diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5414 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.999705680368881,
+  "eval_steps": 500,
+  "global_step": 7644,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.003924261748258609,
+      "grad_norm": 202.0,
+      "learning_rate": 8.695652173913044e-06,
+      "loss": 2.912,
+      "step": 10
+    },
+    {
+      "epoch": 0.007848523496517218,
+      "grad_norm": 138.0,
+      "learning_rate": 1.739130434782609e-05,
+      "loss": 3.3655,
+      "step": 20
+    },
+    {
+      "epoch": 0.011772785244775827,
+      "grad_norm": 156.0,
+      "learning_rate": 2.608695652173913e-05,
+      "loss": 3.1613,
+      "step": 30
+    },
+    {
+      "epoch": 0.015697046993034435,
+      "grad_norm": 252.0,
+      "learning_rate": 3.478260869565218e-05,
+      "loss": 2.9072,
+      "step": 40
+    },
+    {
+      "epoch": 0.019621308741293045,
+      "grad_norm": 366.0,
+      "learning_rate": 4.347826086956522e-05,
+      "loss": 3.1332,
+      "step": 50
+    },
+    {
+      "epoch": 0.023545570489551654,
+      "grad_norm": 189.0,
+      "learning_rate": 5.217391304347826e-05,
+      "loss": 2.7202,
+      "step": 60
+    },
+    {
+      "epoch": 0.02746983223781026,
+      "grad_norm": 382.0,
+      "learning_rate": 6.086956521739131e-05,
+      "loss": 3.0965,
+      "step": 70
+    },
+    {
+      "epoch": 0.03139409398606887,
+      "grad_norm": 424.0,
+      "learning_rate": 6.956521739130436e-05,
+      "loss": 2.8475,
+      "step": 80
+    },
+    {
+      "epoch": 0.035318355734327476,
+      "grad_norm": 388.0,
+      "learning_rate": 7.82608695652174e-05,
+      "loss": 3.0979,
+      "step": 90
+    },
+    {
+      "epoch": 0.03924261748258609,
+      "grad_norm": 482.0,
+      "learning_rate": 8.695652173913044e-05,
+      "loss": 3.3986,
+      "step": 100
+    },
+    {
+      "epoch": 0.043166879230844696,
+      "grad_norm": 202.0,
+      "learning_rate": 9.565217391304348e-05,
+      "loss": 4.1828,
+      "step": 110
+    },
+    {
+      "epoch": 0.04709114097910331,
+      "grad_norm": 486.0,
+      "learning_rate": 0.00010434782608695653,
+      "loss": 4.313,
+      "step": 120
+    },
+    {
+      "epoch": 0.051015402727361915,
+      "grad_norm": 143.0,
+      "learning_rate": 0.00011304347826086956,
+      "loss": 4.1294,
+      "step": 130
+    },
+    {
+      "epoch": 0.05493966447562052,
+      "grad_norm": 300.0,
+      "learning_rate": 0.00012173913043478263,
+      "loss": 4.0368,
+      "step": 140
+    },
+    {
+      "epoch": 0.058863926223879134,
+      "grad_norm": 222.0,
+      "learning_rate": 0.00013043478260869567,
+      "loss": 4.2798,
+      "step": 150
+    },
+    {
+      "epoch": 0.06278818797213774,
+      "grad_norm": 118.0,
+      "learning_rate": 0.0001391304347826087,
+      "loss": 7.3375,
+      "step": 160
+    },
+    {
+      "epoch": 0.06671244972039635,
+      "grad_norm": 166.0,
+      "learning_rate": 0.00014782608695652173,
+      "loss": 4.602,
+      "step": 170
+    },
+    {
+      "epoch": 0.07063671146865495,
+      "grad_norm": 173.0,
+      "learning_rate": 0.0001565217391304348,
+      "loss": 4.8516,
+      "step": 180
+    },
+    {
+      "epoch": 0.07456097321691357,
+      "grad_norm": 48.5,
+      "learning_rate": 0.00016521739130434784,
+      "loss": 4.9902,
+      "step": 190
+    },
+    {
+      "epoch": 0.07848523496517218,
+      "grad_norm": 63.25,
+      "learning_rate": 0.00017391304347826088,
+      "loss": 4.4469,
+      "step": 200
+    },
+    {
+      "epoch": 0.08240949671343079,
+      "grad_norm": 50.0,
+      "learning_rate": 0.00018260869565217392,
+      "loss": 7.7786,
+      "step": 210
+    },
+    {
+      "epoch": 0.08633375846168939,
+      "grad_norm": 66.0,
+      "learning_rate": 0.00019130434782608697,
+      "loss": 5.1004,
+      "step": 220
+    },
+    {
+      "epoch": 0.090258020209948,
+      "grad_norm": 136.0,
+      "learning_rate": 0.0002,
+      "loss": 4.6939,
+      "step": 230
+    },
+    {
+      "epoch": 0.09418228195820662,
+      "grad_norm": 73.0,
+      "learning_rate": 0.00019999910223238217,
+      "loss": 5.0741,
+      "step": 240
+    },
+    {
+      "epoch": 0.09810654370646522,
+      "grad_norm": 34.0,
+      "learning_rate": 0.00019999640894564831,
+      "loss": 4.2687,
+      "step": 250
+    },
+    {
+      "epoch": 0.10203080545472383,
+      "grad_norm": 120.5,
+      "learning_rate": 0.00019999192018815736,
+      "loss": 5.4613,
+      "step": 260
+    },
+    {
+      "epoch": 0.10595506720298244,
+      "grad_norm": 146.0,
+      "learning_rate": 0.0001999856360405066,
+      "loss": 5.0583,
+      "step": 270
+    },
+    {
+      "epoch": 0.10987932895124104,
+      "grad_norm": 78.0,
+      "learning_rate": 0.00019997755661553006,
+      "loss": 4.8757,
+      "step": 280
+    },
+    {
+      "epoch": 0.11380359069949966,
+      "grad_norm": 25.25,
+      "learning_rate": 0.00019996768205829665,
+      "loss": 5.2871,
+      "step": 290
+    },
+    {
+      "epoch": 0.11772785244775827,
+      "grad_norm": 44.25,
+      "learning_rate": 0.00019995601254610757,
+      "loss": 4.4138,
+      "step": 300
+    },
+    {
+      "epoch": 0.12165211419601688,
+      "grad_norm": 25.375,
+      "learning_rate": 0.000199942548288493,
+      "loss": 5.5204,
+      "step": 310
+    },
+    {
+      "epoch": 0.12557637594427548,
+      "grad_norm": 29.75,
+      "learning_rate": 0.0001999272895272084,
+      "loss": 4.7227,
+      "step": 320
+    },
+    {
+      "epoch": 0.1295006376925341,
+      "grad_norm": 22.375,
+      "learning_rate": 0.0001999102365362303,
+      "loss": 4.465,
+      "step": 330
+    },
+    {
+      "epoch": 0.1334248994407927,
+      "grad_norm": 28.25,
+      "learning_rate": 0.00019989138962175104,
+      "loss": 4.6431,
+      "step": 340
+    },
+    {
+      "epoch": 0.1373491611890513,
+      "grad_norm": 40.0,
+      "learning_rate": 0.00019987074912217369,
+      "loss": 4.0119,
+      "step": 350
+    },
+    {
+      "epoch": 0.1412734229373099,
+      "grad_norm": 54.5,
+      "learning_rate": 0.00019984831540810567,
+      "loss": 4.6763,
+      "step": 360
+    },
+    {
+      "epoch": 0.14519768468556854,
+      "grad_norm": 18.375,
+      "learning_rate": 0.00019982408888235225,
+      "loss": 4.4112,
+      "step": 370
+    },
+    {
+      "epoch": 0.14912194643382715,
+      "grad_norm": 12.8125,
+      "learning_rate": 0.00019979806997990918,
+      "loss": 4.2684,
+      "step": 380
+    },
+    {
+      "epoch": 0.15304620818208575,
+      "grad_norm": 50.5,
+      "learning_rate": 0.00019977025916795502,
+      "loss": 4.5111,
+      "step": 390
+    },
+    {
+      "epoch": 0.15697046993034436,
+      "grad_norm": 23.75,
+      "learning_rate": 0.00019974065694584279,
+      "loss": 4.0444,
+      "step": 400
+    },
+    {
+      "epoch": 0.16089473167860296,
+      "grad_norm": 20.375,
+      "learning_rate": 0.00019970926384509072,
+      "loss": 5.0474,
+      "step": 410
+    },
+    {
+      "epoch": 0.16481899342686157,
+      "grad_norm": 24.375,
+      "learning_rate": 0.00019967608042937302,
+      "loss": 4.3519,
+      "step": 420
+    },
+    {
+      "epoch": 0.16874325517512018,
+      "grad_norm": 19.625,
+      "learning_rate": 0.00019964110729450964,
+      "loss": 4.0591,
+      "step": 430
+    },
+    {
+      "epoch": 0.17266751692337878,
+      "grad_norm": 25.0,
+      "learning_rate": 0.00019960434506845553,
+      "loss": 4.4677,
+      "step": 440
+    },
+    {
+      "epoch": 0.1765917786716374,
+      "grad_norm": 11.1875,
+      "learning_rate": 0.00019956579441128943,
+      "loss": 3.7723,
+      "step": 450
+    },
+    {
+      "epoch": 0.180516040419896,
+      "grad_norm": 25.625,
+      "learning_rate": 0.0001995254560152019,
+      "loss": 4.4753,
+      "step": 460
+    },
+    {
+      "epoch": 0.18444030216815463,
+      "grad_norm": 10.4375,
+      "learning_rate": 0.0001994833306044831,
+      "loss": 4.0153,
+      "step": 470
+    },
+    {
+      "epoch": 0.18836456391641324,
+      "grad_norm": 20.75,
+      "learning_rate": 0.00019943941893550967,
+      "loss": 3.704,
+      "step": 480
+    },
+    {
+      "epoch": 0.19228882566467184,
+      "grad_norm": 6.875,
+      "learning_rate": 0.00019939372179673105,
+      "loss": 4.2521,
+      "step": 490
+    },
+    {
+      "epoch": 0.19621308741293045,
+      "grad_norm": 13.8125,
+      "learning_rate": 0.00019934624000865543,
+      "loss": 3.4719,
+      "step": 500
+    },
+    {
+      "epoch": 0.20013734916118905,
+      "grad_norm": 8.5625,
+      "learning_rate": 0.00019929697442383512,
+      "loss": 4.1322,
+      "step": 510
+    },
+    {
+      "epoch": 0.20406161090944766,
+      "grad_norm": 8.5,
+      "learning_rate": 0.00019924592592685103,
+      "loss": 3.874,
+      "step": 520
+    },
+    {
+      "epoch": 0.20798587265770627,
+      "grad_norm": 14.8125,
+      "learning_rate": 0.0001991930954342969,
+      "loss": 3.503,
+      "step": 530
+    },
+    {
+      "epoch": 0.21191013440596487,
+      "grad_norm": 10.0,
+      "learning_rate": 0.00019913848389476285,
+      "loss": 4.0009,
+      "step": 540
+    },
+    {
+      "epoch": 0.21583439615422348,
+      "grad_norm": 12.8125,
+      "learning_rate": 0.00019908209228881827,
+      "loss": 3.3383,
+      "step": 550
+    },
+    {
+      "epoch": 0.21975865790248209,
+      "grad_norm": 24.5,
+      "learning_rate": 0.0001990239216289944,
+      "loss": 4.0019,
+      "step": 560
+    },
+    {
+      "epoch": 0.2236829196507407,
+      "grad_norm": 15.25,
+      "learning_rate": 0.00019896397295976584,
+      "loss": 3.8817,
+      "step": 570
+    },
+    {
+      "epoch": 0.22760718139899933,
+      "grad_norm": 7.78125,
+      "learning_rate": 0.00019890224735753215,
+      "loss": 3.4545,
+      "step": 580
+    },
+    {
+      "epoch": 0.23153144314725793,
+      "grad_norm": 8.125,
+      "learning_rate": 0.00019883874593059825,
+      "loss": 3.885,
+      "step": 590
+    },
+    {
+      "epoch": 0.23545570489551654,
+      "grad_norm": 24.375,
+      "learning_rate": 0.0001987734698191546,
+      "loss": 3.2593,
+      "step": 600
+    },
+    {
+      "epoch": 0.23937996664377514,
+      "grad_norm": 18.75,
+      "learning_rate": 0.00019870642019525678,
+      "loss": 4.0151,
+      "step": 610
+    },
+    {
+      "epoch": 0.24330422839203375,
+      "grad_norm": 18.5,
+      "learning_rate": 0.00019863759826280447,
+      "loss": 3.8175,
+      "step": 620
+    },
+    {
+      "epoch": 0.24722849014029236,
+      "grad_norm": 27.75,
+      "learning_rate": 0.00019856700525751965,
+      "loss": 3.3797,
+      "step": 630
+    },
+    {
+      "epoch": 0.25115275188855096,
+      "grad_norm": 31.25,
+      "learning_rate": 0.00019849464244692467,
+      "loss": 4.0016,
+      "step": 640
+    },
+    {
+      "epoch": 0.25507701363680957,
+      "grad_norm": 8.9375,
+      "learning_rate": 0.00019842051113031924,
+      "loss": 3.3371,
+      "step": 650
+    },
+    {
+      "epoch": 0.2590012753850682,
+      "grad_norm": 21.875,
+      "learning_rate": 0.00019834461263875728,
+      "loss": 4.0255,
+      "step": 660
+    },
+    {
+      "epoch": 0.2629255371333268,
+      "grad_norm": 16.375,
+      "learning_rate": 0.00019826694833502295,
+      "loss": 3.8169,
+      "step": 670
+    },
+    {
+      "epoch": 0.2668497988815854,
+      "grad_norm": 8.9375,
+      "learning_rate": 0.00019818751961360622,
+      "loss": 3.3529,
+      "step": 680
+    },
+    {
+      "epoch": 0.270774060629844,
+      "grad_norm": 12.8125,
+      "learning_rate": 0.00019810632790067772,
+      "loss": 3.9756,
+      "step": 690
+    },
+    {
+      "epoch": 0.2746983223781026,
+      "grad_norm": 11.1875,
+      "learning_rate": 0.0001980233746540633,
+      "loss": 3.3567,
+      "step": 700
+    },
+    {
+      "epoch": 0.2786225841263612,
+      "grad_norm": 7.25,
+      "learning_rate": 0.00019793866136321773,
+      "loss": 3.9975,
+      "step": 710
+    },
+    {
+      "epoch": 0.2825468458746198,
+      "grad_norm": 7.125,
+      "learning_rate": 0.00019785218954919797,
+      "loss": 3.6567,
+      "step": 720
+    },
+    {
+      "epoch": 0.2864711076228785,
+      "grad_norm": 3.75,
+      "learning_rate": 0.00019776396076463595,
+      "loss": 3.1378,
+      "step": 730
+    },
+    {
+      "epoch": 0.2903953693711371,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00019767397659371057,
+      "loss": 3.769,
+      "step": 740
+    },
+    {
+      "epoch": 0.2943196311193957,
+      "grad_norm": 8.3125,
+      "learning_rate": 0.0001975822386521193,
+      "loss": 3.1857,
+      "step": 750
+    },
+    {
+      "epoch": 0.2982438928676543,
+      "grad_norm": 6.03125,
+      "learning_rate": 0.00019748874858704923,
+      "loss": 3.6753,
+      "step": 760
+    },
+    {
+      "epoch": 0.3021681546159129,
+      "grad_norm": 4.96875,
+      "learning_rate": 0.0001973935080771474,
+      "loss": 3.5701,
+      "step": 770
+    },
+    {
+      "epoch": 0.3060924163641715,
+      "grad_norm": 6.28125,
+      "learning_rate": 0.00019729651883249074,
+      "loss": 3.1422,
+      "step": 780
+    },
+    {
+      "epoch": 0.3100166781124301,
+      "grad_norm": 17.5,
+      "learning_rate": 0.00019719778259455532,
+      "loss": 3.6516,
+      "step": 790
+    },
+    {
+      "epoch": 0.3139409398606887,
+      "grad_norm": 7.625,
+      "learning_rate": 0.00019709730113618508,
+      "loss": 3.1236,
+      "step": 800
+    },
+    {
+      "epoch": 0.3178652016089473,
+      "grad_norm": 8.875,
+      "learning_rate": 0.00019699507626156,
+      "loss": 3.6037,
+      "step": 810
+    },
+    {
+      "epoch": 0.32178946335720593,
+      "grad_norm": 7.5625,
+      "learning_rate": 0.0001968911098061637,
+      "loss": 3.5716,
+      "step": 820
+    },
+    {
+      "epoch": 0.32571372510546454,
+      "grad_norm": 6.84375,
+      "learning_rate": 0.0001967854036367506,
+      "loss": 3.067,
+      "step": 830
+    },
+    {
+      "epoch": 0.32963798685372314,
+      "grad_norm": 7.875,
+      "learning_rate": 0.00019667795965131214,
+      "loss": 3.6351,
+      "step": 840
+    },
+    {
+      "epoch": 0.33356224860198175,
+      "grad_norm": 9.5625,
+      "learning_rate": 0.00019656877977904298,
+      "loss": 3.0861,
+      "step": 850
+    },
+    {
+      "epoch": 0.33748651035024035,
+      "grad_norm": 9.8125,
+      "learning_rate": 0.00019645786598030616,
+      "loss": 3.5302,
+      "step": 860
+    },
+    {
+      "epoch": 0.34141077209849896,
+      "grad_norm": 13.8125,
+      "learning_rate": 0.00019634522024659804,
+      "loss": 3.4696,
+      "step": 870
+    },
+    {
+      "epoch": 0.34533503384675757,
+      "grad_norm": 11.9375,
+      "learning_rate": 0.00019623084460051248,
+      "loss": 2.9622,
+      "step": 880
+    },
+    {
+      "epoch": 0.3492592955950162,
+      "grad_norm": 7.5625,
+      "learning_rate": 0.00019611474109570444,
+      "loss": 3.6285,
+      "step": 890
+    },
+    {
+      "epoch": 0.3531835573432748,
+      "grad_norm": 7.9375,
+      "learning_rate": 0.00019599691181685334,
+      "loss": 3.0531,
+      "step": 900
+    },
+    {
+      "epoch": 0.3571078190915334,
+      "grad_norm": 10.0,
+      "learning_rate": 0.00019587735887962533,
+      "loss": 3.5168,
+      "step": 910
+    },
+    {
+      "epoch": 0.361032080839792,
+      "grad_norm": 8.4375,
+      "learning_rate": 0.00019575608443063558,
+      "loss": 3.4281,
+      "step": 920
+    },
+    {
+      "epoch": 0.3649563425880506,
+      "grad_norm": 45.75,
+      "learning_rate": 0.00019563309064740954,
+      "loss": 2.9524,
+      "step": 930
+    },
+    {
+      "epoch": 0.36888060433630926,
+      "grad_norm": 8.625,
+      "learning_rate": 0.0001955083797383439,
+      "loss": 3.6021,
+      "step": 940
+    },
+    {
+      "epoch": 0.37280486608456787,
+      "grad_norm": 6.40625,
+      "learning_rate": 0.000195381953942667,
+      "loss": 3.0039,
+      "step": 950
+    },
+    {
+      "epoch": 0.37672912783282647,
+      "grad_norm": 7.15625,
+      "learning_rate": 0.00019525381553039852,
+      "loss": 3.5288,
