diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5561 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0323434527256183,
+  "eval_steps": 780,
+  "global_step": 7800,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 220.54913330078125,
+      "learning_rate": 1.282051282051282e-07,
+      "loss": 2.8332,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 136.59881591796875,
+      "learning_rate": 2.564102564102564e-07,
+      "loss": 2.6208,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 40.77531814575195,
+      "learning_rate": 3.846153846153847e-07,
+      "loss": 2.0014,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 16.872848510742188,
+      "learning_rate": 5.128205128205128e-07,
+      "loss": 1.4716,
+      "step": 40
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 15.107361793518066,
+      "learning_rate": 6.41025641025641e-07,
+      "loss": 1.1896,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 43.72957229614258,
+      "learning_rate": 7.692307692307694e-07,
+      "loss": 1.0214,
+      "step": 60
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 22.193782806396484,
+      "learning_rate": 8.974358974358975e-07,
+      "loss": 0.9242,
+      "step": 70
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 39.77405548095703,
+      "learning_rate": 1.0256410256410257e-06,
+      "loss": 0.8996,
+      "step": 80
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 11.561063766479492,
+      "learning_rate": 1.153846153846154e-06,
+      "loss": 0.874,
+      "step": 90
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 15.518475532531738,
+      "learning_rate": 1.282051282051282e-06,
+      "loss": 0.8376,
+      "step": 100
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 7.103811740875244,
+      "learning_rate": 1.4102564102564104e-06,
+      "loss": 0.8272,
+      "step": 110
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 9.517041206359863,
+      "learning_rate": 1.5384615384615387e-06,
+      "loss": 0.8133,
+      "step": 120
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.397263526916504,
+      "learning_rate": 1.6666666666666667e-06,
+      "loss": 0.7843,
+      "step": 130
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.737699508666992,
+      "learning_rate": 1.794871794871795e-06,
+      "loss": 0.7893,
+      "step": 140
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.244195938110352,
+      "learning_rate": 1.9230769230769234e-06,
+      "loss": 0.775,
+      "step": 150
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 7.860988140106201,
+      "learning_rate": 2.0512820512820513e-06,
+      "loss": 0.7527,
+      "step": 160
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 9.824834823608398,
+      "learning_rate": 2.1794871794871797e-06,
+      "loss": 0.739,
+      "step": 170
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 8.463064193725586,
+      "learning_rate": 2.307692307692308e-06,
+      "loss": 0.7432,
+      "step": 180
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.5682692527771,
+      "learning_rate": 2.435897435897436e-06,
+      "loss": 0.7301,
+      "step": 190
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 6.610253810882568,
+      "learning_rate": 2.564102564102564e-06,
+      "loss": 0.7238,
+      "step": 200
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 8.918222427368164,
+      "learning_rate": 2.6923076923076923e-06,
+      "loss": 0.7205,
+      "step": 210
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 8.915346145629883,
+      "learning_rate": 2.8205128205128207e-06,
+      "loss": 0.7171,
+      "step": 220
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 7.223239421844482,
+      "learning_rate": 2.948717948717949e-06,
+      "loss": 0.7259,
+      "step": 230
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 8.313645362854004,
+      "learning_rate": 3.0769230769230774e-06,
+      "loss": 0.7207,
+      "step": 240
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 8.011472702026367,
+      "learning_rate": 3.205128205128206e-06,
+      "loss": 0.7195,
+      "step": 250
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 163.78948974609375,
+      "learning_rate": 3.3333333333333333e-06,
+      "loss": 0.8906,
+      "step": 260
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.948330402374268,
+      "learning_rate": 3.4615384615384617e-06,
+      "loss": 0.9354,
+      "step": 270
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 8.010615348815918,
+      "learning_rate": 3.58974358974359e-06,
+      "loss": 0.7462,
+      "step": 280
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.337215900421143,
+      "learning_rate": 3.7179487179487184e-06,
+      "loss": 0.7211,
+      "step": 290
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.3255696296691895,
+      "learning_rate": 3.846153846153847e-06,
+      "loss": 0.7085,
+      "step": 300
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.787943363189697,
+      "learning_rate": 3.974358974358974e-06,
+      "loss": 0.7035,
+      "step": 310
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 5.677683353424072,
+      "learning_rate": 4.102564102564103e-06,
+      "loss": 0.7026,
+      "step": 320
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 4.641010761260986,
+      "learning_rate": 4.230769230769231e-06,
+      "loss": 0.6932,
+      "step": 330
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 7.157845497131348,
+      "learning_rate": 4.358974358974359e-06,
+      "loss": 0.6837,
+      "step": 340
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.9800801277160645,
+      "learning_rate": 4.487179487179488e-06,
+      "loss": 0.6946,
+      "step": 350
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 6.784861087799072,
+      "learning_rate": 4.615384615384616e-06,
+      "loss": 0.7001,
+      "step": 360
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.393420696258545,
+      "learning_rate": 4.743589743589744e-06,
+      "loss": 0.6867,
+      "step": 370
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 7.812753677368164,
+      "learning_rate": 4.871794871794872e-06,
+      "loss": 0.685,
+      "step": 380
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.967596054077148,
+      "learning_rate": 5e-06,
+      "loss": 0.6793,
+      "step": 390
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.9974517822265625,
+      "learning_rate": 5.128205128205128e-06,
+      "loss": 0.6826,
+      "step": 400
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 5.928895950317383,
+      "learning_rate": 5.256410256410257e-06,
+      "loss": 0.6774,
+      "step": 410
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 7.543839931488037,
+      "learning_rate": 5.384615384615385e-06,
+      "loss": 0.6804,
+      "step": 420
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.843835830688477,
+      "learning_rate": 5.512820512820514e-06,
+      "loss": 0.6798,
+      "step": 430
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.476939678192139,
+      "learning_rate": 5.641025641025641e-06,
+      "loss": 0.68,
+      "step": 440
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.189926624298096,
+      "learning_rate": 5.769230769230769e-06,
+      "loss": 0.6836,
+      "step": 450
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 6.458014965057373,
+      "learning_rate": 5.897435897435898e-06,
+      "loss": 0.6773,
+      "step": 460
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.884819030761719,
+      "learning_rate": 6.025641025641026e-06,
+      "loss": 0.6694,
+      "step": 470
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.87222957611084,
+      "learning_rate": 6.153846153846155e-06,
+      "loss": 0.6632,
+      "step": 480
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 5.767669200897217,
+      "learning_rate": 6.282051282051282e-06,
+      "loss": 0.6831,
+      "step": 490
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.337037563323975,
+      "learning_rate": 6.410256410256412e-06,
+      "loss": 0.6703,
+      "step": 500
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 7.12393856048584,
+      "learning_rate": 6.538461538461539e-06,
+      "loss": 0.6655,
+      "step": 510
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 6.461047172546387,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 0.6691,
+      "step": 520
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.647514820098877,
+      "learning_rate": 6.794871794871796e-06,
+      "loss": 0.6766,
+      "step": 530
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.566062927246094,
+      "learning_rate": 6.923076923076923e-06,
+      "loss": 0.677,
+      "step": 540
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.5864129066467285,
+      "learning_rate": 7.051282051282053e-06,
+      "loss": 0.6494,
+      "step": 550
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 5.37799596786499,
+      "learning_rate": 7.17948717948718e-06,
+      "loss": 0.6745,
+      "step": 560
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.159377098083496,
+      "learning_rate": 7.307692307692308e-06,
+      "loss": 0.6645,
+      "step": 570
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.143188953399658,
+      "learning_rate": 7.435897435897437e-06,
+      "loss": 0.6643,
+      "step": 580
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.492277145385742,
+      "learning_rate": 7.564102564102564e-06,
+      "loss": 0.6646,
+      "step": 590
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.831822872161865,
+      "learning_rate": 7.692307692307694e-06,
+      "loss": 0.6717,
+      "step": 600
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.297182083129883,
+      "learning_rate": 7.820512820512822e-06,
+      "loss": 0.6806,
+      "step": 610
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 6.297981262207031,
+      "learning_rate": 7.948717948717949e-06,
+      "loss": 0.6578,
+      "step": 620
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.828994274139404,
+      "learning_rate": 8.076923076923077e-06,
+      "loss": 0.659,
+      "step": 630
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 5.524103164672852,
+      "learning_rate": 8.205128205128205e-06,
+      "loss": 0.6615,
+      "step": 640
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 5.9705729484558105,
+      "learning_rate": 8.333333333333334e-06,
+      "loss": 0.6473,
+      "step": 650
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.803117275238037,
+      "learning_rate": 8.461538461538462e-06,
+      "loss": 0.6658,
+      "step": 660
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.929677486419678,
+      "learning_rate": 8.58974358974359e-06,
+      "loss": 0.6685,
+      "step": 670
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 7.229376792907715,
+      "learning_rate": 8.717948717948719e-06,
+      "loss": 0.6586,
+      "step": 680
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.211330413818359,
+      "learning_rate": 8.846153846153847e-06,
+      "loss": 0.6705,
+      "step": 690
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 8.68479061126709,
+      "learning_rate": 8.974358974358976e-06,
+      "loss": 0.6634,
+      "step": 700
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 6.314587593078613,
+      "learning_rate": 9.102564102564104e-06,
+      "loss": 0.6595,
+      "step": 710
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 24.571941375732422,
+      "learning_rate": 9.230769230769232e-06,
+      "loss": 0.7864,
+      "step": 720
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 82.88648986816406,
+      "learning_rate": 9.358974358974359e-06,
+      "loss": 0.7688,
+      "step": 730
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 16.867046356201172,
+      "learning_rate": 9.487179487179487e-06,
+      "loss": 3.5579,
+      "step": 740
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 46.3172721862793,
+      "learning_rate": 9.615384615384616e-06,
+      "loss": 1.5314,
+      "step": 750
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 4635.1376953125,
+      "learning_rate": 9.743589743589744e-06,
+      "loss": 1.3744,
+      "step": 760
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 8.688736915588379,
+      "learning_rate": 9.871794871794872e-06,
+      "loss": 0.9862,
+      "step": 770
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 7.020205497741699,
+      "learning_rate": 1e-05,
+      "loss": 0.6837,
+      "step": 780
+    },
+    {
+      "epoch": 0.1,
+      "eval_loss": 0.6935335397720337,
+      "eval_runtime": 169.5292,
+      "eval_samples_per_second": 64.886,
+      "eval_steps_per_second": 8.111,
+      "step": 780
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 9.475116729736328,
+      "learning_rate": 9.99994993147413e-06,
+      "loss": 0.6691,
+      "step": 790
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.155986785888672,
+      "learning_rate": 9.999799726899261e-06,
+      "loss": 0.6729,
+      "step": 800
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.841563701629639,
+      "learning_rate": 9.999549389283605e-06,
+      "loss": 0.6576,
+      "step": 810
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.626774311065674,
+      "learning_rate": 9.999198923640774e-06,
+      "loss": 0.6493,
+      "step": 820
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 7.392406940460205,
+      "learning_rate": 9.998748336989687e-06,
+      "loss": 0.6409,
+      "step": 830
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 7.925800800323486,
+      "learning_rate": 9.998197638354428e-06,
+      "loss": 0.6716,
+      "step": 840
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 6.4082231521606445,
+      "learning_rate": 9.997546838764066e-06,
+      "loss": 0.6415,
+      "step": 850
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 5.968785762786865,
+      "learning_rate": 9.996795951252427e-06,
+      "loss": 0.6516,
+      "step": 860
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.396018981933594,
+      "learning_rate": 9.995944990857848e-06,
+      "loss": 0.6449,
+      "step": 870
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.221564292907715,
+      "learning_rate": 9.994993974622863e-06,
+      "loss": 0.6453,
+      "step": 880
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.979955673217773,
+      "learning_rate": 9.993942921593858e-06,
+      "loss": 0.6659,
+      "step": 890
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 5.904052734375,
+      "learning_rate": 9.992791852820709e-06,
+      "loss": 0.66,
+      "step": 900
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 17.975086212158203,
+      "learning_rate": 9.991540791356342e-06,
+      "loss": 0.6866,
+      "step": 910
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.759973049163818,
+      "learning_rate": 9.990189762256275e-06,
+      "loss": 0.7491,
+      "step": 920
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 13.584983825683594,
+      "learning_rate": 9.988738792578126e-06,
+      "loss": 0.6689,
+      "step": 930
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 6.420701503753662,
+      "learning_rate": 9.987187911381059e-06,
+      "loss": 0.6539,
+      "step": 940
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.14314603805542,
+      "learning_rate": 9.985537149725207e-06,
+      "loss": 0.6546,
+      "step": 950
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.1515679359436035,
+      "learning_rate": 9.983786540671052e-06,
+      "loss": 0.6539,
+      "step": 960
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.025599956512451,
+      "learning_rate": 9.981936119278758e-06,
+      "loss": 0.6487,
+      "step": 970
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 7.110809803009033,
+      "learning_rate": 9.979985922607476e-06,
+      "loss": 0.6494,
+      "step": 980
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 6.404927730560303,
