diff --git "a/checkpoint-712/trainer_state.json" "b/checkpoint-712/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-712/trainer_state.json"
@@ -0,0 +1,5005 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.8011251758087201,
+  "eval_steps": 500,
+  "global_step": 712,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.33328112959861755,
+      "learning_rate": 2e-05,
+      "loss": 0.9029,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.341234415769577,
+      "learning_rate": 4e-05,
+      "loss": 0.9276,
+      "step": 2
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.3182106614112854,
+      "learning_rate": 6e-05,
+      "loss": 0.9148,
+      "step": 3
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 0.22865141928195953,
+      "learning_rate": 8e-05,
+      "loss": 0.8421,
+      "step": 4
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.29670122265815735,
+      "learning_rate": 0.0001,
+      "loss": 0.7984,
+      "step": 5
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.28762516379356384,
+      "learning_rate": 0.00012,
+      "loss": 0.7625,
+      "step": 6
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.23907965421676636,
+      "learning_rate": 0.00014,
+      "loss": 0.7226,
+      "step": 7
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.19532263278961182,
+      "learning_rate": 0.00016,
+      "loss": 0.7043,
+      "step": 8
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.1425202488899231,
+      "learning_rate": 0.00018,
+      "loss": 0.6784,
+      "step": 9
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.10882167518138885,
+      "learning_rate": 0.0002,
+      "loss": 0.6579,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.11275648325681686,
+      "learning_rate": 0.00019999935985220405,
+      "loss": 0.6592,
+      "step": 11
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.10155748575925827,
+      "learning_rate": 0.00019999743941701188,
+      "loss": 0.6554,
+      "step": 12
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.0845816433429718,
+      "learning_rate": 0.0001999942387190108,
+      "loss": 0.6513,
+      "step": 13
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.09046202898025513,
+      "learning_rate": 0.0001999897577991792,
+      "loss": 0.6267,
+      "step": 14
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.08296829462051392,
+      "learning_rate": 0.00019998399671488612,
+      "loss": 0.6434,
+      "step": 15
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.07594181597232819,
+      "learning_rate": 0.00019997695553989042,
+      "loss": 0.6096,
+      "step": 16
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.0683172270655632,
+      "learning_rate": 0.00019996863436433997,
+      "loss": 0.6143,
+      "step": 17
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.05627186596393585,
+      "learning_rate": 0.0001999590332947704,
+      "loss": 0.6024,
+      "step": 18
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.05644279345870018,
+      "learning_rate": 0.00019994815245410384,
+      "loss": 0.595,
+      "step": 19
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.05661479011178017,
+      "learning_rate": 0.00019993599198164715,
+      "loss": 0.5759,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.04077177122235298,
+      "learning_rate": 0.00019992255203309033,
+      "loss": 0.582,
+      "step": 21
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 0.04510512948036194,
+      "learning_rate": 0.00019990783278050448,
+      "loss": 0.5751,
+      "step": 22
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.0470162108540535,
+      "learning_rate": 0.00019989183441233952,
+      "loss": 0.5716,
+      "step": 23
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.04402562975883484,
+      "learning_rate": 0.00019987455713342187,
+      "loss": 0.564,
+      "step": 24
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.045594893395900726,
+      "learning_rate": 0.00019985600116495173,
+      "loss": 0.5657,
+      "step": 25
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.037670280784368515,
+      "learning_rate": 0.0001998361667445004,
+      "loss": 0.5619,
+      "step": 26
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.034366946667432785,
+      "learning_rate": 0.00019981505412600706,
+      "loss": 0.554,
+      "step": 27
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.044084370136260986,
+      "learning_rate": 0.00019979266357977564,
+      "loss": 0.5527,
+      "step": 28
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.04839107394218445,
+      "learning_rate": 0.0001997689953924713,
+      "loss": 0.5534,
+      "step": 29
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.04943666234612465,
+      "learning_rate": 0.0001997440498671168,
+      "loss": 0.5357,
+      "step": 30
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 0.05644814297556877,
+      "learning_rate": 0.00019971782732308867,
+      "loss": 0.5388,
+      "step": 31
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.05796538665890694,
+      "learning_rate": 0.00019969032809611287,
+      "loss": 0.5327,
+      "step": 32
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.05399211496114731,
+      "learning_rate": 0.0001996615525382609,
+      "loss": 0.5447,
+      "step": 33
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.03574652597308159,
+      "learning_rate": 0.0001996315010179449,
+      "loss": 0.5213,
+      "step": 34
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.04394914582371712,
+      "learning_rate": 0.00019960017391991314,
+      "loss": 0.5247,
+      "step": 35
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.05197073519229889,
+      "learning_rate": 0.00019956757164524516,
+      "loss": 0.5253,
+      "step": 36
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.05605654790997505,
+      "learning_rate": 0.00019953369461134634,
+      "loss": 0.5289,
+      "step": 37
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.05167644843459129,
+      "learning_rate": 0.00019949854325194294,
+      "loss": 0.5223,
+      "step": 38
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 0.04118728265166283,
+      "learning_rate": 0.0001994621180170762,
+      "loss": 0.52,
+      "step": 39
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.0373263917863369,
+      "learning_rate": 0.00019942441937309684,
+      "loss": 0.5076,
+      "step": 40
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.04870600998401642,
+      "learning_rate": 0.0001993854478026589,
+      "loss": 0.5287,
+      "step": 41
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.05014103651046753,
+      "learning_rate": 0.00019934520380471372,
+      "loss": 0.5048,
+      "step": 42
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.04807833209633827,
+      "learning_rate": 0.0001993036878945034,
+      "loss": 0.5075,
+      "step": 43
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.040045421570539474,
+      "learning_rate": 0.0001992609006035543,
+      "loss": 0.4954,
+      "step": 44
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.04202349856495857,
+      "learning_rate": 0.00019921684247967028,
+      "loss": 0.4953,
+      "step": 45
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.041629109531641006,
+      "learning_rate": 0.0001991715140869255,
+      "loss": 0.501,
+      "step": 46
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.04767894372344017,
+      "learning_rate": 0.0001991249160056574,
+      "loss": 0.4878,
+      "step": 47
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 0.05395263060927391,
+      "learning_rate": 0.00019907704883245916,
+      "loss": 0.5014,
+      "step": 48
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.06040235981345177,
+      "learning_rate": 0.00019902791318017205,
+      "loss": 0.5043,
+      "step": 49
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.06749273091554642,
+      "learning_rate": 0.0001989775096778777,
+      "loss": 0.4931,
+      "step": 50
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.06482889503240585,
+      "learning_rate": 0.00019892583897088994,
+      "loss": 0.4869,
+      "step": 51
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.045358914881944656,
+      "learning_rate": 0.0001988729017207465,
+      "loss": 0.479,
+      "step": 52
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.043360061943531036,
+      "learning_rate": 0.00019881869860520073,
+      "loss": 0.4953,
+      "step": 53
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.060205183923244476,
+      "learning_rate": 0.00019876323031821266,
+      "loss": 0.4705,
+      "step": 54
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.05729120969772339,
+      "learning_rate": 0.00019870649756994037,
+      "loss": 0.4887,
+      "step": 55
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.03843148052692413,
+      "learning_rate": 0.00019864850108673073,
+      "loss": 0.4737,
+      "step": 56
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 0.053673889487981796,
+      "learning_rate": 0.00019858924161111015,
+      "loss": 0.4817,
+      "step": 57
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.05148368701338768,
+      "learning_rate": 0.00019852871990177503,
+      "loss": 0.4763,
+      "step": 58
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.05371672287583351,
+      "learning_rate": 0.00019846693673358226,
+      "loss": 0.4751,
+      "step": 59
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.05490916967391968,
+      "learning_rate": 0.00019840389289753896,
+      "loss": 0.457,
+      "step": 60
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.04629400372505188,
+      "learning_rate": 0.00019833958920079255,
+      "loss": 0.4692,
+      "step": 61
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.051137253642082214,
+      "learning_rate": 0.00019827402646662047,
+      "loss": 0.4614,
+      "step": 62
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.051790811121463776,
+      "learning_rate": 0.0001982072055344195,
+      "loss": 0.4594,
+      "step": 63
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.0445956289768219,
+      "learning_rate": 0.00019813912725969509,
+      "loss": 0.4601,
+      "step": 64
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.04766576737165451,
+      "learning_rate": 0.0001980697925140504,
+      "loss": 0.4631,
+      "step": 65
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 0.04839074984192848,
+      "learning_rate": 0.0001979992021851751,
+      "loss": 0.4605,
+      "step": 66
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.04736727103590965,
+      "learning_rate": 0.0001979273571768341,
+      "loss": 0.4617,
+      "step": 67
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.057293377816677094,
+      "learning_rate": 0.0001978542584088558,
+      "loss": 0.4621,
+      "step": 68
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.05025665834546089,
+      "learning_rate": 0.0001977799068171206,
+      "loss": 0.4671,
+      "step": 69
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.057366687804460526,
+      "learning_rate": 0.0001977043033535486,
+      "loss": 0.4521,
+      "step": 70
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.07595837116241455,
+      "learning_rate": 0.00019762744898608762,
+      "loss": 0.4671,
+      "step": 71
