-05,
+      "loss": 0.613,
+      "step": 2950
+    },
+    {
+      "epoch": 0.11864202974067097,
+      "grad_norm": 0.35685157775878906,
+      "learning_rate": 2e-05,
+      "loss": 0.5553,
+      "step": 2960
+    },
+    {
+      "epoch": 0.11904284740871378,
+      "grad_norm": 0.3450596332550049,
+      "learning_rate": 2e-05,
+      "loss": 0.5538,
+      "step": 2970
+    },
+    {
+      "epoch": 0.11944366507675658,
+      "grad_norm": 0.36301591992378235,
+      "learning_rate": 2e-05,
+      "loss": 0.5696,
+      "step": 2980
+    },
+    {
+      "epoch": 0.11984448274479939,
+      "grad_norm": 0.38117361068725586,
+      "learning_rate": 2e-05,
+      "loss": 0.5949,
+      "step": 2990
+    },
+    {
+      "epoch": 0.1202453004128422,
+      "grad_norm": 0.39063242077827454,
+      "learning_rate": 2e-05,
+      "loss": 0.571,
+      "step": 3000
+    },
+    {
+      "epoch": 0.120646118080885,
+      "grad_norm": 0.37124958634376526,
+      "learning_rate": 2e-05,
+      "loss": 0.5638,
+      "step": 3010
+    },
+    {
+      "epoch": 0.12104693574892782,
+      "grad_norm": 0.36760056018829346,
+      "learning_rate": 2e-05,
+      "loss": 0.5463,
+      "step": 3020
+    },
+    {
+      "epoch": 0.12144775341697062,
+      "grad_norm": 0.38047149777412415,
+      "learning_rate": 2e-05,
+      "loss": 0.5883,
+      "step": 3030
+    },
+    {
+      "epoch": 0.12184857108501343,
+      "grad_norm": 0.366414874792099,
+      "learning_rate": 2e-05,
+      "loss": 0.5626,
+      "step": 3040
+    },
+    {
+      "epoch": 0.12224938875305623,
+      "grad_norm": 0.6604557633399963,
+      "learning_rate": 2e-05,
+      "loss": 0.5845,
+      "step": 3050
+    },
+    {
+      "epoch": 0.12265020642109904,
+      "grad_norm": 0.36171624064445496,
+      "learning_rate": 2e-05,
+      "loss": 0.5455,
+      "step": 3060
+    },
+    {
+      "epoch": 0.12305102408914186,
+      "grad_norm": 0.36407172679901123,
+      "learning_rate": 2e-05,
+      "loss": 0.5672,
+      "step": 3070
+    },
+    {
+      "epoch": 0.12345184175718465,
+      "grad_norm": 0.3666282296180725,
+      "learning_rate": 2e-05,
+      "loss": 0.5636,
+      "step": 3080
+    },
+    {
+      "epoch": 0.12385265942522747,
+      "grad_norm": 0.3591126501560211,
+      "learning_rate": 2e-05,
+      "loss": 0.5561,
+      "step": 3090
+    },
+    {
+      "epoch": 0.12425347709327027,
+      "grad_norm": 0.5449425578117371,
+      "learning_rate": 2e-05,
+      "loss": 0.5575,
+      "step": 3100
+    },
+    {
+      "epoch": 0.12465429476131308,
+      "grad_norm": 0.35526731610298157,
+      "learning_rate": 2e-05,
+      "loss": 0.5565,
+      "step": 3110
+    },
+    {
+      "epoch": 0.12505511242935588,
+      "grad_norm": 0.401933878660202,
+      "learning_rate": 2e-05,
+      "loss": 0.6013,
+      "step": 3120
+    },
+    {
+      "epoch": 0.1254559300973987,
+      "grad_norm": 0.35224899649620056,
+      "learning_rate": 2e-05,
+      "loss": 0.5502,
+      "step": 3130
+    },
+    {
+      "epoch": 0.1258567477654415,
+      "grad_norm": 0.38501036167144775,
+      "learning_rate": 2e-05,
+      "loss": 0.5908,
+      "step": 3140
+    },
+    {
+      "epoch": 0.12625756543348432,
+      "grad_norm": 0.37840890884399414,
+      "learning_rate": 2e-05,
+      "loss": 0.5408,
+      "step": 3150
+    },
+    {
+      "epoch": 0.1266583831015271,
+      "grad_norm": 0.8629856705665588,
+      "learning_rate": 2e-05,
+      "loss": 0.5785,
+      "step": 3160
+    },
+    {
+      "epoch": 0.12705920076956992,
+      "grad_norm": 0.3897327482700348,
+      "learning_rate": 2e-05,
+      "loss": 0.5699,
+      "step": 3170
+    },
+    {
+      "epoch": 0.12746001843761273,
+      "grad_norm": 0.32331597805023193,
+      "learning_rate": 2e-05,
+      "loss": 0.5867,
+      "step": 3180
+    },
+    {
+      "epoch": 0.12786083610565555,
+      "grad_norm": 0.3642168939113617,
+      "learning_rate": 2e-05,
+      "loss": 0.521,
+      "step": 3190
+    },
+    {
+      "epoch": 0.12826165377369833,
+      "grad_norm": 0.49447524547576904,
+      "learning_rate": 2e-05,
+      "loss": 0.5716,
+      "step": 3200
+    },
+    {
+      "epoch": 0.12866247144174114,
+      "grad_norm": 0.3670274317264557,
+      "learning_rate": 2e-05,
+      "loss": 0.587,
+      "step": 3210
+    },
+    {
+      "epoch": 0.12906328910978396,
+      "grad_norm": 0.36232125759124756,
+      "learning_rate": 2e-05,
+      "loss": 0.5616,
+      "step": 3220
+    },
+    {
+      "epoch": 0.12946410677782677,
+      "grad_norm": 0.3495625853538513,
+      "learning_rate": 2e-05,
+      "loss": 0.5484,
+      "step": 3230
+    },
+    {
+      "epoch": 0.12986492444586958,
+      "grad_norm": 0.3833344578742981,
+      "learning_rate": 2e-05,
+      "loss": 0.5517,
+      "step": 3240
+    },
+    {
+      "epoch": 0.13026574211391237,
+      "grad_norm": 0.3367772102355957,
+      "learning_rate": 2e-05,
+      "loss": 0.5767,
+      "step": 3250
+    },
+    {
+      "epoch": 0.13066655978195518,
+      "grad_norm": 0.4046475291252136,
+      "learning_rate": 2e-05,
+      "loss": 0.5342,
+      "step": 3260
+    },
+    {
+      "epoch": 0.131067377449998,
+      "grad_norm": 0.4214325249195099,
+      "learning_rate": 2e-05,
+      "loss": 0.5268,
+      "step": 3270
+    },
+    {
+      "epoch": 0.1314681951180408,
+      "grad_norm": 0.4093036651611328,
+      "learning_rate": 2e-05,
+      "loss": 0.5912,
+      "step": 3280
+    },
+    {
+      "epoch": 0.13186901278608362,
+      "grad_norm": 0.45690345764160156,
+      "learning_rate": 2e-05,
+      "loss": 0.5754,
+      "step": 3290
+    },
+    {
+      "epoch": 0.1322698304541264,
+      "grad_norm": 0.3545232117176056,
+      "learning_rate": 2e-05,
+      "loss": 0.5598,
+      "step": 3300
+    },
+    {
+      "epoch": 0.13267064812216922,
+      "grad_norm": 0.3392185568809509,
+      "learning_rate": 2e-05,
+      "loss": 0.5602,
+      "step": 3310
+    },
+    {
+      "epoch": 0.13307146579021203,
+      "grad_norm": 0.4475401043891907,
+      "learning_rate": 2e-05,
+      "loss": 0.5655,
+      "step": 3320
+    },
+    {
+      "epoch": 0.13347228345825485,
+      "grad_norm": 0.4275551736354828,
+      "learning_rate": 2e-05,
+      "loss": 0.56,