+      "learning_rate": 9.977935989714594e-06,
+      "loss": 0.6504,
+      "step": 990
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.934031963348389,
+      "learning_rate": 9.975786361654959e-06,
+      "loss": 0.637,
+      "step": 1000
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.632777690887451,
+      "learning_rate": 9.973537081480056e-06,
+      "loss": 0.635,
+      "step": 1010
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 5.7026777267456055,
+      "learning_rate": 9.971188194237141e-06,
+      "loss": 0.6388,
+      "step": 1020
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 4.8541388511657715,
+      "learning_rate": 9.96873974696834e-06,
+      "loss": 0.6225,
+      "step": 1030
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.6458210945129395,
+      "learning_rate": 9.966191788709716e-06,
+      "loss": 0.6373,
+      "step": 1040
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 17.634096145629883,
+      "learning_rate": 9.96354437049027e-06,
+      "loss": 0.645,
+      "step": 1050
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 9.264060974121094,
+      "learning_rate": 9.960797545330936e-06,
+      "loss": 0.6341,
+      "step": 1060
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 6.132219314575195,
+      "learning_rate": 9.95795136824351e-06,
+      "loss": 0.6302,
+      "step": 1070
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.637242794036865,
+      "learning_rate": 9.955005896229543e-06,
+      "loss": 0.6363,
+      "step": 1080
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 5.741671085357666,
+      "learning_rate": 9.951961188279216e-06,
+      "loss": 0.6415,
+      "step": 1090
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.340778827667236,
+      "learning_rate": 9.948817305370145e-06,
+      "loss": 0.6324,
+      "step": 1100
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.51924467086792,
+      "learning_rate": 9.94557431046616e-06,
+      "loss": 0.6301,
+      "step": 1110
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 82.26183319091797,
+      "learning_rate": 9.942232268516051e-06,
+      "loss": 0.6402,
+      "step": 1120
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 7.635751247406006,
+      "learning_rate": 9.938791246452267e-06,
+      "loss": 0.6295,
+      "step": 1130
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 6.102968215942383,
+      "learning_rate": 9.935251313189564e-06,
+      "loss": 0.6322,
+      "step": 1140
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.711689472198486,
+      "learning_rate": 9.931612539623643e-06,
+      "loss": 0.6433,
+      "step": 1150
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 7.467568874359131,
+      "learning_rate": 9.927874998629714e-06,
+      "loss": 0.6415,
+      "step": 1160
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 5.802877426147461,
+      "learning_rate": 9.924038765061042e-06,
+      "loss": 0.6469,
+      "step": 1170
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.440192699432373,
+      "learning_rate": 9.920103915747452e-06,
+      "loss": 0.6352,
+      "step": 1180
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.665472984313965,
+      "learning_rate": 9.916070529493785e-06,
+      "loss": 0.628,
+      "step": 1190
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.986988067626953,
+      "learning_rate": 9.911938687078324e-06,
+      "loss": 0.6292,
+      "step": 1200
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.383033752441406,
+      "learning_rate": 9.907708471251173e-06,
+      "loss": 0.6214,
+      "step": 1210
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 49.42042541503906,
+      "learning_rate": 9.9033799667326e-06,
+      "loss": 0.6292,
+      "step": 1220
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 5.570804595947266,
+      "learning_rate": 9.89895326021134e-06,
+      "loss": 0.6371,
+      "step": 1230
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 6.779324531555176,
+      "learning_rate": 9.89442844034286e-06,
+      "loss": 0.6396,
+      "step": 1240
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.269560813903809,
+      "learning_rate": 9.889805597747588e-06,
+      "loss": 0.6405,
+      "step": 1250
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.74788236618042,
+      "learning_rate": 9.885084825009085e-06,
+      "loss": 0.6302,
+      "step": 1260
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.128660202026367,
+      "learning_rate": 9.88026621667221e-06,
+      "loss": 0.6307,
+      "step": 1270
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.70562744140625,
+      "learning_rate": 9.875349869241202e-06,
+      "loss": 0.6275,
+      "step": 1280
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.813882827758789,
+      "learning_rate": 9.870335881177774e-06,
+      "loss": 0.6238,
+      "step": 1290
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.28445291519165,
+      "learning_rate": 9.86522435289912e-06,
+      "loss": 0.6433,
+      "step": 1300
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 6.097434043884277,
+      "learning_rate": 9.860015386775915e-06,
+      "loss": 0.6237,
+      "step": 1310
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 5.640145778656006,
+      "learning_rate": 9.854709087130261e-06,
+      "loss": 0.6294,
+      "step": 1320
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.536041259765625,
+      "learning_rate": 9.849305560233598e-06,
+      "loss": 0.6262,
+      "step": 1330
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 13.29723834991455,
+      "learning_rate": 9.843804914304578e-06,
+      "loss": 0.6426,
+      "step": 1340
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.7919230461120605,
+      "learning_rate": 9.838207259506891e-06,
+      "loss": 0.6376,
+      "step": 1350
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.46877908706665,
+      "learning_rate": 9.83251270794707e-06,
+      "loss": 0.6111,
+      "step": 1360
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.7596516609191895,
+      "learning_rate": 9.826721373672235e-06,
+      "loss": 0.6405,
+      "step": 1370
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 5.5958428382873535,
+      "learning_rate": 9.820833372667813e-06,
+      "loss": 0.6311,
+      "step": 1380
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 6.962647438049316,
+      "learning_rate": 9.814848822855216e-06,
+      "loss": 0.6336,
+      "step": 1390
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.83704948425293,
+      "learning_rate": 9.80876784408948e-06,
+      "loss": 0.6271,
+      "step": 1400
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.530167579650879,
+      "learning_rate": 9.802590558156863e-06,
+      "loss": 0.6261,
+      "step": 1410
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.446717739105225,
+      "learning_rate": 9.796317088772402e-06,
+      "loss": 0.618,
+      "step": 1420
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.7587080001831055,
+      "learning_rate": 9.789947561577445e-06,
+      "loss": 0.6295,
+      "step": 1430
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 6.391341686248779,
+      "learning_rate": 9.783482104137127e-06,
+      "loss": 0.6232,
+      "step": 1440
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.596224308013916,
+      "learning_rate": 9.776920845937816e-06,
+      "loss": 0.6175,
+      "step": 1450
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.266319751739502,
+      "learning_rate": 9.770263918384523e-06,
+      "loss": 0.6237,
+      "step": 1460
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 5.594700813293457,
+      "learning_rate": 9.763511454798268e-06,
+      "loss": 0.6164,
+      "step": 1470
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.352781772613525,
+      "learning_rate": 9.75666359041341e-06,
+      "loss": 0.6521,
+      "step": 1480
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.978342533111572,
+      "learning_rate": 9.749720462374939e-06,
+      "loss": 0.6247,
+      "step": 1490
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.177164554595947,
+      "learning_rate": 9.742682209735727e-06,
+      "loss": 0.6387,
+      "step": 1500
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.946715354919434,
+      "learning_rate": 9.735548973453753e-06,
+      "loss": 0.6445,
+      "step": 1510
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.191423416137695,
+      "learning_rate": 9.728320896389263e-06,
+      "loss": 0.6353,
+      "step": 1520
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 5.473761558532715,
+      "learning_rate": 9.720998123301924e-06,
+      "loss": 0.6223,
+      "step": 1530
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 6.137279033660889,
+      "learning_rate": 9.713580800847917e-06,
+      "loss": 0.6345,
+      "step": 1540
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.810376167297363,
+      "learning_rate": 9.706069077577e-06,
+      "loss": 0.6096,
+      "step": 1550
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.427652835845947,
+      "learning_rate": 9.698463103929542e-06,
+      "loss": 0.6133,
+      "step": 1560
+    },
+    {
+      "epoch": 0.21,
+      "eval_loss": 0.6688335537910461,
+      "eval_runtime": 169.0314,
+      "eval_samples_per_second": 65.077,
+      "eval_steps_per_second": 8.135,
+      "step": 1560
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.631854057312012,
+      "learning_rate": 9.690763032233498e-06,
+      "loss": 0.629,
+      "step": 1570
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 6.538657188415527,
+      "learning_rate": 9.682969016701357e-06,
+      "loss": 0.6359,
+      "step": 1580
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 10.069457054138184,
+      "learning_rate": 9.675081213427076e-06,
+      "loss": 0.6302,
+      "step": 1590
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 12.996840476989746,
+      "learning_rate": 9.66709978038292e-06,
+      "loss": 0.6616,
+      "step": 1600
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.648403644561768,
+      "learning_rate": 9.659024877416328e-06,
+      "loss": 0.6615,
+      "step": 1610
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 5.897425174713135,
+      "learning_rate": 9.650856666246693e-06,
+      "loss": 0.6362,
+      "step": 1620
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.605139255523682,
+      "learning_rate": 9.642595310462133e-06,
+      "loss": 0.6448,
+      "step": 1630
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.556352138519287,
+      "learning_rate": 9.63424097551621e-06,
+      "loss": 0.6246,
+      "step": 1640
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 6.256978511810303,
+      "learning_rate": 9.62579382872462e-06,
+      "loss": 0.6282,
+      "step": 1650
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.5795793533325195,
+      "learning_rate": 9.617254039261835e-06,
+      "loss": 0.6331,
+      "step": 1660
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.844163417816162,
+      "learning_rate": 9.608621778157722e-06,
+      "loss": 0.624,
+      "step": 1670
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 6.517988204956055,
+      "learning_rate": 9.599897218294122e-06,
+      "loss": 0.6327,
+      "step": 1680
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.43684720993042,
+      "learning_rate": 9.591080534401371e-06,
+      "loss": 0.6252,
+      "step": 1690
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 5.179018020629883,
+      "learning_rate": 9.582171903054815e-06,
+      "loss": 0.6149,
+      "step": 1700
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.8681793212890625,
+      "learning_rate": 9.573171502671273e-06,
+      "loss": 0.6292,
+      "step": 1710
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.400745391845703,
+      "learning_rate": 9.564079513505455e-06,
+      "loss": 0.6207,
+      "step": 1720
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.303137302398682,
+      "learning_rate": 9.554896117646357e-06,
+      "loss": 0.6162,
+      "step": 1730
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.112619876861572,
+      "learning_rate": 9.54562149901362e-06,
+      "loss": 0.6301,
+      "step": 1740
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.014323711395264,
+      "learning_rate": 9.536255843353832e-06,
+      "loss": 0.6144,
+      "step": 1750
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.302499294281006,
+      "learning_rate": 9.526799338236828e-06,
+      "loss": 0.637,
+      "step": 1760
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 5.237521171569824,
+      "learning_rate": 9.517252173051912e-06,
+      "loss": 0.6092,
+      "step": 1770
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.327495098114014,
+      "learning_rate": 9.507614539004082e-06,
+      "loss": 0.6364,
+      "step": 1780
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.794159412384033,
+      "learning_rate": 9.497886629110187e-06,
+      "loss": 0.6382,
+      "step": 1790
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.391251564025879,
+      "learning_rate": 9.488068638195072e-06,
+      "loss": 0.62,
+      "step": 1800
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.799778461456299,
+      "learning_rate": 9.478160762887668e-06,
+      "loss": 0.6196,
+      "step": 1810
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.956018924713135,
+      "learning_rate": 9.468163201617063e-06,
+      "loss": 0.6286,
+      "step": 1820
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.173859596252441,
+      "learning_rate": 9.458076154608515e-06,
+      "loss": 0.6208,
+      "step": 1830
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 6.374012470245361,
+      "learning_rate": 9.447899823879456e-06,
+      "loss": 0.625,
+      "step": 1840
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 5.741364479064941,
+      "learning_rate": 9.437634413235437e-06,
+      "loss": 0.6869,
+      "step": 1850
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.758192539215088,
+      "learning_rate": 9.427280128266049e-06,
+      "loss": 0.6102,
+      "step": 1860
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.713798522949219,
+      "learning_rate": 9.416837176340807e-06,
+      "loss": 0.6215,
+      "step": 1870
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 5.849081516265869,
+      "learning_rate": 9.406305766604996e-06,
+      "loss": 0.6158,
+      "step": 1880
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.048788547515869,
+      "learning_rate": 9.395686109975475e-06,
+      "loss": 0.6282,
+      "step": 1890
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 9.530996322631836,
+      "learning_rate": 9.384978419136469e-06,
+      "loss": 0.6135,
+      "step": 1900
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.369543075561523,
+      "learning_rate": 9.374182908535293e-06,
+      "loss": 0.6104,
+      "step": 1910
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 6.199366569519043,
+      "learning_rate": 9.363299794378072e-06,
+      "loss": 0.6199,
+      "step": 1920
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.299448013305664,
+      "learning_rate": 9.352329294625397e-06,
+      "loss": 0.6286,
+      "step": 1930
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.6097917556762695,
+      "learning_rate": 9.34127162898797e-06,
+      "loss": 0.6196,
+      "step": 1940
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.62219762802124,
+      "learning_rate": 9.330127018922195e-06,
+      "loss": 0.6244,
+      "step": 1950
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.205328464508057,
+      "learning_rate": 9.318895687625752e-06,
+      "loss": 0.6089,
+      "step": 1960
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.849984169006348,
+      "learning_rate": 9.307577860033123e-06,
+      "loss": 0.6213,