+      "step": 960
+    },
+    {
+      "epoch": 0.3806533895810851,
+      "grad_norm": 6.09375,
+      "learning_rate": 0.00019512396680230882,
+      "loss": 3.4231,
+      "step": 970
+    },
+    {
+      "epoch": 0.3845776513293437,
+      "grad_norm": 10.1875,
+      "learning_rate": 0.0001949924100898776,
+      "loss": 2.9039,
+      "step": 980
+    },
+    {
+      "epoch": 0.3885019130776023,
+      "grad_norm": 21.375,
+      "learning_rate": 0.00019485914775525193,
+      "loss": 3.5415,
+      "step": 990
+    },
+    {
+      "epoch": 0.3924261748258609,
+      "grad_norm": 9.25,
+      "learning_rate": 0.000194724182191204,
+      "loss": 2.9421,
+      "step": 1000
+    },
+    {
+      "epoch": 0.3963504365741195,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.0001945875158210881,
+      "loss": 3.3961,
+      "step": 1010
+    },
+    {
+      "epoch": 0.4002746983223781,
+      "grad_norm": 10.5625,
+      "learning_rate": 0.00019444915109879704,
+      "loss": 3.3578,
+      "step": 1020
+    },
+    {
+      "epoch": 0.4041989600706367,
+      "grad_norm": 4.625,
+      "learning_rate": 0.00019430909050871815,
+      "loss": 2.8669,
+      "step": 1030
+    },
+    {
+      "epoch": 0.4081232218188953,
+      "grad_norm": 4.9375,
+      "learning_rate": 0.00019416733656568868,
+      "loss": 3.531,
+      "step": 1040
+    },
+    {
+      "epoch": 0.4120474835671539,
+      "grad_norm": 9.0,
+      "learning_rate": 0.00019402389181495062,
+      "loss": 2.8848,
+      "step": 1050
+    },
+    {
+      "epoch": 0.41597174531541253,
+      "grad_norm": 6.40625,
+      "learning_rate": 0.00019387875883210507,
+      "loss": 3.4386,
+      "step": 1060
+    },
+    {
+      "epoch": 0.41989600706367114,
+      "grad_norm": 6.4375,
+      "learning_rate": 0.00019373194022306576,
+      "loss": 3.2987,
+      "step": 1070
+    },
+    {
+      "epoch": 0.42382026881192975,
+      "grad_norm": 7.0,
+      "learning_rate": 0.0001935834386240127,
+      "loss": 2.8508,
+      "step": 1080
+    },
+    {
+      "epoch": 0.42774453056018835,
+      "grad_norm": 4.40625,
+      "learning_rate": 0.00019343325670134428,
+      "loss": 3.4435,
+      "step": 1090
+    },
+    {
+      "epoch": 0.43166879230844696,
+      "grad_norm": 7.5,
+      "learning_rate": 0.00019328139715162995,
+      "loss": 2.8863,
+      "step": 1100
+    },
+    {
+      "epoch": 0.43559305405670556,
+      "grad_norm": 5.34375,
+      "learning_rate": 0.00019312786270156136,
+      "loss": 3.3325,
+      "step": 1110
+    },
+    {
+      "epoch": 0.43951731580496417,
+      "grad_norm": 7.6875,
+      "learning_rate": 0.0001929726561079037,
+      "loss": 3.2968,
+      "step": 1120
+    },
+    {
+      "epoch": 0.4434415775532228,
+      "grad_norm": 5.9375,
+      "learning_rate": 0.00019281578015744604,
+      "loss": 2.8418,
+      "step": 1130
+    },
+    {
+      "epoch": 0.4473658393014814,
+      "grad_norm": 6.28125,
+      "learning_rate": 0.00019265723766695133,
+      "loss": 3.3859,
+      "step": 1140
+    },
+    {
+      "epoch": 0.45129010104974004,
+      "grad_norm": 6.1875,
+      "learning_rate": 0.00019249703148310586,
+      "loss": 2.8663,
+      "step": 1150
+    },
+    {
+      "epoch": 0.45521436279799865,
+      "grad_norm": 3.84375,
+      "learning_rate": 0.00019233516448246814,
+      "loss": 3.3532,
+      "step": 1160
+    },
+    {
+      "epoch": 0.45913862454625726,
+      "grad_norm": 4.3125,
+      "learning_rate": 0.00019217163957141716,
+      "loss": 3.203,
+      "step": 1170
+    },
+    {
+      "epoch": 0.46306288629451586,
+      "grad_norm": 6.09375,
+      "learning_rate": 0.00019200645968610035,
+      "loss": 2.6933,
+      "step": 1180
+    },
+    {
+      "epoch": 0.46698714804277447,
+      "grad_norm": 8.75,
+      "learning_rate": 0.0001918396277923807,
+      "loss": 3.3249,
+      "step": 1190
+    },
+    {
+      "epoch": 0.4709114097910331,
+      "grad_norm": 7.40625,
+      "learning_rate": 0.0001916711468857837,
+      "loss": 2.808,
+      "step": 1200
+    },
+    {
+      "epoch": 0.4748356715392917,
+      "grad_norm": 12.625,
+      "learning_rate": 0.00019150101999144338,
+      "loss": 3.4173,
+      "step": 1210
+    },
+    {
+      "epoch": 0.4787599332875503,
+      "grad_norm": 7.03125,
+      "learning_rate": 0.00019132925016404804,
+      "loss": 3.3611,
+      "step": 1220
+    },
+    {
+      "epoch": 0.4826841950358089,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.0001911558404877855,
+      "loss": 2.8389,
+      "step": 1230
+    },
+    {
+      "epoch": 0.4866084567840675,
+      "grad_norm": 4.5625,
+      "learning_rate": 0.0001909807940762876,
+      "loss": 3.3609,
+      "step": 1240
+    },
+    {
+      "epoch": 0.4905327185323261,
+      "grad_norm": 6.65625,
+      "learning_rate": 0.00019080411407257431,
+      "loss": 2.7412,
+      "step": 1250
+    },
+    {
+      "epoch": 0.4944569802805847,
+      "grad_norm": 4.5625,
+      "learning_rate": 0.00019062580364899732,
+      "loss": 3.2928,
+      "step": 1260
+    },
+    {
+      "epoch": 0.4983812420288433,
+      "grad_norm": 5.6875,
+      "learning_rate": 0.00019044586600718322,
+      "loss": 3.269,
+      "step": 1270
+    },
+    {
+      "epoch": 0.5023055037771019,
+      "grad_norm": 139.0,
+      "learning_rate": 0.00019026430437797567,
+      "loss": 2.6846,
+      "step": 1280
+    },
+    {
+      "epoch": 0.5062297655253606,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.00019008112202137777,
+      "loss": 3.3716,
+      "step": 1290
+    },
+    {
+      "epoch": 0.5101540272736191,
+      "grad_norm": 10.125,
+      "learning_rate": 0.0001898963222264932,
+      "loss": 2.7933,
+      "step": 1300
+    },
+    {
+      "epoch": 0.5140782890218778,
+      "grad_norm": 5.3125,
+      "learning_rate": 0.00018970990831146744,
+      "loss": 3.2652,
+      "step": 1310
+    },
+    {
+      "epoch": 0.5180025507701363,
+      "grad_norm": 9.9375,
+      "learning_rate": 0.00018952188362342804,
+      "loss": 3.2512,
+      "step": 1320
+    },
+    {
+      "epoch": 0.521926812518395,
+      "grad_norm": 7.6875,
+      "learning_rate": 0.00018933225153842446,
+      "loss": 2.71,
+      "step": 1330
+    },
+    {
+      "epoch": 0.5258510742666536,
+      "grad_norm": 3.640625,
+      "learning_rate": 0.00018914101546136766,
+      "loss": 3.2908,
+      "step": 1340
+    },
+    {
+      "epoch": 0.5297753360149122,
+      "grad_norm": 5.46875,
+      "learning_rate": 0.00018894817882596877,
+      "loss": 2.7722,
+      "step": 1350
+    },
+    {
+      "epoch": 0.5336995977631708,
+      "grad_norm": 6.40625,
+      "learning_rate": 0.00018875374509467755,
+      "loss": 3.2667,
+      "step": 1360
+    },
+    {
+      "epoch": 0.5376238595114294,
+      "grad_norm": 7.5625,
+      "learning_rate": 0.00018855771775862012,
+      "loss": 3.1974,
+      "step": 1370
+    },
+    {
+      "epoch": 0.541548121259688,
+      "grad_norm": 4.71875,
+      "learning_rate": 0.00018836010033753637,
+      "loss": 2.6784,
+      "step": 1380
+    },
+    {
+      "epoch": 0.5454723830079466,
+      "grad_norm": 7.21875,
+      "learning_rate": 0.00018816089637971675,
+      "loss": 3.3083,
+      "step": 1390
+    },
+    {
+      "epoch": 0.5493966447562052,
+      "grad_norm": 7.3125,
+      "learning_rate": 0.0001879601094619385,
+      "loss": 2.7173,
+      "step": 1400
+    },
+    {
+      "epoch": 0.5533209065044639,
+      "grad_norm": 6.59375,
+      "learning_rate": 0.0001877577431894015,
+      "loss": 3.1636,
+      "step": 1410
+    },
+    {
+      "epoch": 0.5572451682527224,
+      "grad_norm": 9.4375,
+      "learning_rate": 0.00018755380119566341,
+      "loss": 3.1862,
+      "step": 1420
+    },
+    {
+      "epoch": 0.5611694300009811,
+      "grad_norm": 5.78125,
+      "learning_rate": 0.0001873482871425747,
+      "loss": 2.689,
+      "step": 1430
+    },
+    {
+      "epoch": 0.5650936917492396,
+      "grad_norm": 5.25,
+      "learning_rate": 0.0001871412047202125,
+      "loss": 3.2723,
+      "step": 1440
+    },
+    {
+      "epoch": 0.5690179534974983,
+      "grad_norm": 6.53125,
+      "learning_rate": 0.00018693255764681474,
+      "loss": 2.7397,
+      "step": 1450
+    },
+    {
+      "epoch": 0.572942215245757,
+      "grad_norm": 6.8125,
+      "learning_rate": 0.0001867223496687131,
+      "loss": 3.1995,
+      "step": 1460
+    },
+    {
+      "epoch": 0.5768664769940155,
+      "grad_norm": 13.875,
+      "learning_rate": 0.00018651058456026592,
+      "loss": 3.1304,
+      "step": 1470
+    },
+    {
+      "epoch": 0.5807907387422742,
+      "grad_norm": 5.90625,
+      "learning_rate": 0.00018629726612379033,
+      "loss": 2.6297,
+      "step": 1480
+    },
+    {
+      "epoch": 0.5847150004905327,
+      "grad_norm": 5.5,
+      "learning_rate": 0.000186082398189494,
+      "loss": 3.32,
+      "step": 1490
+    },
+    {
+      "epoch": 0.5886392622387914,
+      "grad_norm": 6.03125,
+      "learning_rate": 0.00018586598461540644,
+      "loss": 2.667,
+      "step": 1500
+    },
+    {
+      "epoch": 0.5925635239870499,
+      "grad_norm": 5.28125,
+      "learning_rate": 0.00018564802928730965,
+      "loss": 3.1841,
+      "step": 1510
+    },
+    {
+      "epoch": 0.5964877857353086,
+      "grad_norm": 8.0625,
+      "learning_rate": 0.00018542853611866825,
+      "loss": 3.152,
+      "step": 1520
+    },
+    {
+      "epoch": 0.6004120474835671,
+      "grad_norm": 8.375,
+      "learning_rate": 0.00018520750905055947,
+      "loss": 2.6081,
+      "step": 1530
+    },
+    {
+      "epoch": 0.6043363092318258,
+      "grad_norm": 10.125,
+      "learning_rate": 0.00018498495205160226,
+      "loss": 3.2542,
+      "step": 1540
+    },
+    {
+      "epoch": 0.6082605709800843,
+      "grad_norm": 5.5,
+      "learning_rate": 0.00018476086911788585,
+      "loss": 2.721,
+      "step": 1550
+    },
+    {
+      "epoch": 0.612184832728343,
+      "grad_norm": 9.1875,
+      "learning_rate": 0.00018453526427289835,
+      "loss": 3.1691,
+      "step": 1560
+    },
+    {
+      "epoch": 0.6161090944766016,
+      "grad_norm": 7.40625,
+      "learning_rate": 0.0001843081415674542,
+      "loss": 3.1468,
+      "step": 1570
+    },
+    {
+      "epoch": 0.6200333562248602,
+      "grad_norm": 5.84375,
+      "learning_rate": 0.00018407950507962165,
+      "loss": 2.6076,
+      "step": 1580
+    },
+    {
+      "epoch": 0.6239576179731188,
+      "grad_norm": 5.96875,
+      "learning_rate": 0.0001838493589146494,
+      "loss": 3.2776,
+      "step": 1590
+    },
+    {
+      "epoch": 0.6278818797213774,
+      "grad_norm": 6.34375,
+      "learning_rate": 0.00018361770720489286,
+      "loss": 2.6292,
+      "step": 1600
+    },
+    {
+      "epoch": 0.631806141469636,
+      "grad_norm": 5.40625,
+      "learning_rate": 0.0001833845541097402,
+      "loss": 3.1379,
+      "step": 1610
+    },
+    {
+      "epoch": 0.6357304032178946,
+      "grad_norm": 5.65625,
+      "learning_rate": 0.00018314990381553727,
+      "loss": 3.1267,
+      "step": 1620
+    },
+    {
+      "epoch": 0.6396546649661532,
+      "grad_norm": 6.625,
+      "learning_rate": 0.00018291376053551292,
+      "loss": 2.6291,
+      "step": 1630
+    },
+    {
+      "epoch": 0.6435789267144119,
+      "grad_norm": 6.59375,
+      "learning_rate": 0.0001826761285097029,
+      "loss": 3.2707,
+      "step": 1640
+    },
+    {
+      "epoch": 0.6475031884626704,
+      "grad_norm": 7.375,
+      "learning_rate": 0.0001824370120048739,
+      "loss": 2.698,
+      "step": 1650
+    },
+    {
+      "epoch": 0.6514274502109291,
+      "grad_norm": 4.71875,
+      "learning_rate": 0.00018219641531444714,
+      "loss": 3.1409,
+      "step": 1660
+    },
+    {
+      "epoch": 0.6553517119591877,
+      "grad_norm": 4.75,
+      "learning_rate": 0.0001819543427584209,
+      "loss": 3.1461,
+      "step": 1670
+    },
+    {
+      "epoch": 0.6592759737074463,
+      "grad_norm": 8.8125,
+      "learning_rate": 0.0001817107986832932,
+      "loss": 2.6312,
+      "step": 1680
+    },
+    {
+      "epoch": 0.663200235455705,
+      "grad_norm": 6.46875,
+      "learning_rate": 0.00018146578746198376,
+      "loss": 3.3014,
+      "step": 1690
+    },
+    {
+      "epoch": 0.6671244972039635,
+      "grad_norm": 6.15625,
+      "learning_rate": 0.00018121931349375538,
+      "loss": 2.7297,
+      "step": 1700
+    },
+    {
+      "epoch": 0.6710487589522222,
+      "grad_norm": 8.9375,
+      "learning_rate": 0.000180971381204135,
+      "loss": 3.2115,
+      "step": 1710
+    },
+    {
+      "epoch": 0.6749730207004807,
+      "grad_norm": 9.5,
+      "learning_rate": 0.00018072199504483427,
+      "loss": 3.1043,
+      "step": 1720
+    },
+    {
+      "epoch": 0.6788972824487394,
+      "grad_norm": 4.84375,
+      "learning_rate": 0.00018047115949366952,
+      "loss": 2.6526,
+      "step": 1730
+    },
+    {
+      "epoch": 0.6828215441969979,
+      "grad_norm": 5.5625,
+      "learning_rate": 0.00018021887905448147,
+      "loss": 3.247,
+      "step": 1740
+    },
+    {
+      "epoch": 0.6867458059452566,
+      "grad_norm": 5.8125,
+      "learning_rate": 0.0001799651582570543,
+      "loss": 2.6606,
+      "step": 1750
+    },
+    {
+      "epoch": 0.6906700676935151,
+      "grad_norm": 10.25,
+      "learning_rate": 0.00017971000165703434,
+      "loss": 3.1366,
+      "step": 1760
+    },
+    {
+      "epoch": 0.6945943294417738,
+      "grad_norm": 5.0,
+      "learning_rate": 0.00017945341383584818,
+      "loss": 3.0867,
+      "step": 1770
+    },
+    {
+      "epoch": 0.6985185911900323,
+      "grad_norm": 7.0625,
+      "learning_rate": 0.00017919539940062067,
+      "loss": 2.6189,
+      "step": 1780
+    },
+    {
+      "epoch": 0.702442852938291,
+      "grad_norm": 9.375,
+      "learning_rate": 0.00017893596298409182,
+      "loss": 3.2992,
+      "step": 1790
+    },
+    {
+      "epoch": 0.7063671146865496,
+      "grad_norm": 8.125,
+      "learning_rate": 0.00017867510924453393,
+      "loss": 2.6632,
+      "step": 1800
+    },
+    {
+      "epoch": 0.7102913764348082,
+      "grad_norm": 14.0,
+      "learning_rate": 0.00017841284286566783,
+      "loss": 3.1754,
+      "step": 1810
+    },
+    {
+      "epoch": 0.7142156381830668,
+      "grad_norm": 11.0625,
+      "learning_rate": 0.00017814916855657872,
+      "loss": 3.1631,
+      "step": 1820
+    },
+    {
+      "epoch": 0.7181398999313254,
+      "grad_norm": 7.71875,
+      "learning_rate": 0.00017788409105163176,
+      "loss": 2.6381,
+      "step": 1830
+    },
+    {
+      "epoch": 0.722064161679584,
+      "grad_norm": 8.9375,
+      "learning_rate": 0.00017761761511038695,
+      "loss": 3.223,
+      "step": 1840
+    },
+    {
+      "epoch": 0.7259884234278426,
+      "grad_norm": 15.125,
+      "learning_rate": 0.0001773497455175137,
+      "loss": 2.6939,
+      "step": 1850
+    },
+    {
+      "epoch": 0.7299126851761012,
+      "grad_norm": 9.1875,
+      "learning_rate": 0.00017708048708270496,
+      "loss": 3.1695,
+      "step": 1860
+    },
+    {
+      "epoch": 0.7338369469243599,
+      "grad_norm": 5.46875,
+      "learning_rate": 0.00017680984464059078,
+      "loss": 3.1646,
+      "step": 1870
+    },
+    {
+      "epoch": 0.7377612086726185,
+      "grad_norm": 4.125,
+      "learning_rate": 0.00017653782305065158,
+      "loss": 2.6036,
+      "step": 1880
+    },
+    {
+      "epoch": 0.7416854704208771,
+      "grad_norm": 6.21875,
+      "learning_rate": 0.00017626442719713084,
+      "loss": 3.2936,
+      "step": 1890
+    },
+    {
+      "epoch": 0.7456097321691357,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.00017598966198894745,
+      "loss": 2.6312,
+      "step": 1900
+    },
+    {
+      "epoch": 0.7495339939173943,
+      "grad_norm": 17.875,
+      "learning_rate": 0.00017571353235960752,
+      "loss": 3.1444,
+      "step": 1910
+    },
+    {
+      "epoch": 0.7534582556656529,
+      "grad_norm": 4.875,
+      "learning_rate": 0.00017543604326711592,
+      "loss": 3.1033,
+      "step": 1920
+    },
+    {
+      "epoch": 0.7573825174139115,