+      "step": 3330
+    },
+    {
+      "epoch": 0.13387310112629763,
+      "grad_norm": 0.4727897346019745,
+      "learning_rate": 2e-05,
+      "loss": 0.5706,
+      "step": 3340
+    },
+    {
+      "epoch": 0.13427391879434045,
+      "grad_norm": 0.43208369612693787,
+      "learning_rate": 2e-05,
+      "loss": 0.5734,
+      "step": 3350
+    },
+    {
+      "epoch": 0.13467473646238326,
+      "grad_norm": 0.33621519804000854,
+      "learning_rate": 2e-05,
+      "loss": 0.5431,
+      "step": 3360
+    },
+    {
+      "epoch": 0.13507555413042607,
+      "grad_norm": 0.42468395829200745,
+      "learning_rate": 2e-05,
+      "loss": 0.6185,
+      "step": 3370
+    },
+    {
+      "epoch": 0.1354763717984689,
+      "grad_norm": 0.4151937663555145,
+      "learning_rate": 2e-05,
+      "loss": 0.5393,
+      "step": 3380
+    },
+    {
+      "epoch": 0.13587718946651167,
+      "grad_norm": 0.38833171129226685,
+      "learning_rate": 2e-05,
+      "loss": 0.5525,
+      "step": 3390
+    },
+    {
+      "epoch": 0.13627800713455449,
+      "grad_norm": 0.5158320665359497,
+      "learning_rate": 2e-05,
+      "loss": 0.5895,
+      "step": 3400
+    },
+    {
+      "epoch": 0.1366788248025973,
+      "grad_norm": 0.41796988248825073,
+      "learning_rate": 2e-05,
+      "loss": 0.554,
+      "step": 3410
+    },
+    {
+      "epoch": 0.1370796424706401,
+      "grad_norm": 0.42714154720306396,
+      "learning_rate": 2e-05,
+      "loss": 0.5294,
+      "step": 3420
+    },
+    {
+      "epoch": 0.13748046013868293,
+      "grad_norm": 0.38487645983695984,
+      "learning_rate": 2e-05,
+      "loss": 0.5707,
+      "step": 3430
+    },
+    {
+      "epoch": 0.1378812778067257,
+      "grad_norm": 0.33553385734558105,
+      "learning_rate": 2e-05,
+      "loss": 0.5772,
+      "step": 3440
+    },
+    {
+      "epoch": 0.13828209547476852,
+      "grad_norm": 0.3741719424724579,
+      "learning_rate": 2e-05,
+      "loss": 0.5593,
+      "step": 3450
+    },
+    {
+      "epoch": 0.13868291314281134,
+      "grad_norm": 0.4149104952812195,
+      "learning_rate": 2e-05,
+      "loss": 0.5677,
+      "step": 3460
+    },
+    {
+      "epoch": 0.13908373081085415,
+      "grad_norm": 0.33128297328948975,
+      "learning_rate": 2e-05,
+      "loss": 0.5684,
+      "step": 3470
+    },
+    {
+      "epoch": 0.13948454847889696,
+      "grad_norm": 0.35428762435913086,
+      "learning_rate": 2e-05,
+      "loss": 0.5665,
+      "step": 3480
+    },
+    {
+      "epoch": 0.13988536614693975,
+      "grad_norm": 0.39158743619918823,
+      "learning_rate": 2e-05,
+      "loss": 0.5705,
+      "step": 3490
+    },
+    {
+      "epoch": 0.14028618381498256,
+      "grad_norm": 0.3916440010070801,
+      "learning_rate": 2e-05,
+      "loss": 0.547,
+      "step": 3500
+    },
+    {
+      "epoch": 0.14068700148302538,
+      "grad_norm": 0.4144287407398224,
+      "learning_rate": 2e-05,
+      "loss": 0.5539,
+      "step": 3510
+    },
+    {
+      "epoch": 0.1410878191510682,
+      "grad_norm": 0.41197875142097473,
+      "learning_rate": 2e-05,
+      "loss": 0.5764,
+      "step": 3520
+    },
+    {
+      "epoch": 0.14148863681911097,
+      "grad_norm": 0.4137912392616272,
+      "learning_rate": 2e-05,
+      "loss": 0.5362,
+      "step": 3530
+    },
+    {
+      "epoch": 0.1418894544871538,
+      "grad_norm": 0.413318008184433,
+      "learning_rate": 2e-05,
+      "loss": 0.5404,
+      "step": 3540
+    },
+    {
+      "epoch": 0.1422902721551966,
+      "grad_norm": 0.3670721650123596,
+      "learning_rate": 2e-05,
+      "loss": 0.5541,
+      "step": 3550
+    },
+    {
+      "epoch": 0.14269108982323941,
+      "grad_norm": 0.3639812171459198,
+      "learning_rate": 2e-05,
+      "loss": 0.5438,
+      "step": 3560
+    },
+    {
+      "epoch": 0.14309190749128223,
+      "grad_norm": 0.39278337359428406,
+      "learning_rate": 2e-05,
+      "loss": 0.5451,
+      "step": 3570
+    },
+    {
+      "epoch": 0.143492725159325,
+      "grad_norm": 0.3662254512310028,
+      "learning_rate": 2e-05,
+      "loss": 0.5813,
+      "step": 3580
+    },
+    {
+      "epoch": 0.14389354282736783,
+      "grad_norm": 0.393451452255249,
+      "learning_rate": 2e-05,
+      "loss": 0.5461,
+      "step": 3590
+    },
+    {
+      "epoch": 0.14429436049541064,
+      "grad_norm": 0.40145015716552734,
+      "learning_rate": 2e-05,
+      "loss": 0.5427,
+      "step": 3600
+    },
+    {
+      "epoch": 0.14469517816345345,
+      "grad_norm": 0.39159318804740906,
+      "learning_rate": 2e-05,
+      "loss": 0.5701,
+      "step": 3610
+    },
+    {
+      "epoch": 0.14509599583149627,
+      "grad_norm": 0.32462480664253235,
+      "learning_rate": 2e-05,
+      "loss": 0.547,
+      "step": 3620
+    },
+    {
+      "epoch": 0.14549681349953905,
+      "grad_norm": 0.3443101644515991,
+      "learning_rate": 2e-05,
+      "loss": 0.5462,
+      "step": 3630
+    },
+    {
+      "epoch": 0.14589763116758186,
+      "grad_norm": 0.3691333830356598,
+      "learning_rate": 2e-05,
+      "loss": 0.5755,
+      "step": 3640
+    },
+    {
+      "epoch": 0.14629844883562468,
+      "grad_norm": 0.3583635687828064,
+      "learning_rate": 2e-05,
+      "loss": 0.5476,
+      "step": 3650
+    },
+    {
+      "epoch": 0.1466992665036675,
+      "grad_norm": 0.41858625411987305,
+      "learning_rate": 2e-05,
+      "loss": 0.5432,
+      "step": 3660
+    },
+    {
+      "epoch": 0.14710008417171028,
+      "grad_norm": 0.37000134587287903,
+      "learning_rate": 2e-05,
+      "loss": 0.5893,
+      "step": 3670
+    },
+    {
+      "epoch": 0.1475009018397531,
+      "grad_norm": 0.3400837481021881,
+      "learning_rate": 2e-05,
+      "loss": 0.5558,
+      "step": 3680
+    },
+    {
+      "epoch": 0.1479017195077959,
+      "grad_norm": 0.3961534798145294,
+      "learning_rate": 2e-05,
+      "loss": 0.5749,
+      "step": 3690
+    },
+    {
+      "epoch": 0.14830253717583872,
+      "grad_norm": 0.5130637884140015,
+      "learning_rate": 2e-05,
+      "loss": 0.5261,
+      "step": 3700
+    },
+    {
+      "epoch": 0.14870335484388153,
+      "grad_norm": 0.41944175958633423,
+      "learning_rate": 2e-05,
+      "loss": 0.6251,
+      "step": 3710
+    },