+      "step": 1970
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 6.027492523193359,
+      "learning_rate": 9.296173762811084e-06,
+      "loss": 0.6192,
+      "step": 1980
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 7.360931873321533,
+      "learning_rate": 9.284683624354172e-06,
+      "loss": 0.6313,
+      "step": 1990
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 5.995561599731445,
+      "learning_rate": 9.273107674780102e-06,
+      "loss": 0.6273,
+      "step": 2000
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.6998701095581055,
+      "learning_rate": 9.261446145925167e-06,
+      "loss": 0.624,
+      "step": 2010
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 7.960203647613525,
+      "learning_rate": 9.249699271339594e-06,
+      "loss": 0.6119,
+      "step": 2020
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.833003997802734,
+      "learning_rate": 9.237867286282855e-06,
+      "loss": 0.6615,
+      "step": 2030
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.727327346801758,
+      "learning_rate": 9.225950427718974e-06,
+      "loss": 0.6363,
+      "step": 2040
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.32903528213501,
+      "learning_rate": 9.213948934311767e-06,
+      "loss": 0.6382,
+      "step": 2050
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 6.892453670501709,
+      "learning_rate": 9.201863046420065e-06,
+      "loss": 0.6196,
+      "step": 2060
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.348752498626709,
+      "learning_rate": 9.189693006092907e-06,
+      "loss": 0.6104,
+      "step": 2070
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.757750034332275,
+      "learning_rate": 9.177439057064684e-06,
+      "loss": 0.6153,
+      "step": 2080
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 22.043140411376953,
+      "learning_rate": 9.165101444750259e-06,
+      "loss": 0.6178,
+      "step": 2090
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.820219993591309,
+      "learning_rate": 9.152680416240059e-06,
+      "loss": 2.4575,
+      "step": 2100
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.317713737487793,
+      "learning_rate": 9.140176220295118e-06,
+      "loss": 1.9869,
+      "step": 2110
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 5.32202672958374,
+      "learning_rate": 9.1275891073421e-06,
+      "loss": 0.6366,
+      "step": 2120
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.378438949584961,
+      "learning_rate": 9.114919329468283e-06,
+      "loss": 0.6138,
+      "step": 2130
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 6.984612941741943,
+      "learning_rate": 9.102167140416503e-06,
+      "loss": 0.6239,
+      "step": 2140
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 9.644604682922363,
+      "learning_rate": 9.089332795580085e-06,
+      "loss": 0.6094,
+      "step": 2150
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.99331521987915,
+      "learning_rate": 9.076416551997721e-06,
+      "loss": 0.6358,
+      "step": 2160
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.56105899810791,
+      "learning_rate": 9.063418668348323e-06,
+      "loss": 0.6118,
+      "step": 2170
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.86183500289917,
+      "learning_rate": 9.050339404945834e-06,
+      "loss": 0.6402,
+      "step": 2180
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 7.894756317138672,
+      "learning_rate": 9.037179023734036e-06,
+      "loss": 0.6323,
+      "step": 2190
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.089447975158691,
+      "learning_rate": 9.023937788281278e-06,
+      "loss": 0.6279,
+      "step": 2200
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 5.405038833618164,
+      "learning_rate": 9.01061596377522e-06,
+      "loss": 0.6088,
+      "step": 2210
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 6.3252949714660645,
+      "learning_rate": 8.997213817017508e-06,
+      "loss": 0.618,
+      "step": 2220
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.129386901855469,
+      "learning_rate": 8.98373161641843e-06,
+      "loss": 0.6124,
+      "step": 2230
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 7.108587265014648,
+      "learning_rate": 8.970169631991556e-06,
+      "loss": 0.627,
+      "step": 2240
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.341206073760986,
+      "learning_rate": 8.95652813534831e-06,
+      "loss": 0.6217,
+      "step": 2250
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 7.05128812789917,
+      "learning_rate": 8.942807399692543e-06,
+      "loss": 0.6335,
+      "step": 2260
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.670904636383057,
+      "learning_rate": 8.92900769981506e-06,
+      "loss": 0.6246,
+      "step": 2270
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.543381690979004,
+      "learning_rate": 8.915129312088112e-06,
+      "loss": 0.6193,
+      "step": 2280
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 5.513144016265869,
+      "learning_rate": 8.901172514459864e-06,
+      "loss": 0.5986,
+      "step": 2290
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 6.7181854248046875,
+      "learning_rate": 8.88713758644883e-06,
+      "loss": 0.5983,
+      "step": 2300
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 6.5360002517700195,
+      "learning_rate": 8.873024809138272e-06,
+      "loss": 0.6202,
+      "step": 2310
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 6.717379093170166,
+      "learning_rate": 8.858834465170576e-06,
+      "loss": 0.6232,
+      "step": 2320
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 8.741230964660645,
+      "learning_rate": 8.84456683874158e-06,
+      "loss": 0.6154,
+      "step": 2330
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 6.7567338943481445,
+      "learning_rate": 8.83022221559489e-06,
+      "loss": 0.607,
+      "step": 2340
+    },
+    {
+      "epoch": 0.31,
+      "eval_loss": 0.6728433966636658,
+      "eval_runtime": 169.7753,
+      "eval_samples_per_second": 64.792,
+      "eval_steps_per_second": 8.099,
+      "step": 2340
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.661694526672363,
+      "learning_rate": 8.815800883016168e-06,
+      "loss": 0.6176,
+      "step": 2350
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 6.330115795135498,
+      "learning_rate": 8.801303129827352e-06,
+      "loss": 0.6089,
+      "step": 2360
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.605167388916016,
+      "learning_rate": 8.786729246380901e-06,
+      "loss": 0.6127,
+      "step": 2370
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 5.510767459869385,
+      "learning_rate": 8.772079524553951e-06,
+      "loss": 0.6199,
+      "step": 2380
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 5.994042873382568,
+      "learning_rate": 8.757354257742501e-06,
+      "loss": 0.6245,
+      "step": 2390
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.51716947555542,
+      "learning_rate": 8.742553740855507e-06,
+      "loss": 0.6305,
+      "step": 2400
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.130660533905029,
+      "learning_rate": 8.727678270308994e-06,
+      "loss": 0.616,
+      "step": 2410
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.110134601593018,
+      "learning_rate": 8.712728144020118e-06,
+      "loss": 0.6277,
+      "step": 2420
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.650820255279541,
+      "learning_rate": 8.697703661401187e-06,
+      "loss": 0.6219,
+      "step": 2430
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 6.013575553894043,
+      "learning_rate": 8.682605123353685e-06,
+      "loss": 0.6268,
+      "step": 2440
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 7.728272914886475,
+      "learning_rate": 8.66743283226223e-06,
+      "loss": 0.6186,
+      "step": 2450
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.504591464996338,
+      "learning_rate": 8.652187091988516e-06,
+      "loss": 0.6007,
+      "step": 2460
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.214788436889648,
+      "learning_rate": 8.636868207865244e-06,
+      "loss": 0.6032,
+      "step": 2470
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.597141265869141,
+      "learning_rate": 8.621476486689991e-06,
+      "loss": 0.6086,
+      "step": 2480
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 7.287806034088135,
+      "learning_rate": 8.606012236719073e-06,
+      "loss": 0.6171,
+      "step": 2490
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.4462714195251465,
+      "learning_rate": 8.590475767661371e-06,
+      "loss": 0.6078,
+      "step": 2500
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 5.836272716522217,
+      "learning_rate": 8.574867390672124e-06,
+      "loss": 0.6125,
+      "step": 2510
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 8.247157096862793,
+      "learning_rate": 8.559187418346703e-06,
+      "loss": 0.6307,
+      "step": 2520
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 6.056578159332275,
+      "learning_rate": 8.543436164714351e-06,
+      "loss": 0.6274,
+      "step": 2530
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.924829483032227,
+      "learning_rate": 8.527613945231886e-06,
+      "loss": 0.6155,
+      "step": 2540
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.2663469314575195,
+      "learning_rate": 8.511721076777388e-06,
+      "loss": 0.6092,
+      "step": 2550
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.494378566741943,
+      "learning_rate": 8.495757877643857e-06,
+      "loss": 0.6178,
+      "step": 2560
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 6.344980239868164,
+      "learning_rate": 8.479724667532836e-06,
+      "loss": 0.6144,
+      "step": 2570
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.618322849273682,
+      "learning_rate": 8.463621767547998e-06,
+      "loss": 0.6056,
+      "step": 2580
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.677299499511719,
+      "learning_rate": 8.447449500188731e-06,
+      "loss": 0.6094,
+      "step": 2590
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.477785110473633,
+      "learning_rate": 8.43120818934367e-06,
+      "loss": 0.6279,
+      "step": 2600
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.992845058441162,
+      "learning_rate": 8.414898160284208e-06,
+      "loss": 0.6112,
+      "step": 2610
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.6498847007751465,
+      "learning_rate": 8.398519739657997e-06,
+      "loss": 0.6196,
+      "step": 2620
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.752150535583496,
+      "learning_rate": 8.382073255482381e-06,
+      "loss": 0.606,
+      "step": 2630
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.367984294891357,
+      "learning_rate": 8.36555903713785e-06,
+      "loss": 0.6123,
+      "step": 2640
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.714753150939941,
+      "learning_rate": 8.348977415361435e-06,
+      "loss": 0.6064,
+      "step": 2650
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 7.572054386138916,
+      "learning_rate": 8.332328722240072e-06,
+      "loss": 0.6109,
+      "step": 2660
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.684463024139404,
+      "learning_rate": 8.315613291203977e-06,
+      "loss": 0.6341,
+      "step": 2670
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 5.910693645477295,
+      "learning_rate": 8.298831457019943e-06,
+      "loss": 0.6156,
+      "step": 2680
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.144363880157471,
+      "learning_rate": 8.28198355578465e-06,
+      "loss": 0.6085,
+      "step": 2690
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.353940963745117,
+      "learning_rate": 8.265069924917925e-06,
+      "loss": 0.6048,
+      "step": 2700
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.588951587677002,
+      "learning_rate": 8.248090903156003e-06,
+      "loss": 0.6272,
+      "step": 2710
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.791905879974365,
+      "learning_rate": 8.231046830544716e-06,
+      "loss": 0.6183,
+      "step": 2720
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 6.0394744873046875,
+      "learning_rate": 8.213938048432697e-06,
+      "loss": 0.6069,
+      "step": 2730
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 8.518630027770996,
+      "learning_rate": 8.196764899464552e-06,
+      "loss": 0.6111,
+      "step": 2740
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 5.609602451324463,
+      "learning_rate": 8.179527727573975e-06,
+      "loss": 0.6239,
+      "step": 2750
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 10.055850982666016,
+      "learning_rate": 8.162226877976886e-06,
+      "loss": 0.6115,
+      "step": 2760
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.720135688781738,
+      "learning_rate": 8.144862697164499e-06,
+      "loss": 0.6025,
+      "step": 2770
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 7.587475299835205,
+      "learning_rate": 8.127435532896388e-06,
+      "loss": 0.6135,
+      "step": 2780
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 8.063986778259277,
+      "learning_rate": 8.10994573419352e-06,
+      "loss": 0.5999,
+      "step": 2790
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 7.293218612670898,
+      "learning_rate": 8.092393651331275e-06,
+      "loss": 0.6108,
+      "step": 2800
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.801172733306885,
+      "learning_rate": 8.074779635832417e-06,
+      "loss": 0.608,
+      "step": 2810
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 6.789213180541992,
+      "learning_rate": 8.057104040460062e-06,
+      "loss": 0.6105,
+      "step": 2820
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 5.721503734588623,
+      "learning_rate": 8.03936721921061e-06,
+      "loss": 0.6214,
+      "step": 2830
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.258848190307617,
+      "learning_rate": 8.021569527306663e-06,
+      "loss": 0.6174,
+      "step": 2840
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.338409900665283,
+      "learning_rate": 8.003711321189895e-06,
+      "loss": 0.6159,
+      "step": 2850
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.426341533660889,
+      "learning_rate": 7.985792958513932e-06,
+      "loss": 0.6152,
+      "step": 2860
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 7.031759262084961,
+      "learning_rate": 7.967814798137173e-06,
+      "loss": 0.6261,
+      "step": 2870
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 5.78716516494751,
+      "learning_rate": 7.949777200115617e-06,
+      "loss": 0.6173,
+      "step": 2880
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 20.726688385009766,
+      "learning_rate": 7.931680525695634e-06,
+      "loss": 0.6178,
+      "step": 2890
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 6.066954612731934,
+      "learning_rate": 7.913525137306756e-06,
+      "loss": 0.6176,
+      "step": 2900
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.461245536804199,
+      "learning_rate": 7.895311398554395e-06,
+      "loss": 0.6169,
+      "step": 2910
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.37878942489624,
+      "learning_rate": 7.877039674212569e-06,
+      "loss": 0.6105,
+      "step": 2920
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 47.42095947265625,
+      "learning_rate": 7.8587103302166e-06,
+      "loss": 0.617,
+      "step": 2930
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.993823528289795,
+      "learning_rate": 7.84032373365578e-06,
+      "loss": 0.6286,
+      "step": 2940
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.397260665893555,
+      "learning_rate": 7.821880252766025e-06,
+      "loss": 0.6031,
+      "step": 2950
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.4610915184021,