+      "grad_norm": 2.78125,
+      "learning_rate": 0.00017515719969388697,
+      "loss": 2.6238,
+      "step": 1930
+    },
+    {
+      "epoch": 0.7613067791621702,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.00017487700664665535,
+      "loss": 3.204,
+      "step": 1940
+    },
+    {
+      "epoch": 0.7652310409104287,
+      "grad_norm": 5.15625,
+      "learning_rate": 0.00017459546915638593,
+      "loss": 2.6456,
+      "step": 1950
+    },
+    {
+      "epoch": 0.7691553026586874,
+      "grad_norm": 4.375,
+      "learning_rate": 0.0001743125922781836,
+      "loss": 3.1186,
+      "step": 1960
+    },
+    {
+      "epoch": 0.7730795644069459,
+      "grad_norm": 4.625,
+      "learning_rate": 0.00017402838109120228,
+      "loss": 3.0447,
+      "step": 1970
+    },
+    {
+      "epoch": 0.7770038261552046,
+      "grad_norm": 4.96875,
+      "learning_rate": 0.0001737428406985541,
+      "loss": 2.5082,
+      "step": 1980
+    },
+    {
+      "epoch": 0.7809280879034631,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.00017345597622721727,
+      "loss": 3.1698,
+      "step": 1990
+    },
+    {
+      "epoch": 0.7848523496517218,
+      "grad_norm": 5.21875,
+      "learning_rate": 0.00017316779282794458,
+      "loss": 2.5909,
+      "step": 2000
+    },
+    {
+      "epoch": 0.7887766113999803,
+      "grad_norm": 4.5625,
+      "learning_rate": 0.00017287829567517047,
+      "loss": 3.0503,
+      "step": 2010
+    },
+    {
+      "epoch": 0.792700873148239,
+      "grad_norm": 4.3125,
+      "learning_rate": 0.00017258748996691828,
+      "loss": 2.9786,
+      "step": 2020
+    },
+    {
+      "epoch": 0.7966251348964976,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.00017229538092470708,
+      "loss": 2.5165,
+      "step": 2030
+    },
+    {
+      "epoch": 0.8005493966447562,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00017200197379345751,
+      "loss": 3.1147,
+      "step": 2040
+    },
+    {
+      "epoch": 0.8044736583930148,
+      "grad_norm": 7.84375,
+      "learning_rate": 0.00017170727384139807,
+      "loss": 2.5776,
+      "step": 2050
+    },
+    {
+      "epoch": 0.8083979201412734,
+      "grad_norm": 5.84375,
+      "learning_rate": 0.00017141128635997027,
+      "loss": 3.0113,
+      "step": 2060
+    },
+    {
+      "epoch": 0.812322181889532,
+      "grad_norm": 4.5,
+      "learning_rate": 0.00017111401666373356,
+      "loss": 2.995,
+      "step": 2070
+    },
+    {
+      "epoch": 0.8162464436377906,
+      "grad_norm": 6.125,
+      "learning_rate": 0.00017081547009027012,
+      "loss": 2.5039,
+      "step": 2080
+    },
+    {
+      "epoch": 0.8201707053860493,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.00017051565200008886,
+      "loss": 3.0807,
+      "step": 2090
+    },
+    {
+      "epoch": 0.8240949671343079,
+      "grad_norm": 7.0625,
+      "learning_rate": 0.00017021456777652926,
+      "loss": 2.5534,
+      "step": 2100
+    },
+    {
+      "epoch": 0.8280192288825665,
+      "grad_norm": 5.25,
+      "learning_rate": 0.00016991222282566464,
+      "loss": 2.9733,
+      "step": 2110
+    },
+    {
+      "epoch": 0.8319434906308251,
+      "grad_norm": 4.84375,
+      "learning_rate": 0.00016960862257620507,
+      "loss": 2.9873,
+      "step": 2120
+    },
+    {
+      "epoch": 0.8358677523790837,
+      "grad_norm": 6.25,
+      "learning_rate": 0.00016930377247940004,
+      "loss": 2.4886,
+      "step": 2130
+    },
+    {
+      "epoch": 0.8397920141273423,
+      "grad_norm": 3.703125,
+      "learning_rate": 0.0001689976780089405,
+      "loss": 3.0768,
+      "step": 2140
+    },
+    {
+      "epoch": 0.8437162758756009,
+      "grad_norm": 6.5,
+      "learning_rate": 0.00016869034466086044,
+      "loss": 2.5609,
+      "step": 2150
+    },
+    {
+      "epoch": 0.8476405376238595,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00016838177795343846,
+      "loss": 3.0276,
+      "step": 2160
+    },
+    {
+      "epoch": 0.8515647993721182,
+      "grad_norm": 6.9375,
+      "learning_rate": 0.00016807198342709857,
+      "loss": 3.0004,
+      "step": 2170
+    },
+    {
+      "epoch": 0.8554890611203767,
+      "grad_norm": 7.28125,
+      "learning_rate": 0.0001677609666443105,
+      "loss": 2.4783,
+      "step": 2180
+    },
+    {
+      "epoch": 0.8594133228686354,
+      "grad_norm": 8.5,
+      "learning_rate": 0.00016744873318949032,
+      "loss": 3.195,
+      "step": 2190
+    },
+    {
+      "epoch": 0.8633375846168939,
+      "grad_norm": 6.25,
+      "learning_rate": 0.00016713528866889966,
+      "loss": 2.628,
+      "step": 2200
+    },
+    {
+      "epoch": 0.8672618463651526,
+      "grad_norm": 6.90625,
+      "learning_rate": 0.00016682063871054534,
+      "loss": 3.0609,
+      "step": 2210
+    },
+    {
+      "epoch": 0.8711861081134111,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00016650478896407825,
+      "loss": 2.9864,
+      "step": 2220
+    },
+    {
+      "epoch": 0.8751103698616698,
+      "grad_norm": 5.0,
+      "learning_rate": 0.00016618774510069185,
+      "loss": 2.5038,
+      "step": 2230
+    },
+    {
+      "epoch": 0.8790346316099283,
+      "grad_norm": 7.21875,
+      "learning_rate": 0.00016586951281302045,
+      "loss": 3.1454,
+      "step": 2240
+    },
+    {
+      "epoch": 0.882958893358187,
+      "grad_norm": 5.0,
+      "learning_rate": 0.0001655500978150369,
+      "loss": 2.5633,
+      "step": 2250
+    },
+    {
+      "epoch": 0.8868831551064456,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.00016522950584195,
+      "loss": 3.0048,
+      "step": 2260
+    },
+    {
+      "epoch": 0.8908074168547042,
+      "grad_norm": 4.25,
+      "learning_rate": 0.00016490774265010168,
+      "loss": 2.946,
+      "step": 2270
+    },
+    {
+      "epoch": 0.8947316786029628,
+      "grad_norm": 3.890625,
+      "learning_rate": 0.0001645848140168633,
+      "loss": 2.4716,
+      "step": 2280
+    },
+    {
+      "epoch": 0.8986559403512214,
+      "grad_norm": 8.0,
+      "learning_rate": 0.00016426072574053237,
+      "loss": 3.0791,
+      "step": 2290
+    },
+    {
+      "epoch": 0.9025802020994801,
+      "grad_norm": 5.75,
+      "learning_rate": 0.00016393548364022803,
+      "loss": 2.5295,
+      "step": 2300
+    },
+    {
+      "epoch": 0.9065044638477386,
+      "grad_norm": 4.59375,
+      "learning_rate": 0.0001636090935557868,
+      "loss": 2.9866,
+      "step": 2310
+    },
+    {
+      "epoch": 0.9104287255959973,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.00016328156134765763,
+      "loss": 2.9469,
+      "step": 2320
+    },
+    {
+      "epoch": 0.9143529873442559,
+      "grad_norm": 3.609375,
+      "learning_rate": 0.00016295289289679674,
+      "loss": 2.4791,
+      "step": 2330
+    },
+    {
+      "epoch": 0.9182772490925145,
+      "grad_norm": 6.28125,
+      "learning_rate": 0.000162623094104562,
+      "loss": 3.0446,
+      "step": 2340
+    },
+    {
+      "epoch": 0.9222015108407731,
+      "grad_norm": 4.75,
+      "learning_rate": 0.00016229217089260692,
+      "loss": 2.4908,
+      "step": 2350
+    },
+    {
+      "epoch": 0.9261257725890317,
+      "grad_norm": 5.1875,
+      "learning_rate": 0.00016196012920277438,
+      "loss": 2.9905,
+      "step": 2360
+    },
+    {
+      "epoch": 0.9300500343372903,
+      "grad_norm": 5.125,
+      "learning_rate": 0.00016162697499698986,
+      "loss": 2.9641,
+      "step": 2370
+    },
+    {
+      "epoch": 0.9339742960855489,
+      "grad_norm": 2.984375,
+      "learning_rate": 0.0001612927142571546,
+      "loss": 2.4331,
+      "step": 2380
+    },
+    {
+      "epoch": 0.9378985578338075,
+      "grad_norm": 4.0,
+      "learning_rate": 0.00016095735298503788,
+      "loss": 3.0465,
+      "step": 2390
+    },
+    {
+      "epoch": 0.9418228195820662,
+      "grad_norm": 5.71875,
+      "learning_rate": 0.00016062089720216956,
+      "loss": 2.4809,
+      "step": 2400
+    },
+    {
+      "epoch": 0.9457470813303247,
+      "grad_norm": 3.671875,
+      "learning_rate": 0.0001602833529497318,
+      "loss": 2.9407,
+      "step": 2410
+    },
+    {
+      "epoch": 0.9496713430785834,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.00015994472628845055,
+      "loss": 2.9277,
+      "step": 2420
+    },
+    {
+      "epoch": 0.9535956048268419,
+      "grad_norm": 3.46875,
+      "learning_rate": 0.0001596050232984868,
+      "loss": 2.3998,
+      "step": 2430
+    },
+    {
+      "epoch": 0.9575198665751006,
+      "grad_norm": 3.203125,
+      "learning_rate": 0.00015926425007932747,
+      "loss": 2.9873,
+      "step": 2440
+    },
+    {
+      "epoch": 0.9614441283233591,
+      "grad_norm": 5.40625,
+      "learning_rate": 0.00015892241274967578,
+      "loss": 2.4555,
+      "step": 2450
+    },
+    {
+      "epoch": 0.9653683900716178,
+      "grad_norm": 7.9375,
+      "learning_rate": 0.00015857951744734146,
+      "loss": 2.9727,
+      "step": 2460
+    },
+    {
+      "epoch": 0.9692926518198763,
+      "grad_norm": 118.0,
+      "learning_rate": 0.00015823557032913044,
+      "loss": 2.9336,
+      "step": 2470
+    },
+    {
+      "epoch": 0.973216913568135,
+      "grad_norm": 3.109375,
+      "learning_rate": 0.00015789057757073445,
+      "loss": 2.4269,
+      "step": 2480
+    },
+    {
+      "epoch": 0.9771411753163936,
+      "grad_norm": 3.671875,
+      "learning_rate": 0.00015754454536662,
+      "loss": 3.0237,
+      "step": 2490
+    },
+    {
+      "epoch": 0.9810654370646522,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.0001571974799299172,
+      "loss": 2.4744,
+      "step": 2500
+    },
+    {
+      "epoch": 0.9849896988129109,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.0001568493874923084,
+      "loss": 2.911,
+      "step": 2510
+    },
+    {
+      "epoch": 0.9889139605611694,
+      "grad_norm": 6.03125,
+      "learning_rate": 0.00015650027430391585,
+      "loss": 2.853,
+      "step": 2520
+    },
+    {
+      "epoch": 0.9928382223094281,
+      "grad_norm": 4.3125,
+      "learning_rate": 0.0001561501466331899,
+      "loss": 2.4222,
+      "step": 2530
+    },
+    {
+      "epoch": 0.9967624840576866,
+      "grad_norm": 3.484375,
+      "learning_rate": 0.00015579901076679623,
+      "loss": 2.9976,
+      "step": 2540
+    },
+    {
+      "epoch": 0.9999018934562935,
+      "eval_loss": 2.8136746883392334,
+      "eval_runtime": 156.5437,
+      "eval_samples_per_second": 159.7,
+      "eval_steps_per_second": 39.925,
+      "step": 2548
+    },
+    {
+      "epoch": 1.0006867458059452,
+      "grad_norm": 3.375,
+      "learning_rate": 0.00015544687300950307,
+      "loss": 2.4788,
+      "step": 2550
+    },
+    {
+      "epoch": 1.0046110075542038,
+      "grad_norm": 8.75,
+      "learning_rate": 0.0001550937396840679,
+      "loss": 2.9544,
+      "step": 2560
+    },
+    {
+      "epoch": 1.0085352693024625,
+      "grad_norm": 5.34375,
+      "learning_rate": 0.00015473961713112405,
+      "loss": 2.7297,
+      "step": 2570
+    },
+    {
+      "epoch": 1.0124595310507212,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.00015438451170906673,
+      "loss": 2.5198,
+      "step": 2580
+    },
+    {
+      "epoch": 1.0163837927989796,
+      "grad_norm": 3.15625,
+      "learning_rate": 0.00015402842979393884,
+      "loss": 2.8894,
+      "step": 2590
+    },
+    {
+      "epoch": 1.0203080545472383,
+      "grad_norm": 4.09375,
+      "learning_rate": 0.00015367137777931673,
+      "loss": 2.4052,
+      "step": 2600
+    },
+    {
+      "epoch": 1.024232316295497,
+      "grad_norm": 8.0625,
+      "learning_rate": 0.00015331336207619506,
+      "loss": 2.9147,
+      "step": 2610
+    },
+    {
+      "epoch": 1.0281565780437556,
+      "grad_norm": 5.09375,
+      "learning_rate": 0.000152954389112872,
+      "loss": 2.7529,
+      "step": 2620
+    },
+    {
+      "epoch": 1.032080839792014,
+      "grad_norm": 3.203125,
+      "learning_rate": 0.00015259446533483356,
+      "loss": 2.5033,
+      "step": 2630
+    },
+    {
+      "epoch": 1.0360051015402727,
+      "grad_norm": 3.78125,
+      "learning_rate": 0.00015223359720463795,
+      "loss": 2.9189,
+      "step": 2640
+    },
+    {
+      "epoch": 1.0399293632885314,
+      "grad_norm": 3.296875,
+      "learning_rate": 0.0001518717912017997,
+      "loss": 2.373,
+      "step": 2650
+    },
+    {
+      "epoch": 1.04385362503679,
+      "grad_norm": 3.40625,
+      "learning_rate": 0.00015150905382267296,
+      "loss": 2.8885,
+      "step": 2660
+    },
+    {
+      "epoch": 1.0477778867850485,
+      "grad_norm": 36.5,
+      "learning_rate": 0.00015114539158033529,
+      "loss": 2.7508,
+      "step": 2670
+    },
+    {
+      "epoch": 1.0517021485333071,
+      "grad_norm": 3.0,
+      "learning_rate": 0.00015078081100447036,
+      "loss": 2.5126,
+      "step": 2680
+    },
+    {
+      "epoch": 1.0556264102815658,
+      "grad_norm": 4.3125,
+      "learning_rate": 0.00015041531864125082,
+      "loss": 2.8859,
+      "step": 2690
+    },
+    {
+      "epoch": 1.0595506720298244,
+      "grad_norm": 3.703125,
+      "learning_rate": 0.0001500489210532209,
+      "loss": 2.3348,
+      "step": 2700
+    },
+    {
+      "epoch": 1.063474933778083,
+      "grad_norm": 4.6875,
+      "learning_rate": 0.00014968162481917836,
+      "loss": 2.8947,
+      "step": 2710
+    },
+    {
+      "epoch": 1.0673991955263415,
+      "grad_norm": 4.03125,
+      "learning_rate": 0.00014931343653405653,
+      "loss": 2.7665,
+      "step": 2720
+    },
+    {
+      "epoch": 1.0713234572746002,
+      "grad_norm": 4.46875,
+      "learning_rate": 0.00014894436280880576,
+      "loss": 2.484,
+      "step": 2730
+    },
+    {
+      "epoch": 1.0752477190228589,
+      "grad_norm": 3.65625,
+      "learning_rate": 0.00014857441027027484,
+      "loss": 2.8724,
+      "step": 2740
+    },
+    {
+      "epoch": 1.0791719807711175,
+      "grad_norm": 3.25,
+      "learning_rate": 0.000148203585561092,
+      "loss": 2.356,
+      "step": 2750
+    },
+    {
+      "epoch": 1.083096242519376,
+      "grad_norm": 3.1875,
+      "learning_rate": 0.00014783189533954555,
+      "loss": 2.8666,
+      "step": 2760
+    },
+    {
+      "epoch": 1.0870205042676346,
+      "grad_norm": 5.0625,
+      "learning_rate": 0.00014745934627946432,
+      "loss": 2.7198,
+      "step": 2770
+    },
+    {
+      "epoch": 1.0909447660158933,
+      "grad_norm": 6.0,
+      "learning_rate": 0.00014708594507009806,
+      "loss": 2.4611,
+      "step": 2780
+    },
+    {
+      "epoch": 1.094869027764152,
+      "grad_norm": 4.5625,
+      "learning_rate": 0.00014671169841599694,
+      "loss": 2.9093,
+      "step": 2790
+    },
+    {
+      "epoch": 1.0987932895124104,
+      "grad_norm": 4.3125,
+      "learning_rate": 0.00014633661303689157,
+      "loss": 2.3454,
+      "step": 2800
+    },
+    {
+      "epoch": 1.102717551260669,
+      "grad_norm": 3.765625,
+      "learning_rate": 0.00014596069566757207,
+      "loss": 2.8689,
+      "step": 2810
+    },
+    {
+      "epoch": 1.1066418130089277,
+      "grad_norm": 4.28125,
+      "learning_rate": 0.0001455839530577673,
+      "loss": 2.724,
+      "step": 2820
+    },
+    {
+      "epoch": 1.1105660747571864,
+      "grad_norm": 5.90625,
+      "learning_rate": 0.00014520639197202354,
+      "loss": 2.4845,
+      "step": 2830
+    },
+    {
+      "epoch": 1.1144903365054448,
+      "grad_norm": 5.625,
+      "learning_rate": 0.0001448280191895831,
+      "loss": 2.8606,
+      "step": 2840
+    },
+    {
+      "epoch": 1.1184145982537035,
+      "grad_norm": 3.484375,
+      "learning_rate": 0.00014444884150426269,
+      "loss": 2.3385,
+      "step": 2850
+    },
+    {
+      "epoch": 1.1223388600019621,
+      "grad_norm": 4.09375,
+      "learning_rate": 0.00014406886572433113,
+      "loss": 2.876,
+      "step": 2860
+    },
+    {
+      "epoch": 1.1262631217502208,
+      "grad_norm": 3.546875,
+      "learning_rate": 0.00014368809867238753,
+      "loss": 2.679,
+      "step": 2870
+    },
+    {
+      "epoch": 1.1301873834984795,
+      "grad_norm": 2.890625,