+    {
+      "epoch": 0.14910417251192432,
+      "grad_norm": 0.37696608901023865,
+      "learning_rate": 2e-05,
+      "loss": 0.578,
+      "step": 3720
+    },
+    {
+      "epoch": 0.14950499017996713,
+      "grad_norm": 0.35710376501083374,
+      "learning_rate": 2e-05,
+      "loss": 0.5317,
+      "step": 3730
+    },
+    {
+      "epoch": 0.14990580784800994,
+      "grad_norm": 0.39019304513931274,
+      "learning_rate": 2e-05,
+      "loss": 0.568,
+      "step": 3740
+    },
+    {
+      "epoch": 0.15030662551605276,
+      "grad_norm": 0.36526137590408325,
+      "learning_rate": 2e-05,
+      "loss": 0.539,
+      "step": 3750
+    },
+    {
+      "epoch": 0.15070744318409557,
+      "grad_norm": 0.39668208360671997,
+      "learning_rate": 2e-05,
+      "loss": 0.5691,
+      "step": 3760
+    },
+    {
+      "epoch": 0.15110826085213835,
+      "grad_norm": 0.3485737144947052,
+      "learning_rate": 2e-05,
+      "loss": 0.5579,
+      "step": 3770
+    },
+    {
+      "epoch": 0.15150907852018117,
+      "grad_norm": 0.4233294725418091,
+      "learning_rate": 2e-05,
+      "loss": 0.5503,
+      "step": 3780
+    },
+    {
+      "epoch": 0.15190989618822398,
+      "grad_norm": 0.3929007947444916,
+      "learning_rate": 2e-05,
+      "loss": 0.5794,
+      "step": 3790
+    },
+    {
+      "epoch": 0.1523107138562668,
+      "grad_norm": 0.7184209227561951,
+      "learning_rate": 2e-05,
+      "loss": 0.5272,
+      "step": 3800
+    },
+    {
+      "epoch": 0.15271153152430958,
+      "grad_norm": 0.3964136838912964,
+      "learning_rate": 2e-05,
+      "loss": 0.5294,
+      "step": 3810
+    },
+    {
+      "epoch": 0.1531123491923524,
+      "grad_norm": 0.48776546120643616,
+      "learning_rate": 2e-05,
+      "loss": 0.562,
+      "step": 3820
+    },
+    {
+      "epoch": 0.1535131668603952,
+      "grad_norm": 0.41285786032676697,
+      "learning_rate": 2e-05,
+      "loss": 0.5546,
+      "step": 3830
+    },
+    {
+      "epoch": 0.15391398452843802,
+      "grad_norm": 0.33892032504081726,
+      "learning_rate": 2e-05,
+      "loss": 0.5638,
+      "step": 3840
+    },
+    {
+      "epoch": 0.15431480219648083,
+      "grad_norm": 0.365267813205719,
+      "learning_rate": 2e-05,
+      "loss": 0.557,
+      "step": 3850
+    },
+    {
+      "epoch": 0.15471561986452362,
+      "grad_norm": 0.36361292004585266,
+      "learning_rate": 2e-05,
+      "loss": 0.5403,
+      "step": 3860
+    },
+    {
+      "epoch": 0.15511643753256643,
+      "grad_norm": 0.35188499093055725,
+      "learning_rate": 2e-05,
+      "loss": 0.5161,
+      "step": 3870
+    },
+    {
+      "epoch": 0.15551725520060924,
+      "grad_norm": 0.3540356755256653,
+      "learning_rate": 2e-05,
+      "loss": 0.5593,
+      "step": 3880
+    },
+    {
+      "epoch": 0.15591807286865206,
+      "grad_norm": 0.3354050815105438,
+      "learning_rate": 2e-05,
+      "loss": 0.5553,
+      "step": 3890
+    },
+    {
+      "epoch": 0.15631889053669487,
+      "grad_norm": 0.36081844568252563,
+      "learning_rate": 2e-05,
+      "loss": 0.5633,
+      "step": 3900
+    },
+    {
+      "epoch": 0.15671970820473766,
+      "grad_norm": 0.3673272430896759,
+      "learning_rate": 2e-05,
+      "loss": 0.565,
+      "step": 3910
+    },
+    {
+      "epoch": 0.15712052587278047,
+      "grad_norm": 0.3504160940647125,
+      "learning_rate": 2e-05,
+      "loss": 0.5637,
+      "step": 3920
+    },
+    {
+      "epoch": 0.15752134354082328,
+      "grad_norm": 0.3604793846607208,
+      "learning_rate": 2e-05,
+      "loss": 0.5542,
+      "step": 3930
+    },
+    {
+      "epoch": 0.1579221612088661,
+      "grad_norm": 0.389984667301178,
+      "learning_rate": 2e-05,
+      "loss": 0.5333,
+      "step": 3940
+    },
+    {
+      "epoch": 0.15832297887690888,
+      "grad_norm": 0.4087458848953247,
+      "learning_rate": 2e-05,
+      "loss": 0.5887,
+      "step": 3950
+    },
+    {
+      "epoch": 0.1587237965449517,
+      "grad_norm": 0.4695092439651489,
+      "learning_rate": 2e-05,
+      "loss": 0.5821,
+      "step": 3960
+    },
+    {
+      "epoch": 0.1591246142129945,
+      "grad_norm": 0.4853971004486084,
+      "learning_rate": 2e-05,
+      "loss": 0.5412,
+      "step": 3970
+    },
+    {
+      "epoch": 0.15952543188103732,
+      "grad_norm": 0.3997495472431183,
+      "learning_rate": 2e-05,
+      "loss": 0.5536,
+      "step": 3980
+    },
+    {
+      "epoch": 0.15992624954908014,
+      "grad_norm": 0.40451139211654663,
+      "learning_rate": 2e-05,
+      "loss": 0.534,
+      "step": 3990
+    },
+    {
+      "epoch": 0.16032706721712292,
+      "grad_norm": 0.39277586340904236,
+      "learning_rate": 2e-05,
+      "loss": 0.538,
+      "step": 4000
+    },
+    {
+      "epoch": 0.16072788488516573,
+      "grad_norm": 0.3622714877128601,
+      "learning_rate": 2e-05,
+      "loss": 0.5459,
+      "step": 4010
+    },
+    {
+      "epoch": 0.16112870255320855,
+      "grad_norm": 0.3667004108428955,
+      "learning_rate": 2e-05,
+      "loss": 0.5495,
+      "step": 4020
+    },
+    {
+      "epoch": 0.16152952022125136,
+      "grad_norm": 0.3324030041694641,
+      "learning_rate": 2e-05,
+      "loss": 0.548,
+      "step": 4030
+    },
+    {
+      "epoch": 0.16193033788929417,
+      "grad_norm": 0.3541210889816284,
+      "learning_rate": 2e-05,
+      "loss": 0.5511,
+      "step": 4040
+    },
+    {
+      "epoch": 0.16233115555733696,
+      "grad_norm": 0.3800106942653656,
+      "learning_rate": 2e-05,
+      "loss": 0.5828,
+      "step": 4050
+    },
+    {
+      "epoch": 0.16273197322537977,
+      "grad_norm": 0.4944571256637573,
+      "learning_rate": 2e-05,
+      "loss": 0.5362,
+      "step": 4060
+    },
+    {
+      "epoch": 0.16313279089342259,
+      "grad_norm": 0.4176241159439087,
+      "learning_rate": 2e-05,
+      "loss": 0.535,
+      "step": 4070
+    },
+    {
+      "epoch": 0.1635336085614654,
+      "grad_norm": 0.4892776310443878,
+      "learning_rate": 2e-05,
+      "loss": 0.5831,
+      "step": 4080
+    },
+    {
+      "epoch": 0.16393442622950818,
+      "grad_norm": 0.3938373327255249,
+      "learning_rate": 2e-05,
+      "loss": 0.5624,
+      "step": 4090
+    },
+    {
+      "epoch": 0.164335243897551,