+      "learning_rate": 7.803380256922495e-06,
+      "loss": 0.6303,
+      "step": 2960
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 6.266001224517822,
+      "learning_rate": 7.784824116632198e-06,
+      "loss": 0.613,
+      "step": 2970
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 5.821497917175293,
+      "learning_rate": 7.76621220352657e-06,
+      "loss": 0.6156,
+      "step": 2980
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.601454257965088,
+      "learning_rate": 7.747544890354031e-06,
+      "loss": 0.6169,
+      "step": 2990
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.371436595916748,
+      "learning_rate": 7.728822550972523e-06,
+      "loss": 0.6099,
+      "step": 3000
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.537062168121338,
+      "learning_rate": 7.710045560342021e-06,
+      "loss": 0.6175,
+      "step": 3010
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.325994491577148,
+      "learning_rate": 7.69121429451702e-06,
+      "loss": 0.6005,
+      "step": 3020
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 5.711528778076172,
+      "learning_rate": 7.672329130639007e-06,
+      "loss": 0.612,
+      "step": 3030
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.956279754638672,
+      "learning_rate": 7.65339044692891e-06,
+      "loss": 0.6123,
+      "step": 3040
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 6.838433265686035,
+      "learning_rate": 7.634398622679517e-06,
+      "loss": 0.5872,
+      "step": 3050
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 10.948911666870117,
+      "learning_rate": 7.615354038247889e-06,
+      "loss": 0.6257,
+      "step": 3060
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 7.123111248016357,
+      "learning_rate": 7.59625707504773e-06,
+      "loss": 0.619,
+      "step": 3070
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 7.197707653045654,
+      "learning_rate": 7.577108115541761e-06,
+      "loss": 0.6077,
+      "step": 3080
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5471.7568359375,
+      "learning_rate": 7.557907543234051e-06,
+      "loss": 0.6283,
+      "step": 3090
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 6.704062461853027,
+      "learning_rate": 7.53865574266234e-06,
+      "loss": 0.6213,
+      "step": 3100
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.542397499084473,
+      "learning_rate": 7.519353099390336e-06,
+      "loss": 0.6223,
+      "step": 3110
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.533992767333984,
+      "learning_rate": 7.500000000000001e-06,
+      "loss": 0.6051,
+      "step": 3120
+    },
+    {
+      "epoch": 0.41,
+      "eval_loss": 0.6693721413612366,
+      "eval_runtime": 169.48,
+      "eval_samples_per_second": 64.904,
+      "eval_steps_per_second": 8.113,
+      "step": 3120
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 5.968539714813232,
+      "learning_rate": 7.480596832083795e-06,
+      "loss": 0.6143,
+      "step": 3130
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.992849349975586,
+      "learning_rate": 7.461143984236925e-06,
+      "loss": 0.616,
+      "step": 3140
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.622570037841797,
+      "learning_rate": 7.441641846049557e-06,
+      "loss": 0.609,
+      "step": 3150
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.672114849090576,
+      "learning_rate": 7.422090808099014e-06,
+      "loss": 0.6175,
+      "step": 3160
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.839463710784912,
+      "learning_rate": 7.402491261941958e-06,
+      "loss": 0.6085,
+      "step": 3170
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.008167266845703,
+      "learning_rate": 7.382843600106539e-06,
+      "loss": 0.6069,
+      "step": 3180
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 6.840440273284912,
+      "learning_rate": 7.363148216084548e-06,
+      "loss": 0.6132,
+      "step": 3190
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.5630693435668945,
+      "learning_rate": 7.343405504323519e-06,
+      "loss": 0.6065,
+      "step": 3200
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 5.013092994689941,
+      "learning_rate": 7.323615860218844e-06,
+      "loss": 0.6123,
+      "step": 3210
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 48.6692008972168,
+      "learning_rate": 7.303779680105844e-06,
+      "loss": 0.6109,
+      "step": 3220
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.752770900726318,
+      "learning_rate": 7.28389736125184e-06,
+      "loss": 0.606,
+      "step": 3230
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.404486656188965,
+      "learning_rate": 7.263969301848188e-06,
+      "loss": 0.6109,
+      "step": 3240
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.718954563140869,
+      "learning_rate": 7.243995901002312e-06,
+      "loss": 0.6046,
+      "step": 3250
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 6.1195831298828125,
+      "learning_rate": 7.223977558729707e-06,
+      "loss": 0.6072,
+      "step": 3260
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.4051337242126465,
+      "learning_rate": 7.203914675945929e-06,
+      "loss": 0.6212,
+      "step": 3270
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 5.423203945159912,
+      "learning_rate": 7.183807654458565e-06,
+      "loss": 0.6204,
+      "step": 3280
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.746906757354736,
+      "learning_rate": 7.163656896959181e-06,
+      "loss": 0.6023,
+      "step": 3290
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.6502156257629395,
+      "learning_rate": 7.143462807015271e-06,
+      "loss": 0.6123,
+      "step": 3300
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 7.021644592285156,
+      "learning_rate": 7.1232257890621605e-06,
+      "loss": 0.6094,
+      "step": 3310
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.808534145355225,
+      "learning_rate": 7.102946248394908e-06,
+      "loss": 0.628,
+      "step": 3320
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.422878265380859,
+      "learning_rate": 7.082624591160201e-06,
+      "loss": 0.5978,
+      "step": 3330
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 5.411374568939209,
+      "learning_rate": 7.0622612243482035e-06,
+      "loss": 0.611,
+      "step": 3340
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 6.385402679443359,
+      "learning_rate": 7.041856555784421e-06,
+      "loss": 0.5982,
+      "step": 3350
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 4.959177494049072,
+      "learning_rate": 7.021410994121525e-06,
+      "loss": 0.5913,
+      "step": 3360
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 7.832031726837158,
+      "learning_rate": 7.0009249488311685e-06,
+      "loss": 0.615,
+      "step": 3370
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 7.2229485511779785,
+      "learning_rate": 6.980398830195785e-06,
+      "loss": 0.619,
+      "step": 3380
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 5.132372856140137,
+      "learning_rate": 6.959833049300376e-06,
+      "loss": 0.6236,
+      "step": 3390
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 5.803369045257568,
+      "learning_rate": 6.939228018024275e-06,
+      "loss": 0.6091,
+      "step": 3400
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 5.913552284240723,
+      "learning_rate": 6.918584149032899e-06,
+      "loss": 0.61,
+      "step": 3410
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 5.712104797363281,
+      "learning_rate": 6.897901855769483e-06,
+      "loss": 0.6139,
+      "step": 3420
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 6.360735893249512,
+      "learning_rate": 6.8771815524468e-06,
+      "loss": 0.606,
+      "step": 3430
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 7.527002334594727,
+      "learning_rate": 6.856423654038868e-06,
+      "loss": 0.5892,
+      "step": 3440
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.143304824829102,
+      "learning_rate": 6.8356285762726385e-06,
+      "loss": 0.6041,
+      "step": 3450
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.828638553619385,
+      "learning_rate": 6.814796735619664e-06,
+      "loss": 0.5932,
+      "step": 3460
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.889917373657227,
+      "learning_rate": 6.793928549287767e-06,
+      "loss": 0.6117,
+      "step": 3470
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.243914604187012,
+      "learning_rate": 6.773024435212678e-06,
+      "loss": 0.6034,
+      "step": 3480
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 6.9411420822143555,
+      "learning_rate": 6.75208481204967e-06,
+      "loss": 0.6003,
+      "step": 3490
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.287111282348633,
+      "learning_rate": 6.731110099165165e-06,
+      "loss": 0.6208,
+      "step": 3500
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 5.86116886138916,
+      "learning_rate": 6.710100716628345e-06,
+      "loss": 0.597,
+      "step": 3510
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.526810646057129,
+      "learning_rate": 6.689057085202737e-06,
+      "loss": 0.6052,
+      "step": 3520
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.855119228363037,
+      "learning_rate": 6.667979626337783e-06,
+      "loss": 0.6147,
+      "step": 3530
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.467052936553955,
+      "learning_rate": 6.646868762160399e-06,
+      "loss": 0.6221,
+      "step": 3540
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.706518650054932,
+      "learning_rate": 6.625724915466526e-06,
+      "loss": 0.6083,
+      "step": 3550
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 5.94938325881958,
+      "learning_rate": 6.6045485097126585e-06,
+      "loss": 0.6067,
+      "step": 3560
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.971024990081787,
+      "learning_rate": 6.583339969007364e-06,
+      "loss": 0.6172,
+      "step": 3570
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.673549652099609,
+      "learning_rate": 6.562099718102788e-06,
+      "loss": 0.5947,
+      "step": 3580
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.783125400543213,
+      "learning_rate": 6.540828182386154e-06,
+      "loss": 0.6101,
+      "step": 3590
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.100213527679443,
+      "learning_rate": 6.519525787871235e-06,
+      "loss": 0.6201,
+      "step": 3600
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.297542095184326,
+      "learning_rate": 6.498192961189829e-06,
+      "loss": 0.6125,
+      "step": 3610
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 7.120528697967529,
+      "learning_rate": 6.476830129583207e-06,
+      "loss": 0.6224,
+      "step": 3620
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 6.501312255859375,
+      "learning_rate": 6.455437720893565e-06,
+      "loss": 0.6167,
+      "step": 3630
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.54337739944458,
+      "learning_rate": 6.434016163555452e-06,
+      "loss": 0.6052,
+      "step": 3640
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 7.030393123626709,
+      "learning_rate": 6.412565886587186e-06,
+      "loss": 0.6056,
+      "step": 3650
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 5.686173915863037,
+      "learning_rate": 6.391087319582264e-06,
+      "loss": 0.6069,
+      "step": 3660
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.340301990509033,
+      "learning_rate": 6.369580892700763e-06,
+      "loss": 0.6117,
+      "step": 3670
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 7.223570823669434,
+      "learning_rate": 6.34804703666072e-06,
+      "loss": 0.6086,
+      "step": 3680
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.784290790557861,
+      "learning_rate": 6.326486182729504e-06,
+      "loss": 0.603,
+      "step": 3690
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 6.71796178817749,
+      "learning_rate": 6.304898762715187e-06,
+      "loss": 0.6162,
+      "step": 3700
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.618252277374268,
+      "learning_rate": 6.283285208957884e-06,
+      "loss": 0.6067,
+      "step": 3710
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.947731971740723,
+      "learning_rate": 6.261645954321109e-06,
+      "loss": 0.5991,
+      "step": 3720
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.969636917114258,
+      "learning_rate": 6.239981432183093e-06,
+      "loss": 0.5944,
+      "step": 3730
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.365227222442627,
+      "learning_rate": 6.21829207642811e-06,
+      "loss": 0.5966,
+      "step": 3740
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.247452735900879,
+      "learning_rate": 6.1965783214377895e-06,
+      "loss": 0.611,
+      "step": 3750
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.950301170349121,
+      "learning_rate": 6.1748406020824115e-06,
+      "loss": 0.6248,
+      "step": 3760
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.410281658172607,
+      "learning_rate": 6.153079353712201e-06,
+      "loss": 0.6153,
+      "step": 3770
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 5.590805530548096,
+      "learning_rate": 6.131295012148613e-06,
+      "loss": 0.5987,
+      "step": 3780
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 5.457326412200928,
+      "learning_rate": 6.1094880136755886e-06,
+      "loss": 0.6026,
+      "step": 3790
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.030434608459473,
+      "learning_rate": 6.087658795030838e-06,
+      "loss": 0.6042,
+      "step": 3800
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 6.162191390991211,
+      "learning_rate": 6.065807793397081e-06,
+      "loss": 0.6068,
+      "step": 3810
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 5.414748191833496,
+      "learning_rate": 6.043935446393294e-06,
+      "loss": 0.6108,
+      "step": 3820
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.97568941116333,
+      "learning_rate": 6.022042192065946e-06,
+      "loss": 0.6181,
+      "step": 3830
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 5.456526756286621,
+      "learning_rate": 6.000128468880223e-06,
+      "loss": 0.602,
+      "step": 3840
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 5.599926471710205,
+      "learning_rate": 5.978194715711254e-06,
+      "loss": 0.6007,
+      "step": 3850
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.381017684936523,
+      "learning_rate": 5.956241371835312e-06,
+      "loss": 0.6228,
+      "step": 3860
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.424213409423828,
+      "learning_rate": 5.934268876921026e-06,
+      "loss": 0.6139,
+      "step": 3870
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 6.539239406585693,
+      "learning_rate": 5.912277671020564e-06,
+      "loss": 0.6137,
+      "step": 3880
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 5.962428569793701,
+      "learning_rate": 5.890268194560834e-06,
+      "loss": 0.6146,
+      "step": 3890
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 7.252692222595215,
+      "learning_rate": 5.8682408883346535e-06,
+      "loss": 0.6106,
+      "step": 3900
+    },
+    {
+      "epoch": 0.52,
+      "eval_loss": 0.665640652179718,
+      "eval_runtime": 169.5554,
+      "eval_samples_per_second": 64.876,
+      "eval_steps_per_second": 8.109,
+      "step": 3900
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 5.630374431610107,
+      "learning_rate": 5.846196193491919e-06,
+      "loss": 0.6113,
+      "step": 3910
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.424310684204102,
+      "learning_rate": 5.824134551530783e-06,
+      "loss": 0.6183,
+      "step": 3920
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.524292469024658,