+      "learning_rate": 0.00014330654718523847,
+      "loss": 2.4405,
+      "step": 2880
+    },
+    {
+      "epoch": 1.134111645246738,
+      "grad_norm": 3.40625,
+      "learning_rate": 0.00014292421811377533,
+      "loss": 2.8675,
+      "step": 2890
+    },
+    {
+      "epoch": 1.1380359069949966,
+      "grad_norm": 4.15625,
+      "learning_rate": 0.00014254111832285126,
+      "loss": 2.3588,
+      "step": 2900
+    },
+    {
+      "epoch": 1.1419601687432552,
+      "grad_norm": 3.453125,
+      "learning_rate": 0.00014215725469115807,
+      "loss": 2.8569,
+      "step": 2910
+    },
+    {
+      "epoch": 1.1458844304915137,
+      "grad_norm": 3.25,
+      "learning_rate": 0.0001417726341111025,
+      "loss": 2.6694,
+      "step": 2920
+    },
+    {
+      "epoch": 1.1498086922397723,
+      "grad_norm": 2.4375,
+      "learning_rate": 0.00014138726348868249,
+      "loss": 2.4397,
+      "step": 2930
+    },
+    {
+      "epoch": 1.153732953988031,
+      "grad_norm": 3.453125,
+      "learning_rate": 0.00014100114974336352,
+      "loss": 2.8587,
+      "step": 2940
+    },
+    {
+      "epoch": 1.1576572157362897,
+      "grad_norm": 3.6875,
+      "learning_rate": 0.00014061429980795383,
+      "loss": 2.3305,
+      "step": 2950
+    },
+    {
+      "epoch": 1.1615814774845483,
+      "grad_norm": 3.265625,
+      "learning_rate": 0.00014022672062848036,
+      "loss": 2.8314,
+      "step": 2960
+    },
+    {
+      "epoch": 1.1655057392328068,
+      "grad_norm": 3.65625,
+      "learning_rate": 0.00013983841916406382,
+      "loss": 2.6601,
+      "step": 2970
+    },
+    {
+      "epoch": 1.1694300009810654,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.00013944940238679382,
+      "loss": 2.4379,
+      "step": 2980
+    },
+    {
+      "epoch": 1.173354262729324,
+      "grad_norm": 3.484375,
+      "learning_rate": 0.00013905967728160368,
+      "loss": 2.8203,
+      "step": 2990
+    },
+    {
+      "epoch": 1.1772785244775827,
+      "grad_norm": 2.359375,
+      "learning_rate": 0.00013866925084614503,
+      "loss": 2.3221,
+      "step": 3000
+    },
+    {
+      "epoch": 1.1812027862258412,
+      "grad_norm": 3.703125,
+      "learning_rate": 0.00013827813009066202,
+      "loss": 2.7772,
+      "step": 3010
+    },
+    {
+      "epoch": 1.1851270479740998,
+      "grad_norm": 3.6875,
+      "learning_rate": 0.00013788632203786565,
+      "loss": 2.6651,
+      "step": 3020
+    },
+    {
+      "epoch": 1.1890513097223585,
+      "grad_norm": 3.171875,
+      "learning_rate": 0.0001374938337228076,
+      "loss": 2.445,
+      "step": 3030
+    },
+    {
+      "epoch": 1.1929755714706172,
+      "grad_norm": 4.0,
+      "learning_rate": 0.00013710067219275382,
+      "loss": 2.8027,
+      "step": 3040
+    },
+    {
+      "epoch": 1.1968998332188756,
+      "grad_norm": 3.3125,
+      "learning_rate": 0.00013670684450705812,
+      "loss": 2.2656,
+      "step": 3050
+    },
+    {
+      "epoch": 1.2008240949671343,
+      "grad_norm": 3.34375,
+      "learning_rate": 0.00013631235773703534,
+      "loss": 2.8038,
+      "step": 3060
+    },
+    {
+      "epoch": 1.204748356715393,
+      "grad_norm": 3.875,
+      "learning_rate": 0.00013591721896583453,
+      "loss": 2.6447,
+      "step": 3070
+    },
+    {
+      "epoch": 1.2086726184636516,
+      "grad_norm": 3.296875,
+      "learning_rate": 0.0001355214352883115,
+      "loss": 2.4303,
+      "step": 3080
+    },
+    {
+      "epoch": 1.21259688021191,
+      "grad_norm": 2.890625,
+      "learning_rate": 0.00013512501381090158,
+      "loss": 2.7888,
+      "step": 3090
+    },
+    {
+      "epoch": 1.2165211419601687,
+      "grad_norm": 3.375,
+      "learning_rate": 0.00013472796165149217,
+      "loss": 2.2864,
+      "step": 3100
+    },
+    {
+      "epoch": 1.2204454037084274,
+      "grad_norm": 7.5625,
+      "learning_rate": 0.00013433028593929467,
+      "loss": 2.7817,
+      "step": 3110
+    },
+    {
+      "epoch": 1.224369665456686,
+      "grad_norm": 3.796875,
+      "learning_rate": 0.00013393199381471656,
+      "loss": 2.5721,
+      "step": 3120
+    },
+    {
+      "epoch": 1.2282939272049447,
+      "grad_norm": 2.953125,
+      "learning_rate": 0.00013353309242923334,
+      "loss": 2.4053,
+      "step": 3130
+    },
+    {
+      "epoch": 1.2322181889532031,
+      "grad_norm": 4.15625,
+      "learning_rate": 0.00013313358894525995,
+      "loss": 2.7832,
+      "step": 3140
+    },
+    {
+      "epoch": 1.2361424507014618,
+      "grad_norm": 5.53125,
+      "learning_rate": 0.0001327334905360222,
+      "loss": 2.2835,
+      "step": 3150
+    },
+    {
+      "epoch": 1.2400667124497204,
+      "grad_norm": 21.375,
+      "learning_rate": 0.00013233280438542796,
+      "loss": 2.8109,
+      "step": 3160
+    },
+    {
+      "epoch": 1.2439909741979789,
+      "grad_norm": 12.375,
+      "learning_rate": 0.0001319315376879383,
+      "loss": 2.6082,
+      "step": 3170
+    },
+    {
+      "epoch": 1.2479152359462375,
+      "grad_norm": 5.78125,
+      "learning_rate": 0.00013152969764843812,
+      "loss": 2.4347,
+      "step": 3180
+    },
+    {
+      "epoch": 1.2518394976944962,
+      "grad_norm": 10.0625,
+      "learning_rate": 0.00013112729148210693,
+      "loss": 2.8174,
+      "step": 3190
+    },
+    {
+      "epoch": 1.2557637594427549,
+      "grad_norm": 5.78125,
+      "learning_rate": 0.00013072432641428932,
+      "loss": 2.2879,
+      "step": 3200
+    },
+    {
+      "epoch": 1.2596880211910135,
+      "grad_norm": 6.875,
+      "learning_rate": 0.00013032080968036498,
+      "loss": 2.7919,
+      "step": 3210
+    },
+    {
+      "epoch": 1.263612282939272,
+      "grad_norm": 5.3125,
+      "learning_rate": 0.00012991674852561903,
+      "loss": 2.6585,
+      "step": 3220
+    },
+    {
+      "epoch": 1.2675365446875306,
+      "grad_norm": 4.25,
+      "learning_rate": 0.00012951215020511197,
+      "loss": 2.4156,
+      "step": 3230
+    },
+    {
+      "epoch": 1.2714608064357893,
+      "grad_norm": 6.78125,
+      "learning_rate": 0.00012910702198354915,
+      "loss": 2.7929,
+      "step": 3240
+    },
+    {
+      "epoch": 1.275385068184048,
+      "grad_norm": 3.984375,
+      "learning_rate": 0.00012870137113515053,
+      "loss": 2.2606,
+      "step": 3250
+    },
+    {
+      "epoch": 1.2793093299323064,
+      "grad_norm": 3.875,
+      "learning_rate": 0.00012829520494352003,
+      "loss": 2.7775,
+      "step": 3260
+    },
+    {
+      "epoch": 1.283233591680565,
+      "grad_norm": 4.84375,
+      "learning_rate": 0.00012788853070151476,
+      "loss": 2.5604,
+      "step": 3270
+    },
+    {
+      "epoch": 1.2871578534288237,
+      "grad_norm": 4.0,
+      "learning_rate": 0.00012748135571111403,
+      "loss": 2.4264,
+      "step": 3280
+    },
+    {
+      "epoch": 1.2910821151770824,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.00012707368728328826,
+      "loss": 2.7452,
+      "step": 3290
+    },
+    {
+      "epoch": 1.295006376925341,
+      "grad_norm": 4.0,
+      "learning_rate": 0.0001266655327378677,
+      "loss": 2.2741,
+      "step": 3300
+    },
+    {
+      "epoch": 1.2989306386735995,
+      "grad_norm": 11.125,
+      "learning_rate": 0.000126256899403411,
+      "loss": 2.768,
+      "step": 3310
+    },
+    {
+      "epoch": 1.3028549004218581,
+      "grad_norm": 8.25,
+      "learning_rate": 0.00012584779461707375,
+      "loss": 2.6628,
+      "step": 3320
+    },
+    {
+      "epoch": 1.3067791621701168,
+      "grad_norm": 3.359375,
+      "learning_rate": 0.0001254382257244765,
+      "loss": 2.402,
+      "step": 3330
+    },
+    {
+      "epoch": 1.3107034239183752,
+      "grad_norm": 4.65625,
+      "learning_rate": 0.000125028200079573,
+      "loss": 2.7731,
+      "step": 3340
+    },
+    {
+      "epoch": 1.314627685666634,
+      "grad_norm": 2.34375,
+      "learning_rate": 0.0001246177250445182,
+      "loss": 2.2911,
+      "step": 3350
+    },
+    {
+      "epoch": 1.3185519474148926,
+      "grad_norm": 3.34375,
+      "learning_rate": 0.00012420680798953603,
+      "loss": 2.7716,
+      "step": 3360
+    },
+    {
+      "epoch": 1.3224762091631512,
+      "grad_norm": 3.1875,
+      "learning_rate": 0.00012379545629278692,
+      "loss": 2.6154,
+      "step": 3370
+    },
+    {
+      "epoch": 1.32640047091141,
+      "grad_norm": 2.671875,
+      "learning_rate": 0.0001233836773402356,
+      "loss": 2.3851,
+      "step": 3380
+    },
+    {
+      "epoch": 1.3303247326596683,
+      "grad_norm": 3.5625,
+      "learning_rate": 0.00012297147852551821,
+      "loss": 2.7638,
+      "step": 3390
+    },
+    {
+      "epoch": 1.334248994407927,
+      "grad_norm": 4.90625,
+      "learning_rate": 0.00012255886724980974,
+      "loss": 2.243,
+      "step": 3400
+    },
+    {
+      "epoch": 1.3381732561561857,
+      "grad_norm": 3.203125,
+      "learning_rate": 0.000122145850921691,
+      "loss": 2.6826,
+      "step": 3410
+    },
+    {
+      "epoch": 1.342097517904444,
+      "grad_norm": 3.890625,
+      "learning_rate": 0.00012173243695701576,
+      "loss": 2.5981,
+      "step": 3420
+    },
+    {
+      "epoch": 1.3460217796527028,
+      "grad_norm": 2.703125,
+      "learning_rate": 0.00012131863277877731,
+      "loss": 2.3953,
+      "step": 3430
+    },
+    {
+      "epoch": 1.3499460414009614,
+      "grad_norm": 3.375,
+      "learning_rate": 0.0001209044458169756,
+      "loss": 2.746,
+      "step": 3440
+    },
+    {
+      "epoch": 1.35387030314922,
+      "grad_norm": 5.15625,
+      "learning_rate": 0.00012048988350848338,
+      "loss": 2.2994,
+      "step": 3450
+    },
+    {
+      "epoch": 1.3577945648974787,
+      "grad_norm": 4.46875,
+      "learning_rate": 0.00012007495329691301,
+      "loss": 2.751,
+      "step": 3460
+    },
+    {
+      "epoch": 1.3617188266457374,
+      "grad_norm": 3.625,
+      "learning_rate": 0.00011965966263248267,
+      "loss": 2.6565,
+      "step": 3470
+    },
+    {
+      "epoch": 1.3656430883939958,
+      "grad_norm": 3.0,
+      "learning_rate": 0.00011924401897188248,
+      "loss": 2.3981,
+      "step": 3480
+    },
+    {
+      "epoch": 1.3695673501422545,
+      "grad_norm": 4.59375,
+      "learning_rate": 0.00011882802977814091,
+      "loss": 2.7372,
+      "step": 3490
+    },
+    {
+      "epoch": 1.3734916118905132,
+      "grad_norm": 2.3125,
+      "learning_rate": 0.0001184117025204905,
+      "loss": 2.1724,
+      "step": 3500
+    },
+    {
+      "epoch": 1.3774158736387716,
+      "grad_norm": 36.5,
+      "learning_rate": 0.0001179950446742338,
+      "loss": 2.7682,
+      "step": 3510
+    },
+    {
+      "epoch": 1.3813401353870303,
+      "grad_norm": 3.921875,
+      "learning_rate": 0.00011757806372060933,
+      "loss": 2.5904,
+      "step": 3520
+    },
+    {
+      "epoch": 1.385264397135289,
+      "grad_norm": 3.59375,
+      "learning_rate": 0.000117160767146657,
+      "loss": 2.3617,
+      "step": 3530
+    },
+    {
+      "epoch": 1.3891886588835476,
+      "grad_norm": 2.8125,
+      "learning_rate": 0.00011674316244508381,
+      "loss": 2.7573,
+      "step": 3540
+    },
+    {
+      "epoch": 1.3931129206318063,
+      "grad_norm": 2.546875,
+      "learning_rate": 0.00011632525711412937,
+      "loss": 2.2589,
+      "step": 3550
+    },
+    {
+      "epoch": 1.3970371823800647,
+      "grad_norm": 2.46875,
+      "learning_rate": 0.00011590705865743107,
+      "loss": 2.6838,
+      "step": 3560
+    },
+    {
+      "epoch": 1.4009614441283234,
+      "grad_norm": 3.359375,
+      "learning_rate": 0.00011548857458388967,
+      "loss": 2.5688,
+      "step": 3570
+    },
+    {
+      "epoch": 1.404885705876582,
+      "grad_norm": 2.453125,
+      "learning_rate": 0.00011506981240753405,
+      "loss": 2.3612,
+      "step": 3580
+    },
+    {
+      "epoch": 1.4088099676248405,
+      "grad_norm": 3.953125,
+      "learning_rate": 0.00011465077964738673,
+      "loss": 2.6975,
+      "step": 3590
+    },
+    {
+      "epoch": 1.4127342293730991,
+      "grad_norm": 4.84375,
+      "learning_rate": 0.00011423148382732853,
+      "loss": 2.2364,
+      "step": 3600
+    },
+    {
+      "epoch": 1.4166584911213578,
+      "grad_norm": 3.6875,
+      "learning_rate": 0.00011381193247596364,
+      "loss": 2.7031,
+      "step": 3610
+    },
+    {
+      "epoch": 1.4205827528696164,
+      "grad_norm": 3.421875,
+      "learning_rate": 0.0001133921331264844,
+      "loss": 2.5324,
+      "step": 3620
+    },
+    {
+      "epoch": 1.424507014617875,
+      "grad_norm": 9.4375,
+      "learning_rate": 0.00011297209331653606,
+      "loss": 2.3539,
+      "step": 3630
+    },
+    {
+      "epoch": 1.4284312763661335,
+      "grad_norm": 7.15625,
+      "learning_rate": 0.00011255182058808143,
+      "loss": 2.7526,
+      "step": 3640
+    },
+    {
+      "epoch": 1.4323555381143922,
+      "grad_norm": 3.609375,
+      "learning_rate": 0.00011213132248726541,
+      "loss": 2.2623,
+      "step": 3650
+    },
+    {
+      "epoch": 1.4362797998626509,
+      "grad_norm": 3.359375,
+      "learning_rate": 0.00011171060656427956,
+      "loss": 2.6897,
+      "step": 3660
+    },
+    {
+      "epoch": 1.4402040616109095,
+      "grad_norm": 3.78125,
+      "learning_rate": 0.00011128968037322653,
+      "loss": 2.6037,
+      "step": 3670
+    },
+    {
+      "epoch": 1.444128323359168,
+      "grad_norm": 3.046875,
+      "learning_rate": 0.00011086855147198441,
+      "loss": 2.3846,
+      "step": 3680
+    },
+    {
+      "epoch": 1.4480525851074266,
+      "grad_norm": 3.09375,
+      "learning_rate": 0.00011044722742207101,
+      "loss": 2.7154,
+      "step": 3690
+    },
+    {
+      "epoch": 1.4519768468556853,
+      "grad_norm": 2.5,
+      "learning_rate": 0.00011002571578850806,
+      "loss": 2.2069,
+      "step": 3700
+    },
+    {
+      "epoch": 1.455901108603944,
+      "grad_norm": 4.5625,
+      "learning_rate": 0.0001096040241396855,
+      "loss": 2.738,
+      "step": 3710
+    },
+    {
+      "epoch": 1.4598253703522026,
+      "grad_norm": 3.453125,
+      "learning_rate": 0.0001091821600472255,
+      "loss": 2.5221,
+      "step": 3720
+    },
+    {
+      "epoch": 1.463749632100461,
+      "grad_norm": 3.359375,
+      "learning_rate": 0.00010876013108584644,
+      "loss": 2.3419,
+      "step": 3730
+    },
+    {
+      "epoch": 1.4676738938487197,
+      "grad_norm": 3.09375,
+      "learning_rate": 0.00010833794483322699,
+      "loss": 2.7389,
+      "step": 3740
+    },
+    {
+      "epoch": 1.4715981555969784,
+      "grad_norm": 2.515625,
+      "learning_rate": 0.00010791560886987016,
+      "loss": 2.2295,
+      "step": 3750
+    },
+    {
+      "epoch": 1.4755224173452368,
+      "grad_norm": 3.390625,
+      "learning_rate": 0.00010749313077896697,
+      "loss": 2.7152,
+      "step": 3760
+    },
+    {
+      "epoch": 1.4794466790934955,
+      "grad_norm": 3.609375,
+      "learning_rate": 0.00010707051814626034,
+      "loss": 2.5429,
+      "step": 3770
+    },
+    {
+      "epoch": 1.4833709408417541,
+      "grad_norm": 2.65625,
+      "learning_rate": 0.00010664777855990909,
+      "loss": 2.3185,
+      "step": 3780
+    },
+    {
+      "epoch": 1.4872952025900128,
+      "grad_norm": 3.0,
+      "learning_rate": 0.00010622491961035139,
+      "loss": 2.7324,
+      "step": 3790
+    },
+    {
+      "epoch": 1.4912194643382715,
+      "grad_norm": 2.453125,
+      "learning_rate": 0.00010580194889016865,
+      "loss": 2.2062,
+      "step": 3800
+    },
+    {
+      "epoch": 1.49514372608653,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.00010537887399394925,
+      "loss": 2.68,
+      "step": 3810
+    },
+    {
+      "epoch": 1.4990679878347886,
+      "grad_norm": 4.8125,
+      "learning_rate": 0.00010495570251815204,
+      "loss": 2.5305,
+      "step": 3820
+    },
+    {
+      "epoch": 1.5029922495830472,
+      "grad_norm": 4.25,
+      "learning_rate": 0.00010453244206096992,
+      "loss": 2.3289,
+      "step": 3830
+    },
+    {
+      "epoch": 1.5069165113313057,
+      "grad_norm": 3.625,
+      "learning_rate": 0.00010410910022219356,