+      "grad_norm": 0.3661962151527405,
+      "learning_rate": 2e-05,
+      "loss": 0.5444,
+      "step": 4100
+    },
+    {
+      "epoch": 0.1647360615655938,
+      "grad_norm": 0.38597309589385986,
+      "learning_rate": 2e-05,
+      "loss": 0.5476,
+      "step": 4110
+    },
+    {
+      "epoch": 0.16513687923363662,
+      "grad_norm": 0.3524716794490814,
+      "learning_rate": 2e-05,
+      "loss": 0.5446,
+      "step": 4120
+    },
+    {
+      "epoch": 0.16553769690167944,
+      "grad_norm": 0.5786010026931763,
+      "learning_rate": 2e-05,
+      "loss": 0.5485,
+      "step": 4130
+    },
+    {
+      "epoch": 0.16593851456972222,
+      "grad_norm": 0.3923215866088867,
+      "learning_rate": 2e-05,
+      "loss": 0.5843,
+      "step": 4140
+    },
+    {
+      "epoch": 0.16633933223776504,
+      "grad_norm": 0.4153671860694885,
+      "learning_rate": 2e-05,
+      "loss": 0.52,
+      "step": 4150
+    },
+    {
+      "epoch": 0.16674014990580785,
+      "grad_norm": 0.4191442131996155,
+      "learning_rate": 2e-05,
+      "loss": 0.5613,
+      "step": 4160
+    },
+    {
+      "epoch": 0.16714096757385066,
+      "grad_norm": 0.34973934292793274,
+      "learning_rate": 2e-05,
+      "loss": 0.5421,
+      "step": 4170
+    },
+    {
+      "epoch": 0.16754178524189348,
+      "grad_norm": 0.41015172004699707,
+      "learning_rate": 2e-05,
+      "loss": 0.5525,
+      "step": 4180
+    },
+    {
+      "epoch": 0.16794260290993626,
+      "grad_norm": 0.34978100657463074,
+      "learning_rate": 2e-05,
+      "loss": 0.5246,
+      "step": 4190
+    },
+    {
+      "epoch": 0.16834342057797907,
+      "grad_norm": 0.434487521648407,
+      "learning_rate": 2e-05,
+      "loss": 0.543,
+      "step": 4200
+    },
+    {
+      "epoch": 0.1687442382460219,
+      "grad_norm": 0.4001392722129822,
+      "learning_rate": 2e-05,
+      "loss": 0.545,
+      "step": 4210
+    },
+    {
+      "epoch": 0.1691450559140647,
+      "grad_norm": 0.3071456551551819,
+      "learning_rate": 2e-05,
+      "loss": 0.5445,
+      "step": 4220
+    },
+    {
+      "epoch": 0.1695458735821075,
+      "grad_norm": 0.46033722162246704,
+      "learning_rate": 2e-05,
+      "loss": 0.5518,
+      "step": 4230
+    },
+    {
+      "epoch": 0.1699466912501503,
+      "grad_norm": 0.3467840254306793,
+      "learning_rate": 2e-05,
+      "loss": 0.5808,
+      "step": 4240
+    },
+    {
+      "epoch": 0.1703475089181931,
+      "grad_norm": 0.3539920151233673,
+      "learning_rate": 2e-05,
+      "loss": 0.5735,
+      "step": 4250
+    },
+    {
+      "epoch": 0.17074832658623593,
+      "grad_norm": 0.41687360405921936,
+      "learning_rate": 2e-05,
+      "loss": 0.6231,
+      "step": 4260
+    },
+    {
+      "epoch": 0.17114914425427874,
+      "grad_norm": 0.4145330488681793,
+      "learning_rate": 2e-05,
+      "loss": 0.5863,
+      "step": 4270
+    },
+    {
+      "epoch": 0.17154996192232153,
+      "grad_norm": 0.39799070358276367,
+      "learning_rate": 2e-05,
+      "loss": 0.5676,
+      "step": 4280
+    },
+    {
+      "epoch": 0.17195077959036434,
+      "grad_norm": 0.39829158782958984,
+      "learning_rate": 2e-05,
+      "loss": 0.5341,
+      "step": 4290
+    },
+    {
+      "epoch": 0.17235159725840715,
+      "grad_norm": 0.4105652868747711,
+      "learning_rate": 2e-05,
+      "loss": 0.5437,
+      "step": 4300
+    },
+    {
+      "epoch": 0.17275241492644997,
+      "grad_norm": 0.38942575454711914,
+      "learning_rate": 2e-05,
+      "loss": 0.5519,
+      "step": 4310
+    },
+    {
+      "epoch": 0.17315323259449278,
+      "grad_norm": 0.4069030284881592,
+      "learning_rate": 2e-05,
+      "loss": 0.5624,
+      "step": 4320
+    },
+    {
+      "epoch": 0.17355405026253556,
+      "grad_norm": 0.35332298278808594,
+      "learning_rate": 2e-05,
+      "loss": 0.5426,
+      "step": 4330
+    },
+    {
+      "epoch": 0.17395486793057838,
+      "grad_norm": 0.48292481899261475,
+      "learning_rate": 2e-05,
+      "loss": 0.5423,
+      "step": 4340
+    },
+    {
+      "epoch": 0.1743556855986212,
+      "grad_norm": 0.3841131925582886,
+      "learning_rate": 2e-05,
+      "loss": 0.5407,
+      "step": 4350
+    },
+    {
+      "epoch": 0.174756503266664,
+      "grad_norm": 0.3174104392528534,
+      "learning_rate": 2e-05,
+      "loss": 0.5365,
+      "step": 4360
+    },
+    {
+      "epoch": 0.1751573209347068,
+      "grad_norm": 0.4082847535610199,
+      "learning_rate": 2e-05,
+      "loss": 0.55,
+      "step": 4370
+    },
+    {
+      "epoch": 0.1755581386027496,
+      "grad_norm": 0.7956055402755737,
+      "learning_rate": 2e-05,
+      "loss": 0.5348,
+      "step": 4380
+    },
+    {
+      "epoch": 0.17595895627079242,
+      "grad_norm": 0.405898779630661,
+      "learning_rate": 2e-05,
+      "loss": 0.5361,
+      "step": 4390
+    },
+    {
+      "epoch": 0.17635977393883523,
+      "grad_norm": 0.46592554450035095,
+      "learning_rate": 2e-05,
+      "loss": 0.593,
+      "step": 4400
+    },
+    {
+      "epoch": 0.17676059160687804,
+      "grad_norm": 0.407651424407959,
+      "learning_rate": 2e-05,
+      "loss": 0.5275,
+      "step": 4410
+    },
+    {
+      "epoch": 0.17716140927492083,
+      "grad_norm": 0.4172528386116028,
+      "learning_rate": 2e-05,
+      "loss": 0.5633,
+      "step": 4420
+    },
+    {
+      "epoch": 0.17756222694296364,
+      "grad_norm": 0.4054419696331024,
+      "learning_rate": 2e-05,
+      "loss": 0.5695,
+      "step": 4430
+    },
+    {
+      "epoch": 0.17796304461100645,
+      "grad_norm": 0.3614487946033478,
+      "learning_rate": 2e-05,
+      "loss": 0.5078,
+      "step": 4440
+    },
+    {
+      "epoch": 0.17836386227904927,
+      "grad_norm": 0.3634093403816223,
+      "learning_rate": 2e-05,
+      "loss": 0.5367,
+      "step": 4450
+    },
+    {
+      "epoch": 0.17876467994709208,
+      "grad_norm": 0.39328640699386597,
+      "learning_rate": 2e-05,
+      "loss": 0.5739,
+      "step": 4460
+    },
+    {
+      "epoch": 0.17916549761513487,
+      "grad_norm": 0.4024289548397064,
+      "learning_rate": 2e-05,
+      "loss": 0.5347,
+      "step": 4470
+    },
+    {
+      "epoch": 0.17956631528317768,
+      "grad_norm": 0.37119895219802856,