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.07574213296175003,
+      "learning_rate": 0.0001975493446987007,
+      "loss": 0.4664,
+      "step": 72
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.06472938507795334,
+      "learning_rate": 0.00019746999149135362,
+      "loss": 0.456,
+      "step": 73
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.05983012542128563,
+      "learning_rate": 0.00019738939038000205,
+      "loss": 0.4459,
+      "step": 74
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.05136057734489441,
+      "learning_rate": 0.00019730754239657842,
+      "loss": 0.4486,
+      "step": 75
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.06191498041152954,
+      "learning_rate": 0.00019722444858897878,
+      "loss": 0.4424,
+      "step": 76
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.06742191314697266,
+      "learning_rate": 0.0001971401100210496,
+      "loss": 0.458,
+      "step": 77
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.06019548326730728,
+      "learning_rate": 0.00019705452777257377,
+      "loss": 0.4423,
+      "step": 78
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.05012982338666916,
+      "learning_rate": 0.0001969677029392571,
+      "loss": 0.4466,
+      "step": 79
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.0552060566842556,
+      "learning_rate": 0.00019687963663271409,
+      "loss": 0.4534,
+      "step": 80
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.05883748456835747,
+      "learning_rate": 0.00019679032998045376,
+      "loss": 0.4409,
+      "step": 81
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.07146705687046051,
+      "learning_rate": 0.00019669978412586528,
+      "loss": 0.4582,
+      "step": 82
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.054095230996608734,
+      "learning_rate": 0.00019660800022820317,
+      "loss": 0.4487,
+      "step": 83
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 0.04927053675055504,
+      "learning_rate": 0.00019651497946257266,
+      "loss": 0.4429,
+      "step": 84
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.06037526577711105,
+      "learning_rate": 0.00019642072301991455,
+      "loss": 0.4456,
+      "step": 85
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.05555957555770874,
+      "learning_rate": 0.00019632523210698987,
+      "loss": 0.4382,
+      "step": 86
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.04606284573674202,
+      "learning_rate": 0.00019622850794636455,
+      "loss": 0.4411,
+      "step": 87
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.04605920985341072,
+      "learning_rate": 0.00019613055177639384,
+      "loss": 0.4326,
+      "step": 88
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.050325632095336914,
+      "learning_rate": 0.0001960313648512062,
+      "loss": 0.4338,
+      "step": 89
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.04921424016356468,
+      "learning_rate": 0.00019593094844068748,
+      "loss": 0.4316,
+      "step": 90
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.04333706200122833,
+      "learning_rate": 0.00019582930383046457,
+      "loss": 0.4441,
+      "step": 91
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.048454612493515015,
+      "learning_rate": 0.0001957264323218889,
+      "loss": 0.4382,
+      "step": 92
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 0.0541059784591198,
+      "learning_rate": 0.00019562233523201986,
+      "loss": 0.4328,
+      "step": 93
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.043696511536836624,
+      "learning_rate": 0.00019551701389360795,
+      "loss": 0.4335,
+      "step": 94
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.04407835751771927,
+      "learning_rate": 0.00019541046965507758,
+      "loss": 0.4327,
+      "step": 95
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.05477238819003105,
+      "learning_rate": 0.00019530270388050998,
+      "loss": 0.4294,
+      "step": 96
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.05609311908483505,
+      "learning_rate": 0.00019519371794962556,
+      "loss": 0.4305,
+      "step": 97
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.045145273208618164,
+      "learning_rate": 0.00019508351325776642,
+      "loss": 0.4395,
+      "step": 98
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.04475285857915878,
+      "learning_rate": 0.00019497209121587837,
+      "loss": 0.4284,
+      "step": 99
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.04405711591243744,
+      "learning_rate": 0.00019485945325049288,
+      "loss": 0.4214,
+      "step": 100
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.04461454227566719,
+      "learning_rate": 0.0001947456008037089,
+      "loss": 0.4154,
+      "step": 101
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 0.04791221395134926,
+      "learning_rate": 0.00019463053533317425,
+      "loss": 0.4248,
+      "step": 102
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.05543987452983856,
+      "learning_rate": 0.00019451425831206706,
+      "loss": 0.4303,
+      "step": 103
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.06330578774213791,
+      "learning_rate": 0.00019439677122907697,
+      "loss": 0.4274,
+      "step": 104
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.05569112300872803,
+      "learning_rate": 0.00019427807558838588,
+      "loss": 0.4234,
+      "step": 105
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.047680530697107315,
+      "learning_rate": 0.00019415817290964883,
+      "loss": 0.4155,
+      "step": 106
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.05214262008666992,
+      "learning_rate": 0.0001940370647279746,
+      "loss": 0.4224,
+      "step": 107
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.06332990527153015,
+      "learning_rate": 0.00019391475259390584,
+      "loss": 0.4233,
+      "step": 108
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.05726313218474388,
+      "learning_rate": 0.00019379123807339942,
+      "loss": 0.4118,
+      "step": 109
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.044936031103134155,
+      "learning_rate": 0.00019366652274780628,
+      "loss": 0.4296,
+      "step": 110
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.05117325484752655,
+      "learning_rate": 0.0001935406082138513,
+      "loss": 0.4287,
+      "step": 111
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.058542776852846146,
+      "learning_rate": 0.00019341349608361267,
+      "loss": 0.4213,
+      "step": 112
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.056066304445266724,
+      "learning_rate": 0.00019328518798450138,
+      "loss": 0.4174,
+      "step": 113
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.049762677401304245,
+      "learning_rate": 0.00019315568555924035,
+      "loss": 0.418,
+      "step": 114
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.043821126222610474,
+      "learning_rate": 0.00019302499046584348,
+      "loss": 0.4012,
+      "step": 115
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.05036221817135811,
+      "learning_rate": 0.00019289310437759427,
+      "loss": 0.4237,
+      "step": 116
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.050889529287815094,
+      "learning_rate": 0.00019276002898302447,
+      "loss": 0.4144,
+      "step": 117
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.04269757494330406,
+      "learning_rate": 0.0001926257659858925,
+      "loss": 0.4078,
+      "step": 118
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 0.04927165433764458,
+      "learning_rate": 0.00019249031710516162,
+      "loss": 0.4155,
+      "step": 119
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.05124311521649361,
+      "learning_rate": 0.00019235368407497788,
+      "loss": 0.3966,
+      "step": 120
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.04073040187358856,
+      "learning_rate": 0.00019221586864464786,
+      "loss": 0.4064,
+      "step": 121
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.04988453537225723,
+      "learning_rate": 0.00019207687257861655,
+      "loss": 0.4197,
+      "step": 122
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.05227258801460266,
+      "learning_rate": 0.0001919366976564444,
+      "loss": 0.414,
+      "step": 123
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.0466819666326046,
+      "learning_rate": 0.00019179534567278475,
+      "loss": 0.4173,
+      "step": 124
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.047633491456508636,
+      "learning_rate": 0.00019165281843736085,
+      "loss": 0.4085,
+      "step": 125
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.05280464142560959,
+      "learning_rate": 0.00019150911777494258,
+      "loss": 0.4051,
+      "step": 126
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.052302148193120956,
+      "learning_rate": 0.00019136424552532318,
+      "loss": 0.42,
+      "step": 127
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.04875241965055466,
+      "learning_rate": 0.00019121820354329577,
+      "loss": 0.4258,
+      "step": 128
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.04654408246278763,
+      "learning_rate": 0.0001910709936986293,
+      "loss": 0.409,
+      "step": 129
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.05745020881295204,
+      "learning_rate": 0.00019092261787604492,
+      "loss": 0.4059,
+      "step": 130
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.06945241987705231,
+      "learning_rate": 0.00019077307797519183,
+      "loss": 0.4038,
+      "step": 131
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.06346461176872253,
+      "learning_rate": 0.00019062237591062272,
+      "loss": 0.4031,
+      "step": 132
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.058026187121868134,
+      "learning_rate": 0.00019047051361176953,
+      "loss": 0.4126,
+      "step": 133
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.04755179584026337,
+      "learning_rate": 0.0001903174930229185,
+      "loss": 0.4209,
+      "step": 134
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.05765068158507347,
+      "learning_rate": 0.0001901633161031856,
+      "loss": 0.4067,
+      "step": 135
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.05687811225652695,
+      "learning_rate": 0.000190007984826491,
+      "loss": 0.3975,
+      "step": 136
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 0.04930473491549492,
+      "learning_rate": 0.0001898515011815343,
+      "loss": 0.4146,
+      "step": 137
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.05147051811218262,
+      "learning_rate": 0.0001896938671717687,
+      "loss": 0.4035,
+      "step": 138
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.05680418014526367,
+      "learning_rate": 0.0001895350848153754,
+      "loss": 0.4049,
+      "step": 139
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.0444297268986702,
+      "learning_rate": 0.00018937515614523797,
+      "loss": 0.4085,
+      "step": 140
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.05083802342414856,
+      "learning_rate": 0.00018921408320891612,
+      "loss": 0.4036,
+      "step": 141
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.04978756606578827,
+      "learning_rate": 0.00018905186806861957,
+      "loss": 0.4058,
+      "step": 142