+      "loss": 2.6917,
+      "step": 3840
+    },
+    {
+      "epoch": 1.5108407730795643,
+      "grad_norm": 2.984375,
+      "learning_rate": 0.00010368568460307481,
+      "loss": 2.2382,
+      "step": 3850
+    },
+    {
+      "epoch": 1.514765034827823,
+      "grad_norm": 2.59375,
+      "learning_rate": 0.00010326220280619037,
+      "loss": 2.6929,
+      "step": 3860
+    },
+    {
+      "epoch": 1.5186892965760816,
+      "grad_norm": 3.34375,
+      "learning_rate": 0.00010283866243530505,
+      "loss": 2.5137,
+      "step": 3870
+    },
+    {
+      "epoch": 1.5226135583243403,
+      "grad_norm": 2.296875,
+      "learning_rate": 0.00010241507109523552,
+      "loss": 2.3452,
+      "step": 3880
+    },
+    {
+      "epoch": 1.526537820072599,
+      "grad_norm": 2.96875,
+      "learning_rate": 0.00010199143639171349,
+      "loss": 2.729,
+      "step": 3890
+    },
+    {
+      "epoch": 1.5304620818208574,
+      "grad_norm": 2.84375,
+      "learning_rate": 0.00010156776593124933,
+      "loss": 2.2073,
+      "step": 3900
+    },
+    {
+      "epoch": 1.534386343569116,
+      "grad_norm": 2.25,
+      "learning_rate": 0.00010114406732099548,
+      "loss": 2.6876,
+      "step": 3910
+    },
+    {
+      "epoch": 1.5383106053173745,
+      "grad_norm": 2.921875,
+      "learning_rate": 0.00010072034816860979,
+      "loss": 2.5714,
+      "step": 3920
+    },
+    {
+      "epoch": 1.5422348670656332,
+      "grad_norm": 2.359375,
+      "learning_rate": 0.00010029661608211885,
+      "loss": 2.3232,
+      "step": 3930
+    },
+    {
+      "epoch": 1.5461591288138918,
+      "grad_norm": 3.765625,
+      "learning_rate": 9.987287866978169e-05,
+      "loss": 2.6951,
+      "step": 3940
+    },
+    {
+      "epoch": 1.5500833905621505,
+      "grad_norm": 1.8515625,
+      "learning_rate": 9.944914353995277e-05,
+      "loss": 2.2313,
+      "step": 3950
+    },
+    {
+      "epoch": 1.5540076523104092,
+      "grad_norm": 3.75,
+      "learning_rate": 9.902541830094569e-05,
+      "loss": 2.6744,
+      "step": 3960
+    },
+    {
+      "epoch": 1.5579319140586678,
+      "grad_norm": 3.265625,
+      "learning_rate": 9.860171056089646e-05,
+      "loss": 2.4961,
+      "step": 3970
+    },
+    {
+      "epoch": 1.5618561758069265,
+      "grad_norm": 2.984375,
+      "learning_rate": 9.817802792762676e-05,
+      "loss": 2.31,
+      "step": 3980
+    },
+    {
+      "epoch": 1.565780437555185,
+      "grad_norm": 3.34375,
+      "learning_rate": 9.775437800850763e-05,
+      "loss": 2.6866,
+      "step": 3990
+    },
+    {
+      "epoch": 1.5697046993034436,
+      "grad_norm": 2.09375,
+      "learning_rate": 9.73307684103226e-05,
+      "loss": 2.2269,
+      "step": 4000
+    },
+    {
+      "epoch": 1.573628961051702,
+      "grad_norm": 2.421875,
+      "learning_rate": 9.690720673913133e-05,
+      "loss": 2.6199,
+      "step": 4010
+    },
+    {
+      "epoch": 1.5775532227999607,
+      "grad_norm": 4.34375,
+      "learning_rate": 9.648370060013279e-05,
+      "loss": 2.5303,
+      "step": 4020
+    },
+    {
+      "epoch": 1.5814774845482193,
+      "grad_norm": 2.578125,
+      "learning_rate": 9.606025759752895e-05,
+      "loss": 2.3054,
+      "step": 4030
+    },
+    {
+      "epoch": 1.585401746296478,
+      "grad_norm": 3.25,
+      "learning_rate": 9.563688533438819e-05,
+      "loss": 2.6912,
+      "step": 4040
+    },
+    {
+      "epoch": 1.5893260080447367,
+      "grad_norm": 2.484375,
+      "learning_rate": 9.521359141250859e-05,
+      "loss": 2.1854,
+      "step": 4050
+    },
+    {
+      "epoch": 1.5932502697929953,
+      "grad_norm": 2.96875,
+      "learning_rate": 9.479038343228173e-05,
+      "loss": 2.6575,
+      "step": 4060
+    },
+    {
+      "epoch": 1.5971745315412538,
+      "grad_norm": 3.21875,
+      "learning_rate": 9.436726899255597e-05,
+      "loss": 2.5612,
+      "step": 4070
+    },
+    {
+      "epoch": 1.6010987932895124,
+      "grad_norm": 2.375,
+      "learning_rate": 9.394425569050017e-05,
+      "loss": 2.3302,
+      "step": 4080
+    },
+    {
+      "epoch": 1.6050230550377709,
+      "grad_norm": 3.28125,
+      "learning_rate": 9.352135112146726e-05,
+      "loss": 2.6851,
+      "step": 4090
+    },
+    {
+      "epoch": 1.6089473167860295,
+      "grad_norm": 2.1875,
+      "learning_rate": 9.309856287885774e-05,
+      "loss": 2.2078,
+      "step": 4100
+    },
+    {
+      "epoch": 1.6128715785342882,
+      "grad_norm": 2.484375,
+      "learning_rate": 9.267589855398356e-05,
+      "loss": 2.6356,
+      "step": 4110
+    },
+    {
+      "epoch": 1.6167958402825469,
+      "grad_norm": 3.546875,
+      "learning_rate": 9.225336573593149e-05,
+      "loss": 2.5357,
+      "step": 4120
+    },
+    {
+      "epoch": 1.6207201020308055,
+      "grad_norm": 3.0,
+      "learning_rate": 9.183097201142722e-05,
+      "loss": 2.3012,
+      "step": 4130
+    },
+    {
+      "epoch": 1.6246443637790642,
+      "grad_norm": 2.828125,
+      "learning_rate": 9.14087249646989e-05,
+      "loss": 2.6688,
+      "step": 4140
+    },
+    {
+      "epoch": 1.6285686255273226,
+      "grad_norm": 2.15625,
+      "learning_rate": 9.098663217734102e-05,
+      "loss": 2.2073,
+      "step": 4150
+    },
+    {
+      "epoch": 1.6324928872755813,
+      "grad_norm": 2.625,
+      "learning_rate": 9.056470122817836e-05,
+      "loss": 2.6221,
+      "step": 4160
+    },
+    {
+      "epoch": 1.63641714902384,
+      "grad_norm": 3.96875,
+      "learning_rate": 9.01429396931297e-05,
+      "loss": 2.5187,
+      "step": 4170
+    },
+    {
+      "epoch": 1.6403414107720984,
+      "grad_norm": 2.234375,
+      "learning_rate": 8.972135514507211e-05,
+      "loss": 2.2868,
+      "step": 4180
+    },
+    {
+      "epoch": 1.644265672520357,
+      "grad_norm": 3.296875,
+      "learning_rate": 8.92999551537046e-05,
+      "loss": 2.705,
+      "step": 4190
+    },
+    {
+      "epoch": 1.6481899342686157,
+      "grad_norm": 2.125,
+      "learning_rate": 8.887874728541258e-05,
+      "loss": 2.1701,
+      "step": 4200
+    },
+    {
+      "epoch": 1.6521141960168744,
+      "grad_norm": 3.21875,
+      "learning_rate": 8.845773910313168e-05,
+      "loss": 2.6072,
+      "step": 4210
+    },
+    {
+      "epoch": 1.656038457765133,
+      "grad_norm": 13.0,
+      "learning_rate": 8.803693816621217e-05,
+      "loss": 2.4848,
+      "step": 4220
+    },
+    {
+      "epoch": 1.6599627195133917,
+      "grad_norm": 4.5,
+      "learning_rate": 8.761635203028318e-05,
+      "loss": 2.2841,
+      "step": 4230
+    },
+    {
+      "epoch": 1.6638869812616501,
+      "grad_norm": 3.6875,
+      "learning_rate": 8.719598824711693e-05,
+      "loss": 2.6995,
+      "step": 4240
+    },
+    {
+      "epoch": 1.6678112430099088,
+      "grad_norm": 2.4375,
+      "learning_rate": 8.677585436449333e-05,
+      "loss": 2.1557,
+      "step": 4250
+    },
+    {
+      "epoch": 1.6717355047581672,
+      "grad_norm": 3.34375,
+      "learning_rate": 8.635595792606419e-05,
+      "loss": 2.6367,
+      "step": 4260
+    },
+    {
+      "epoch": 1.675659766506426,
+      "grad_norm": 3.125,
+      "learning_rate": 8.593630647121809e-05,
+      "loss": 2.5122,
+      "step": 4270
+    },
+    {
+      "epoch": 1.6795840282546846,
+      "grad_norm": 2.59375,
+      "learning_rate": 8.551690753494476e-05,
+      "loss": 2.3082,
+      "step": 4280
+    },
+    {
+      "epoch": 1.6835082900029432,
+      "grad_norm": 3.125,
+      "learning_rate": 8.509776864769981e-05,
+      "loss": 2.6515,
+      "step": 4290
+    },
+    {
+      "epoch": 1.6874325517512019,
+      "grad_norm": 2.21875,
+      "learning_rate": 8.467889733526977e-05,
+      "loss": 2.1493,
+      "step": 4300
+    },
+    {
+      "epoch": 1.6913568134994605,
+      "grad_norm": 2.8125,
+      "learning_rate": 8.426030111863653e-05,
+      "loss": 2.675,
+      "step": 4310
+    },
+    {
+      "epoch": 1.695281075247719,
+      "grad_norm": 3.5625,
+      "learning_rate": 8.384198751384272e-05,
+      "loss": 2.5057,
+      "step": 4320
+    },
+    {
+      "epoch": 1.6992053369959776,
+      "grad_norm": 2.265625,
+      "learning_rate": 8.342396403185648e-05,
+      "loss": 2.2782,
+      "step": 4330
+    },
+    {
+      "epoch": 1.703129598744236,
+      "grad_norm": 2.78125,
+      "learning_rate": 8.300623817843673e-05,
+      "loss": 2.6237,
+      "step": 4340
+    },
+    {
+      "epoch": 1.7070538604924947,
+      "grad_norm": 1.859375,
+      "learning_rate": 8.258881745399837e-05,
+      "loss": 2.1529,
+      "step": 4350
+    },
+    {
+      "epoch": 1.7109781222407534,
+      "grad_norm": 2.546875,
+      "learning_rate": 8.217170935347756e-05,
+      "loss": 2.6025,
+      "step": 4360
+    },
+    {
+      "epoch": 1.714902383989012,
+      "grad_norm": 3.0,
+      "learning_rate": 8.175492136619729e-05,
+      "loss": 2.4563,
+      "step": 4370
+    },
+    {
+      "epoch": 1.7188266457372707,
+      "grad_norm": 2.5625,
+      "learning_rate": 8.133846097573263e-05,
+      "loss": 2.304,
+      "step": 4380
+    },
+    {
+      "epoch": 1.7227509074855294,
+      "grad_norm": 3.25,
+      "learning_rate": 8.092233565977669e-05,
+      "loss": 2.6518,
+      "step": 4390
+    },
+    {
+      "epoch": 1.726675169233788,
+      "grad_norm": 2.265625,
+      "learning_rate": 8.050655289000612e-05,
+      "loss": 2.1644,
+      "step": 4400
+    },
+    {
+      "epoch": 1.7305994309820465,
+      "grad_norm": 3.0,
+      "learning_rate": 8.009112013194706e-05,
+      "loss": 2.6419,
+      "step": 4410
+    },
+    {
+      "epoch": 1.7345236927303052,
+      "grad_norm": 3.1875,
+      "learning_rate": 7.96760448448411e-05,
+      "loss": 2.4603,
+      "step": 4420
+    },
+    {
+      "epoch": 1.7384479544785636,
+      "grad_norm": 4.84375,
+      "learning_rate": 7.926133448151121e-05,
+      "loss": 2.293,
+      "step": 4430
+    },
+    {
+      "epoch": 1.7423722162268223,
+      "grad_norm": 3.0625,
+      "learning_rate": 7.884699648822815e-05,
+      "loss": 2.6509,
+      "step": 4440
+    },
+    {
+      "epoch": 1.746296477975081,
+      "grad_norm": 2.03125,
+      "learning_rate": 7.843303830457655e-05,
+      "loss": 2.117,
+      "step": 4450
+    },
+    {
+      "epoch": 1.7502207397233396,
+      "grad_norm": 2.359375,
+      "learning_rate": 7.801946736332144e-05,
+      "loss": 2.6209,
+      "step": 4460
+    },
+    {
+      "epoch": 1.7541450014715982,
+      "grad_norm": 3.859375,
+      "learning_rate": 7.760629109027488e-05,
+      "loss": 2.46,
+      "step": 4470
+    },
+    {
+      "epoch": 1.758069263219857,
+      "grad_norm": 2.4375,
+      "learning_rate": 7.719351690416233e-05,
+      "loss": 2.3012,
+      "step": 4480
+    },
+    {
+      "epoch": 1.7619935249681153,
+      "grad_norm": 2.640625,
+      "learning_rate": 7.678115221648983e-05,
+      "loss": 2.6127,
+      "step": 4490
+    },
+    {
+      "epoch": 1.765917786716374,
+      "grad_norm": 2.09375,
+      "learning_rate": 7.636920443141057e-05,
+      "loss": 2.1809,
+      "step": 4500
+    },
+    {
+      "epoch": 1.7698420484646324,
+      "grad_norm": 2.484375,
+      "learning_rate": 7.595768094559226e-05,
+      "loss": 2.6198,
+      "step": 4510
+    },
+    {
+      "epoch": 1.773766310212891,
+      "grad_norm": 3.484375,
+      "learning_rate": 7.554658914808404e-05,
+      "loss": 2.4936,
+      "step": 4520
+    },
+    {
+      "epoch": 1.7776905719611498,
+      "grad_norm": 3.5625,
+      "learning_rate": 7.513593642018398e-05,
+      "loss": 2.3168,
+      "step": 4530
+    },
+    {
+      "epoch": 1.7816148337094084,
+      "grad_norm": 3.21875,
+      "learning_rate": 7.472573013530657e-05,
+      "loss": 2.6327,
+      "step": 4540
+    },
+    {
+      "epoch": 1.785539095457667,
+      "grad_norm": 2.4375,
+      "learning_rate": 7.431597765885013e-05,
+      "loss": 2.1957,
+      "step": 4550
+    },
+    {
+      "epoch": 1.7894633572059258,
+      "grad_norm": 3.109375,
+      "learning_rate": 7.39066863480648e-05,
+      "loss": 2.5922,
+      "step": 4560
+    },
+    {
+      "epoch": 1.7933876189541842,
+      "grad_norm": 4.625,
+      "learning_rate": 7.349786355192023e-05,
+      "loss": 2.4749,
+      "step": 4570
+    },
+    {
+      "epoch": 1.7973118807024429,
+      "grad_norm": 3.203125,
+      "learning_rate": 7.308951661097379e-05,
+      "loss": 2.2285,
+      "step": 4580
+    },
+    {
+      "epoch": 1.8012361424507015,
+      "grad_norm": 2.6875,
+      "learning_rate": 7.268165285723875e-05,
+      "loss": 2.6691,
+      "step": 4590
+    },
+    {
+      "epoch": 1.80516040419896,
+      "grad_norm": 2.171875,
+      "learning_rate": 7.227427961405245e-05,
+      "loss": 2.1628,
+      "step": 4600
+    },
+    {
+      "epoch": 1.8090846659472186,
+      "grad_norm": 2.59375,
+      "learning_rate": 7.186740419594505e-05,
+      "loss": 2.6104,
+      "step": 4610
+    },
+    {
+      "epoch": 1.8130089276954773,
+      "grad_norm": 3.171875,
+      "learning_rate": 7.1461033908508e-05,
+      "loss": 2.4951,
+      "step": 4620
+    },
+    {
+      "epoch": 1.816933189443736,
+      "grad_norm": 2.4375,
+      "learning_rate": 7.105517604826309e-05,
+      "loss": 2.3001,
+      "step": 4630
+    },
+    {
+      "epoch": 1.8208574511919946,
+      "grad_norm": 2.625,
+      "learning_rate": 7.064983790253108e-05,
+      "loss": 2.6023,
+      "step": 4640
+    },
+    {
+      "epoch": 1.8247817129402533,
+      "grad_norm": 2.265625,
+      "learning_rate": 7.024502674930132e-05,
+      "loss": 2.1788,
+      "step": 4650
+    },
+    {
+      "epoch": 1.8287059746885117,
+      "grad_norm": 2.234375,
+      "learning_rate": 6.984074985710067e-05,
+      "loss": 2.5881,
+      "step": 4660
+    },
+    {
+      "epoch": 1.8326302364367704,
+      "grad_norm": 4.125,
+      "learning_rate": 6.943701448486313e-05,
+      "loss": 2.5108,
+      "step": 4670
+    },
+    {
+      "epoch": 1.8365544981850288,
+      "grad_norm": 2.1875,
+      "learning_rate": 6.903382788179961e-05,
+      "loss": 2.3231,
+      "step": 4680
+    },
+    {
+      "epoch": 1.8404787599332875,
+      "grad_norm": 3.765625,
+      "learning_rate": 6.863119728726763e-05,
+      "loss": 2.6404,
+      "step": 4690
+    },
+    {
+      "epoch": 1.8444030216815461,
+      "grad_norm": 1.7890625,
+      "learning_rate": 6.82291299306414e-05,
+      "loss": 2.1706,
+      "step": 4700
+    },
+    {
+      "epoch": 1.8483272834298048,
+      "grad_norm": 2.21875,
+      "learning_rate": 6.782763303118193e-05,
+      "loss": 2.6509,
+      "step": 4710
+    },
+    {
+      "epoch": 1.8522515451780635,
+      "grad_norm": 3.265625,
+      "learning_rate": 6.742671379790756e-05,
+      "loss": 2.4585,
+      "step": 4720
+    },
+    {
+      "epoch": 1.8561758069263221,
+      "grad_norm": 2.546875,
+      "learning_rate": 6.702637942946442e-05,
+      "loss": 2.2775,
+      "step": 4730
+    },
+    {
+      "epoch": 1.8601000686745806,
+      "grad_norm": 3.640625,
+      "learning_rate": 6.662663711399705e-05,
+      "loss": 2.5993,
+      "step": 4740
+    },
+    {
+      "epoch": 1.8640243304228392,
+      "grad_norm": 1.859375,
+      "learning_rate": 6.622749402901971e-05,
+      "loss": 2.1516,
+      "step": 4750
+    },
+    {
+      "epoch": 1.8679485921710977,
+      "grad_norm": 3.359375,
+      "learning_rate": 6.582895734128702e-05,
+      "loss": 2.6089,
+      "step": 4760
+    },
+    {
+      "epoch": 1.8718728539193563,
+      "grad_norm": 3.296875,
+      "learning_rate": 6.543103420666569e-05,
+      "loss": 2.4808,
+      "step": 4770
+    },
+    {
+      "epoch": 1.875797115667615,
+      "grad_norm": 2.4375,
+      "learning_rate": 6.503373177000581e-05,
+      "loss": 2.2604,
+      "step": 4780
+    },
+    {
+      "epoch": 1.8797213774158736,
+      "grad_norm": 3.265625,
+      "learning_rate": 6.46370571650126e-05,
+      "loss": 2.6436,
+      "step": 4790
+    },
+    {
+      "epoch": 1.8836456391641323,
+      "grad_norm": 2.03125,
+      "learning_rate": 6.424101751411842e-05,
+      "loss": 2.1435,
+      "step": 4800
+    },
+    {
+      "epoch": 1.887569900912391,