+      "learning_rate": 2e-05,
+      "loss": 0.585,
+      "step": 4480
+    },
+    {
+      "epoch": 0.1799671329512205,
+      "grad_norm": 0.38500264286994934,
+      "learning_rate": 2e-05,
+      "loss": 0.5529,
+      "step": 4490
+    },
+    {
+      "epoch": 0.1803679506192633,
+      "grad_norm": 0.4702656865119934,
+      "learning_rate": 2e-05,
+      "loss": 0.5283,
+      "step": 4500
+    },
+    {
+      "epoch": 0.1807687682873061,
+      "grad_norm": 0.6010221838951111,
+      "learning_rate": 2e-05,
+      "loss": 0.5547,
+      "step": 4510
+    },
+    {
+      "epoch": 0.1811695859553489,
+      "grad_norm": 0.38243117928504944,
+      "learning_rate": 2e-05,
+      "loss": 0.5219,
+      "step": 4520
+    },
+    {
+      "epoch": 0.18157040362339172,
+      "grad_norm": 0.3752060830593109,
+      "learning_rate": 2e-05,
+      "loss": 0.5374,
+      "step": 4530
+    },
+    {
+      "epoch": 0.18197122129143453,
+      "grad_norm": 0.5424579977989197,
+      "learning_rate": 2e-05,
+      "loss": 0.5597,
+      "step": 4540
+    },
+    {
+      "epoch": 0.18237203895947734,
+      "grad_norm": 0.36557134985923767,
+      "learning_rate": 2e-05,
+      "loss": 0.5449,
+      "step": 4550
+    },
+    {
+      "epoch": 0.18277285662752013,
+      "grad_norm": 0.3883345425128937,
+      "learning_rate": 2e-05,
+      "loss": 0.5741,
+      "step": 4560
+    },
+    {
+      "epoch": 0.18317367429556294,
+      "grad_norm": 0.4248678982257843,
+      "learning_rate": 2e-05,
+      "loss": 0.5697,
+      "step": 4570
+    },
+    {
+      "epoch": 0.18357449196360576,
+      "grad_norm": 0.40503668785095215,
+      "learning_rate": 2e-05,
+      "loss": 0.5374,
+      "step": 4580
+    },
+    {
+      "epoch": 0.18397530963164857,
+      "grad_norm": 0.4111357629299164,
+      "learning_rate": 2e-05,
+      "loss": 0.5375,
+      "step": 4590
+    },
+    {
+      "epoch": 0.18437612729969138,
+      "grad_norm": 0.36233001947402954,
+      "learning_rate": 2e-05,
+      "loss": 0.5305,
+      "step": 4600
+    },
+    {
+      "epoch": 0.18477694496773417,
+      "grad_norm": 0.45881518721580505,
+      "learning_rate": 2e-05,
+      "loss": 0.5189,
+      "step": 4610
+    },
+    {
+      "epoch": 0.18517776263577698,
+      "grad_norm": 0.43970099091529846,
+      "learning_rate": 2e-05,
+      "loss": 0.5345,
+      "step": 4620
+    },
+    {
+      "epoch": 0.1855785803038198,
+      "grad_norm": 0.5237261056900024,
+      "learning_rate": 2e-05,
+      "loss": 0.5192,
+      "step": 4630
+    },
+    {
+      "epoch": 0.1859793979718626,
+      "grad_norm": 0.3353913724422455,
+      "learning_rate": 2e-05,
+      "loss": 0.511,
+      "step": 4640
+    },
+    {
+      "epoch": 0.1863802156399054,
+      "grad_norm": 0.40313178300857544,
+      "learning_rate": 2e-05,
+      "loss": 0.5524,
+      "step": 4650
+    },
+    {
+      "epoch": 0.1867810333079482,
+      "grad_norm": 0.38814249634742737,
+      "learning_rate": 2e-05,
+      "loss": 0.5621,
+      "step": 4660
+    },
+    {
+      "epoch": 0.18718185097599102,
+      "grad_norm": 0.37966763973236084,
+      "learning_rate": 2e-05,
+      "loss": 0.5435,
+      "step": 4670
+    },
+    {
+      "epoch": 0.18758266864403383,
+      "grad_norm": 0.4094284772872925,
+      "learning_rate": 2e-05,
+      "loss": 0.548,
+      "step": 4680
+    },
+    {
+      "epoch": 0.18798348631207665,
+      "grad_norm": 0.3713163733482361,
+      "learning_rate": 2e-05,
+      "loss": 0.5718,
+      "step": 4690
+    },
+    {
+      "epoch": 0.18838430398011943,
+      "grad_norm": 0.441877156496048,
+      "learning_rate": 2e-05,
+      "loss": 0.5588,
+      "step": 4700
+    },
+    {
+      "epoch": 0.18878512164816225,
+      "grad_norm": 0.4034399688243866,
+      "learning_rate": 2e-05,
+      "loss": 0.5166,
+      "step": 4710
+    },
+    {
+      "epoch": 0.18918593931620506,
+      "grad_norm": 0.4547659754753113,
+      "learning_rate": 2e-05,
+      "loss": 0.5562,
+      "step": 4720
+    },
+    {
+      "epoch": 0.18958675698424787,
+      "grad_norm": 0.3528577983379364,
+      "learning_rate": 2e-05,
+      "loss": 0.5681,
+      "step": 4730
+    },
+    {
+      "epoch": 0.18998757465229069,
+      "grad_norm": 0.31532567739486694,
+      "learning_rate": 2e-05,
+      "loss": 0.5578,
+      "step": 4740
+    },
+    {
+      "epoch": 0.19038839232033347,
+      "grad_norm": 0.44949790835380554,
+      "learning_rate": 2e-05,
+      "loss": 0.5696,
+      "step": 4750
+    },
+    {
+      "epoch": 0.19078920998837628,
+      "grad_norm": 0.41615206003189087,
+      "learning_rate": 2e-05,
+      "loss": 0.5387,
+      "step": 4760
+    },
+    {
+      "epoch": 0.1911900276564191,
+      "grad_norm": 0.41089311242103577,
+      "learning_rate": 2e-05,
+      "loss": 0.5432,
+      "step": 4770
+    },
+    {
+      "epoch": 0.1915908453244619,
+      "grad_norm": 0.4379238784313202,
+      "learning_rate": 2e-05,
+      "loss": 0.5288,
+      "step": 4780
+    },
+    {
+      "epoch": 0.1919916629925047,
+      "grad_norm": 0.40153923630714417,
+      "learning_rate": 2e-05,
+      "loss": 0.5587,
+      "step": 4790
+    },
+    {
+      "epoch": 0.1923924806605475,
+      "grad_norm": 0.40166398882865906,
+      "learning_rate": 2e-05,
+      "loss": 0.5348,
+      "step": 4800
+    },
+    {
+      "epoch": 0.19279329832859032,
+      "grad_norm": 0.4740196168422699,
+      "learning_rate": 2e-05,
+      "loss": 0.5194,
+      "step": 4810
+    },
+    {
+      "epoch": 0.19319411599663314,
+      "grad_norm": 0.3766801357269287,
+      "learning_rate": 2e-05,
+      "loss": 0.5414,
+      "step": 4820
+    },
+    {
+      "epoch": 0.19359493366467595,
+      "grad_norm": 0.3910306692123413,
+      "learning_rate": 2e-05,
+      "loss": 0.5289,
+      "step": 4830
+    },
+    {
+      "epoch": 0.19399575133271874,
+      "grad_norm": 0.3671242892742157,
+      "learning_rate": 2e-05,
+      "loss": 0.5542,
+      "step": 4840
+    },
+    {
+      "epoch": 0.19439656900076155,
+      "grad_norm": 0.3537273406982422,
+      "learning_rate": 2e-05,
+      "loss": 0.5538,
+      "step": 4850
+    },
+    {
+      "epoch": 0.19479738666880436,
+      "grad_norm": 0.391032338142395,