+      "learning_rate": 5.8020564042888015e-06,
+      "loss": 0.5956,
+      "step": 3930
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.760373592376709,
+      "learning_rate": 5.77996219393409e-06,
+      "loss": 0.6046,
+      "step": 3940
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 17.418819427490234,
+      "learning_rate": 5.757852362956463e-06,
+      "loss": 0.5965,
+      "step": 3950
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 6.237477779388428,
+      "learning_rate": 5.735727354158581e-06,
+      "loss": 0.608,
+      "step": 3960
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 6.2491536140441895,
+      "learning_rate": 5.713587610647073e-06,
+      "loss": 0.6216,
+      "step": 3970
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 5.632169723510742,
+      "learning_rate": 5.6914335758236665e-06,
+      "loss": 0.6086,
+      "step": 3980
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 10.672119140625,
+      "learning_rate": 5.669265693376309e-06,
+      "loss": 0.6306,
+      "step": 3990
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 7.666195869445801,
+      "learning_rate": 5.647084407270277e-06,
+      "loss": 0.6212,
+      "step": 4000
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 7.483991622924805,
+      "learning_rate": 5.624890161739292e-06,
+      "loss": 0.6169,
+      "step": 4010
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 6.328516960144043,
+      "learning_rate": 5.6026834012766155e-06,
+      "loss": 0.6156,
+      "step": 4020
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 6.012834072113037,
+      "learning_rate": 5.5804645706261515e-06,
+      "loss": 0.5989,
+      "step": 4030
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 6.906214714050293,
+      "learning_rate": 5.5582341147735396e-06,
+      "loss": 0.5973,
+      "step": 4040
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 5.286527156829834,
+      "learning_rate": 5.53599247893724e-06,
+      "loss": 0.6118,
+      "step": 4050
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.227565288543701,
+      "learning_rate": 5.5137401085596224e-06,
+      "loss": 0.6164,
+      "step": 4060
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.331086158752441,
+      "learning_rate": 5.491477449298036e-06,
+      "loss": 0.599,
+      "step": 4070
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 5.511313438415527,
+      "learning_rate": 5.469204947015897e-06,
+      "loss": 0.6144,
+      "step": 4080
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.1719136238098145,
+      "learning_rate": 5.4469230477737466e-06,
+      "loss": 0.6062,
+      "step": 4090
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.096289157867432,
+      "learning_rate": 5.424632197820325e-06,
+      "loss": 0.5993,
+      "step": 4100
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 6.381209850311279,
+      "learning_rate": 5.402332843583631e-06,
+      "loss": 0.5882,
+      "step": 4110
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 5.978997230529785,
+      "learning_rate": 5.380025431661981e-06,
+      "loss": 0.6173,
+      "step": 4120
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 7.058841705322266,
+      "learning_rate": 5.3577104088150685e-06,
+      "loss": 0.6216,
+      "step": 4130
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 29.12560272216797,
+      "learning_rate": 5.335388221955012e-06,
+      "loss": 0.6031,
+      "step": 4140
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 5.8936357498168945,
+      "learning_rate": 5.31305931813741e-06,
+      "loss": 0.609,
+      "step": 4150
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 5.908172130584717,
+      "learning_rate": 5.290724144552379e-06,
+      "loss": 0.5983,
+      "step": 4160
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 5.777021408081055,
+      "learning_rate": 5.268383148515608e-06,
+      "loss": 0.6044,
+      "step": 4170
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 6.296756744384766,
+      "learning_rate": 5.246036777459391e-06,
+      "loss": 0.5939,
+      "step": 4180
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 6.158642292022705,
+      "learning_rate": 5.223685478923671e-06,
+      "loss": 0.6138,
+      "step": 4190
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 6.421185493469238,
+      "learning_rate": 5.201329700547077e-06,
+      "loss": 0.6052,
+      "step": 4200
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 6.061400890350342,
+      "learning_rate": 5.178969890057953e-06,
+      "loss": 0.5955,
+      "step": 4210
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 6.597458362579346,
+      "learning_rate": 5.156606495265402e-06,
+      "loss": 0.602,
+      "step": 4220
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 6.535460948944092,
+      "learning_rate": 5.1342399640503074e-06,
+      "loss": 0.5859,
+      "step": 4230
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 7.226145267486572,
+      "learning_rate": 5.111870744356366e-06,
+      "loss": 0.623,
+      "step": 4240
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 5.355672836303711,
+      "learning_rate": 5.0894992841811216e-06,
+      "loss": 0.6009,
+      "step": 4250
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 9.383895874023438,
+      "learning_rate": 5.067126031566988e-06,
+      "loss": 0.6037,
+      "step": 4260
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 5.733826160430908,
+      "learning_rate": 5.044751434592274e-06,
+      "loss": 0.5935,
+      "step": 4270
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 6.465234279632568,
+      "learning_rate": 5.022375941362218e-06,
+      "loss": 0.6264,
+      "step": 4280
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 5.9884514808654785,
+      "learning_rate": 5e-06,
+      "loss": 0.605,
+      "step": 4290
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 5.8463029861450195,
+      "learning_rate": 4.977624058637783e-06,
+      "loss": 0.6032,
+      "step": 4300
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 6.030627250671387,
+      "learning_rate": 4.955248565407727e-06,
+      "loss": 0.5974,
+      "step": 4310
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 7.388688564300537,
+      "learning_rate": 4.932873968433014e-06,
+      "loss": 0.6036,
+      "step": 4320
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 11.483485221862793,
+      "learning_rate": 4.910500715818879e-06,
+      "loss": 0.5931,
+      "step": 4330
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 12.173521995544434,
+      "learning_rate": 4.8881292556436355e-06,
+      "loss": 0.6069,
+      "step": 4340
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 5.8268866539001465,
+      "learning_rate": 4.865760035949695e-06,
+      "loss": 0.6134,
+      "step": 4350
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 7.893507957458496,
+      "learning_rate": 4.8433935047346e-06,
+      "loss": 0.6227,
+      "step": 4360
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 5.854642391204834,
+      "learning_rate": 4.821030109942048e-06,
+      "loss": 0.5971,
+      "step": 4370
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 5.798580169677734,
+      "learning_rate": 4.798670299452926e-06,
+      "loss": 0.6025,
+      "step": 4380
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 5.5879011154174805,
+      "learning_rate": 4.77631452107633e-06,
+      "loss": 0.6108,
+      "step": 4390
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 5.21212911605835,
+      "learning_rate": 4.75396322254061e-06,
+      "loss": 0.6054,
+      "step": 4400
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 6.118653774261475,
+      "learning_rate": 4.731616851484392e-06,
+      "loss": 0.6178,
+      "step": 4410
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 6.629573345184326,
+      "learning_rate": 4.7092758554476215e-06,
+      "loss": 0.5872,
+      "step": 4420
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 6.795328617095947,
+      "learning_rate": 4.686940681862591e-06,
+      "loss": 0.6138,
+      "step": 4430
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 7.000875473022461,
+      "learning_rate": 4.664611778044988e-06,
+      "loss": 0.5995,
+      "step": 4440
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 6.718314170837402,
+      "learning_rate": 4.642289591184934e-06,
+      "loss": 0.6128,
+      "step": 4450
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 7.343639373779297,
+      "learning_rate": 4.619974568338021e-06,
+      "loss": 0.5903,
+      "step": 4460
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 6.8176069259643555,
+      "learning_rate": 4.597667156416371e-06,
+      "loss": 0.6094,
+      "step": 4470
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 6.549642562866211,
+      "learning_rate": 4.575367802179675e-06,
+      "loss": 0.6127,
+      "step": 4480
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 6.30284309387207,
+      "learning_rate": 4.553076952226255e-06,
+      "loss": 0.6077,
+      "step": 4490
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 6.47008752822876,
+      "learning_rate": 4.530795052984104e-06,
+      "loss": 0.6188,
+      "step": 4500
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 6.808812618255615,
+      "learning_rate": 4.508522550701965e-06,
+      "loss": 0.5933,
+      "step": 4510
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 8.047310829162598,
+      "learning_rate": 4.48625989144038e-06,
+      "loss": 0.615,
+      "step": 4520
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 5.543693542480469,
+      "learning_rate": 4.464007521062761e-06,
+      "loss": 0.6225,
+      "step": 4530
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 5.247865200042725,
+      "learning_rate": 4.441765885226462e-06,
+      "loss": 0.6078,
+      "step": 4540
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 14.305082321166992,
+      "learning_rate": 4.4195354293738484e-06,
+      "loss": 0.6163,
+      "step": 4550
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 8.802489280700684,
+      "learning_rate": 4.397316598723385e-06,
+      "loss": 0.5915,
+      "step": 4560
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 5.3068952560424805,
+      "learning_rate": 4.3751098382607084e-06,
+      "loss": 0.6135,
+      "step": 4570
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 6.380529403686523,
+      "learning_rate": 4.352915592729723e-06,
+      "loss": 0.6061,
+      "step": 4580
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 6.269029140472412,
+      "learning_rate": 4.330734306623694e-06,
+      "loss": 0.5981,
+      "step": 4590
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 6.364799499511719,
+      "learning_rate": 4.308566424176336e-06,
+      "loss": 0.5953,
+      "step": 4600
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 5.895086288452148,
+      "learning_rate": 4.286412389352929e-06,
+      "loss": 0.613,
+      "step": 4610
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 7.223175048828125,
+      "learning_rate": 4.264272645841419e-06,
+      "loss": 0.6082,
+      "step": 4620
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 5.315123558044434,
+      "learning_rate": 4.242147637043539e-06,
+      "loss": 0.5997,
+      "step": 4630
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 6.662210941314697,
+      "learning_rate": 4.220037806065911e-06,
+      "loss": 0.6159,
+      "step": 4640
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 8.199719429016113,
+      "learning_rate": 4.1979435957111984e-06,
+      "loss": 0.6022,
+      "step": 4650
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 7.3503499031066895,
+      "learning_rate": 4.175865448469219e-06,
+      "loss": 0.6111,
+      "step": 4660
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 5.498391151428223,
+      "learning_rate": 4.153803806508083e-06,
+      "loss": 0.5987,
+      "step": 4670
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 6.277636528015137,
+      "learning_rate": 4.131759111665349e-06,
+      "loss": 0.5946,
+      "step": 4680
+    },
+    {
+      "epoch": 0.62,
+      "eval_loss": 0.6720156073570251,
+      "eval_runtime": 169.1493,
+      "eval_samples_per_second": 65.031,
+      "eval_steps_per_second": 8.129,
+      "step": 4680
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 7.248624801635742,
+      "learning_rate": 4.109731805439168e-06,
+      "loss": 0.6196,
+      "step": 4690
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 7.02220344543457,
+      "learning_rate": 4.087722328979437e-06,
+      "loss": 0.6122,
+      "step": 4700
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 6.665443420410156,
+      "learning_rate": 4.065731123078977e-06,
+      "loss": 0.5931,
+      "step": 4710
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 7.180152416229248,
+      "learning_rate": 4.043758628164688e-06,
+      "loss": 0.6053,
+      "step": 4720
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 5.866796970367432,
+      "learning_rate": 4.021805284288749e-06,
+      "loss": 0.6113,
+      "step": 4730
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 6.837026596069336,
+      "learning_rate": 3.999871531119779e-06,
+      "loss": 0.5973,
+      "step": 4740
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 7.780344486236572,
+      "learning_rate": 3.977957807934055e-06,
+      "loss": 0.6065,
+      "step": 4750
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 7.0064568519592285,
+      "learning_rate": 3.956064553606708e-06,
+      "loss": 0.5954,
+      "step": 4760
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 5.387651443481445,
+      "learning_rate": 3.934192206602921e-06,
+      "loss": 0.6239,
+      "step": 4770
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 5.932905673980713,
+      "learning_rate": 3.912341204969164e-06,
+      "loss": 0.6012,
+      "step": 4780
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 6.533616065979004,
+      "learning_rate": 3.890511986324413e-06,
+      "loss": 0.6034,
+      "step": 4790
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 6.714369297027588,
+      "learning_rate": 3.86870498785139e-06,
+      "loss": 0.6067,
+      "step": 4800
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 6.622659206390381,
+      "learning_rate": 3.8469206462878e-06,
+      "loss": 0.6018,
+      "step": 4810
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 6.005035400390625,
+      "learning_rate": 3.825159397917589e-06,
+      "loss": 0.6064,
+      "step": 4820
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 5.99045991897583,
+      "learning_rate": 3.803421678562213e-06,
+      "loss": 0.5974,
+      "step": 4830
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 6.072996616363525,
+      "learning_rate": 3.781707923571891e-06,
+      "loss": 0.5925,
+      "step": 4840
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 5.99989652633667,
+      "learning_rate": 3.7600185678169083e-06,
+      "loss": 0.6085,
+      "step": 4850
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 7.072287082672119,
+      "learning_rate": 3.7383540456788915e-06,
+      "loss": 0.5999,
+      "step": 4860
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 28.138442993164062,
+      "learning_rate": 3.7167147910421165e-06,
+      "loss": 0.6033,
+      "step": 4870
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.020947456359863,
+      "learning_rate": 3.695101237284815e-06,
+      "loss": 0.6274,
+      "step": 4880
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.451099872589111,
+      "learning_rate": 3.6735138172704967e-06,
+      "loss": 0.6164,
+      "step": 4890
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.501095294952393,
+      "learning_rate": 3.6519529633392825e-06,
+      "loss": 0.6001,
+      "step": 4900