+      "grad_norm": 3.140625,
+      "learning_rate": 6.384561992835467e-05,
+      "loss": 2.5906,
+      "step": 4810
+    },
+    {
+      "epoch": 1.8914941626606496,
+      "grad_norm": 3.0625,
+      "learning_rate": 6.34508715072244e-05,
+      "loss": 2.4619,
+      "step": 4820
+    },
+    {
+      "epoch": 1.895418424408908,
+      "grad_norm": 4.15625,
+      "learning_rate": 6.305677933857455e-05,
+      "loss": 2.278,
+      "step": 4830
+    },
+    {
+      "epoch": 1.8993426861571667,
+      "grad_norm": 2.8125,
+      "learning_rate": 6.266335049846885e-05,
+      "loss": 2.6018,
+      "step": 4840
+    },
+    {
+      "epoch": 1.9032669479054252,
+      "grad_norm": 1.8359375,
+      "learning_rate": 6.227059205106084e-05,
+      "loss": 2.1545,
+      "step": 4850
+    },
+    {
+      "epoch": 1.9071912096536838,
+      "grad_norm": 2.671875,
+      "learning_rate": 6.187851104846676e-05,
+      "loss": 2.6046,
+      "step": 4860
+    },
+    {
+      "epoch": 1.9111154714019425,
+      "grad_norm": 3.09375,
+      "learning_rate": 6.148711453063921e-05,
+      "loss": 2.4506,
+      "step": 4870
+    },
+    {
+      "epoch": 1.9150397331502012,
+      "grad_norm": 2.171875,
+      "learning_rate": 6.109640952524051e-05,
+      "loss": 2.2826,
+      "step": 4880
+    },
+    {
+      "epoch": 1.9189639948984598,
+      "grad_norm": 3.046875,
+      "learning_rate": 6.0706403047516766e-05,
+      "loss": 2.6344,
+      "step": 4890
+    },
+    {
+      "epoch": 1.9228882566467185,
+      "grad_norm": 1.890625,
+      "learning_rate": 6.0317102100171706e-05,
+      "loss": 2.1654,
+      "step": 4900
+    },
+    {
+      "epoch": 1.926812518394977,
+      "grad_norm": 2.84375,
+      "learning_rate": 5.992851367324097e-05,
+      "loss": 2.5765,
+      "step": 4910
+    },
+    {
+      "epoch": 1.9307367801432356,
+      "grad_norm": 3.71875,
+      "learning_rate": 5.954064474396674e-05,
+      "loss": 2.4731,
+      "step": 4920
+    },
+    {
+      "epoch": 1.934661041891494,
+      "grad_norm": 2.203125,
+      "learning_rate": 5.915350227667225e-05,
+      "loss": 2.2479,
+      "step": 4930
+    },
+    {
+      "epoch": 1.9385853036397527,
+      "grad_norm": 3.484375,
+      "learning_rate": 5.876709322263696e-05,
+      "loss": 2.6404,
+      "step": 4940
+    },
+    {
+      "epoch": 1.9425095653880113,
+      "grad_norm": 1.8671875,
+      "learning_rate": 5.838142451997155e-05,
+      "loss": 2.1397,
+      "step": 4950
+    },
+    {
+      "epoch": 1.94643382713627,
+      "grad_norm": 3.15625,
+      "learning_rate": 5.799650309349348e-05,
+      "loss": 2.6034,
+      "step": 4960
+    },
+    {
+      "epoch": 1.9503580888845287,
+      "grad_norm": 3.40625,
+      "learning_rate": 5.761233585460265e-05,
+      "loss": 2.4958,
+      "step": 4970
+    },
+    {
+      "epoch": 1.9542823506327873,
+      "grad_norm": 2.53125,
+      "learning_rate": 5.722892970115712e-05,
+      "loss": 2.2344,
+      "step": 4980
+    },
+    {
+      "epoch": 1.9582066123810458,
+      "grad_norm": 2.875,
+      "learning_rate": 5.684629151734948e-05,
+      "loss": 2.634,
+      "step": 4990
+    },
+    {
+      "epoch": 1.9621308741293044,
+      "grad_norm": 1.8828125,
+      "learning_rate": 5.646442817358317e-05,
+      "loss": 2.1363,
+      "step": 5000
+    },
+    {
+      "epoch": 1.966055135877563,
+      "grad_norm": 2.9375,
+      "learning_rate": 5.608334652634913e-05,
+      "loss": 2.568,
+      "step": 5010
+    },
+    {
+      "epoch": 1.9699793976258215,
+      "grad_norm": 3.96875,
+      "learning_rate": 5.5703053418102514e-05,
+      "loss": 2.4782,
+      "step": 5020
+    },
+    {
+      "epoch": 1.9739036593740802,
+      "grad_norm": 2.59375,
+      "learning_rate": 5.5323555677140124e-05,
+      "loss": 2.2704,
+      "step": 5030
+    },
+    {
+      "epoch": 1.9778279211223389,
+      "grad_norm": 3.875,
+      "learning_rate": 5.4944860117477615e-05,
+      "loss": 2.5752,
+      "step": 5040
+    },
+    {
+      "epoch": 1.9817521828705975,
+      "grad_norm": 2.140625,
+      "learning_rate": 5.456697353872722e-05,
+      "loss": 2.1329,
+      "step": 5050
+    },
+    {
+      "epoch": 1.9856764446188562,
+      "grad_norm": 2.71875,
+      "learning_rate": 5.418990272597561e-05,
+      "loss": 2.5846,
+      "step": 5060
+    },
+    {
+      "epoch": 1.9896007063671148,
+      "grad_norm": 3.6875,
+      "learning_rate": 5.381365444966204e-05,
+      "loss": 2.444,
+      "step": 5070
+    },
+    {
+      "epoch": 1.9935249681153733,
+      "grad_norm": 2.90625,
+      "learning_rate": 5.343823546545692e-05,
+      "loss": 2.242,
+      "step": 5080
+    },
+    {
+      "epoch": 1.997449229863632,
+      "grad_norm": 3.828125,
+      "learning_rate": 5.306365251414043e-05,
+      "loss": 2.5794,
+      "step": 5090
+    },
+    {
+      "epoch": 1.999803786912587,
+      "eval_loss": 2.4651312828063965,
+      "eval_runtime": 156.2939,
+      "eval_samples_per_second": 159.955,
+      "eval_steps_per_second": 39.989,
+      "step": 5096
+    },
+    {
+      "epoch": 2.0013734916118904,
+      "grad_norm": 2.140625,
+      "learning_rate": 5.268991232148137e-05,
+      "loss": 2.104,
+      "step": 5100
+    },
+    {
+      "epoch": 2.005297753360149,
+      "grad_norm": 7.8125,
+      "learning_rate": 5.2317021598116634e-05,
+      "loss": 2.6752,
+      "step": 5110
+    },
+    {
+      "epoch": 2.0092220151084077,
+      "grad_norm": 4.78125,
+      "learning_rate": 5.194498703943054e-05,
+      "loss": 2.2376,
+      "step": 5120
+    },
+    {
+      "epoch": 2.0131462768566664,
+      "grad_norm": 2.46875,
+      "learning_rate": 5.1573815325434726e-05,
+      "loss": 2.3855,
+      "step": 5130
+    },
+    {
+      "epoch": 2.017070538604925,
+      "grad_norm": 6.40625,
+      "learning_rate": 5.120351312064802e-05,
+      "loss": 2.5248,
+      "step": 5140
+    },
+    {
+      "epoch": 2.0209948003531837,
+      "grad_norm": 2.171875,
+      "learning_rate": 5.083408707397703e-05,
+      "loss": 2.0569,
+      "step": 5150
+    },
+    {
+      "epoch": 2.0249190621014423,
+      "grad_norm": 2.875,
+      "learning_rate": 5.0465543818596626e-05,
+      "loss": 2.6757,
+      "step": 5160
+    },
+    {
+      "epoch": 2.0288433238497006,
+      "grad_norm": 3.578125,
+      "learning_rate": 5.009788997183073e-05,
+      "loss": 2.2656,
+      "step": 5170
+    },
+    {
+      "epoch": 2.0327675855979592,
+      "grad_norm": 2.9375,
+      "learning_rate": 4.9731132135033786e-05,
+      "loss": 2.3929,
+      "step": 5180
+    },
+    {
+      "epoch": 2.036691847346218,
+      "grad_norm": 3.640625,
+      "learning_rate": 4.936527689347194e-05,
+      "loss": 2.5691,
+      "step": 5190
+    },
+    {
+      "epoch": 2.0406161090944765,
+      "grad_norm": 2.546875,
+      "learning_rate": 4.9000330816204995e-05,
+      "loss": 2.0556,
+      "step": 5200
+    },
+    {
+      "epoch": 2.044540370842735,
+      "grad_norm": 3.015625,
+      "learning_rate": 4.863630045596837e-05,
+      "loss": 2.6262,
+      "step": 5210
+    },
+    {
+      "epoch": 2.048464632590994,
+      "grad_norm": 3.921875,
+      "learning_rate": 4.827319234905541e-05,
+      "loss": 2.2537,
+      "step": 5220
+    },
+    {
+      "epoch": 2.0523888943392525,
+      "grad_norm": 2.765625,
+      "learning_rate": 4.791101301520016e-05,
+      "loss": 2.3967,
+      "step": 5230
+    },
+    {
+      "epoch": 2.056313156087511,
+      "grad_norm": 3.640625,
+      "learning_rate": 4.754976895746007e-05,
+      "loss": 2.5324,
+      "step": 5240
+    },
+    {
+      "epoch": 2.06023741783577,
+      "grad_norm": 2.25,
+      "learning_rate": 4.7189466662099654e-05,
+      "loss": 2.1252,
+      "step": 5250
+    },
+    {
+      "epoch": 2.064161679584028,
+      "grad_norm": 3.4375,
+      "learning_rate": 4.683011259847345e-05,
+      "loss": 2.6362,
+      "step": 5260
+    },
+    {
+      "epoch": 2.0680859413322867,
+      "grad_norm": 3.953125,
+      "learning_rate": 4.647171321891034e-05,
+      "loss": 2.2765,
+      "step": 5270
+    },
+    {
+      "epoch": 2.0720102030805454,
+      "grad_norm": 3.09375,
+      "learning_rate": 4.61142749585975e-05,
+      "loss": 2.3645,
+      "step": 5280
+    },
+    {
+      "epoch": 2.075934464828804,
+      "grad_norm": 4.09375,
+      "learning_rate": 4.575780423546476e-05,
+      "loss": 2.5522,
+      "step": 5290
+    },
+    {
+      "epoch": 2.0798587265770627,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.540230745006961e-05,
+      "loss": 2.0538,
+      "step": 5300
+    },
+    {
+      "epoch": 2.0837829883253214,
+      "grad_norm": 3.03125,
+      "learning_rate": 4.504779098548209e-05,
+      "loss": 2.6162,
+      "step": 5310
+    },
+    {
+      "epoch": 2.08770725007358,
+      "grad_norm": 4.34375,
+      "learning_rate": 4.469426120717025e-05,
+      "loss": 2.2484,
+      "step": 5320
+    },
+    {
+      "epoch": 2.0916315118218387,
+      "grad_norm": 2.484375,
+      "learning_rate": 4.4341724462885794e-05,
+      "loss": 2.3575,
+      "step": 5330
+    },
+    {
+      "epoch": 2.095555773570097,
+      "grad_norm": 3.546875,
+      "learning_rate": 4.399018708255018e-05,
+      "loss": 2.5575,
+      "step": 5340
+    },
+    {
+      "epoch": 2.0994800353183556,
+      "grad_norm": 1.7734375,
+      "learning_rate": 4.363965537814102e-05,
+      "loss": 2.069,
+      "step": 5350
+    },
+    {
+      "epoch": 2.1034042970666142,
+      "grad_norm": 3.703125,
+      "learning_rate": 4.3290135643578476e-05,
+      "loss": 2.6762,
+      "step": 5360
+    },
+    {
+      "epoch": 2.107328558814873,
+      "grad_norm": 4.375,
+      "learning_rate": 4.294163415461258e-05,
+      "loss": 2.2998,
+      "step": 5370
+    },
+    {
+      "epoch": 2.1112528205631316,
+      "grad_norm": 2.65625,
+      "learning_rate": 4.259415716871037e-05,
+      "loss": 2.3821,
+      "step": 5380
+    },
+    {
+      "epoch": 2.1151770823113902,
+      "grad_norm": 3.375,
+      "learning_rate": 4.224771092494355e-05,
+      "loss": 2.5521,
+      "step": 5390
+    },
+    {
+      "epoch": 2.119101344059649,
+      "grad_norm": 1.8046875,
+      "learning_rate": 4.190230164387655e-05,
+      "loss": 2.0596,
+      "step": 5400
+    },
+    {
+      "epoch": 2.1230256058079076,
+      "grad_norm": 10.0,
+      "learning_rate": 4.155793552745465e-05,
+      "loss": 2.6444,
+      "step": 5410
+    },
+    {
+      "epoch": 2.126949867556166,
+      "grad_norm": 5.21875,
+      "learning_rate": 4.1214618758892866e-05,
+      "loss": 2.2774,
+      "step": 5420
+    },
+    {
+      "epoch": 2.1308741293044244,
+      "grad_norm": 2.421875,
+      "learning_rate": 4.0872357502564686e-05,
+      "loss": 2.4077,
+      "step": 5430
+    },
+    {
+      "epoch": 2.134798391052683,
+      "grad_norm": 4.8125,
+      "learning_rate": 4.053115790389159e-05,
+      "loss": 2.5567,
+      "step": 5440
+    },
+    {
+      "epoch": 2.1387226528009418,
+      "grad_norm": 2.3125,
+      "learning_rate": 4.019102608923262e-05,
+      "loss": 2.0606,
+      "step": 5450
+    },
+    {
+      "epoch": 2.1426469145492004,
+      "grad_norm": 3.09375,
+      "learning_rate": 3.985196816577433e-05,
+      "loss": 2.6365,
+      "step": 5460
+    },
+    {
+      "epoch": 2.146571176297459,
+      "grad_norm": 4.3125,
+      "learning_rate": 3.951399022142127e-05,
+      "loss": 2.2485,
+      "step": 5470
+    },
+    {
+      "epoch": 2.1504954380457177,
+      "grad_norm": 2.578125,
+      "learning_rate": 3.9177098324686413e-05,
+      "loss": 2.4047,
+      "step": 5480
+    },
+    {
+      "epoch": 2.1544196997939764,
+      "grad_norm": 3.8125,
+      "learning_rate": 3.884129852458254e-05,
+      "loss": 2.5754,
+      "step": 5490
+    },
+    {
+      "epoch": 2.158343961542235,
+      "grad_norm": 2.015625,
+      "learning_rate": 3.8506596850513355e-05,
+      "loss": 2.1057,
+      "step": 5500
+    },
+    {
+      "epoch": 2.1622682232904933,
+      "grad_norm": 2.703125,
+      "learning_rate": 3.817299931216537e-05,
+      "loss": 2.6122,
+      "step": 5510
+    },
+    {
+      "epoch": 2.166192485038752,
+      "grad_norm": 6.53125,
+      "learning_rate": 3.7840511899399954e-05,
+      "loss": 2.2762,
+      "step": 5520
+    },
+    {
+      "epoch": 2.1701167467870106,
+      "grad_norm": 2.46875,
+      "learning_rate": 3.7509140582145706e-05,
+      "loss": 2.3975,
+      "step": 5530
+    },
+    {
+      "epoch": 2.1740410085352693,
+      "grad_norm": 3.421875,
+      "learning_rate": 3.7178891310291444e-05,
+      "loss": 2.5686,
+      "step": 5540
+    },
+    {
+      "epoch": 2.177965270283528,
+      "grad_norm": 1.796875,
+      "learning_rate": 3.684977001357913e-05,
+      "loss": 2.1248,
+      "step": 5550
+    },
+    {
+      "epoch": 2.1818895320317866,
+      "grad_norm": 3.671875,
+      "learning_rate": 3.6521782601497677e-05,
+      "loss": 2.6592,
+      "step": 5560
+    },
+    {
+      "epoch": 2.1858137937800453,
+      "grad_norm": 4.4375,
+      "learning_rate": 3.619493496317662e-05,
+      "loss": 2.2166,
+      "step": 5570
+    },
+    {
+      "epoch": 2.189738055528304,
+      "grad_norm": 3.84375,
+      "learning_rate": 3.586923296728046e-05,
+      "loss": 2.3321,
+      "step": 5580
+    },
+    {
+      "epoch": 2.193662317276562,
+      "grad_norm": 4.15625,
+      "learning_rate": 3.5544682461903366e-05,
+      "loss": 2.5543,
+      "step": 5590
+    },
+    {
+      "epoch": 2.197586579024821,
+      "grad_norm": 1.890625,
+      "learning_rate": 3.522128927446392e-05,
+      "loss": 2.0798,
+      "step": 5600
+    },
+    {
+      "epoch": 2.2015108407730795,
+      "grad_norm": 3.578125,
+      "learning_rate": 3.4899059211600825e-05,
+      "loss": 2.6359,
+      "step": 5610
+    },
+    {
+      "epoch": 2.205435102521338,
+      "grad_norm": 4.125,
+      "learning_rate": 3.457799805906835e-05,
+      "loss": 2.2062,
+      "step": 5620
+    },
+    {
+      "epoch": 2.209359364269597,
+      "grad_norm": 2.734375,
+      "learning_rate": 3.4258111581632636e-05,
+      "loss": 2.4114,
+      "step": 5630
+    },
+    {
+      "epoch": 2.2132836260178554,
+      "grad_norm": 3.75,
+      "learning_rate": 3.3939405522968105e-05,
+      "loss": 2.5709,
+      "step": 5640
+    },
+    {
+      "epoch": 2.217207887766114,
+      "grad_norm": 2.03125,
+      "learning_rate": 3.362188560555434e-05,
+      "loss": 2.0567,
+      "step": 5650
+    },
+    {
+      "epoch": 2.2211321495143728,
+      "grad_norm": 3.15625,
+      "learning_rate": 3.330555753057336e-05,
+      "loss": 2.6326,
+      "step": 5660
+    },
+    {
+      "epoch": 2.2250564112626314,
+      "grad_norm": 4.375,
+      "learning_rate": 3.2990426977807156e-05,
+      "loss": 2.2731,
+      "step": 5670
+    },
+    {
+      "epoch": 2.2289806730108896,
+      "grad_norm": 2.90625,
+      "learning_rate": 3.2676499605535914e-05,
+      "loss": 2.3522,
+      "step": 5680
+    },
+    {
+      "epoch": 2.2329049347591483,
+      "grad_norm": 3.46875,
+      "learning_rate": 3.23637810504361e-05,
+      "loss": 2.5525,
+      "step": 5690
+    },
+    {
+      "epoch": 2.236829196507407,
+      "grad_norm": 2.265625,
+      "learning_rate": 3.205227692747968e-05,
+      "loss": 2.0494,
+      "step": 5700
+    },
+    {
+      "epoch": 2.2407534582556656,
+      "grad_norm": 2.953125,
+      "learning_rate": 3.174199282983292e-05,
+      "loss": 2.64,
+      "step": 5710
+    },
+    {
+      "epoch": 2.2446777200039243,
+      "grad_norm": 4.09375,
+      "learning_rate": 3.143293432875607e-05,
+      "loss": 2.2469,
+      "step": 5720
+    },
+    {
+      "epoch": 2.248601981752183,
+      "grad_norm": 2.828125,
+      "learning_rate": 3.1125106973503485e-05,
+      "loss": 2.4091,
+      "step": 5730
+    },
+    {
+      "epoch": 2.2525262435004416,
+      "grad_norm": 3.375,
+      "learning_rate": 3.081851629122372e-05,
+      "loss": 2.592,
+      "step": 5740
+    },
+    {
+      "epoch": 2.2564505052487003,
+      "grad_norm": 2.375,
+      "learning_rate": 3.051316778686055e-05,
+      "loss": 2.0698,
+      "step": 5750
+    },
+    {