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.04963681101799011,
+      "learning_rate": 0.00018888851280118155,
+      "loss": 0.3977,
+      "step": 143
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.0466095507144928,
+      "learning_rate": 0.00018872401949803237,
+      "loss": 0.3945,
+      "step": 144
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.04972768574953079,
+      "learning_rate": 0.00018855839026517257,
+      "loss": 0.4151,
+      "step": 145
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.054370637983083725,
+      "learning_rate": 0.0001883916272231459,
+      "loss": 0.3944,
+      "step": 146
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.054699357599020004,
+      "learning_rate": 0.00018822373250701224,
+      "loss": 0.3989,
+      "step": 147
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.054452769458293915,
+      "learning_rate": 0.00018805470826632024,
+      "loss": 0.3984,
+      "step": 148
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.04596908017992973,
+      "learning_rate": 0.00018788455666507981,
+      "loss": 0.4018,
+      "step": 149
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.054354868829250336,
+      "learning_rate": 0.00018771327988173435,
+      "loss": 0.3985,
+      "step": 150
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.05570242181420326,
+      "learning_rate": 0.00018754088010913304,
+      "loss": 0.3818,
+      "step": 151
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.054722048342227936,
+      "learning_rate": 0.00018736735955450251,
+      "loss": 0.4111,
+      "step": 152
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.04620000347495079,
+      "learning_rate": 0.00018719272043941882,
+      "loss": 0.3949,
+      "step": 153
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.048443205654621124,
+      "learning_rate": 0.00018701696499977884,
+      "loss": 0.3856,
+      "step": 154
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 0.06628945469856262,
+      "learning_rate": 0.00018684009548577168,
+      "loss": 0.4048,
+      "step": 155
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.05339967459440231,
+      "learning_rate": 0.00018666211416184999,
+      "loss": 0.3894,
+      "step": 156
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.04650304839015007,
+      "learning_rate": 0.00018648302330670082,
+      "loss": 0.4004,
+      "step": 157
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.05634591728448868,
+      "learning_rate": 0.00018630282521321645,
+      "loss": 0.4033,
+      "step": 158
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.048666685819625854,
+      "learning_rate": 0.00018612152218846513,
+      "loss": 0.399,
+      "step": 159
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.04597772657871246,
+      "learning_rate": 0.0001859391165536615,
+      "loss": 0.3931,
+      "step": 160
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.0526028610765934,
+      "learning_rate": 0.00018575561064413689,
+      "loss": 0.3879,
+      "step": 161
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.05867009237408638,
+      "learning_rate": 0.00018557100680930937,
+      "loss": 0.3905,
+      "step": 162
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.05077454075217247,
+      "learning_rate": 0.00018538530741265364,
+      "loss": 0.395,
+      "step": 163
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 0.0461389385163784,
+      "learning_rate": 0.00018519851483167097,
+      "loss": 0.4016,
+      "step": 164
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.059010252356529236,
+      "learning_rate": 0.00018501063145785846,
+      "loss": 0.3823,
+      "step": 165
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.06437338888645172,
+      "learning_rate": 0.00018482165969667874,
+      "loss": 0.3918,
+      "step": 166
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.04585932940244675,
+      "learning_rate": 0.00018463160196752887,
+      "loss": 0.3808,
+      "step": 167
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.05361521616578102,
+      "learning_rate": 0.00018444046070370963,
+      "loss": 0.3858,
+      "step": 168
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.05653822794556618,
+      "learning_rate": 0.00018424823835239417,
+      "loss": 0.3785,
+      "step": 169
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.04439689964056015,
+      "learning_rate": 0.0001840549373745968,
+      "loss": 0.3894,
+      "step": 170
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.05564529448747635,
+      "learning_rate": 0.00018386056024514137,
+      "loss": 0.3883,
+      "step": 171
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.06035888195037842,
+      "learning_rate": 0.00018366510945262972,
+      "loss": 0.3855,
+      "step": 172
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 0.044238511472940445,
+      "learning_rate": 0.0001834685874994098,
+      "loss": 0.3934,
+      "step": 173
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.050235260277986526,
+      "learning_rate": 0.00018327099690154344,
+      "loss": 0.3819,
+      "step": 174
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.051336683332920074,
+      "learning_rate": 0.00018307234018877434,
+      "loss": 0.3897,
+      "step": 175
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.045052576810121536,
+      "learning_rate": 0.0001828726199044957,
+      "loss": 0.3822,
+      "step": 176
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.05162283405661583,
+      "learning_rate": 0.00018267183860571753,
+      "loss": 0.4047,
+      "step": 177
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.0488157793879509,
+      "learning_rate": 0.00018246999886303383,
+      "loss": 0.3947,
+      "step": 178
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.04454487934708595,
+      "learning_rate": 0.00018226710326059006,
+      "loss": 0.3942,
+      "step": 179
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.0500001423060894,
+      "learning_rate": 0.0001820631543960496,
+      "loss": 0.3826,
+      "step": 180
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.04919865354895592,
+      "learning_rate": 0.00018185815488056076,
+      "loss": 0.3791,
+      "step": 181
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 0.04547140747308731,
+      "learning_rate": 0.00018165210733872336,
+      "loss": 0.3879,
+      "step": 182
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.044622063636779785,
+      "learning_rate": 0.00018144501440855496,
+      "loss": 0.3778,
+      "step": 183
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.04467932507395744,
+      "learning_rate": 0.00018123687874145721,
+      "loss": 0.3994,
+      "step": 184
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.04281982406973839,
+      "learning_rate": 0.0001810277030021819,
+      "loss": 0.3817,
+      "step": 185
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.05303504317998886,
+      "learning_rate": 0.00018081748986879679,
+      "loss": 0.3749,
+      "step": 186
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.046573616564273834,
+      "learning_rate": 0.00018060624203265134,
+      "loss": 0.3866,
+      "step": 187
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.044320229440927505,
+      "learning_rate": 0.00018039396219834237,
+      "loss": 0.3732,
+      "step": 188
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.05708359181880951,
+      "learning_rate": 0.00018018065308367912,
+      "loss": 0.3863,
+      "step": 189
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.045029208064079285,
+      "learning_rate": 0.00017996631741964888,
+      "loss": 0.3862,
+      "step": 190
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 0.055195923894643784,
+      "learning_rate": 0.00017975095795038165,
+      "loss": 0.3835,
+      "step": 191
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.048293352127075195,
+      "learning_rate": 0.00017953457743311523,
+      "loss": 0.374,
+      "step": 192
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.04677055403590202,
+      "learning_rate": 0.00017931717863815987,
+      "loss": 0.377,
+      "step": 193
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.04955766722559929,
+      "learning_rate": 0.00017909876434886273,
+      "loss": 0.3808,
+      "step": 194
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.04526973515748978,
+      "learning_rate": 0.00017887933736157233,
+      "loss": 0.3796,
+      "step": 195
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.043622203171253204,
+      "learning_rate": 0.00017865890048560277,
+      "loss": 0.376,
+      "step": 196
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.046581387519836426,
+      "learning_rate": 0.0001784374565431976,
+      "loss": 0.3716,
+      "step": 197
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.04433497413992882,
+      "learning_rate": 0.00017821500836949386,
+      "loss": 0.3715,
+      "step": 198
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 0.04146367311477661,
+      "learning_rate": 0.00017799155881248572,
+      "loss": 0.3809,
+      "step": 199
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.045288585126399994,
+      "learning_rate": 0.000177767110732988,
+      "loss": 0.3885,
+      "step": 200
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.04070120304822922,
+      "learning_rate": 0.00017754166700459958,
+      "loss": 0.3713,
+      "step": 201
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.042820919305086136,
+      "learning_rate": 0.00017731523051366658,
+      "loss": 0.3839,
+      "step": 202
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.04416365176439285,
+      "learning_rate": 0.00017708780415924539,
+      "loss": 0.3728,
+      "step": 203
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.04461952671408653,
+      "learning_rate": 0.00017685939085306562,
+      "loss": 0.373,
+      "step": 204
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.04675828292965889,
+      "learning_rate": 0.00017662999351949278,
+      "loss": 0.3711,
+      "step": 205
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.04258272796869278,
+      "learning_rate": 0.00017639961509549078,
+      "loss": 0.3782,
+      "step": 206
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.04638506844639778,
+      "learning_rate": 0.00017616825853058443,
+      "loss": 0.3592,
+      "step": 207
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 0.04781416058540344,
+      "learning_rate": 0.00017593592678682166,
+      "loss": 0.383,
+      "step": 208
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.04813629388809204,
+      "learning_rate": 0.00017570262283873552,
+      "loss": 0.3775,
+      "step": 209
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.046996332705020905,
+      "learning_rate": 0.00017546834967330617,
+      "loss": 0.3815,
+      "step": 210
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.04889595881104469,
+      "learning_rate": 0.00017523311028992268,
+      "loss": 0.3636,
+      "step": 211
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.04298345744609833,
+      "learning_rate": 0.00017499690770034443,
+      "loss": 0.3672,
+      "step": 212
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.04219110682606697,
+      "learning_rate": 0.00017475974492866278,
+      "loss": 0.3801,