+      "learning_rate": 2e-05,
+      "loss": 0.5279,
+      "step": 4860
+    },
+    {
+      "epoch": 0.19519820433684718,
+      "grad_norm": 0.42511165142059326,
+      "learning_rate": 2e-05,
+      "loss": 0.5848,
+      "step": 4870
+    },
+    {
+      "epoch": 0.19559902200489,
+      "grad_norm": 0.4249589741230011,
+      "learning_rate": 2e-05,
+      "loss": 0.5361,
+      "step": 4880
+    },
+    {
+      "epoch": 0.19599983967293277,
+      "grad_norm": 0.4259021580219269,
+      "learning_rate": 2e-05,
+      "loss": 0.5253,
+      "step": 4890
+    },
+    {
+      "epoch": 0.1964006573409756,
+      "grad_norm": 0.3779090344905853,
+      "learning_rate": 2e-05,
+      "loss": 0.5453,
+      "step": 4900
+    },
+    {
+      "epoch": 0.1968014750090184,
+      "grad_norm": 0.43570026755332947,
+      "learning_rate": 2e-05,
+      "loss": 0.5414,
+      "step": 4910
+    },
+    {
+      "epoch": 0.19720229267706121,
+      "grad_norm": 0.3836417496204376,
+      "learning_rate": 2e-05,
+      "loss": 0.5511,
+      "step": 4920
+    },
+    {
+      "epoch": 0.197603110345104,
+      "grad_norm": 0.4048263728618622,
+      "learning_rate": 2e-05,
+      "loss": 0.5697,
+      "step": 4930
+    },
+    {
+      "epoch": 0.1980039280131468,
+      "grad_norm": 0.36949628591537476,
+      "learning_rate": 2e-05,
+      "loss": 0.5188,
+      "step": 4940
+    },
+    {
+      "epoch": 0.19840474568118963,
+      "grad_norm": 0.37402185797691345,
+      "learning_rate": 2e-05,
+      "loss": 0.5235,
+      "step": 4950
+    },
+    {
+      "epoch": 0.19880556334923244,
+      "grad_norm": 0.4147647023200989,
+      "learning_rate": 2e-05,
+      "loss": 0.5145,
+      "step": 4960
+    },
+    {
+      "epoch": 0.19920638101727525,
+      "grad_norm": 0.358914315700531,
+      "learning_rate": 2e-05,
+      "loss": 0.5591,
+      "step": 4970
+    },
+    {
+      "epoch": 0.19960719868531804,
+      "grad_norm": 0.38443484902381897,
+      "learning_rate": 2e-05,
+      "loss": 0.5656,
+      "step": 4980
+    },
+    {
+      "epoch": 0.20000801635336085,
+      "grad_norm": 0.45405998826026917,
+      "learning_rate": 2e-05,
+      "loss": 0.5687,
+      "step": 4990
+    },
+    {
+      "epoch": 0.20040883402140366,
+      "grad_norm": 0.3737223744392395,
+      "learning_rate": 2e-05,
+      "loss": 0.541,
+      "step": 5000
+    },
+    {
+      "epoch": 0.20080965168944648,
+      "grad_norm": 0.46821582317352295,
+      "learning_rate": 2e-05,
+      "loss": 0.5463,
+      "step": 5010
+    },
+    {
+      "epoch": 0.2012104693574893,
+      "grad_norm": 0.41541174054145813,
+      "learning_rate": 2e-05,
+      "loss": 0.5198,
+      "step": 5020
+    },
+    {
+      "epoch": 0.20161128702553208,
+      "grad_norm": 0.4159490764141083,
+      "learning_rate": 2e-05,
+      "loss": 0.5533,
+      "step": 5030
+    },
+    {
+      "epoch": 0.2020121046935749,
+      "grad_norm": 0.41450053453445435,
+      "learning_rate": 2e-05,
+      "loss": 0.5147,
+      "step": 5040
+    },
+    {
+      "epoch": 0.2024129223616177,
+      "grad_norm": 0.3766546845436096,
+      "learning_rate": 2e-05,
+      "loss": 0.5096,
+      "step": 5050
+    },
+    {
+      "epoch": 0.20281374002966052,
+      "grad_norm": 0.3497719168663025,
+      "learning_rate": 2e-05,
+      "loss": 0.5245,
+      "step": 5060
+    },
+    {
+      "epoch": 0.2032145576977033,
+      "grad_norm": 0.43203771114349365,
+      "learning_rate": 2e-05,
+      "loss": 0.5634,
+      "step": 5070
+    },
+    {
+      "epoch": 0.20361537536574612,
+      "grad_norm": 0.39704185724258423,
+      "learning_rate": 2e-05,
+      "loss": 0.5175,
+      "step": 5080
+    },
+    {
+      "epoch": 0.20401619303378893,
+      "grad_norm": 0.3234955370426178,
+      "learning_rate": 2e-05,
+      "loss": 0.5106,
+      "step": 5090
+    },
+    {
+      "epoch": 0.20441701070183174,
+      "grad_norm": 0.35997021198272705,
+      "learning_rate": 2e-05,
+      "loss": 0.5182,
+      "step": 5100
+    },
+    {
+      "epoch": 0.20481782836987455,
+      "grad_norm": 0.40877920389175415,
+      "learning_rate": 2e-05,
+      "loss": 0.5675,
+      "step": 5110
+    },
+    {
+      "epoch": 0.20521864603791734,
+      "grad_norm": 0.4481983184814453,
+      "learning_rate": 2e-05,
+      "loss": 0.4981,
+      "step": 5120
+    },
+    {
+      "epoch": 0.20561946370596015,
+      "grad_norm": 0.3461613357067108,
+      "learning_rate": 2e-05,
+      "loss": 0.5122,
+      "step": 5130
+    },
+    {
+      "epoch": 0.20602028137400297,
+      "grad_norm": 0.3736863434314728,
+      "learning_rate": 2e-05,
+      "loss": 0.5446,
+      "step": 5140
+    },
+    {
+      "epoch": 0.20642109904204578,
+      "grad_norm": 0.42698124051094055,
+      "learning_rate": 2e-05,
+      "loss": 0.5608,
+      "step": 5150
+    },
+    {
+      "epoch": 0.2068219167100886,
+      "grad_norm": 0.3503425717353821,
+      "learning_rate": 2e-05,
+      "loss": 0.5106,
+      "step": 5160
+    },
+    {
+      "epoch": 0.20722273437813138,
+      "grad_norm": 0.34551504254341125,
+      "learning_rate": 2e-05,
+      "loss": 0.5642,
+      "step": 5170
+    },
+    {
+      "epoch": 0.2076235520461742,
+      "grad_norm": 0.37515121698379517,
+      "learning_rate": 2e-05,
+      "loss": 0.5335,
+      "step": 5180
+    },
+    {
+      "epoch": 0.208024369714217,
+      "grad_norm": 0.3841811418533325,
+      "learning_rate": 2e-05,
+      "loss": 0.5558,
+      "step": 5190
+    },
+    {
+      "epoch": 0.20842518738225982,
+      "grad_norm": 0.3704339861869812,
+      "learning_rate": 2e-05,
+      "loss": 0.5381,
+      "step": 5200
+    },
+    {
+      "epoch": 0.2088260050503026,
+      "grad_norm": 0.4769802987575531,
+      "learning_rate": 2e-05,
+      "loss": 0.5079,
+      "step": 5210
+    },
+    {
+      "epoch": 0.20922682271834542,
+      "grad_norm": 0.4107668995857239,
+      "learning_rate": 2e-05,
+      "loss": 0.5266,
+      "step": 5220
+    },
+    {
+      "epoch": 0.20962764038638823,
+      "grad_norm": 0.373456209897995,
+      "learning_rate": 2e-05,
+      "loss": 0.5694,
+      "step": 5230
+    },
+    {
+      "epoch": 0.21002845805443104,
+      "grad_norm": 0.4065524637699127,
+      "learning_rate": 2e-05,
+      "loss": 0.5206,