+      "epoch": 2.260374766996959,
+      "grad_norm": 3.453125,
+      "learning_rate": 3.0209066943053943e-05,
+      "loss": 2.661,
+      "step": 5760
+    },
+    {
+      "epoch": 2.264299028745217,
+      "grad_norm": 4.5,
+      "learning_rate": 2.990621922004172e-05,
+      "loss": 2.2046,
+      "step": 5770
+    },
+    {
+      "epoch": 2.268223290493476,
+      "grad_norm": 2.75,
+      "learning_rate": 2.960463005556149e-05,
+      "loss": 2.4104,
+      "step": 5780
+    },
+    {
+      "epoch": 2.2721475522417345,
+      "grad_norm": 3.828125,
+      "learning_rate": 2.9304304864752884e-05,
+      "loss": 2.5661,
+      "step": 5790
+    },
+    {
+      "epoch": 2.276071813989993,
+      "grad_norm": 2.109375,
+      "learning_rate": 2.9005249040060612e-05,
+      "loss": 2.0535,
+      "step": 5800
+    },
+    {
+      "epoch": 2.279996075738252,
+      "grad_norm": 3.03125,
+      "learning_rate": 2.87074679511373e-05,
+      "loss": 2.6028,
+      "step": 5810
+    },
+    {
+      "epoch": 2.2839203374865105,
+      "grad_norm": 4.34375,
+      "learning_rate": 2.8410966944747376e-05,
+      "loss": 2.2238,
+      "step": 5820
+    },
+    {
+      "epoch": 2.287844599234769,
+      "grad_norm": 2.8125,
+      "learning_rate": 2.811575134467086e-05,
+      "loss": 2.3803,
+      "step": 5830
+    },
+    {
+      "epoch": 2.2917688609830273,
+      "grad_norm": 3.71875,
+      "learning_rate": 2.782182645160789e-05,
+      "loss": 2.555,
+      "step": 5840
+    },
+    {
+      "epoch": 2.295693122731286,
+      "grad_norm": 2.21875,
+      "learning_rate": 2.7529197543083508e-05,
+      "loss": 2.0639,
+      "step": 5850
+    },
+    {
+      "epoch": 2.2996173844795447,
+      "grad_norm": 3.359375,
+      "learning_rate": 2.7237869873352827e-05,
+      "loss": 2.6308,
+      "step": 5860
+    },
+    {
+      "epoch": 2.3035416462278033,
+      "grad_norm": 4.1875,
+      "learning_rate": 2.6947848673306853e-05,
+      "loss": 2.2897,
+      "step": 5870
+    },
+    {
+      "epoch": 2.307465907976062,
+      "grad_norm": 2.578125,
+      "learning_rate": 2.6659139150378377e-05,
+      "loss": 2.3666,
+      "step": 5880
+    },
+    {
+      "epoch": 2.3113901697243207,
+      "grad_norm": 4.65625,
+      "learning_rate": 2.6371746488448613e-05,
+      "loss": 2.5263,
+      "step": 5890
+    },
+    {
+      "epoch": 2.3153144314725793,
+      "grad_norm": 2.015625,
+      "learning_rate": 2.6085675847754153e-05,
+      "loss": 2.0356,
+      "step": 5900
+    },
+    {
+      "epoch": 2.319238693220838,
+      "grad_norm": 3.015625,
+      "learning_rate": 2.580093236479406e-05,
+      "loss": 2.6162,
+      "step": 5910
+    },
+    {
+      "epoch": 2.3231629549690966,
+      "grad_norm": 4.4375,
+      "learning_rate": 2.5517521152237965e-05,
+      "loss": 2.2341,
+      "step": 5920
+    },
+    {
+      "epoch": 2.327087216717355,
+      "grad_norm": 3.015625,
+      "learning_rate": 2.5235447298834004e-05,
+      "loss": 2.3921,
+      "step": 5930
+    },
+    {
+      "epoch": 2.3310114784656135,
+      "grad_norm": 3.8125,
+      "learning_rate": 2.4954715869317603e-05,
+      "loss": 2.5433,
+      "step": 5940
+    },
+    {
+      "epoch": 2.334935740213872,
+      "grad_norm": 1.8515625,
+      "learning_rate": 2.4675331904320532e-05,
+      "loss": 2.0282,
+      "step": 5950
+    },
+    {
+      "epoch": 2.338860001962131,
+      "grad_norm": 3.40625,
+      "learning_rate": 2.43973004202803e-05,
+      "loss": 2.6386,
+      "step": 5960
+    },
+    {
+      "epoch": 2.3427842637103895,
+      "grad_norm": 4.03125,
+      "learning_rate": 2.4120626409350212e-05,
+      "loss": 2.235,
+      "step": 5970
+    },
+    {
+      "epoch": 2.346708525458648,
+      "grad_norm": 2.65625,
+      "learning_rate": 2.3845314839309564e-05,
+      "loss": 2.4052,
+      "step": 5980
+    },
+    {
+      "epoch": 2.350632787206907,
+      "grad_norm": 3.609375,
+      "learning_rate": 2.3571370653474657e-05,
+      "loss": 2.5666,
+      "step": 5990
+    },
+    {
+      "epoch": 2.3545570489551655,
+      "grad_norm": 2.203125,
+      "learning_rate": 2.329879877060981e-05,
+      "loss": 2.0736,
+      "step": 6000
+    },
+    {
+      "epoch": 2.358481310703424,
+      "grad_norm": 4.1875,
+      "learning_rate": 2.302760408483926e-05,
+      "loss": 2.6477,
+      "step": 6010
+    },
+    {
+      "epoch": 2.3624055724516824,
+      "grad_norm": 5.125,
+      "learning_rate": 2.2757791465559153e-05,
+      "loss": 2.2439,
+      "step": 6020
+    },
+    {
+      "epoch": 2.366329834199941,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.2489365757350135e-05,
+      "loss": 2.3662,
+      "step": 6030
+    },
+    {
+      "epoch": 2.3702540959481997,
+      "grad_norm": 4.03125,
+      "learning_rate": 2.2222331779890394e-05,
+      "loss": 2.569,
+      "step": 6040
+    },
+    {
+      "epoch": 2.3741783576964584,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2.1956694327869042e-05,
+      "loss": 2.0887,
+      "step": 6050
+    },
+    {
+      "epoch": 2.378102619444717,
+      "grad_norm": 3.328125,
+      "learning_rate": 2.1692458170900198e-05,
+      "loss": 2.6021,
+      "step": 6060
+    },
+    {
+      "epoch": 2.3820268811929757,
+      "grad_norm": 5.25,
+      "learning_rate": 2.1429628053437078e-05,
+      "loss": 2.2236,
+      "step": 6070
+    },
+    {
+      "epoch": 2.3859511429412343,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.116820869468711e-05,
+      "loss": 2.411,
+      "step": 6080
+    },
+    {
+      "epoch": 2.3898754046894926,
+      "grad_norm": 4.21875,
+      "learning_rate": 2.0908204788526964e-05,
+      "loss": 2.6153,
+      "step": 6090
+    },
+    {
+      "epoch": 2.393799666437751,
+      "grad_norm": 2.09375,
+      "learning_rate": 2.064962100341842e-05,
+      "loss": 2.0543,
+      "step": 6100
+    },
+    {
+      "epoch": 2.39772392818601,
+      "grad_norm": 3.59375,
+      "learning_rate": 2.0392461982324463e-05,
+      "loss": 2.5924,
+      "step": 6110
+    },
+    {
+      "epoch": 2.4016481899342685,
+      "grad_norm": 4.5,
+      "learning_rate": 2.0136732342625875e-05,
+      "loss": 2.2227,
+      "step": 6120
+    },
+    {
+      "epoch": 2.405572451682527,
+      "grad_norm": 2.515625,
+      "learning_rate": 1.9882436676038472e-05,
+      "loss": 2.3773,
+      "step": 6130
+    },
+    {
+      "epoch": 2.409496713430786,
+      "grad_norm": 3.25,
+      "learning_rate": 1.9629579548530553e-05,
+      "loss": 2.5616,
+      "step": 6140
+    },
+    {
+      "epoch": 2.4134209751790445,
+      "grad_norm": 2.5,
+      "learning_rate": 1.9378165500240943e-05,
+      "loss": 2.1299,
+      "step": 6150
+    },
+    {
+      "epoch": 2.417345236927303,
+      "grad_norm": 3.75,
+      "learning_rate": 1.9128199045397488e-05,
+      "loss": 2.6251,
+      "step": 6160
+    },
+    {
+      "epoch": 2.421269498675562,
+      "grad_norm": 4.4375,
+      "learning_rate": 1.887968467223591e-05,
+      "loss": 2.2619,
+      "step": 6170
+    },
+    {
+      "epoch": 2.42519376042382,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.86326268429194e-05,
+      "loss": 2.3959,
+      "step": 6180
+    },
+    {
+      "epoch": 2.4291180221720787,
+      "grad_norm": 3.765625,
+      "learning_rate": 1.838702999345827e-05,
+      "loss": 2.5841,
+      "step": 6190
+    },
+    {
+      "epoch": 2.4330422839203374,
+      "grad_norm": 1.8359375,
+      "learning_rate": 1.8142898533630536e-05,
+      "loss": 2.084,
+      "step": 6200
+    },
+    {
+      "epoch": 2.436966545668596,
+      "grad_norm": 3.3125,
+      "learning_rate": 1.7900236846902572e-05,
+      "loss": 2.6167,
+      "step": 6210
+    },
+    {
+      "epoch": 2.4408908074168547,
+      "grad_norm": 4.65625,
+      "learning_rate": 1.765904929035046e-05,
+      "loss": 2.2582,
+      "step": 6220
+    },
+    {
+      "epoch": 2.4448150691651134,
+      "grad_norm": 2.546875,
+      "learning_rate": 1.74193401945818e-05,
+      "loss": 2.3975,
+      "step": 6230
+    },
+    {
+      "epoch": 2.448739330913372,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.7181113863657805e-05,
+      "loss": 2.5523,
+      "step": 6240
+    },
+    {
+      "epoch": 2.4526635926616307,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.6944374575016253e-05,
+      "loss": 2.1125,
+      "step": 6250
+    },
+    {
+      "epoch": 2.4565878544098894,
+      "grad_norm": 2.953125,
+      "learning_rate": 1.670912657939443e-05,
+      "loss": 2.6167,
+      "step": 6260
+    },
+    {
+      "epoch": 2.4605121161581476,
+      "grad_norm": 4.15625,
+      "learning_rate": 1.6475374100753018e-05,
+      "loss": 2.224,
+      "step": 6270
+    },
+    {
+      "epoch": 2.4644363779064062,
+      "grad_norm": 2.515625,
+      "learning_rate": 1.6243121336200127e-05,
+      "loss": 2.4065,
+      "step": 6280
+    },
+    {
+      "epoch": 2.468360639654665,
+      "grad_norm": 4.03125,
+      "learning_rate": 1.601237245591599e-05,
+      "loss": 2.5741,
+      "step": 6290
+    },
+    {
+      "epoch": 2.4722849014029236,
+      "grad_norm": 2.125,
+      "learning_rate": 1.5783131603078083e-05,
+      "loss": 2.1104,
+      "step": 6300
+    },
+    {
+      "epoch": 2.4762091631511822,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.555540289378663e-05,
+      "loss": 2.6218,
+      "step": 6310
+    },
+    {
+      "epoch": 2.480133424899441,
+      "grad_norm": 4.875,
+      "learning_rate": 1.532919041699089e-05,
+      "loss": 2.2685,
+      "step": 6320
+    },
+    {
+      "epoch": 2.4840576866476995,
+      "grad_norm": 2.609375,
+      "learning_rate": 1.510449823441561e-05,
+      "loss": 2.4088,
+      "step": 6330
+    },
+    {
+      "epoch": 2.4879819483959578,
+      "grad_norm": 3.796875,
+      "learning_rate": 1.4881330380488013e-05,
+      "loss": 2.5315,
+      "step": 6340
+    },
+    {
+      "epoch": 2.4919062101442164,
+      "grad_norm": 2.0,
+      "learning_rate": 1.4659690862265673e-05,
+      "loss": 2.0717,
+      "step": 6350
+    },
+    {
+      "epoch": 2.495830471892475,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.4439583659364153e-05,
+      "loss": 2.6015,
+      "step": 6360
+    },
+    {
+      "epoch": 2.4997547336407338,
+      "grad_norm": 4.375,
+      "learning_rate": 1.4221012723885874e-05,
+      "loss": 2.242,
+      "step": 6370
+    },
+    {
+      "epoch": 2.5036789953889924,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.4003981980348969e-05,
+      "loss": 2.404,
+      "step": 6380
+    },
+    {
+      "epoch": 2.507603257137251,
+      "grad_norm": 3.8125,
+      "learning_rate": 1.3788495325616912e-05,
+      "loss": 2.577,
+      "step": 6390
+    },
+    {
+      "epoch": 2.5115275188855097,
+      "grad_norm": 1.9140625,
+      "learning_rate": 1.357455662882855e-05,
+      "loss": 2.064,
+      "step": 6400
+    },
+    {
+      "epoch": 2.5154517806337684,
+      "grad_norm": 3.78125,
+      "learning_rate": 1.3362169731328534e-05,
+      "loss": 2.6141,
+      "step": 6410
+    },
+    {
+      "epoch": 2.519376042382027,
+      "grad_norm": 4.375,
+      "learning_rate": 1.3151338446598482e-05,
+      "loss": 2.2446,
+      "step": 6420
+    },
+    {
+      "epoch": 2.5233003041302853,
+      "grad_norm": 2.953125,
+      "learning_rate": 1.2942066560188348e-05,
+      "loss": 2.3844,
+      "step": 6430
+    },
+    {
+      "epoch": 2.527224565878544,
+      "grad_norm": 4.21875,
+      "learning_rate": 1.2734357829648625e-05,
+      "loss": 2.5215,
+      "step": 6440
+    },
+    {
+      "epoch": 2.5311488276268026,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.2528215984462765e-05,
+      "loss": 2.0389,
+      "step": 6450
+    },
+    {
+      "epoch": 2.5350730893750613,
+      "grad_norm": 4.1875,
+      "learning_rate": 1.23236447259802e-05,
+      "loss": 2.6438,
+      "step": 6460
+    },
+    {
+      "epoch": 2.53899735112332,
+      "grad_norm": 5.46875,
+      "learning_rate": 1.2120647727349976e-05,
+      "loss": 2.2073,
+      "step": 6470
+    },
+    {
+      "epoch": 2.5429216128715786,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.1919228633454737e-05,
+      "loss": 2.4119,
+      "step": 6480
+    },
+    {
+      "epoch": 2.5468458746198372,
+      "grad_norm": 4.8125,
+      "learning_rate": 1.1719391060845298e-05,
+      "loss": 2.5692,
+      "step": 6490
+    },
+    {
+      "epoch": 2.550770136368096,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.1521138597675651e-05,
+      "loss": 2.0601,
+      "step": 6500
+    },
+    {
+      "epoch": 2.5546943981163546,
+      "grad_norm": 5.6875,
+      "learning_rate": 1.1324474803638652e-05,
+      "loss": 2.6731,
+      "step": 6510
+    },
+    {
+      "epoch": 2.558618659864613,
+      "grad_norm": 12.4375,
+      "learning_rate": 1.1129403209902034e-05,
+      "loss": 2.2524,
+      "step": 6520
+    },
+    {
+      "epoch": 2.5625429216128714,
+      "grad_norm": 2.671875,
+      "learning_rate": 1.0935927319044958e-05,
+      "loss": 2.3624,
+      "step": 6530
+    },
+    {
+      "epoch": 2.56646718336113,
+      "grad_norm": 3.828125,
+      "learning_rate": 1.0744050604995237e-05,
+      "loss": 2.5474,
+      "step": 6540
+    },
+    {
+      "epoch": 2.5703914451093888,
+      "grad_norm": 2.125,
+      "learning_rate": 1.0553776512966885e-05,
+      "loss": 2.0246,
+      "step": 6550
+    },
+    {
+      "epoch": 2.5743157068576474,
+      "grad_norm": 3.28125,
+      "learning_rate": 1.0365108459398277e-05,
+      "loss": 2.6239,
+      "step": 6560
+    },
+    {
+      "epoch": 2.578239968605906,
+      "grad_norm": 4.5625,
+      "learning_rate": 1.0178049831890768e-05,
+      "loss": 2.261,
+      "step": 6570
+    },
+    {
+      "epoch": 2.5821642303541648,
+      "grad_norm": 2.765625,
+      "learning_rate": 9.992603989147942e-06,
+      "loss": 2.4117,
+      "step": 6580
+    },
+    {
+      "epoch": 2.586088492102423,
+      "grad_norm": 3.71875,
+      "learning_rate": 9.808774260915244e-06,
+      "loss": 2.5965,
+      "step": 6590
+    },
+    {
+      "epoch": 2.590012753850682,
+      "grad_norm": 2.09375,
+      "learning_rate": 9.62656394792023e-06,
+      "loss": 2.047,
+      "step": 6600
+    },
+    {
+      "epoch": 2.5939370155989403,
+      "grad_norm": 3.671875,
+      "learning_rate": 9.445976321813277e-06,
+      "loss": 2.5658,
+      "step": 6610
+    },
+    {
+      "epoch": 2.597861277347199,
+      "grad_norm": 4.34375,
+      "learning_rate": 9.267014625108805e-06,
+      "loss": 2.2336,
+      "step": 6620
+    },
+    {
+      "epoch": 2.6017855390954576,
+      "grad_norm": 2.515625,
+      "learning_rate": 9.08968207112717e-06,
+      "loss": 2.4331,
+      "step": 6630
+    },
+    {
+      "epoch": 2.6057098008437163,
+      "grad_norm": 4.125,
+      "learning_rate": 8.91398184393687e-06,
+      "loss": 2.5676,
+      "step": 6640
+    },
+    {
+      "epoch": 2.609634062591975,
+      "grad_norm": 2.203125,
+      "learning_rate": 8.739917098297356e-06,
+      "loss": 2.1017,
+      "step": 6650
+    },
+    {
+      "epoch": 2.6135583243402336,
+      "grad_norm": 3.75,
+      "learning_rate": 8.56749095960251e-06,
+      "loss": 2.6188,
+      "step": 6660
+    },
+    {
+      "epoch": 2.6174825860884923,
+      "grad_norm": 4.46875,
+      "learning_rate": 8.396706523824372e-06,
+      "loss": 2.2505,
+      "step": 6670
+    },
+    {
+      "epoch": 2.6214068478367505,
+      "grad_norm": 2.828125,
+      "learning_rate": 8.227566857457703e-06,
+      "loss": 2.3833,
+      "step": 6680
+    },
+    {
+      "epoch": 2.6253311095850096,
+      "grad_norm": 3.625,
+      "learning_rate": 8.060074997464772e-06,
+      "loss": 2.5766,
+      "step": 6690
+    },
+    {
+      "epoch": 2.629255371333268,
+      "grad_norm": 2.046875,
+      "learning_rate": 7.894233951220953e-06,
+      "loss": 2.057,
+      "step": 6700
+    },
+    {
+      "epoch": 2.6331796330815265,
+      "grad_norm": 3.59375,
+      "learning_rate": 7.73004669646069e-06,
+      "loss": 2.619,
+      "step": 6710
+    },
+    {
+      "epoch": 2.637103894829785,
+      "grad_norm": 3.921875,
+      "learning_rate": 7.567516181223966e-06,