+      "step": 213
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.051573265343904495,
+      "learning_rate": 0.00017452162501126227,
+      "loss": 0.3778,
+      "step": 214
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.048954349011182785,
+      "learning_rate": 0.00017428255099678167,
+      "loss": 0.3849,
+      "step": 215
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.042610183358192444,
+      "learning_rate": 0.0001740425259460751,
+      "loss": 0.3682,
+      "step": 216
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.04517417773604393,
+      "learning_rate": 0.00017380155293217264,
+      "loss": 0.3827,
+      "step": 217
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.04968888312578201,
+      "learning_rate": 0.00017355963504024123,
+      "loss": 0.3821,
+      "step": 218
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.051313381642103195,
+      "learning_rate": 0.0001733167753675449,
+      "loss": 0.381,
+      "step": 219
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.044351302087306976,
+      "learning_rate": 0.0001730729770234054,
+      "loss": 0.381,
+      "step": 220
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.03970547392964363,
+      "learning_rate": 0.00017282824312916218,
+      "loss": 0.3698,
+      "step": 221
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.04822370782494545,
+      "learning_rate": 0.00017258257681813244,
+      "loss": 0.3838,
+      "step": 222
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.045927174389362335,
+      "learning_rate": 0.0001723359812355712,
+      "loss": 0.3662,
+      "step": 223
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.042983219027519226,
+      "learning_rate": 0.00017208845953863076,
+      "loss": 0.3574,
+      "step": 224
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.0422198586165905,
+      "learning_rate": 0.0001718400148963206,
+      "loss": 0.3559,
+      "step": 225
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 0.042307570576667786,
+      "learning_rate": 0.00017159065048946644,
+      "loss": 0.3834,
+      "step": 226
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.04701109230518341,
+      "learning_rate": 0.0001713403695106698,
+      "loss": 0.3718,
+      "step": 227
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.04007503017783165,
+      "learning_rate": 0.00017108917516426704,
+      "loss": 0.3785,
+      "step": 228
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.04560061916708946,
+      "learning_rate": 0.00017083707066628832,
+      "loss": 0.3713,
+      "step": 229
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.04315731301903725,
+      "learning_rate": 0.00017058405924441636,
+      "loss": 0.3702,
+      "step": 230
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.040260497480630875,
+      "learning_rate": 0.0001703301441379453,
+      "loss": 0.367,
+      "step": 231
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.04882992431521416,
+      "learning_rate": 0.000170075328597739,
+      "loss": 0.3737,
+      "step": 232
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.04410382732748985,
+      "learning_rate": 0.0001698196158861896,
+      "loss": 0.3625,
+      "step": 233
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.04889338091015816,
+      "learning_rate": 0.00016956300927717575,
+      "loss": 0.3697,
+      "step": 234
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 0.044603537768125534,
+      "learning_rate": 0.00016930551205602043,
+      "loss": 0.3729,
+      "step": 235
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.0539550743997097,
+      "learning_rate": 0.00016904712751944931,
+      "loss": 0.3625,
+      "step": 236
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.04753349721431732,
+      "learning_rate": 0.00016878785897554818,
+      "loss": 0.3662,
+      "step": 237
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.04425463080406189,
+      "learning_rate": 0.0001685277097437208,
+      "loss": 0.3595,
+      "step": 238
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.04160892590880394,
+      "learning_rate": 0.0001682666831546463,
+      "loss": 0.3679,
+      "step": 239
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.04600003361701965,
+      "learning_rate": 0.0001680047825502366,
+      "loss": 0.3702,
+      "step": 240
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.03887678310275078,
+      "learning_rate": 0.00016774201128359357,
+      "loss": 0.3633,
+      "step": 241
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.0477156862616539,
+      "learning_rate": 0.00016747837271896622,
+      "loss": 0.3702,
+      "step": 242
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.04244072362780571,
+      "learning_rate": 0.00016721387023170737,
+      "loss": 0.3668,
+      "step": 243
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 0.04049496725201607,
+      "learning_rate": 0.0001669485072082308,
+      "loss": 0.3785,
+      "step": 244
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.04432998597621918,
+      "learning_rate": 0.00016668228704596756,
+      "loss": 0.3703,
+      "step": 245
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.0432085320353508,
+      "learning_rate": 0.00016641521315332265,
+      "loss": 0.3615,
+      "step": 246
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.03820549696683884,
+      "learning_rate": 0.00016614728894963135,
+      "loss": 0.3483,
+      "step": 247
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.04436295107007027,
+      "learning_rate": 0.00016587851786511543,
+      "loss": 0.3661,
+      "step": 248
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.04371733218431473,
+      "learning_rate": 0.00016560890334083926,
+      "loss": 0.3503,
+      "step": 249
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.039205193519592285,
+      "learning_rate": 0.00016533844882866568,
+      "loss": 0.3482,
+      "step": 250
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.04308384284377098,
+      "learning_rate": 0.00016506715779121187,
+      "loss": 0.373,
+      "step": 251
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.040143441408872604,
+      "learning_rate": 0.00016479503370180507,
+      "loss": 0.3609,
+      "step": 252
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 0.03845199570059776,
+      "learning_rate": 0.000164522080044438,
+      "loss": 0.3644,
+      "step": 253
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.039730221033096313,
+      "learning_rate": 0.00016424830031372425,
+      "loss": 0.3514,
+      "step": 254
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.04021477699279785,
+      "learning_rate": 0.00016397369801485366,
+      "loss": 0.3566,
+      "step": 255
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.03929613530635834,
+      "learning_rate": 0.00016369827666354745,
+      "loss": 0.3649,
+      "step": 256
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.040966227650642395,
+      "learning_rate": 0.0001634220397860129,
+      "loss": 0.3661,
+      "step": 257
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.036787159740924835,
+      "learning_rate": 0.0001631449909188987,
+      "loss": 0.3572,
+      "step": 258
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.039864350110292435,
+      "learning_rate": 0.00016286713360924918,
+      "loss": 0.3593,
+      "step": 259
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.039622630923986435,
+      "learning_rate": 0.00016258847141445928,
+      "loss": 0.3711,
+      "step": 260
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.03840857744216919,
+      "learning_rate": 0.00016230900790222878,
+      "loss": 0.3537,
+      "step": 261
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 0.03800921142101288,
+      "learning_rate": 0.00016202874665051674,
+      "loss": 0.3662,
+      "step": 262
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.03894530236721039,
+      "learning_rate": 0.0001617476912474956,
+      "loss": 0.3633,
+      "step": 263
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.04486812278628349,
+      "learning_rate": 0.00016146584529150526,
+      "loss": 0.3594,
+      "step": 264
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.0394977331161499,
+      "learning_rate": 0.00016118321239100712,
+      "loss": 0.3473,
+      "step": 265
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.03960977867245674,
+      "learning_rate": 0.0001608997961645377,
+      "loss": 0.363,
+      "step": 266
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.04547852650284767,
+      "learning_rate": 0.00016061560024066248,
+      "loss": 0.3698,
+      "step": 267
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.045081403106451035,
+      "learning_rate": 0.00016033062825792935,
+      "loss": 0.3684,
+      "step": 268
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.03854916989803314,
+      "learning_rate": 0.00016004488386482205,
+      "loss": 0.3467,
+      "step": 269
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.042972736060619354,
+      "learning_rate": 0.0001597583707197134,
+      "loss": 0.3589,
+      "step": 270
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 0.0463341549038887,
+      "learning_rate": 0.0001594710924908186,
+      "loss": 0.3517,
+      "step": 271
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.04469626024365425,
+      "learning_rate": 0.00015918305285614822,
+      "loss": 0.3509,
+      "step": 272
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.037887342274188995,
+      "learning_rate": 0.0001588942555034609,
+      "loss": 0.3622,
+      "step": 273
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.046203695237636566,
+      "learning_rate": 0.00015860470413021642,
+      "loss": 0.3587,
+      "step": 274
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.043443821370601654,
+      "learning_rate": 0.00015831440244352832,
+      "loss": 0.3742,
+      "step": 275
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.037403471767902374,
+      "learning_rate": 0.00015802335416011625,
+      "loss": 0.357,
+      "step": 276
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.041037216782569885,
+      "learning_rate": 0.00015773156300625857,
+      "loss": 0.347,
+      "step": 277
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.04024377465248108,
+      "learning_rate": 0.00015743903271774455,
+      "loss": 0.3584,
+      "step": 278
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 0.04302544146776199,
+      "learning_rate": 0.0001571457670398266,
+      "loss": 0.3754,
+      "step": 279
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.04049861803650856,
+      "learning_rate": 0.00015685176972717223,
+      "loss": 0.3617,
+      "step": 280
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.04286986216902733,
+      "learning_rate": 0.0001565570445438161,
+      "loss": 0.3459,
+      "step": 281
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.04489251226186752,
+      "learning_rate": 0.00015626159526311174,
+      "loss": 0.342,
+      "step": 282
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.039040662348270416,
+      "learning_rate": 0.00015596542566768327,
+      "loss": 0.3579,
+      "step": 283
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.041434600949287415,
+      "learning_rate": 0.00015566853954937694,