+      "step": 5240
+    },
+    {
+      "epoch": 0.21042927572247386,
+      "grad_norm": 0.36516571044921875,
+      "learning_rate": 2e-05,
+      "loss": 0.522,
+      "step": 5250
+    },
+    {
+      "epoch": 0.21083009339051664,
+      "grad_norm": 0.43594372272491455,
+      "learning_rate": 2e-05,
+      "loss": 0.5315,
+      "step": 5260
+    },
+    {
+      "epoch": 0.21123091105855946,
+      "grad_norm": 0.32319360971450806,
+      "learning_rate": 2e-05,
+      "loss": 0.5704,
+      "step": 5270
+    },
+    {
+      "epoch": 0.21163172872660227,
+      "grad_norm": 0.37819454073905945,
+      "learning_rate": 2e-05,
+      "loss": 0.5228,
+      "step": 5280
+    },
+    {
+      "epoch": 0.21203254639464508,
+      "grad_norm": 0.37566232681274414,
+      "learning_rate": 2e-05,
+      "loss": 0.5559,
+      "step": 5290
+    },
+    {
+      "epoch": 0.2124333640626879,
+      "grad_norm": 0.4790727198123932,
+      "learning_rate": 2e-05,
+      "loss": 0.5278,
+      "step": 5300
+    },
+    {
+      "epoch": 0.21283418173073068,
+      "grad_norm": 0.45754361152648926,
+      "learning_rate": 2e-05,
+      "loss": 0.5514,
+      "step": 5310
+    },
+    {
+      "epoch": 0.2132349993987735,
+      "grad_norm": 0.3573928475379944,
+      "learning_rate": 2e-05,
+      "loss": 0.5221,
+      "step": 5320
+    },
+    {
+      "epoch": 0.2136358170668163,
+      "grad_norm": 0.4499244689941406,
+      "learning_rate": 2e-05,
+      "loss": 0.5538,
+      "step": 5330
+    },
+    {
+      "epoch": 0.21403663473485912,
+      "grad_norm": 0.4337273836135864,
+      "learning_rate": 2e-05,
+      "loss": 0.5077,
+      "step": 5340
+    },
+    {
+      "epoch": 0.2144374524029019,
+      "grad_norm": 0.38488462567329407,
+      "learning_rate": 1.999194520388485e-05,
+      "loss": 0.5123,
+      "step": 5350
+    },
+    {
+      "epoch": 0.21483827007094472,
+      "grad_norm": 0.4251571595668793,
+      "learning_rate": 1.9959247303157763e-05,
+      "loss": 0.5152,
+      "step": 5360
+    },
+    {
+      "epoch": 0.21523908773898753,
+      "grad_norm": 0.34759294986724854,
+      "learning_rate": 1.990149424459817e-05,
+      "loss": 0.5191,
+      "step": 5370
+    },
+    {
+      "epoch": 0.21563990540703035,
+      "grad_norm": 0.39413633942604065,
+      "learning_rate": 1.9818847538648688e-05,
+      "loss": 0.5424,
+      "step": 5380
+    },
+    {
+      "epoch": 0.21604072307507316,
+      "grad_norm": 0.41560786962509155,
+      "learning_rate": 1.971153831256721e-05,
+      "loss": 0.5371,
+      "step": 5390
+    },
+    {
+      "epoch": 0.21644154074311595,
+      "grad_norm": 0.4269552528858185,
+      "learning_rate": 1.9579866664063435e-05,
+      "loss": 0.538,
+      "step": 5400
+    },
+    {
+      "epoch": 0.21684235841115876,
+      "grad_norm": 0.39197859168052673,
+      "learning_rate": 1.9424200822054782e-05,
+      "loss": 0.5532,
+      "step": 5410
+    },
+    {
+      "epoch": 0.21724317607920157,
+      "grad_norm": 0.404278039932251,
+      "learning_rate": 1.9244976116888626e-05,
+      "loss": 0.5445,
+      "step": 5420
+    },
+    {
+      "epoch": 0.21764399374724439,
+      "grad_norm": 0.40637174248695374,
+      "learning_rate": 1.904269376291071e-05,
+      "loss": 0.524,
+      "step": 5430
+    },
+    {
+      "epoch": 0.2180448114152872,
+      "grad_norm": 0.4392496645450592,
+      "learning_rate": 1.88179194567844e-05,
+      "loss": 0.5399,
+      "step": 5440
+    },
+    {
+      "epoch": 0.21844562908332998,
+      "grad_norm": 0.3582078218460083,
+      "learning_rate": 1.8571281795480632e-05,
+      "loss": 0.5212,
+      "step": 5450
+    },
+    {
+      "epoch": 0.2188464467513728,
+      "grad_norm": 0.3846853971481323,
+      "learning_rate": 1.8303470518362808e-05,
+      "loss": 0.5409,
+      "step": 5460
+    },
+    {
+      "epoch": 0.2192472644194156,
+      "grad_norm": 0.4698985815048218,
+      "learning_rate": 1.801523457828271e-05,
+      "loss": 0.5632,
+      "step": 5470
+    },
+    {
+      "epoch": 0.21964808208745842,
+      "grad_norm": 0.564365029335022,
+      "learning_rate": 1.7707380047081796e-05,
+      "loss": 0.5165,
+      "step": 5480
+    },
+    {
+      "epoch": 0.2200488997555012,
+      "grad_norm": 0.45134034752845764,
+      "learning_rate": 1.7380767861355252e-05,
+      "loss": 0.5425,
+      "step": 5490
+    },
+    {
+      "epoch": 0.22044971742354402,
+      "grad_norm": 0.3817583918571472,
+      "learning_rate": 1.703631141478299e-05,
+      "loss": 0.5341,
+      "step": 5500
+    },
+    {
+      "epoch": 0.22085053509158684,
+      "grad_norm": 0.36346763372421265,
+      "learning_rate": 1.6674974003760705e-05,
+      "loss": 0.526,
+      "step": 5510
+    },
+    {
+      "epoch": 0.22125135275962965,
+      "grad_norm": 0.39670056104660034,
+      "learning_rate": 1.6297766133474556e-05,
+      "loss": 0.5446,
+      "step": 5520
+    },
+    {
+      "epoch": 0.22165217042767246,
+      "grad_norm": 0.39365604519844055,
+      "learning_rate": 1.5905742691953186e-05,
+      "loss": 0.5624,
+      "step": 5530
+    },
+    {
+      "epoch": 0.22205298809571525,
+      "grad_norm": 0.40811771154403687,
+      "learning_rate": 1.55e-05,
+      "loss": 0.5402,
+      "step": 5540
+    },
+    {
+      "epoch": 0.22245380576375806,
+      "grad_norm": 0.3727797865867615,
+      "learning_rate": 1.5081672745255862e-05,
+      "loss": 0.5374,
+      "step": 5550
+    },
+    {
+      "epoch": 0.22285462343180087,
+      "grad_norm": 0.39216357469558716,
+      "learning_rate": 1.4651930808966206e-05,
+      "loss": 0.5055,
+      "step": 5560
+    },
+    {
+      "epoch": 0.2232554410998437,
+      "grad_norm": 0.32694211602211,
+      "learning_rate": 1.4211975994326848e-05,
+      "loss": 0.5184,
+      "step": 5570
+    },
+    {
+      "epoch": 0.2236562587678865,
+      "grad_norm": 0.38194262981414795,
+      "learning_rate": 1.376303866555779e-05,
+      "loss": 0.5363,
+      "step": 5580
+    },
+    {
+      "epoch": 0.2240570764359293,
+      "grad_norm": 0.435739666223526,
+      "learning_rate": 1.3306374307104164e-05,
+      "loss": 0.5647,
+      "step": 5590
+    },
+    {
+      "epoch": 0.2244578941039721,
+      "grad_norm": 0.34788525104522705,