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.591226100921631,
+      "learning_rate": 3.6304191072992376e-06,
+      "loss": 0.6006,
+      "step": 4910
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.160434246063232,
+      "learning_rate": 3.6089126804177373e-06,
+      "loss": 0.6086,
+      "step": 4920
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 5.883294105529785,
+      "learning_rate": 3.5874341134128156e-06,
+      "loss": 0.5847,
+      "step": 4930
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 6.181455135345459,
+      "learning_rate": 3.5659838364445505e-06,
+      "loss": 0.6045,
+      "step": 4940
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 7.934025764465332,
+      "learning_rate": 3.544562279106436e-06,
+      "loss": 0.5986,
+      "step": 4950
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 5.755185127258301,
+      "learning_rate": 3.523169870416795e-06,
+      "loss": 0.5843,
+      "step": 4960
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.632129669189453,
+      "learning_rate": 3.501807038810174e-06,
+      "loss": 0.6017,
+      "step": 4970
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.323507785797119,
+      "learning_rate": 3.480474212128766e-06,
+      "loss": 0.6,
+      "step": 4980
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.898191928863525,
+      "learning_rate": 3.459171817613847e-06,
+      "loss": 0.6072,
+      "step": 4990
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.425208568572998,
+      "learning_rate": 3.4379002818972122e-06,
+      "loss": 0.6057,
+      "step": 5000
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.319818019866943,
+      "learning_rate": 3.416660030992639e-06,
+      "loss": 0.5943,
+      "step": 5010
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 6.483379364013672,
+      "learning_rate": 3.3954514902873427e-06,
+      "loss": 0.606,
+      "step": 5020
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 5.618423938751221,
+      "learning_rate": 3.3742750845334748e-06,
+      "loss": 0.5912,
+      "step": 5030
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 6.536773681640625,
+      "learning_rate": 3.3531312378396026e-06,
+      "loss": 0.5996,
+      "step": 5040
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 5.867932319641113,
+      "learning_rate": 3.3320203736622185e-06,
+      "loss": 0.6041,
+      "step": 5050
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 5.637106895446777,
+      "learning_rate": 3.310942914797265e-06,
+      "loss": 0.6047,
+      "step": 5060
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 5.646607875823975,
+      "learning_rate": 3.289899283371657e-06,
+      "loss": 0.5951,
+      "step": 5070
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 5.731027603149414,
+      "learning_rate": 3.2688899008348386e-06,
+      "loss": 0.583,
+      "step": 5080
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 6.196643352508545,
+      "learning_rate": 3.2479151879503324e-06,
+      "loss": 0.6097,
+      "step": 5090
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 5.765705108642578,
+      "learning_rate": 3.226975564787322e-06,
+      "loss": 0.6014,
+      "step": 5100
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 6.380090713500977,
+      "learning_rate": 3.206071450712235e-06,
+      "loss": 0.5932,
+      "step": 5110
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 5.803713798522949,
+      "learning_rate": 3.1852032643803377e-06,
+      "loss": 0.5956,
+      "step": 5120
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 5.434231758117676,
+      "learning_rate": 3.1643714237273628e-06,
+      "loss": 0.6111,
+      "step": 5130
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 6.616121292114258,
+      "learning_rate": 3.143576345961132e-06,
+      "loss": 0.5966,
+      "step": 5140
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 5.209160804748535,
+      "learning_rate": 3.1228184475532015e-06,
+      "loss": 0.5979,
+      "step": 5150
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 6.537261486053467,
+      "learning_rate": 3.1020981442305187e-06,
+      "loss": 0.6166,
+      "step": 5160
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 7.26732063293457,
+      "learning_rate": 3.0814158509671015e-06,
+      "loss": 0.6068,
+      "step": 5170
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.632195472717285,
+      "learning_rate": 3.0607719819757264e-06,
+      "loss": 0.598,
+      "step": 5180
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.387509346008301,
+      "learning_rate": 3.040166950699626e-06,
+      "loss": 0.5923,
+      "step": 5190
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.4542975425720215,
+      "learning_rate": 3.019601169804216e-06,
+      "loss": 0.5946,
+      "step": 5200
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 5.96135139465332,
+      "learning_rate": 2.9990750511688323e-06,
+      "loss": 0.604,
+      "step": 5210
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 5.886032581329346,
+      "learning_rate": 2.978589005878476e-06,
+      "loss": 0.6048,
+      "step": 5220
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.059714317321777,
+      "learning_rate": 2.9581434442155798e-06,
+      "loss": 0.6122,
+      "step": 5230
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.174469947814941,
+      "learning_rate": 2.937738775651798e-06,
+      "loss": 0.5996,
+      "step": 5240
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 6.120980262756348,
+      "learning_rate": 2.917375408839803e-06,
+      "loss": 0.6096,
+      "step": 5250
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 6.153501510620117,
+      "learning_rate": 2.8970537516050935e-06,
+      "loss": 0.6014,
+      "step": 5260
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 6.2347211837768555,
+      "learning_rate": 2.876774210937843e-06,
+      "loss": 0.5917,
+      "step": 5270
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 7.352357864379883,
+      "learning_rate": 2.8565371929847286e-06,
+      "loss": 0.6026,
+      "step": 5280
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 6.723049163818359,
+      "learning_rate": 2.836343103040819e-06,
+      "loss": 0.6073,
+      "step": 5290
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 6.070889472961426,
+      "learning_rate": 2.816192345541437e-06,
+      "loss": 0.6028,
+      "step": 5300
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 5.253773212432861,
+      "learning_rate": 2.7960853240540703e-06,
+      "loss": 0.6038,
+      "step": 5310
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 7.381737232208252,
+      "learning_rate": 2.776022441270295e-06,
+      "loss": 0.6045,
+      "step": 5320
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 6.627200126647949,
+      "learning_rate": 2.7560040989976894e-06,
+      "loss": 0.5919,
+      "step": 5330
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 6.173203945159912,
+      "learning_rate": 2.736030698151815e-06,
+      "loss": 0.582,
+      "step": 5340
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 7.563076972961426,
+      "learning_rate": 2.7161026387481636e-06,
+      "loss": 0.5951,
+      "step": 5350
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 7.46902322769165,
+      "learning_rate": 2.6962203198941587e-06,
+      "loss": 0.6025,
+      "step": 5360
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 5.640381336212158,
+      "learning_rate": 2.6763841397811576e-06,
+      "loss": 0.6019,
+      "step": 5370
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 7.339272975921631,
+      "learning_rate": 2.656594495676482e-06,
+      "loss": 0.5874,
+      "step": 5380
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 8.082894325256348,
+      "learning_rate": 2.636851783915454e-06,
+      "loss": 0.6085,
+      "step": 5390
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 6.488770008087158,
+      "learning_rate": 2.6171563998934605e-06,
+      "loss": 0.5936,
+      "step": 5400
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 7.645140647888184,
+      "learning_rate": 2.5975087380580445e-06,
+      "loss": 0.6101,
+      "step": 5410
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 6.867657661437988,
+      "learning_rate": 2.577909191900988e-06,
+      "loss": 0.619,
+      "step": 5420
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 6.3017258644104,
+      "learning_rate": 2.5583581539504464e-06,
+      "loss": 0.5912,
+      "step": 5430
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 7.013880252838135,
+      "learning_rate": 2.5388560157630765e-06,
+      "loss": 0.604,
+      "step": 5440
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 6.177152633666992,
+      "learning_rate": 2.519403167916207e-06,
+      "loss": 0.5953,
+      "step": 5450
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 9.21536922454834,
+      "learning_rate": 2.5000000000000015e-06,
+      "loss": 0.6057,
+      "step": 5460
+    },
+    {
+      "epoch": 0.72,
+      "eval_loss": 0.6694371104240417,
+      "eval_runtime": 169.1763,
+      "eval_samples_per_second": 65.021,
+      "eval_steps_per_second": 8.128,
+      "step": 5460
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 9.104439735412598,
+      "learning_rate": 2.480646900609664e-06,
+      "loss": 0.597,
+      "step": 5470
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 6.437440395355225,
+      "learning_rate": 2.4613442573376625e-06,
+      "loss": 0.5985,
+      "step": 5480
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 7.327053546905518,
+      "learning_rate": 2.4420924567659508e-06,
+      "loss": 0.6095,
+      "step": 5490
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 6.946102142333984,
+      "learning_rate": 2.422891884458241e-06,
+      "loss": 0.5944,
+      "step": 5500
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 7.699533462524414,
+      "learning_rate": 2.4037429249522702e-06,
+      "loss": 0.6156,
+      "step": 5510
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 6.800520896911621,
+      "learning_rate": 2.384645961752113e-06,
+      "loss": 0.6075,
+      "step": 5520
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 6.476116180419922,
+      "learning_rate": 2.3656013773204843e-06,
+      "loss": 0.5989,
+      "step": 5530
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 7.190838813781738,
+      "learning_rate": 2.346609553071093e-06,
+      "loss": 0.5993,
+      "step": 5540
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 6.840661525726318,
+      "learning_rate": 2.3276708693609947e-06,
+      "loss": 0.6068,
+      "step": 5550
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 6.123993396759033,
+      "learning_rate": 2.308785705482982e-06,
+      "loss": 0.6089,
+      "step": 5560
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 6.094503402709961,
+      "learning_rate": 2.289954439657981e-06,
+      "loss": 0.6055,
+      "step": 5570
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 7.03631067276001,
+      "learning_rate": 2.2711774490274767e-06,
+      "loss": 0.6056,
+      "step": 5580
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 7.555922508239746,
+      "learning_rate": 2.2524551096459703e-06,
+      "loss": 0.6018,
+      "step": 5590
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 10.68952465057373,
+      "learning_rate": 2.2337877964734324e-06,
+      "loss": 0.5999,
+      "step": 5600
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 10.688403129577637,
+      "learning_rate": 2.2151758833678044e-06,
+      "loss": 0.6061,
+      "step": 5610
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 8.527271270751953,
+      "learning_rate": 2.1966197430775056e-06,
+      "loss": 0.5914,
+      "step": 5620
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 8.061843872070312,
+      "learning_rate": 2.178119747233976e-06,
+      "loss": 0.5957,
+      "step": 5630
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 7.056491851806641,
+      "learning_rate": 2.159676266344222e-06,
+      "loss": 0.5993,
+      "step": 5640
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 6.01476526260376,
+      "learning_rate": 2.141289669783401e-06,
+      "loss": 0.5988,
+      "step": 5650
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 7.841024875640869,
+      "learning_rate": 2.122960325787432e-06,
+      "loss": 0.6008,
+      "step": 5660
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 6.125739574432373,
+      "learning_rate": 2.104688601445606e-06,
+      "loss": 0.5924,
+      "step": 5670
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 5.347300052642822,
+      "learning_rate": 2.086474862693244e-06,
+      "loss": 0.6024,
+      "step": 5680
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 8.490260124206543,
+      "learning_rate": 2.068319474304365e-06,
+      "loss": 0.5842,
+      "step": 5690
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 5.8698344230651855,
+      "learning_rate": 2.050222799884387e-06,
+      "loss": 0.6001,
+      "step": 5700
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 5.577418327331543,
+      "learning_rate": 2.0321852018628278e-06,
+      "loss": 0.5981,
+      "step": 5710
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 6.239425182342529,
+      "learning_rate": 2.0142070414860704e-06,
+      "loss": 0.5936,
+      "step": 5720
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 5.704605579376221,
+      "learning_rate": 1.996288678810105e-06,
+      "loss": 0.5973,
+      "step": 5730
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 6.664093017578125,
+      "learning_rate": 1.9784304726933384e-06,
+      "loss": 0.5979,
+      "step": 5740
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 5.986905574798584,
+      "learning_rate": 1.9606327807893905e-06,
+      "loss": 0.6006,
+      "step": 5750
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 6.272210121154785,
+      "learning_rate": 1.942895959539939e-06,
+      "loss": 0.5991,
+      "step": 5760
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 6.319902420043945,
+      "learning_rate": 1.9252203641675854e-06,
+      "loss": 0.5921,
+      "step": 5770
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 6.060082912445068,
+      "learning_rate": 1.9076063486687256e-06,
+      "loss": 0.5872,
+      "step": 5780
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 6.724019527435303,
+      "learning_rate": 1.8900542658064807e-06,
+      "loss": 0.5845,
+      "step": 5790
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 5.889808654785156,
+      "learning_rate": 1.8725644671036125e-06,
+      "loss": 0.6051,
+      "step": 5800
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 8.664203643798828,
+      "learning_rate": 1.8551373028355013e-06,
+      "loss": 0.609,
+      "step": 5810
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 6.822065830230713,
+      "learning_rate": 1.8377731220231144e-06,
+      "loss": 0.6033,
+      "step": 5820
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 6.636890888214111,
+      "learning_rate": 1.8204722724260266e-06,
+      "loss": 0.6068,
+      "step": 5830
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 7.8858256340026855,
+      "learning_rate": 1.803235100535452e-06,
+      "loss": 0.5931,
+      "step": 5840
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 7.670845985412598,
+      "learning_rate": 1.7860619515673034e-06,
+      "loss": 0.6111,
+      "step": 5850
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 6.426079750061035,
+      "learning_rate": 1.7689531694552863e-06,
+      "loss": 0.5929,
+      "step": 5860
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 5.926182270050049,
+      "learning_rate": 1.7519090968439966e-06,
+      "loss": 0.587,
+      "step": 5870
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 5.617947101593018,
+      "learning_rate": 1.7349300750820758e-06,