+      "loss": 0.3607,
+      "step": 284
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.040113747119903564,
+      "learning_rate": 0.00015537094070921267,
+      "loss": 0.3581,
+      "step": 285
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.047135982662439346,
+      "learning_rate": 0.00015507263295733528,
+      "loss": 0.355,
+      "step": 286
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.04679419472813606,
+      "learning_rate": 0.00015477362011296575,
+      "loss": 0.3506,
+      "step": 287
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.04256124794483185,
+      "learning_rate": 0.00015447390600435238,
+      "loss": 0.3547,
+      "step": 288
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.041164420545101166,
+      "learning_rate": 0.00015417349446872168,
+      "loss": 0.3533,
+      "step": 289
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.04434530436992645,
+      "learning_rate": 0.00015387238935222927,
+      "loss": 0.3655,
+      "step": 290
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.042684029787778854,
+      "learning_rate": 0.00015357059450991068,
+      "loss": 0.3388,
+      "step": 291
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.03664594888687134,
+      "learning_rate": 0.00015326811380563204,
+      "loss": 0.3415,
+      "step": 292
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.044670287519693375,
+      "learning_rate": 0.0001529649511120404,
+      "loss": 0.3639,
+      "step": 293
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.043318092823028564,
+      "learning_rate": 0.00015266111031051442,
+      "loss": 0.3705,
+      "step": 294
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.03925277665257454,
+      "learning_rate": 0.00015235659529111445,
+      "loss": 0.3452,
+      "step": 295
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.048936985433101654,
+      "learning_rate": 0.00015205140995253283,
+      "loss": 0.3511,
+      "step": 296
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 0.04244200140237808,
+      "learning_rate": 0.00015174555820204408,
+      "loss": 0.3573,
+      "step": 297
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.041375573724508286,
+      "learning_rate": 0.00015143904395545466,
+      "loss": 0.3474,
+      "step": 298
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.04903846234083176,
+      "learning_rate": 0.0001511318711370529,
+      "loss": 0.3677,
+      "step": 299
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.03968316316604614,
+      "learning_rate": 0.0001508240436795589,
+      "loss": 0.3504,
+      "step": 300
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.038755469024181366,
+      "learning_rate": 0.00015051556552407399,
+      "loss": 0.3594,
+      "step": 301
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.03975263237953186,
+      "learning_rate": 0.00015020644062003046,
+      "loss": 0.3536,
+      "step": 302
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.03559111803770065,
+      "learning_rate": 0.0001498966729251408,
+      "loss": 0.3489,
+      "step": 303
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.0421811044216156,
+      "learning_rate": 0.0001495862664053471,
+      "loss": 0.3558,
+      "step": 304
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.04090610146522522,
+      "learning_rate": 0.00014927522503477048,
+      "loss": 0.3532,
+      "step": 305
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 0.04141204059123993,
+      "learning_rate": 0.00014896355279565976,
+      "loss": 0.361,
+      "step": 306
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.039628610014915466,
+      "learning_rate": 0.00014865125367834092,
+      "loss": 0.353,
+      "step": 307
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.040805645287036896,
+      "learning_rate": 0.00014833833168116582,
+      "loss": 0.3442,
+      "step": 308
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.03766808658838272,
+      "learning_rate": 0.00014802479081046102,
+      "loss": 0.3429,
+      "step": 309
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.03960324451327324,
+      "learning_rate": 0.00014771063508047636,
+      "loss": 0.3641,
+      "step": 310
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.03804844617843628,
+      "learning_rate": 0.0001473958685133339,
+      "loss": 0.3462,
+      "step": 311
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.03872362896800041,
+      "learning_rate": 0.0001470804951389761,
+      "loss": 0.3529,
+      "step": 312
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.038420502096414566,
+      "learning_rate": 0.00014676451899511437,
+      "loss": 0.3594,
+      "step": 313
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.038893669843673706,
+      "learning_rate": 0.00014644794412717736,
+      "loss": 0.3461,
+      "step": 314
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 0.039783939719200134,
+      "learning_rate": 0.00014613077458825913,
+      "loss": 0.3641,
+      "step": 315
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.040646251291036606,
+      "learning_rate": 0.0001458130144390673,
+      "loss": 0.3408,
+      "step": 316
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.03997185081243515,
+      "learning_rate": 0.00014549466774787108,
+      "loss": 0.3476,
+      "step": 317
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.03733928129076958,
+      "learning_rate": 0.00014517573859044907,
+      "loss": 0.348,
+      "step": 318
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.04240603372454643,
+      "learning_rate": 0.00014485623105003732,
+      "loss": 0.3526,
+      "step": 319
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.04083820432424545,
+      "learning_rate": 0.00014453614921727668,
+      "loss": 0.358,
+      "step": 320
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.038078900426626205,
+      "learning_rate": 0.00014421549719016081,
+      "loss": 0.3383,
+      "step": 321
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.043162036687135696,
+      "learning_rate": 0.00014389427907398342,
+      "loss": 0.3502,
+      "step": 322
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.04186828061938286,
+      "learning_rate": 0.00014357249898128594,
+      "loss": 0.355,
+      "step": 323
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 0.041309986263513565,
+      "learning_rate": 0.0001432501610318047,
+      "loss": 0.3634,
+      "step": 324
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.03907867148518562,
+      "learning_rate": 0.00014292726935241827,
+      "loss": 0.3431,
+      "step": 325
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.04641426354646683,
+      "learning_rate": 0.00014260382807709457,
+      "loss": 0.3423,
+      "step": 326
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.042708221822977066,
+      "learning_rate": 0.000142279841346838,
+      "loss": 0.3459,
+      "step": 327
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.03900787606835365,
+      "learning_rate": 0.00014195531330963635,
+      "loss": 0.349,
+      "step": 328
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.04202708601951599,
+      "learning_rate": 0.0001416302481204078,
+      "loss": 0.3478,
+      "step": 329
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.04631033539772034,
+      "learning_rate": 0.0001413046499409477,
+      "loss": 0.3495,
+      "step": 330
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.04463302344083786,
+      "learning_rate": 0.00014097852293987507,
+      "loss": 0.3422,
+      "step": 331
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.03706689178943634,
+      "learning_rate": 0.00014065187129257964,
+      "loss": 0.3558,
+      "step": 332
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.04193353280425072,
+      "learning_rate": 0.000140324699181168,
+      "loss": 0.3539,
+      "step": 333
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.03598970174789429,
+      "learning_rate": 0.00013999701079441028,
+      "loss": 0.3412,
+      "step": 334
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.04364067688584328,
+      "learning_rate": 0.00013966881032768643,
+      "loss": 0.3496,
+      "step": 335
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.034331586211919785,
+      "learning_rate": 0.00013934010198293257,
+      "loss": 0.3495,
+      "step": 336
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.04461033269762993,
+      "learning_rate": 0.00013901088996858717,
+      "loss": 0.3555,
+      "step": 337
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.037957970052957535,
+      "learning_rate": 0.0001386811784995371,
+      "loss": 0.3535,
+      "step": 338
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.04143986850976944,
+      "learning_rate": 0.0001383509717970638,
+      "loss": 0.3552,
+      "step": 339
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.04176801070570946,
+      "learning_rate": 0.0001380202740887891,
+      "loss": 0.3462,
+      "step": 340
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.038572393357753754,
+      "learning_rate": 0.00013768908960862123,
+      "loss": 0.3582,
+      "step": 341
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.047118738293647766,
+      "learning_rate": 0.0001373574225967004,
+      "loss": 0.3451,
+      "step": 342
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.042480722069740295,
+      "learning_rate": 0.00013702527729934482,
+      "loss": 0.3355,
+      "step": 343
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.04277138039469719,
+      "learning_rate": 0.00013669265796899607,
+      "loss": 0.3395,
+      "step": 344
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.04838969558477402,
+      "learning_rate": 0.0001363595688641648,
+      "loss": 0.3419,
+      "step": 345
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.04388457536697388,
+      "learning_rate": 0.00013602601424937604,
+      "loss": 0.3412,
+      "step": 346
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.041595809161663055,
+      "learning_rate": 0.00013569199839511494,
+      "loss": 0.3488,
+      "step": 347
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.045120954513549805,
+      "learning_rate": 0.0001353575255777717,
+      "loss": 0.3486,
+      "step": 348
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.04108583182096481,
+      "learning_rate": 0.00013502260007958706,
+      "loss": 0.3522,
+      "step": 349
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.046396516263484955,
+      "learning_rate": 0.00013468722618859743,
+      "loss": 0.3474,
+      "step": 350
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.04111646115779877,
+      "learning_rate": 0.0001343514081985799,
+      "loss": 0.3522,
+      "step": 351
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.03970329090952873,
+      "learning_rate": 0.00013401515040899746,
+      "loss": 0.3504,
+      "step": 352
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.041780147701501846,
+      "learning_rate": 0.00013367845712494372,
+      "loss": 0.3464,
+      "step": 353
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.03978356719017029,
+      "learning_rate": 0.000133341332657088,
+      "loss": 0.3473,
+      "step": 354
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.04411393776535988,