+      "learning_rate": 1.2843260012586718e-05,
+      "loss": 0.5128,
+      "step": 5600
+    },
+    {
+      "epoch": 0.2248587117720149,
+      "grad_norm": 0.40309491753578186,
+      "learning_rate": 1.2374990913320726e-05,
+      "loss": 0.516,
+      "step": 5610
+    },
+    {
+      "epoch": 0.22525952944005773,
+      "grad_norm": 0.48371002078056335,
+      "learning_rate": 1.1902876556391247e-05,
+      "loss": 0.5169,
+      "step": 5620
+    },
+    {
+      "epoch": 0.2256603471081005,
+      "grad_norm": 0.3576868772506714,
+      "learning_rate": 1.1428237242413685e-05,
+      "loss": 0.5195,
+      "step": 5630
+    },
+    {
+      "epoch": 0.22606116477614333,
+      "grad_norm": 0.44343671202659607,
+      "learning_rate": 1.095240033322132e-05,
+      "loss": 0.5445,
+      "step": 5640
+    },
+    {
+      "epoch": 0.22646198244418614,
+      "grad_norm": 0.3886483609676361,
+      "learning_rate": 1.047669653980572e-05,
+      "loss": 0.5538,
+      "step": 5650
+    },
+    {
+      "epoch": 0.22686280011222895,
+      "grad_norm": 0.3778638541698456,
+      "learning_rate": 1.0002456200890902e-05,
+      "loss": 0.5674,
+      "step": 5660
+    },
+    {
+      "epoch": 0.22726361778027176,
+      "grad_norm": 0.3364357352256775,
+      "learning_rate": 9.531005562548668e-06,
+      "loss": 0.542,
+      "step": 5670
+    },
+    {
+      "epoch": 0.22766443544831455,
+      "grad_norm": 0.3202096223831177,
+      "learning_rate": 9.06366306925925e-06,
+      "loss": 0.5385,
+      "step": 5680
+    },
+    {
+      "epoch": 0.22806525311635736,
+      "grad_norm": 0.4113284945487976,
+      "learning_rate": 8.601735676789686e-06,
+      "loss": 0.5822,
+      "step": 5690
+    },
+    {
+      "epoch": 0.22846607078440018,
+      "grad_norm": 0.37083569169044495,
+      "learning_rate": 8.14651519720118e-06,
+      "loss": 0.5335,
+      "step": 5700
+    },
+    {
+      "epoch": 0.228866888452443,
+      "grad_norm": 0.3675459325313568,
+      "learning_rate": 7.699274686206866e-06,
+      "loss": 0.5251,
+      "step": 5710
+    },
+    {
+      "epoch": 0.2292677061204858,
+      "grad_norm": 0.7313494086265564,
+      "learning_rate": 7.261264882983024e-06,
+      "loss": 0.5515,
+      "step": 5720
+    },
+    {
+      "epoch": 0.2296685237885286,
+      "grad_norm": 0.4613094627857208,
+      "learning_rate": 6.8337107123900386e-06,
+      "loss": 0.5106,
+      "step": 5730
+    },
+    {
+      "epoch": 0.2300693414565714,
+      "grad_norm": 0.36956268548965454,
+      "learning_rate": 6.417807859384881e-06,
+      "loss": 0.5444,
+      "step": 5740
+    },
+    {
+      "epoch": 0.23047015912461422,
+      "grad_norm": 0.3318433463573456,
+      "learning_rate": 6.014719425205009e-06,
+      "loss": 0.5182,
+      "step": 5750
+    },
+    {
+      "epoch": 0.23087097679265703,
+      "grad_norm": 0.4842143952846527,
+      "learning_rate": 5.6255726746749275e-06,
+      "loss": 0.5199,
+      "step": 5760
+    },
+    {
+      "epoch": 0.23127179446069981,
+      "grad_norm": 0.4765832722187042,
+      "learning_rate": 5.251455883731728e-06,
+      "loss": 0.5684,
+      "step": 5770
+    },
+    {
+      "epoch": 0.23167261212874263,
+      "grad_norm": 0.37936750054359436,
+      "learning_rate": 4.893415295985812e-06,
+      "loss": 0.4974,
+      "step": 5780
+    },
+    {
+      "epoch": 0.23207342979678544,
+      "grad_norm": 0.3929890990257263,
+      "learning_rate": 4.552452196827885e-06,
+      "loss": 0.5255,
+      "step": 5790
+    },
+    {
+      "epoch": 0.23247424746482825,
+      "grad_norm": 0.3550904095172882,
+      "learning_rate": 4.229520113264785e-06,
+      "loss": 0.5336,
+      "step": 5800
+    },
+    {
+      "epoch": 0.23287506513287107,
+      "grad_norm": 0.41087806224823,
+      "learning_rate": 3.925522147314915e-06,
+      "loss": 0.5283,
+      "step": 5810
+    },
+    {
+      "epoch": 0.23327588280091385,
+      "grad_norm": 0.5142199993133545,
+      "learning_rate": 3.641308450420675e-06,
+      "loss": 0.491,
+      "step": 5820
+    },
+    {
+      "epoch": 0.23367670046895667,
+      "grad_norm": 0.3663991093635559,
+      "learning_rate": 3.37767384594087e-06,
+      "loss": 0.5366,
+      "step": 5830
+    },
+    {
+      "epoch": 0.23407751813699948,
+      "grad_norm": 0.4289434850215912,
+      "learning_rate": 3.1353556063719357e-06,
+      "loss": 0.5217,
+      "step": 5840
+    },
+    {
+      "epoch": 0.2344783358050423,
+      "grad_norm": 0.41637104749679565,
+      "learning_rate": 2.9150313915141474e-06,
+      "loss": 0.5048,
+      "step": 5850
+    },
+    {
+      "epoch": 0.2348791534730851,
+      "grad_norm": 0.4472004473209381,
+      "learning_rate": 2.717317353348891e-06,
+      "loss": 0.538,
+      "step": 5860
+    },
+    {
+      "epoch": 0.2352799711411279,
+      "grad_norm": 0.349174827337265,
+      "learning_rate": 2.5427664129268253e-06,
+      "loss": 0.521,
+      "step": 5870
+    },
+    {
+      "epoch": 0.2356807888091707,
+      "grad_norm": 0.3317427337169647,
+      "learning_rate": 2.391866714085717e-06,
+      "loss": 0.5185,
+      "step": 5880
+    },
+    {
+      "epoch": 0.23608160647721352,
+      "grad_norm": 0.3936678171157837,
+      "learning_rate": 2.2650402583222753e-06,
+      "loss": 0.526,
+      "step": 5890
+    },
+    {
+      "epoch": 0.23648242414525633,
+      "grad_norm": 0.3802640736103058,
+      "learning_rate": 2.1626417246356398e-06,
+      "loss": 0.5172,
+      "step": 5900
+    },
+    {
+      "epoch": 0.23688324181329912,
+      "grad_norm": 0.341263085603714,
+      "learning_rate": 2.084957477642894e-06,
+      "loss": 0.526,
+      "step": 5910
+    },
+    {
+      "epoch": 0.23728405948134193,
+      "grad_norm": 0.4577714800834656,
+      "learning_rate": 2.0322047667405147e-06,
+      "loss": 0.5443,
+      "step": 5920
+    },
+    {
+      "epoch": 0.23768487714938474,
+      "grad_norm": 0.3793877363204956,
+      "learning_rate": 2.0045311185513342e-06,
+      "loss": 0.5152,
+      "step": 5930
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 5935,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 2035,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.6494000762504348e+19,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}