+      "loss": 0.5958,
+      "step": 5880
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 5.876589775085449,
+      "learning_rate": 1.7180164442153529e-06,
+      "loss": 0.6037,
+      "step": 5890
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 6.038319110870361,
+      "learning_rate": 1.7011685429800596e-06,
+      "loss": 0.6023,
+      "step": 5900
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 7.0782318115234375,
+      "learning_rate": 1.6843867087960252e-06,
+      "loss": 0.6068,
+      "step": 5910
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 5.8368144035339355,
+      "learning_rate": 1.6676712777599275e-06,
+      "loss": 0.5919,
+      "step": 5920
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 6.66851806640625,
+      "learning_rate": 1.6510225846385668e-06,
+      "loss": 0.5912,
+      "step": 5930
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.366353988647461,
+      "learning_rate": 1.6344409628621482e-06,
+      "loss": 0.6087,
+      "step": 5940
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 7.111102104187012,
+      "learning_rate": 1.6179267445176206e-06,
+      "loss": 0.6059,
+      "step": 5950
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.061978816986084,
+      "learning_rate": 1.6014802603420044e-06,
+      "loss": 0.5938,
+      "step": 5960
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.1793341636657715,
+      "learning_rate": 1.5851018397157918e-06,
+      "loss": 0.6087,
+      "step": 5970
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.220329761505127,
+      "learning_rate": 1.5687918106563326e-06,
+      "loss": 0.5961,
+      "step": 5980
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.557837009429932,
+      "learning_rate": 1.5525504998112717e-06,
+      "loss": 0.5922,
+      "step": 5990
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 6.448672771453857,
+      "learning_rate": 1.5363782324520033e-06,
+      "loss": 0.5969,
+      "step": 6000
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 6.809442043304443,
+      "learning_rate": 1.520275332467166e-06,
+      "loss": 0.6038,
+      "step": 6010
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 5.7299041748046875,
+      "learning_rate": 1.504242122356143e-06,
+      "loss": 0.5929,
+      "step": 6020
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 5.356698989868164,
+      "learning_rate": 1.4882789232226124e-06,
+      "loss": 0.5916,
+      "step": 6030
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 6.809737682342529,
+      "learning_rate": 1.4723860547681163e-06,
+      "loss": 0.6108,
+      "step": 6040
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 6.65049409866333,
+      "learning_rate": 1.4565638352856504e-06,
+      "loss": 0.5958,
+      "step": 6050
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 6.739075660705566,
+      "learning_rate": 1.4408125816532981e-06,
+      "loss": 0.5805,
+      "step": 6060
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 6.686561584472656,
+      "learning_rate": 1.4251326093278773e-06,
+      "loss": 0.6039,
+      "step": 6070
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 6.082182884216309,
+      "learning_rate": 1.4095242323386305e-06,
+      "loss": 0.5949,
+      "step": 6080
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 5.692592620849609,
+      "learning_rate": 1.3939877632809279e-06,
+      "loss": 0.5944,
+      "step": 6090
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 5.9873480796813965,
+      "learning_rate": 1.3785235133100088e-06,
+      "loss": 0.5853,
+      "step": 6100
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 34.229915618896484,
+      "learning_rate": 1.3631317921347564e-06,
+      "loss": 0.5978,
+      "step": 6110
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 7.1477952003479,
+      "learning_rate": 1.347812908011485e-06,
+      "loss": 0.5803,
+      "step": 6120
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 6.827530384063721,
+      "learning_rate": 1.3325671677377727e-06,
+      "loss": 0.5955,
+      "step": 6130
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 6.848330974578857,
+      "learning_rate": 1.3173948766463146e-06,
+      "loss": 0.5997,
+      "step": 6140
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 6.4514288902282715,
+      "learning_rate": 1.3022963385988153e-06,
+      "loss": 0.59,
+      "step": 6150
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.169240951538086,
+      "learning_rate": 1.2872718559798852e-06,
+      "loss": 0.602,
+      "step": 6160
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.365018367767334,
+      "learning_rate": 1.2723217296910078e-06,
+      "loss": 0.6034,
+      "step": 6170
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.3110785484313965,
+      "learning_rate": 1.257446259144494e-06,
+      "loss": 0.5973,
+      "step": 6180
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.271914958953857,
+      "learning_rate": 1.2426457422575e-06,
+      "loss": 0.5988,
+      "step": 6190
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.755878448486328,
+      "learning_rate": 1.2279204754460494e-06,
+      "loss": 0.5896,
+      "step": 6200
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 5.7621684074401855,
+      "learning_rate": 1.2132707536191008e-06,
+      "loss": 0.5961,
+      "step": 6210
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 5.900276184082031,
+      "learning_rate": 1.1986968701726492e-06,
+      "loss": 0.5944,
+      "step": 6220
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 6.725846290588379,
+      "learning_rate": 1.184199116983834e-06,
+      "loss": 0.5897,
+      "step": 6230
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 8.138350486755371,
+      "learning_rate": 1.1697777844051105e-06,
+      "loss": 0.6036,
+      "step": 6240
+    },
+    {
+      "epoch": 0.83,
+      "eval_loss": 0.6633522510528564,
+      "eval_runtime": 169.2613,
+      "eval_samples_per_second": 64.988,
+      "eval_steps_per_second": 8.124,
+      "step": 6240
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.216264247894287,
+      "learning_rate": 1.1554331612584218e-06,
+      "loss": 0.5926,
+      "step": 6250
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.344448566436768,
+      "learning_rate": 1.141165534829425e-06,
+      "loss": 0.5855,
+      "step": 6260
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 7.457535743713379,
+      "learning_rate": 1.1269751908617277e-06,
+      "loss": 0.5912,
+      "step": 6270
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.311199188232422,
+      "learning_rate": 1.1128624135511712e-06,
+      "loss": 0.5983,
+      "step": 6280
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.246806621551514,
+      "learning_rate": 1.0988274855401377e-06,
+      "loss": 0.5875,
+      "step": 6290
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 6.010527610778809,
+      "learning_rate": 1.0848706879118893e-06,
+      "loss": 0.5905,
+      "step": 6300
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 5.997354984283447,
+      "learning_rate": 1.070992300184941e-06,
+      "loss": 0.608,
+      "step": 6310
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 6.129363536834717,
+      "learning_rate": 1.057192600307456e-06,
+      "loss": 0.6013,
+      "step": 6320
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 6.758306503295898,
+      "learning_rate": 1.0434718646516917e-06,
+      "loss": 0.5961,
+      "step": 6330
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 6.165789604187012,
+      "learning_rate": 1.0298303680084448e-06,
+      "loss": 0.5952,
+      "step": 6340
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 6.356081008911133,
+      "learning_rate": 1.0162683835815706e-06,
+      "loss": 0.595,
+      "step": 6350
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 5.993287563323975,
+      "learning_rate": 1.0027861829824953e-06,
+      "loss": 0.6059,
+      "step": 6360
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 6.2351226806640625,
+      "learning_rate": 9.893840362247809e-07,
+      "loss": 0.5966,
+      "step": 6370
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 6.520230293273926,
+      "learning_rate": 9.760622117187234e-07,
+      "loss": 0.6169,
+      "step": 6380
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 6.322525978088379,
+      "learning_rate": 9.628209762659658e-07,
+      "loss": 0.6066,
+      "step": 6390
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 8.321556091308594,
+      "learning_rate": 9.496605950541676e-07,
+      "loss": 0.5976,
+      "step": 6400
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 7.012268543243408,
+      "learning_rate": 9.365813316516787e-07,
+      "loss": 0.6028,
+      "step": 6410
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 7.8040771484375,
+      "learning_rate": 9.235834480022788e-07,
+      "loss": 0.611,
+      "step": 6420
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 5.389317035675049,
+      "learning_rate": 9.10667204419915e-07,
+      "loss": 0.5917,
+      "step": 6430
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 7.818176746368408,
+      "learning_rate": 8.978328595834984e-07,
+      "loss": 0.6056,
+      "step": 6440
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 6.365832328796387,
+      "learning_rate": 8.850806705317183e-07,
+      "loss": 0.5949,
+      "step": 6450
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 6.176076412200928,
+      "learning_rate": 8.724108926579e-07,
+      "loss": 0.6034,
+      "step": 6460
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 6.508719444274902,
+      "learning_rate": 8.598237797048825e-07,
+      "loss": 0.5851,
+      "step": 6470
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 6.904651165008545,
+      "learning_rate": 8.473195837599419e-07,
+      "loss": 0.5949,
+      "step": 6480
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 6.214872360229492,
+      "learning_rate": 8.348985552497424e-07,
+      "loss": 0.5826,
+      "step": 6490
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 9.990653991699219,
+      "learning_rate": 8.225609429353187e-07,
+      "loss": 0.6072,
+      "step": 6500
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 7.119464874267578,
+      "learning_rate": 8.103069939070945e-07,
+      "loss": 0.6097,
+      "step": 6510
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 7.176183700561523,
+      "learning_rate": 7.981369535799354e-07,
+      "loss": 0.5989,
+      "step": 6520
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 6.2535014152526855,
+      "learning_rate": 7.860510656882342e-07,
+      "loss": 0.602,
+      "step": 6530
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 6.955148696899414,
+      "learning_rate": 7.740495722810271e-07,
+      "loss": 0.6027,
+      "step": 6540
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 6.898377895355225,
+      "learning_rate": 7.621327137171447e-07,
+      "loss": 0.5999,
+      "step": 6550
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 6.001263618469238,
+      "learning_rate": 7.50300728660407e-07,
+      "loss": 0.5719,
+      "step": 6560
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 6.911612510681152,
+      "learning_rate": 7.385538540748327e-07,
+      "loss": 0.5802,
+      "step": 6570
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 5.093260765075684,
+      "learning_rate": 7.26892325219899e-07,
+      "loss": 0.5966,
+      "step": 6580
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 6.112546920776367,
+      "learning_rate": 7.153163756458287e-07,
+      "loss": 0.5949,
+      "step": 6590
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 6.923007011413574,
+      "learning_rate": 7.03826237188916e-07,
+      "loss": 0.5967,
+      "step": 6600
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 8.302557945251465,
+      "learning_rate": 6.924221399668785e-07,
+      "loss": 0.5879,
+      "step": 6610
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.832093238830566,
+      "learning_rate": 6.811043123742494e-07,
+      "loss": 0.596,
+      "step": 6620
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.365548133850098,
+      "learning_rate": 6.698729810778065e-07,
+      "loss": 0.5975,
+      "step": 6630
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.902804851531982,
+      "learning_rate": 6.587283710120324e-07,
+      "loss": 0.5882,
+      "step": 6640
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 7.0722808837890625,
+      "learning_rate": 6.47670705374604e-07,
+      "loss": 0.5945,
+      "step": 6650
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 9.358778953552246,
+      "learning_rate": 6.367002056219285e-07,
+      "loss": 0.5891,
+      "step": 6660
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 7.326229572296143,
+      "learning_rate": 6.258170914647077e-07,
+      "loss": 0.6016,
+      "step": 6670
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 5.735571384429932,
+      "learning_rate": 6.150215808635334e-07,
+      "loss": 0.6066,
+      "step": 6680
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 5.679074764251709,
+      "learning_rate": 6.043138900245277e-07,
+      "loss": 0.5944,
+      "step": 6690
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 7.554742813110352,
+      "learning_rate": 5.936942333950063e-07,
+      "loss": 0.5904,
+      "step": 6700
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 7.517423629760742,
+      "learning_rate": 5.831628236591929e-07,
+      "loss": 0.5885,
+      "step": 6710
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 7.0763959884643555,
+      "learning_rate": 5.727198717339511e-07,
+      "loss": 0.6007,
+      "step": 6720
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 6.555006504058838,
+      "learning_rate": 5.623655867645628e-07,
+      "loss": 0.6059,
+      "step": 6730
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 6.526163101196289,
+      "learning_rate": 5.521001761205441e-07,
+      "loss": 0.5923,
+      "step": 6740
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 7.336793899536133,
+      "learning_rate": 5.41923845391486e-07,
+      "loss": 0.5961,
+      "step": 6750
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 7.833865642547607,
+      "learning_rate": 5.318367983829393e-07,
+      "loss": 0.6139,
+      "step": 6760
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 6.014789581298828,
+      "learning_rate": 5.218392371123326e-07,
+      "loss": 0.5936,
+      "step": 6770
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 5.68282413482666,
+      "learning_rate": 5.119313618049309e-07,
+      "loss": 0.6043,
+      "step": 6780
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 7.465061664581299,
+      "learning_rate": 5.021133708898146e-07,
+      "loss": 0.6057,
+      "step": 6790
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 6.3078083992004395,
+      "learning_rate": 4.9238546099592e-07,
+      "loss": 0.5766,
+      "step": 6800
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 6.257416725158691,
+      "learning_rate": 4.827478269480895e-07,
+      "loss": 0.6008,
+      "step": 6810
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 6.170396327972412,
+      "learning_rate": 4.732006617631729e-07,
+      "loss": 0.588,
+      "step": 6820
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 6.257867813110352,
+      "learning_rate": 4.637441566461681e-07,
+      "loss": 0.5986,
+      "step": 6830
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 6.7348952293396,
+      "learning_rate": 4.54378500986381e-07,
+      "loss": 0.6115,
+      "step": 6840
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 6.10768461227417,
+      "learning_rate": 4.451038823536441e-07,
+      "loss": 0.6063,
+      "step": 6850
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 6.2565226554870605,