+      "learning_rate": 0.00013300378132161992,
+      "loss": 0.3451,
+      "step": 355
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.04426994547247887,
+      "learning_rate": 0.00013266580744019445,
+      "loss": 0.3366,
+      "step": 356
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.037810515612363815,
+      "learning_rate": 0.00013232741533987622,
+      "loss": 0.342,
+      "step": 357
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.04545320197939873,
+      "learning_rate": 0.00013198860935308444,
+      "loss": 0.3553,
+      "step": 358
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.04098277539014816,
+      "learning_rate": 0.00013164939381753713,
+      "loss": 0.3458,
+      "step": 359
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.04471420869231224,
+      "learning_rate": 0.00013130977307619594,
+      "loss": 0.3318,
+      "step": 360
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.038031481206417084,
+      "learning_rate": 0.00013096975147721017,
+      "loss": 0.3533,
+      "step": 361
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.0440196692943573,
+      "learning_rate": 0.00013062933337386142,
+      "loss": 0.3299,
+      "step": 362
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.034114640206098557,
+      "learning_rate": 0.00013028852312450763,
+      "loss": 0.3401,
+      "step": 363
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.042403046041727066,
+      "learning_rate": 0.00012994732509252744,
+      "loss": 0.3458,
+      "step": 364
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.0356055311858654,
+      "learning_rate": 0.00012960574364626412,
+      "loss": 0.3498,
+      "step": 365
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.040617913007736206,
+      "learning_rate": 0.00012926378315896998,
+      "loss": 0.3454,
+      "step": 366
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.03738779574632645,
+      "learning_rate": 0.00012892144800875,
+      "loss": 0.342,
+      "step": 367
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 0.039146389812231064,
+      "learning_rate": 0.00012857874257850605,
+      "loss": 0.35,
+      "step": 368
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.03738423436880112,
+      "learning_rate": 0.0001282356712558806,
+      "loss": 0.3503,
+      "step": 369
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.03822367265820503,
+      "learning_rate": 0.00012789223843320073,
+      "loss": 0.3489,
+      "step": 370
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.03619171679019928,
+      "learning_rate": 0.00012754844850742172,
+      "loss": 0.3392,
+      "step": 371
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.03672062233090401,
+      "learning_rate": 0.00012720430588007077,
+      "loss": 0.3503,
+      "step": 372
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.038667939603328705,
+      "learning_rate": 0.00012685981495719087,
+      "loss": 0.3582,
+      "step": 373
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.033767540007829666,
+      "learning_rate": 0.00012651498014928402,
+      "loss": 0.3465,
+      "step": 374
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.0382758267223835,
+      "learning_rate": 0.00012616980587125512,
+      "loss": 0.3435,
+      "step": 375
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.03528289496898651,
+      "learning_rate": 0.00012582429654235523,
+      "loss": 0.3427,
+      "step": 376
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 0.03565202280879021,
+      "learning_rate": 0.00012547845658612508,
+      "loss": 0.3453,
+      "step": 377
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.03661198914051056,
+      "learning_rate": 0.0001251322904303383,
+      "loss": 0.3341,
+      "step": 378
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.03750570863485336,
+      "learning_rate": 0.00012478580250694504,
+      "loss": 0.3375,
+      "step": 379
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.03714427724480629,
+      "learning_rate": 0.00012443899725201482,
+      "loss": 0.3447,
+      "step": 380
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.03938335180282593,
+      "learning_rate": 0.0001240918791056801,
+      "loss": 0.3551,
+      "step": 381
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.037106383591890335,
+      "learning_rate": 0.00012374445251207914,
+      "loss": 0.3365,
+      "step": 382
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.039744965732097626,
+      "learning_rate": 0.00012339672191929936,
+      "loss": 0.3333,
+      "step": 383
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.04372095316648483,
+      "learning_rate": 0.0001230486917793202,
+      "loss": 0.344,
+      "step": 384
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.03674423322081566,
+      "learning_rate": 0.00012270036654795613,
+      "loss": 0.3269,
+      "step": 385
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 0.03725577890872955,
+      "learning_rate": 0.00012235175068479984,
+      "loss": 0.3394,
+      "step": 386
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.037114035338163376,
+      "learning_rate": 0.00012200284865316475,
+      "loss": 0.3411,
+      "step": 387
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.037100065499544144,
+      "learning_rate": 0.00012165366492002832,
+      "loss": 0.3351,
+      "step": 388
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.042357347905635834,
+      "learning_rate": 0.00012130420395597437,
+      "loss": 0.3458,
+      "step": 389
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.03712886944413185,
+      "learning_rate": 0.0001209544702351363,
+      "loss": 0.3385,
+      "step": 390
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.03989335149526596,
+      "learning_rate": 0.00012060446823513949,
+      "loss": 0.3431,
+      "step": 391
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.03781385347247124,
+      "learning_rate": 0.0001202542024370441,
+      "loss": 0.3444,
+      "step": 392
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.03825078904628754,
+      "learning_rate": 0.00011990367732528773,
+      "loss": 0.3394,
+      "step": 393
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.03969776630401611,
+      "learning_rate": 0.00011955289738762796,
+      "loss": 0.3444,
+      "step": 394
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 0.04011238366365433,
+      "learning_rate": 0.00011920186711508479,
+      "loss": 0.3466,
+      "step": 395
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.03450911492109299,
+      "learning_rate": 0.00011885059100188341,
+      "loss": 0.337,
+      "step": 396
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.038461893796920776,
+      "learning_rate": 0.00011849907354539633,
+      "loss": 0.3347,
+      "step": 397
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.037344567477703094,
+      "learning_rate": 0.00011814731924608616,
+      "loss": 0.3452,
+      "step": 398
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.03558855876326561,
+      "learning_rate": 0.00011779533260744757,
+      "loss": 0.3401,
+      "step": 399
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.04183749854564667,
+      "learning_rate": 0.00011744311813595006,
+      "loss": 0.3421,
+      "step": 400
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.03574857488274574,
+      "learning_rate": 0.00011709068034097997,
+      "loss": 0.3271,
+      "step": 401
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.03993195295333862,
+      "learning_rate": 0.0001167380237347828,
+      "loss": 0.3267,
+      "step": 402
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.03541812300682068,
+      "learning_rate": 0.0001163851528324056,
+      "loss": 0.3347,
+      "step": 403
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 0.037207312881946564,
+      "learning_rate": 0.00011603207215163894,
+      "loss": 0.3418,
+      "step": 404
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.03919894993305206,
+      "learning_rate": 0.00011567878621295922,
+      "loss": 0.3516,
+      "step": 405
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.0382562056183815,
+      "learning_rate": 0.00011532529953947075,
+      "loss": 0.3499,
+      "step": 406
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.03956792503595352,
+      "learning_rate": 0.00011497161665684784,
+      "loss": 0.347,
+      "step": 407
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.03666992112994194,
+      "learning_rate": 0.0001146177420932768,
+      "loss": 0.3292,
+      "step": 408
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.03818726912140846,
+      "learning_rate": 0.00011426368037939813,
+      "loss": 0.346,
+      "step": 409
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.0352170504629612,
+      "learning_rate": 0.00011390943604824826,
+      "loss": 0.3253,
+      "step": 410
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.04157968610525131,
+      "learning_rate": 0.00011355501363520185,
+      "loss": 0.3411,
+      "step": 411
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.037365157157182693,
+      "learning_rate": 0.00011320041767791336,
+      "loss": 0.329,
+      "step": 412
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 0.03966519236564636,
+      "learning_rate": 0.00011284565271625922,
+      "loss": 0.3412,
+      "step": 413
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.037126705050468445,
+      "learning_rate": 0.00011249072329227959,
+      "loss": 0.3387,
+      "step": 414
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.03643345460295677,
+      "learning_rate": 0.00011213563395012026,
+      "loss": 0.3392,
+      "step": 415
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.03495652228593826,
+      "learning_rate": 0.0001117803892359744,
+      "loss": 0.3327,
+      "step": 416
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.03979011997580528,
+      "learning_rate": 0.00011142499369802444,
+      "loss": 0.3298,
+      "step": 417
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.03582910820841789,
+      "learning_rate": 0.00011106945188638378,
+      "loss": 0.3424,
+      "step": 418
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.03996811434626579,
+      "learning_rate": 0.00011071376835303858,
+      "loss": 0.3335,
+      "step": 419
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.039901167154312134,
+      "learning_rate": 0.00011035794765178941,
+      "loss": 0.3333,
+      "step": 420
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.03835071623325348,
+      "learning_rate": 0.00011000199433819305,
+      "loss": 0.328,
+      "step": 421
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 0.03873496130108833,
+      "learning_rate": 0.00010964591296950406,
+      "loss": 0.3414,
+      "step": 422
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.03686109930276871,
+      "learning_rate": 0.00010928970810461652,
+      "loss": 0.3393,
+      "step": 423
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.038315027952194214,
+      "learning_rate": 0.00010893338430400562,
+      "loss": 0.3349,
+      "step": 424
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.036203257739543915,
+      "learning_rate": 0.0001085769461296692,
+      "loss": 0.332,
+      "step": 425
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.037686560302972794,