+      "loss": 2.2492,
+      "step": 6720
+    },
+    {
+      "epoch": 2.641028156578044,
+      "grad_norm": 2.78125,
+      "learning_rate": 7.406645323803463e-06,
+      "loss": 2.4048,
+      "step": 6730
+    },
+    {
+      "epoch": 2.6449524183263025,
+      "grad_norm": 3.859375,
+      "learning_rate": 7.247437012692104e-06,
+      "loss": 2.5512,
+      "step": 6740
+    },
+    {
+      "epoch": 2.648876680074561,
+      "grad_norm": 2.34375,
+      "learning_rate": 7.089894106531214e-06,
+      "loss": 2.0628,
+      "step": 6750
+    },
+    {
+      "epoch": 2.65280094182282,
+      "grad_norm": 3.5,
+      "learning_rate": 6.9340194340592135e-06,
+      "loss": 2.6487,
+      "step": 6760
+    },
+    {
+      "epoch": 2.656725203571078,
+      "grad_norm": 4.0625,
+      "learning_rate": 6.779815794060718e-06,
+      "loss": 2.1519,
+      "step": 6770
+    },
+    {
+      "epoch": 2.6606494653193367,
+      "grad_norm": 2.671875,
+      "learning_rate": 6.627285955316475e-06,
+      "loss": 2.3814,
+      "step": 6780
+    },
+    {
+      "epoch": 2.6645737270675953,
+      "grad_norm": 4.46875,
+      "learning_rate": 6.476432656553411e-06,
+      "loss": 2.5957,
+      "step": 6790
+    },
+    {
+      "epoch": 2.668497988815854,
+      "grad_norm": 1.8671875,
+      "learning_rate": 6.327258606395736e-06,
+      "loss": 2.0273,
+      "step": 6800
+    },
+    {
+      "epoch": 2.6724222505641126,
+      "grad_norm": 3.140625,
+      "learning_rate": 6.179766483316041e-06,
+      "loss": 2.6087,
+      "step": 6810
+    },
+    {
+      "epoch": 2.6763465123123713,
+      "grad_norm": 4.03125,
+      "learning_rate": 6.033958935587369e-06,
+      "loss": 2.2209,
+      "step": 6820
+    },
+    {
+      "epoch": 2.68027077406063,
+      "grad_norm": 2.734375,
+      "learning_rate": 5.88983858123564e-06,
+      "loss": 2.4105,
+      "step": 6830
+    },
+    {
+      "epoch": 2.684195035808888,
+      "grad_norm": 3.828125,
+      "learning_rate": 5.747408007992572e-06,
+      "loss": 2.5615,
+      "step": 6840
+    },
+    {
+      "epoch": 2.6881192975571473,
+      "grad_norm": 2.015625,
+      "learning_rate": 5.606669773249296e-06,
+      "loss": 2.0735,
+      "step": 6850
+    },
+    {
+      "epoch": 2.6920435593054055,
+      "grad_norm": 3.25,
+      "learning_rate": 5.467626404010406e-06,
+      "loss": 2.6125,
+      "step": 6860
+    },
+    {
+      "epoch": 2.695967821053664,
+      "grad_norm": 4.53125,
+      "learning_rate": 5.3302803968486195e-06,
+      "loss": 2.237,
+      "step": 6870
+    },
+    {
+      "epoch": 2.699892082801923,
+      "grad_norm": 3.078125,
+      "learning_rate": 5.19463421785985e-06,
+      "loss": 2.3924,
+      "step": 6880
+    },
+    {
+      "epoch": 2.7038163445501815,
+      "grad_norm": 4.0,
+      "learning_rate": 5.0606903026190535e-06,
+      "loss": 2.5669,
+      "step": 6890
+    },
+    {
+      "epoch": 2.70774060629844,
+      "grad_norm": 2.453125,
+      "learning_rate": 4.928451056136441e-06,
+      "loss": 2.0461,
+      "step": 6900
+    },
+    {
+      "epoch": 2.711664868046699,
+      "grad_norm": 3.234375,
+      "learning_rate": 4.797918852814254e-06,
+      "loss": 2.6294,
+      "step": 6910
+    },
+    {
+      "epoch": 2.7155891297949575,
+      "grad_norm": 3.875,
+      "learning_rate": 4.669096036404197e-06,
+      "loss": 2.2064,
+      "step": 6920
+    },
+    {
+      "epoch": 2.7195133915432157,
+      "grad_norm": 2.953125,
+      "learning_rate": 4.541984919965336e-06,
+      "loss": 2.3805,
+      "step": 6930
+    },
+    {
+      "epoch": 2.723437653291475,
+      "grad_norm": 3.8125,
+      "learning_rate": 4.416587785822568e-06,
+      "loss": 2.5629,
+      "step": 6940
+    },
+    {
+      "epoch": 2.727361915039733,
+      "grad_norm": 2.609375,
+      "learning_rate": 4.2929068855256275e-06,
+      "loss": 2.0674,
+      "step": 6950
+    },
+    {
+      "epoch": 2.7312861767879917,
+      "grad_norm": 3.515625,
+      "learning_rate": 4.170944439808622e-06,
+      "loss": 2.6399,
+      "step": 6960
+    },
+    {
+      "epoch": 2.7352104385362503,
+      "grad_norm": 5.6875,
+      "learning_rate": 4.050702638550275e-06,
+      "loss": 2.2867,
+      "step": 6970
+    },
+    {
+      "epoch": 2.739134700284509,
+      "grad_norm": 2.671875,
+      "learning_rate": 3.932183640734466e-06,
+      "loss": 2.3746,
+      "step": 6980
+    },
+    {
+      "epoch": 2.7430589620327677,
+      "grad_norm": 4.0625,
+      "learning_rate": 3.815389574411576e-06,
+      "loss": 2.5992,
+      "step": 6990
+    },
+    {
+      "epoch": 2.7469832237810263,
+      "grad_norm": 2.1875,
+      "learning_rate": 3.700322536660228e-06,
+      "loss": 2.0533,
+      "step": 7000
+    },
+    {
+      "epoch": 2.750907485529285,
+      "grad_norm": 3.328125,
+      "learning_rate": 3.586984593549614e-06,
+      "loss": 2.6233,
+      "step": 7010
+    },
+    {
+      "epoch": 2.754831747277543,
+      "grad_norm": 4.71875,
+      "learning_rate": 3.4753777801024513e-06,
+      "loss": 2.2222,
+      "step": 7020
+    },
+    {
+      "epoch": 2.758756009025802,
+      "grad_norm": 3.09375,
+      "learning_rate": 3.3655041002583987e-06,
+      "loss": 2.383,
+      "step": 7030
+    },
+    {
+      "epoch": 2.7626802707740605,
+      "grad_norm": 4.4375,
+      "learning_rate": 3.2573655268380743e-06,
+      "loss": 2.5302,
+      "step": 7040
+    },
+    {
+      "epoch": 2.766604532522319,
+      "grad_norm": 2.15625,
+      "learning_rate": 3.1509640015076948e-06,
+      "loss": 2.0733,
+      "step": 7050
+    },
+    {
+      "epoch": 2.770528794270578,
+      "grad_norm": 3.359375,
+      "learning_rate": 3.0463014347441255e-06,
+      "loss": 2.6325,
+      "step": 7060
+    },
+    {
+      "epoch": 2.7744530560188365,
+      "grad_norm": 4.71875,
+      "learning_rate": 2.94337970580062e-06,
+      "loss": 2.2353,
+      "step": 7070
+    },
+    {
+      "epoch": 2.778377317767095,
+      "grad_norm": 3.21875,
+      "learning_rate": 2.8422006626731114e-06,
+      "loss": 2.4174,
+      "step": 7080
+    },
+    {
+      "epoch": 2.782301579515354,
+      "grad_norm": 3.953125,
+      "learning_rate": 2.7427661220669532e-06,
+      "loss": 2.5395,
+      "step": 7090
+    },
+    {
+      "epoch": 2.7862258412636125,
+      "grad_norm": 1.953125,
+      "learning_rate": 2.645077869364354e-06,
+      "loss": 2.0349,
+      "step": 7100
+    },
+    {
+      "epoch": 2.7901501030118707,
+      "grad_norm": 4.71875,
+      "learning_rate": 2.5491376585923266e-06,
+      "loss": 2.6257,
+      "step": 7110
+    },
+    {
+      "epoch": 2.7940743647601294,
+      "grad_norm": 3.671875,
+      "learning_rate": 2.4549472123911566e-06,
+      "loss": 2.2344,
+      "step": 7120
+    },
+    {
+      "epoch": 2.797998626508388,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.362508221983484e-06,
+      "loss": 2.3803,
+      "step": 7130
+    },
+    {
+      "epoch": 2.8019228882566467,
+      "grad_norm": 3.703125,
+      "learning_rate": 2.2718223471439815e-06,
+      "loss": 2.5409,
+      "step": 7140
+    },
+    {
+      "epoch": 2.8058471500049054,
+      "grad_norm": 1.90625,
+      "learning_rate": 2.182891216169447e-06,
+      "loss": 2.07,
+      "step": 7150
+    },
+    {
+      "epoch": 2.809771411753164,
+      "grad_norm": 3.203125,
+      "learning_rate": 2.095716425849703e-06,
+      "loss": 2.624,
+      "step": 7160
+    },
+    {
+      "epoch": 2.8136956735014227,
+      "grad_norm": 4.90625,
+      "learning_rate": 2.010299541438798e-06,
+      "loss": 2.2178,
+      "step": 7170
+    },
+    {
+      "epoch": 2.817619935249681,
+      "grad_norm": 2.640625,
+      "learning_rate": 1.9266420966270184e-06,
+      "loss": 2.3739,
+      "step": 7180
+    },
+    {
+      "epoch": 2.82154419699794,
+      "grad_norm": 3.796875,
+      "learning_rate": 1.8447455935132418e-06,
+      "loss": 2.5512,
+      "step": 7190
+    },
+    {
+      "epoch": 2.8254684587461982,
+      "grad_norm": 3.578125,
+      "learning_rate": 1.7646115025780507e-06,
+      "loss": 2.0392,
+      "step": 7200
+    },
+    {
+      "epoch": 2.829392720494457,
+      "grad_norm": 6.75,
+      "learning_rate": 1.6862412626572842e-06,
+      "loss": 2.6232,
+      "step": 7210
+    },
+    {
+      "epoch": 2.8333169822427156,
+      "grad_norm": 5.375,
+      "learning_rate": 1.6096362809162047e-06,
+      "loss": 2.2863,
+      "step": 7220
+    },
+    {
+      "epoch": 2.837241243990974,
+      "grad_norm": 2.5625,
+      "learning_rate": 1.5347979328242611e-06,
+      "loss": 2.3873,
+      "step": 7230
+    },
+    {
+      "epoch": 2.841165505739233,
+      "grad_norm": 3.96875,
+      "learning_rate": 1.461727562130344e-06,
+      "loss": 2.5569,
+      "step": 7240
+    },
+    {
+      "epoch": 2.8450897674874915,
+      "grad_norm": 5.59375,
+      "learning_rate": 1.3904264808387246e-06,
+      "loss": 2.042,
+      "step": 7250
+    },
+    {
+      "epoch": 2.84901402923575,
+      "grad_norm": 3.28125,
+      "learning_rate": 1.3208959691854539e-06,
+      "loss": 2.6092,
+      "step": 7260
+    },
+    {
+      "epoch": 2.8529382909840084,
+      "grad_norm": 4.28125,
+      "learning_rate": 1.2531372756153458e-06,
+      "loss": 2.203,
+      "step": 7270
+    },
+    {
+      "epoch": 2.856862552732267,
+      "grad_norm": 3.046875,
+      "learning_rate": 1.1871516167596186e-06,
+      "loss": 2.3949,
+      "step": 7280
+    },
+    {
+      "epoch": 2.8607868144805257,
+      "grad_norm": 3.84375,
+      "learning_rate": 1.1229401774140448e-06,
+      "loss": 2.5757,
+      "step": 7290
+    },
+    {
+      "epoch": 2.8647110762287844,
+      "grad_norm": 1.9609375,
+      "learning_rate": 1.0605041105176129e-06,
+      "loss": 2.0566,
+      "step": 7300
+    },
+    {
+      "epoch": 2.868635337977043,
+      "grad_norm": 2.984375,
+      "learning_rate": 9.998445371319332e-07,
+      "loss": 2.6256,
+      "step": 7310
+    },
+    {
+      "epoch": 2.8725595997253017,
+      "grad_norm": 4.34375,
+      "learning_rate": 9.409625464210092e-07,
+      "loss": 2.2386,
+      "step": 7320
+    },
+    {
+      "epoch": 2.8764838614735604,
+      "grad_norm": 2.78125,
+      "learning_rate": 8.838591956317311e-07,
+      "loss": 2.3862,
+      "step": 7330
+    },
+    {
+      "epoch": 2.880408123221819,
+      "grad_norm": 3.640625,
+      "learning_rate": 8.285355100748904e-07,
+      "loss": 2.5551,
+      "step": 7340
+    },
+    {
+      "epoch": 2.8843323849700777,
+      "grad_norm": 1.7890625,
+      "learning_rate": 7.749924831067401e-07,
+      "loss": 2.0778,
+      "step": 7350
+    },
+    {
+      "epoch": 2.888256646718336,
+      "grad_norm": 4.46875,
+      "learning_rate": 7.232310761112082e-07,
+      "loss": 2.5868,
+      "step": 7360
+    },
+    {
+      "epoch": 2.8921809084665946,
+      "grad_norm": 4.03125,
+      "learning_rate": 6.732522184825896e-07,
+      "loss": 2.2131,
+      "step": 7370
+    },
+    {
+      "epoch": 2.8961051702148533,
+      "grad_norm": 3.875,
+      "learning_rate": 6.250568076088815e-07,
+      "loss": 2.3724,
+      "step": 7380
+    },
+    {
+      "epoch": 2.900029431963112,
+      "grad_norm": 4.25,
+      "learning_rate": 5.78645708855674e-07,
+      "loss": 2.5877,
+      "step": 7390
+    },
+    {
+      "epoch": 2.9039536937113706,
+      "grad_norm": 2.203125,
+      "learning_rate": 5.340197555505966e-07,
+      "loss": 2.051,
+      "step": 7400
+    },
+    {
+      "epoch": 2.9078779554596292,
+      "grad_norm": 3.21875,
+      "learning_rate": 4.911797489683733e-07,
+      "loss": 2.628,
+      "step": 7410
+    },
+    {
+      "epoch": 2.911802217207888,
+      "grad_norm": 5.625,
+      "learning_rate": 4.501264583164022e-07,
+      "loss": 2.2205,
+      "step": 7420
+    },
+    {
+      "epoch": 2.915726478956146,
+      "grad_norm": 2.578125,
+      "learning_rate": 4.108606207209875e-07,
+      "loss": 2.3974,
+      "step": 7430
+    },
+    {
+      "epoch": 2.9196507407044052,
+      "grad_norm": 6.71875,
+      "learning_rate": 3.7338294121407323e-07,
+      "loss": 2.5681,
+      "step": 7440
+    },
+    {
+      "epoch": 2.9235750024526634,
+      "grad_norm": 2.1875,
+      "learning_rate": 3.376940927206196e-07,
+      "loss": 2.0398,
+      "step": 7450
+    },
+    {
+      "epoch": 2.927499264200922,
+      "grad_norm": 3.375,
+      "learning_rate": 3.0379471604645717e-07,
+      "loss": 2.6153,
+      "step": 7460
+    },
+    {
+      "epoch": 2.9314235259491808,
+      "grad_norm": 4.46875,
+      "learning_rate": 2.716854198668517e-07,
+      "loss": 2.2301,
+      "step": 7470
+    },
+    {
+      "epoch": 2.9353477876974394,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.41366780715524e-07,
+      "loss": 2.3844,
+      "step": 7480
+    },
+    {
+      "epoch": 2.939272049445698,
+      "grad_norm": 5.40625,
+      "learning_rate": 2.1283934297432473e-07,
+      "loss": 2.528,
+      "step": 7490
+    },
+    {
+      "epoch": 2.9431963111939567,
+      "grad_norm": 2.25,
+      "learning_rate": 1.861036188634424e-07,
+      "loss": 2.0426,
+      "step": 7500
+    },
+    {
+      "epoch": 2.9471205729422154,
+      "grad_norm": 3.609375,
+      "learning_rate": 1.6116008843224395e-07,
+      "loss": 2.6279,
+      "step": 7510
+    },
+    {
+      "epoch": 2.9510448346904736,
+      "grad_norm": 5.5625,
+      "learning_rate": 1.3800919955058167e-07,
+      "loss": 2.2091,
+      "step": 7520
+    },
+    {
+      "epoch": 2.9549690964387327,
+      "grad_norm": 3.140625,
+      "learning_rate": 1.1665136790084407e-07,
+      "loss": 2.4096,
+      "step": 7530
+    },
+    {
+      "epoch": 2.958893358186991,
+      "grad_norm": 4.84375,
+      "learning_rate": 9.708697697040636e-08,
+      "loss": 2.5678,
+      "step": 7540
+    },
+    {
+      "epoch": 2.9628176199352496,
+      "grad_norm": 2.09375,
+      "learning_rate": 7.931637804481362e-08,
+      "loss": 2.0519,
+      "step": 7550
+    },
+    {
+      "epoch": 2.9667418816835083,
+      "grad_norm": 6.34375,
+      "learning_rate": 6.333989020143039e-08,
+      "loss": 2.6151,
+      "step": 7560
+    },
+    {
+      "epoch": 2.970666143431767,
+      "grad_norm": 4.40625,
+      "learning_rate": 4.9157800303722965e-08,
+      "loss": 2.2428,
+      "step": 7570
+    },
+    {
+      "epoch": 2.9745904051800256,
+      "grad_norm": 2.703125,
+      "learning_rate": 3.6770362996108033e-08,
+      "loss": 2.3933,
+      "step": 7580
+    },
+    {
+      "epoch": 2.9785146669282843,
+      "grad_norm": 4.15625,
+      "learning_rate": 2.617780069940068e-08,
+      "loss": 2.5695,
+      "step": 7590
+    },
+    {
+      "epoch": 2.982438928676543,
+      "grad_norm": 2.578125,
+      "learning_rate": 1.738030360677323e-08,
+      "loss": 2.0302,
+      "step": 7600
+    },
+    {
+      "epoch": 2.986363190424801,
+      "grad_norm": 3.125,
+      "learning_rate": 1.0378029680391255e-08,
+      "loss": 2.6026,
+      "step": 7610
+    },
+    {
+      "epoch": 2.99028745217306,
+      "grad_norm": 4.40625,
+      "learning_rate": 5.171104648549196e-09,
+      "loss": 2.3131,
+      "step": 7620
+    },
+    {
+      "epoch": 2.9942117139213185,
+      "grad_norm": 3.84375,
+      "learning_rate": 1.7596220034277188e-09,
+      "loss": 2.4259,
+      "step": 7630
+    },
+    {
+      "epoch": 2.998135975669577,
+      "grad_norm": 4.0625,
+      "learning_rate": 1.436429993950661e-10,
+      "loss": 2.5149,
+      "step": 7640
+    },
+    {
+      "epoch": 2.999705680368881,
+      "eval_loss": 2.4481096267700195,
+      "eval_runtime": 156.2821,
+      "eval_samples_per_second": 159.967,
+      "eval_steps_per_second": 39.992,
+      "step": 7644
+    },
+    {
+      "epoch": 2.999705680368881,
+      "step": 7644,
+      "total_flos": 5.656192905023324e+17,
+      "train_loss": 2.742037247513552,
+      "train_runtime": 9381.445,
+      "train_samples_per_second": 104.296,
+      "train_steps_per_second": 0.815
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 7644,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.656192905023324e+17,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}