+      "learning_rate": 4.35920486494546e-07,
+      "loss": 0.5842,
+      "step": 6860
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 7.341768741607666,
+      "learning_rate": 4.268284973287273e-07,
+      "loss": 0.5969,
+      "step": 6870
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 5.864892959594727,
+      "learning_rate": 4.1782809694518533e-07,
+      "loss": 0.597,
+      "step": 6880
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 5.842800617218018,
+      "learning_rate": 4.089194655986306e-07,
+      "loss": 0.5968,
+      "step": 6890
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 6.370209217071533,
+      "learning_rate": 4.001027817058789e-07,
+      "loss": 0.6031,
+      "step": 6900
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 6.6684489250183105,
+      "learning_rate": 3.9137822184227845e-07,
+      "loss": 0.5959,
+      "step": 6910
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 6.804083824157715,
+      "learning_rate": 3.8274596073816784e-07,
+      "loss": 0.5985,
+      "step": 6920
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 6.737246990203857,
+      "learning_rate": 3.742061712753825e-07,
+      "loss": 0.5926,
+      "step": 6930
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 7.578712463378906,
+      "learning_rate": 3.657590244837911e-07,
+      "loss": 0.5812,
+      "step": 6940
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 7.795099258422852,
+      "learning_rate": 3.5740468953786854e-07,
+      "loss": 0.5836,
+      "step": 6950
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 8.640656471252441,
+      "learning_rate": 3.49143333753309e-07,
+      "loss": 0.598,
+      "step": 6960
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 7.3872971534729,
+      "learning_rate": 3.4097512258367385e-07,
+      "loss": 0.5771,
+      "step": 6970
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 6.033993244171143,
+      "learning_rate": 3.3290021961708163e-07,
+      "loss": 0.5866,
+      "step": 6980
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 7.115429401397705,
+      "learning_rate": 3.2491878657292643e-07,
+      "loss": 0.5888,
+      "step": 6990
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 7.127076148986816,
+      "learning_rate": 3.1703098329864237e-07,
+      "loss": 0.5962,
+      "step": 7000
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 5.615909576416016,
+      "learning_rate": 3.0923696776650414e-07,
+      "loss": 0.5977,
+      "step": 7010
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 7.339127540588379,
+      "learning_rate": 3.015368960704584e-07,
+      "loss": 0.5952,
+      "step": 7020
+    },
+    {
+      "epoch": 0.93,
+      "eval_loss": 0.6624879240989685,
+      "eval_runtime": 169.2633,
+      "eval_samples_per_second": 64.988,
+      "eval_steps_per_second": 8.123,
+      "step": 7020
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 6.424869060516357,
+      "learning_rate": 2.9393092242300026e-07,
+      "loss": 0.6031,
+      "step": 7030
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 7.278805255889893,
+      "learning_rate": 2.864191991520848e-07,
+      "loss": 0.582,
+      "step": 7040
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 7.123143672943115,
+      "learning_rate": 2.790018766980773e-07,
+      "loss": 0.6012,
+      "step": 7050
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 7.571608543395996,
+      "learning_rate": 2.71679103610738e-07,
+      "loss": 0.5986,
+      "step": 7060
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 9.16704273223877,
+      "learning_rate": 2.64451026546248e-07,
+      "loss": 0.5973,
+      "step": 7070
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 21.925724029541016,
+      "learning_rate": 2.573177902642726e-07,
+      "loss": 0.6081,
+      "step": 7080
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 8.148933410644531,
+      "learning_rate": 2.502795376250622e-07,
+      "loss": 0.5902,
+      "step": 7090
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 7.818910598754883,
+      "learning_rate": 2.4333640958659144e-07,
+      "loss": 0.6025,
+      "step": 7100
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 8.272276878356934,
+      "learning_rate": 2.3648854520173237e-07,
+      "loss": 0.5759,
+      "step": 7110
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 7.721363544464111,
+      "learning_rate": 2.2973608161547755e-07,
+      "loss": 0.5904,
+      "step": 7120
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 6.874349117279053,
+      "learning_rate": 2.2307915406218517e-07,
+      "loss": 0.5935,
+      "step": 7130
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 7.524652481079102,
+      "learning_rate": 2.1651789586287442e-07,
+      "loss": 0.5943,
+      "step": 7140
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 7.76678991317749,
+      "learning_rate": 2.1005243842255552e-07,
+      "loss": 0.5978,
+      "step": 7150
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 7.988404750823975,
+      "learning_rate": 2.0368291122759898e-07,
+      "loss": 0.6036,
+      "step": 7160
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 10.661617279052734,
+      "learning_rate": 1.9740944184313882e-07,
+      "loss": 0.5818,
+      "step": 7170
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 8.164312362670898,
+      "learning_rate": 1.9123215591052014e-07,
+      "loss": 0.5947,
+      "step": 7180
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 6.9098639488220215,
+      "learning_rate": 1.8515117714478447e-07,
+      "loss": 0.5905,
+      "step": 7190
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 6.581024646759033,
+      "learning_rate": 1.7916662733218848e-07,
+      "loss": 0.5863,
+      "step": 7200
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 7.17993688583374,
+      "learning_rate": 1.7327862632776638e-07,
+      "loss": 0.5896,
+      "step": 7210
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 8.833417892456055,
+      "learning_rate": 1.6748729205293024e-07,
+      "loss": 0.5968,
+      "step": 7220
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 9.489375114440918,
+      "learning_rate": 1.6179274049310966e-07,
+      "loss": 0.5915,
+      "step": 7230
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 9.514772415161133,
+      "learning_rate": 1.5619508569542363e-07,
+      "loss": 0.6042,
+      "step": 7240
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 8.74714183807373,
+      "learning_rate": 1.5069443976640287e-07,
+      "loss": 0.5893,
+      "step": 7250
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 9.094064712524414,
+      "learning_rate": 1.4529091286973994e-07,
+      "loss": 0.5952,
+      "step": 7260
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 7.5646748542785645,
+      "learning_rate": 1.3998461322408563e-07,
+      "loss": 0.5858,
+      "step": 7270
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 6.778071880340576,
+      "learning_rate": 1.3477564710088097e-07,
+      "loss": 0.5909,
+      "step": 7280
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 8.356637954711914,
+      "learning_rate": 1.2966411882222695e-07,
+      "loss": 0.587,
+      "step": 7290
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 30.982200622558594,
+      "learning_rate": 1.2465013075879884e-07,
+      "loss": 0.5991,
+      "step": 7300
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 9.689143180847168,
+      "learning_rate": 1.1973378332779229e-07,
+      "loss": 0.5941,
+      "step": 7310
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 9.626836776733398,
+      "learning_rate": 1.1491517499091498e-07,
+      "loss": 0.6101,
+      "step": 7320
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 8.071940422058105,
+      "learning_rate": 1.1019440225241317e-07,
+      "loss": 0.6048,
+      "step": 7330
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 8.815977096557617,
+      "learning_rate": 1.055715596571405e-07,
+      "loss": 0.5978,
+      "step": 7340
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 8.037093162536621,
+      "learning_rate": 1.0104673978866164e-07,
+      "loss": 0.5957,
+      "step": 7350
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 10.43423080444336,
+      "learning_rate": 9.662003326740166e-08,
+      "loss": 0.5852,
+      "step": 7360
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 11.2993803024292,
+      "learning_rate": 9.22915287488274e-08,
+      "loss": 0.5991,
+      "step": 7370
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 8.106484413146973,
+      "learning_rate": 8.80613129216762e-08,
+      "loss": 0.5826,
+      "step": 7380
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 7.548207759857178,
+      "learning_rate": 8.392947050621603e-08,
+      "loss": 0.6091,
+      "step": 7390
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 9.398821830749512,
+      "learning_rate": 7.989608425254924e-08,
+      "loss": 0.6062,
+      "step": 7400
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 9.462056159973145,
+      "learning_rate": 7.59612349389599e-08,
+      "loss": 0.5883,
+      "step": 7410
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 7.41684627532959,
+      "learning_rate": 7.212500137028789e-08,
+      "loss": 0.5713,
+      "step": 7420
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 8.549308776855469,
+      "learning_rate": 6.838746037635735e-08,
+      "loss": 0.5913,
+      "step": 7430
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 9.199472427368164,
+      "learning_rate": 6.474868681043578e-08,
+      "loss": 0.5937,
+      "step": 7440
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 9.204155921936035,
+      "learning_rate": 6.120875354773459e-08,
+      "loss": 0.6123,
+      "step": 7450
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 11.123385429382324,
+      "learning_rate": 5.776773148394976e-08,
+      "loss": 0.601,
+      "step": 7460
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 10.552239418029785,
+      "learning_rate": 5.4425689533841864e-08,
+      "loss": 0.5879,
+      "step": 7470
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 9.744135856628418,
+      "learning_rate": 5.1182694629857145e-08,
+      "loss": 0.5935,
+      "step": 7480
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 8.253293991088867,
+      "learning_rate": 4.803881172078473e-08,
+      "loss": 0.5794,
+      "step": 7490
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 10.403715133666992,
+      "learning_rate": 4.499410377045765e-08,
+      "loss": 0.5859,
+      "step": 7500
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 10.5465669631958,
+      "learning_rate": 4.2048631756492206e-08,
+      "loss": 0.5915,
+      "step": 7510
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 10.897085189819336,
+      "learning_rate": 3.9202454669063915e-08,
+      "loss": 0.5948,
+      "step": 7520
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 10.889774322509766,
+      "learning_rate": 3.645562950973014e-08,
+      "loss": 0.5855,
+      "step": 7530
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 14.552692413330078,
+      "learning_rate": 3.3808211290284886e-08,
+      "loss": 0.5792,
+      "step": 7540
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 14.617317199707031,
+      "learning_rate": 3.1260253031660247e-08,
+      "loss": 0.5728,
+      "step": 7550
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 9.893068313598633,
+      "learning_rate": 2.8811805762860578e-08,
+      "loss": 0.5764,
+      "step": 7560
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 12.069340705871582,
+      "learning_rate": 2.6462918519944425e-08,
+      "loss": 0.5868,
+      "step": 7570
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 11.886162757873535,
+      "learning_rate": 2.4213638345040868e-08,
+      "loss": 0.5765,
+      "step": 7580
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 15.018707275390625,
+      "learning_rate": 2.206401028540639e-08,
+      "loss": 0.5941,
+      "step": 7590
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 13.571678161621094,
+      "learning_rate": 2.0014077392525035e-08,
+      "loss": 0.5735,
+      "step": 7600
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 10.342925071716309,
+      "learning_rate": 1.8063880721242986e-08,
+      "loss": 0.5858,
+      "step": 7610
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 12.00433349609375,
+      "learning_rate": 1.6213459328950355e-08,
+      "loss": 0.5805,
+      "step": 7620
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 11.714844703674316,
+      "learning_rate": 1.4462850274794548e-08,
+      "loss": 0.5769,
+      "step": 7630
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 14.100560188293457,
+      "learning_rate": 1.2812088618942009e-08,
+      "loss": 0.5837,
+      "step": 7640
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 11.162514686584473,
+      "learning_rate": 1.1261207421874309e-08,
+      "loss": 0.5901,
+      "step": 7650
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 13.700465202331543,
+      "learning_rate": 9.810237743724805e-09,
+      "loss": 0.6027,
+      "step": 7660
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 10.646740913391113,
+      "learning_rate": 8.459208643659122e-09,
+      "loss": 0.5797,
+      "step": 7670
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 14.71796703338623,
+      "learning_rate": 7.2081471792911914e-09,
+      "loss": 0.5724,
+      "step": 7680
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 14.192370414733887,
+      "learning_rate": 6.057078406142003e-09,
+      "loss": 0.5751,
+      "step": 7690
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 13.943014144897461,
+      "learning_rate": 5.006025377138901e-09,
+      "loss": 0.5801,
+      "step": 7700
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 16.19605827331543,
+      "learning_rate": 4.055009142152066e-09,
+      "loss": 0.5887,
+      "step": 7710
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 15.496548652648926,
+      "learning_rate": 3.204048747573185e-09,
+      "loss": 0.5885,
+      "step": 7720
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 14.580850601196289,
+      "learning_rate": 2.4531612359363077e-09,
+      "loss": 0.5806,
+      "step": 7730
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 11.053496360778809,
+      "learning_rate": 1.8023616455731253e-09,
+      "loss": 0.5916,
+      "step": 7740
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 12.858407020568848,
+      "learning_rate": 1.2516630103137638e-09,
+      "loss": 0.5805,
+      "step": 7750
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 12.221976280212402,
+      "learning_rate": 8.010763592264381e-10,
+      "loss": 0.5747,
+      "step": 7760
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 14.73697566986084,
+      "learning_rate": 4.506107163948503e-10,
+      "loss": 0.5811,
+      "step": 7770
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 11.176998138427734,
+      "learning_rate": 2.0027310073833516e-10,
+      "loss": 0.5929,
+      "step": 7780
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 17.99811363220215,
+      "learning_rate": 5.0068525870305974e-11,
+      "loss": 0.5805,
+      "step": 7790
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 12.911456108093262,
+      "learning_rate": 0.0,
+      "loss": 0.5872,
+      "step": 7800
+    },
+    {
+      "epoch": 1.03,
+      "eval_loss": 0.6635136008262634,
+      "eval_runtime": 169.3272,
+      "eval_samples_per_second": 64.963,
+      "eval_steps_per_second": 8.12,
+      "step": 7800
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 7800,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 1,
+  "total_flos": 6.410944662073875e+19,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}