+      "learning_rate": 0.00010822039814506964,
+      "loss": 0.3437,
+      "step": 426
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.0374578982591629,
+      "learning_rate": 0.00010786374491507494,
+      "loss": 0.3424,
+      "step": 427
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.03514384478330612,
+      "learning_rate": 0.00010750699100590076,
+      "loss": 0.3317,
+      "step": 428
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.040791869163513184,
+      "learning_rate": 0.00010715014098505162,
+      "loss": 0.3389,
+      "step": 429
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.03556138649582863,
+      "learning_rate": 0.00010679319942126264,
+      "loss": 0.3424,
+      "step": 430
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.040588121861219406,
+      "learning_rate": 0.00010643617088444094,
+      "loss": 0.3418,
+      "step": 431
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.03443857282400131,
+      "learning_rate": 0.0001060790599456071,
+      "loss": 0.3395,
+      "step": 432
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.038418177515268326,
+      "learning_rate": 0.00010572187117683674,
+      "loss": 0.3325,
+      "step": 433
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.03434626758098602,
+      "learning_rate": 0.0001053646091512019,
+      "loss": 0.3378,
+      "step": 434
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.03498569130897522,
+      "learning_rate": 0.0001050072784427126,
+      "loss": 0.3375,
+      "step": 435
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.03690873831510544,
+      "learning_rate": 0.00010464988362625812,
+      "loss": 0.331,
+      "step": 436
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.040180571377277374,
+      "learning_rate": 0.00010429242927754854,
+      "loss": 0.3394,
+      "step": 437
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.03546585142612457,
+      "learning_rate": 0.00010393491997305613,
+      "loss": 0.3368,
+      "step": 438
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 0.03656655550003052,
+      "learning_rate": 0.00010357736028995677,
+      "loss": 0.332,
+      "step": 439
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.03666246309876442,
+      "learning_rate": 0.00010321975480607129,
+      "loss": 0.3342,
+      "step": 440
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.03500663861632347,
+      "learning_rate": 0.00010286210809980697,
+      "loss": 0.3389,
+      "step": 441
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.03510987013578415,
+      "learning_rate": 0.0001025044247500988,
+      "loss": 0.3338,
+      "step": 442
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.037753257900476456,
+      "learning_rate": 0.00010214670933635095,
+      "loss": 0.333,
+      "step": 443
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.03440658003091812,
+      "learning_rate": 0.00010178896643837809,
+      "loss": 0.3456,
+      "step": 444
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.0362187922000885,
+      "learning_rate": 0.00010143120063634681,
+      "loss": 0.3484,
+      "step": 445
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.034865692257881165,
+      "learning_rate": 0.00010107341651071684,
+      "loss": 0.3392,
+      "step": 446
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.035264838486909866,
+      "learning_rate": 0.00010071561864218262,
+      "loss": 0.3267,
+      "step": 447
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.035725150257349014,
+      "learning_rate": 0.00010035781161161446,
+      "loss": 0.3454,
+      "step": 448
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.03839932009577751,
+      "learning_rate": 0.0001,
+      "loss": 0.3304,
+      "step": 449
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.03366202488541603,
+      "learning_rate": 9.964218838838554e-05,
+      "loss": 0.3288,
+      "step": 450
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.03588353097438812,
+      "learning_rate": 9.92843813578174e-05,
+      "loss": 0.3406,
+      "step": 451
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.034711651504039764,
+      "learning_rate": 9.892658348928316e-05,
+      "loss": 0.3289,
+      "step": 452
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.03611205145716667,
+      "learning_rate": 9.856879936365321e-05,
+      "loss": 0.3294,
+      "step": 453
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.03515039384365082,
+      "learning_rate": 9.821103356162189e-05,
+      "loss": 0.3324,
+      "step": 454
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.03375555947422981,
+      "learning_rate": 9.785329066364906e-05,
+      "loss": 0.3214,
+      "step": 455
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.03730206564068794,
+      "learning_rate": 9.749557524990121e-05,
+      "loss": 0.3258,
+      "step": 456
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 0.033822208642959595,
+      "learning_rate": 9.713789190019304e-05,
+      "loss": 0.3353,
+      "step": 457
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.03683510422706604,
+      "learning_rate": 9.678024519392871e-05,
+      "loss": 0.3264,
+      "step": 458
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.037592969834804535,
+      "learning_rate": 9.642263971004324e-05,
+      "loss": 0.329,
+      "step": 459
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.03537925332784653,
+      "learning_rate": 9.606508002694386e-05,
+      "loss": 0.3448,
+      "step": 460
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.033904388546943665,
+      "learning_rate": 9.570757072245147e-05,
+      "loss": 0.3472,
+      "step": 461
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.036394666880369186,
+      "learning_rate": 9.535011637374189e-05,
+      "loss": 0.3288,
+      "step": 462
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.03391895070672035,
+      "learning_rate": 9.499272155728742e-05,
+      "loss": 0.3368,
+      "step": 463
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.037903424352407455,
+      "learning_rate": 9.463539084879809e-05,
+      "loss": 0.3443,
+      "step": 464
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.033752232789993286,
+      "learning_rate": 9.427812882316329e-05,
+      "loss": 0.3316,
+      "step": 465
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.03620493784546852,
+      "learning_rate": 9.392094005439291e-05,
+      "loss": 0.3318,
+      "step": 466
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.0346236452460289,
+      "learning_rate": 9.356382911555907e-05,
+      "loss": 0.3438,
+      "step": 467
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.03393540903925896,
+      "learning_rate": 9.320680057873735e-05,
+      "loss": 0.3459,
+      "step": 468
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.03790227323770523,
+      "learning_rate": 9.28498590149484e-05,
+      "loss": 0.3405,
+      "step": 469
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.03373325988650322,
+      "learning_rate": 9.249300899409924e-05,
+      "loss": 0.3329,
+      "step": 470
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.03829934075474739,
+      "learning_rate": 9.213625508492508e-05,
+      "loss": 0.3321,
+      "step": 471
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.03592018410563469,
+      "learning_rate": 9.177960185493036e-05,
+      "loss": 0.3186,
+      "step": 472
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.03398677706718445,
+      "learning_rate": 9.142305387033081e-05,
+      "loss": 0.3465,
+      "step": 473
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.03493371605873108,
+      "learning_rate": 9.106661569599442e-05,
+      "loss": 0.3358,
+      "step": 474
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 0.03722028434276581,
+      "learning_rate": 9.071029189538353e-05,
+      "loss": 0.3278,
+      "step": 475
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.033628251403570175,
+      "learning_rate": 9.035408703049596e-05,
+      "loss": 0.3315,
+      "step": 476
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.036100320518016815,
+      "learning_rate": 8.9998005661807e-05,
+      "loss": 0.322,
+      "step": 477
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.03419538959860802,
+      "learning_rate": 8.96420523482106e-05,
+      "loss": 0.3241,
+      "step": 478
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.03728823736310005,
+      "learning_rate": 8.928623164696146e-05,
+      "loss": 0.3312,
+      "step": 479
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.03538035228848457,
+      "learning_rate": 8.893054811361624e-05,
+      "loss": 0.3244,
+      "step": 480
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.036142975091934204,
+      "learning_rate": 8.85750063019756e-05,
+      "loss": 0.3401,
+      "step": 481
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.0373205728828907,
+      "learning_rate": 8.821961076402563e-05,
+      "loss": 0.3334,
+      "step": 482
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.035391587764024734,
+      "learning_rate": 8.786436604987978e-05,
+      "loss": 0.341,
+      "step": 483
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.03749712184071541,
+      "learning_rate": 8.750927670772044e-05,
+      "loss": 0.3179,
+      "step": 484
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.037055715918540955,
+      "learning_rate": 8.715434728374083e-05,
+      "loss": 0.3226,
+      "step": 485
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.03762289509177208,
+      "learning_rate": 8.679958232208668e-05,
+      "loss": 0.337,
+      "step": 486
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.03537070378661156,
+      "learning_rate": 8.644498636479819e-05,
+      "loss": 0.3306,
+      "step": 487
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.03710364177823067,
+      "learning_rate": 8.609056395175175e-05,
+      "loss": 0.3256,
+      "step": 488
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.03595058619976044,
+      "learning_rate": 8.573631962060192e-05,
+      "loss": 0.3337,
+      "step": 489
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.03377842903137207,
+      "learning_rate": 8.538225790672322e-05,
+      "loss": 0.3411,
+      "step": 490
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.036737628281116486,
+      "learning_rate": 8.50283833431522e-05,
+      "loss": 0.3286,
+      "step": 491
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.034495435655117035,
+      "learning_rate": 8.467470046052927e-05,
+      "loss": 0.3345,
+      "step": 492
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 0.036502912640571594,
+      "learning_rate": 8.432121378704081e-05,
+      "loss": 0.3237,
+      "step": 493
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.03458232432603836,
+      "learning_rate": 8.396792784836108e-05,
+      "loss": 0.3215,
+      "step": 494
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.036796458065509796,
+      "learning_rate": 8.361484716759445e-05,
+      "loss": 0.3375,
+      "step": 495
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.03392602503299713,
+      "learning_rate": 8.326197626521723e-05,
+      "loss": 0.335,
+      "step": 496
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.03623703867197037,
+      "learning_rate": 8.290931965902008e-05,
+      "loss": 0.3388,
+      "step": 497
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.03508217632770538,
+      "learning_rate": 8.255688186404996e-05,
+      "loss": 0.3267,
+      "step": 498
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.03563014790415764,
+      "learning_rate": 8.220466739255244e-05,
+      "loss": 0.3277,
+      "step": 499
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.035729020833969116,
+      "learning_rate": 8.185268075391388e-05,
+      "loss": 0.3348,
+      "step": 500
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.034020714461803436,
+      "learning_rate": 8.150092645460366e-05,
+      "loss": 0.32,
+      "step": 501
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.033443402498960495,
+      "learning_rate": 8.114940899811662e-05,
+      "loss": 0.3402,
+      "step": 502
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.03387276828289032,
+      "learning_rate": 8.079813288491521e-05,
+      "loss": 0.3319,
+      "step": 503
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.033886682242155075,
+      "learning_rate": 8.044710261237207e-05,
+      "loss": 0.3307,
+      "step": 504
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.03618338331580162,
+      "learning_rate": 8.009632267471227e-05,
+      "loss": 0.3286,
+      "step": 505
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.03527137264609337,
+      "learning_rate": 7.974579756295591e-05,
+      "loss": 0.3171,
+      "step": 506
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.036220818758010864,
+      "learning_rate": 7.939553176486052e-05,
+      "loss": 0.3396,
+      "step": 507
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.035215239971876144,
+      "learning_rate": 7.904552976486372e-05,
+      "loss": 0.3226,
+      "step": 508
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.0399019829928875,
+      "learning_rate": 7.869579604402562e-05,
+      "loss": 0.3245,
+      "step": 509
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.03600882738828659,
+      "learning_rate": 7.83463350799717e-05,
+      "loss": 0.3217,
+      "step": 510
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 0.04228314757347107,
+      "learning_rate": 7.799715134683523e-05,
+      "loss": 0.3282,
+      "step": 511
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.03412134200334549,
+      "learning_rate": 7.764824931520018e-05,
+      "loss": 0.3297,
+      "step": 512
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.03876680135726929,
+      "learning_rate": 7.729963345204386e-05,
+      "loss": 0.3232,
+      "step": 513
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.034336164593696594,
+      "learning_rate": 7.695130822067984e-05,
+      "loss": 0.32,
+      "step": 514
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.0393151193857193,
+      "learning_rate": 7.660327808070064e-05,
+      "loss": 0.3321,
+      "step": 515
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.03484569489955902,
+      "learning_rate": 7.625554748792085e-05,
+      "loss": 0.3315,
+      "step": 516
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.042867034673690796,
+      "learning_rate": 7.59081208943199e-05,
+      "loss": 0.3294,
+      "step": 517
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.03290384262800217,
+      "learning_rate": 7.556100274798519e-05,
+      "loss": 0.3284,
+      "step": 518
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 0.042282383888959885,
+      "learning_rate": 7.521419749305497e-05,
+      "loss": 0.3304,
+      "step": 519
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.03494780510663986,
+      "learning_rate": 7.486770956966171e-05,
+      "loss": 0.3378,
+      "step": 520
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.04240773990750313,
+      "learning_rate": 7.452154341387493e-05,
+      "loss": 0.3325,
+      "step": 521
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.037106744945049286,
+      "learning_rate": 7.417570345764481e-05,
+      "loss": 0.3267,
+      "step": 522
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.0382254458963871,
+      "learning_rate": 7.383019412874489e-05,
+      "loss": 0.3225,
+      "step": 523
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.037587832659482956,
+      "learning_rate": 7.348501985071603e-05,
+      "loss": 0.3197,
+      "step": 524
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.03936656191945076,
+      "learning_rate": 7.314018504280917e-05,
+      "loss": 0.3268,
+      "step": 525
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.037244342267513275,
+      "learning_rate": 7.279569411992926e-05,
+      "loss": 0.3254,
+      "step": 526
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.03609207645058632,
+      "learning_rate": 7.24515514925783e-05,
+      "loss": 0.3388,
+      "step": 527
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 0.03564932942390442,
+      "learning_rate": 7.210776156679931e-05,
+      "loss": 0.3307,
+      "step": 528
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.03292916342616081,
+      "learning_rate": 7.176432874411941e-05,
+      "loss": 0.3257,
+      "step": 529
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.0374549925327301,
+      "learning_rate": 7.1421257421494e-05,
+      "loss": 0.3417,
+      "step": 530
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.03253106027841568,
+      "learning_rate": 7.107855199125002e-05,
+      "loss": 0.3397,
+      "step": 531
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.035386502742767334,
+      "learning_rate": 7.073621684103007e-05,
+      "loss": 0.3329,
+      "step": 532
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.032128430902957916,
+      "learning_rate": 7.039425635373589e-05,
+      "loss": 0.3314,
+      "step": 533
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.03548587113618851,
+      "learning_rate": 7.005267490747263e-05,
+      "loss": 0.3267,
+      "step": 534
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.034080877900123596,
+      "learning_rate": 6.971147687549239e-05,
+      "loss": 0.3276,
+      "step": 535
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.03397037461400032,
+      "learning_rate": 6.937066662613863e-05,
+      "loss": 0.3326,
+      "step": 536
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 0.03224964439868927,
+      "learning_rate": 6.903024852278985e-05,
+      "loss": 0.3159,
+      "step": 537
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.0338578037917614,
+      "learning_rate": 6.869022692380411e-05,
+      "loss": 0.3276,
+      "step": 538
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.03182509168982506,
+      "learning_rate": 6.835060618246289e-05,
+      "loss": 0.3293,
+      "step": 539
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.03487447649240494,
+      "learning_rate": 6.801139064691562e-05,
+      "loss": 0.3266,
+      "step": 540
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.03520505875349045,
+      "learning_rate": 6.76725846601238e-05,
+      "loss": 0.3172,
+      "step": 541
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.0320887491106987,
+      "learning_rate": 6.733419255980559e-05,
+      "loss": 0.3267,
+      "step": 542
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.03487367928028107,
+      "learning_rate": 6.699621867838008e-05,
+      "loss": 0.3243,
+      "step": 543
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.03125692531466484,
+      "learning_rate": 6.665866734291205e-05,
+      "loss": 0.3196,
+      "step": 544
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.0348886102437973,
+      "learning_rate": 6.63215428750563e-05,
+      "loss": 0.3331,
+      "step": 545
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 0.03260529041290283,
+      "learning_rate": 6.598484959100257e-05,
+      "loss": 0.3217,
+      "step": 546
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.03670089319348335,
+      "learning_rate": 6.564859180142012e-05,
+      "loss": 0.313,
+      "step": 547
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.03376466780900955,
+      "learning_rate": 6.53127738114026e-05,
+      "loss": 0.332,
+      "step": 548
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.0370427742600441,
+      "learning_rate": 6.497739992041296e-05,
+      "loss": 0.3107,
+      "step": 549
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.0331474132835865,
+      "learning_rate": 6.46424744222283e-05,
+      "loss": 0.3226,
+      "step": 550
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.03344728797674179,
+      "learning_rate": 6.430800160488508e-05,
+      "loss": 0.3164,
+      "step": 551
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.03475061431527138,
+      "learning_rate": 6.397398575062396e-05,
+      "loss": 0.3283,
+      "step": 552
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.03373516723513603,
+      "learning_rate": 6.364043113583524e-05,
+      "loss": 0.3245,
+      "step": 553
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.03636647015810013,
+      "learning_rate": 6.330734203100394e-05,
+      "loss": 0.3344,
+      "step": 554
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 0.03600539267063141,
+      "learning_rate": 6.297472270065519e-05,
+      "loss": 0.3211,
+      "step": 555
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.03470810502767563,
+      "learning_rate": 6.26425774032996e-05,
+      "loss": 0.3227,
+      "step": 556
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.03501768782734871,
+      "learning_rate": 6.231091039137881e-05,
+      "loss": 0.3402,
+      "step": 557
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.03468218818306923,
+      "learning_rate": 6.19797259112109e-05,
+      "loss": 0.3404,
+      "step": 558
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.03609640896320343,
+      "learning_rate": 6.164902820293621e-05,
+      "loss": 0.3288,
+      "step": 559
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.03670278191566467,
+      "learning_rate": 6.131882150046291e-05,
+      "loss": 0.3202,
+      "step": 560
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.036678191274404526,
+      "learning_rate": 6.0989110031412856e-05,
+      "loss": 0.3256,
+      "step": 561
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.03913621976971626,
+      "learning_rate": 6.065989801706744e-05,
+      "loss": 0.3219,
+      "step": 562
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.03633202239871025,
+      "learning_rate": 6.033118967231359e-05,
+      "loss": 0.3227,
+      "step": 563
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 0.03726629912853241,
+      "learning_rate": 6.0002989205589734e-05,
+      "loss": 0.3221,
+      "step": 564
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.03927043080329895,
+      "learning_rate": 5.9675300818832016e-05,
+      "loss": 0.333,
+      "step": 565
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.032321300357580185,
+      "learning_rate": 5.934812870742036e-05,
+      "loss": 0.3251,
+      "step": 566
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.039472486823797226,
+      "learning_rate": 5.9021477060124954e-05,
+      "loss": 0.3333,
+      "step": 567
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.03381982445716858,
+      "learning_rate": 5.869535005905232e-05,
+      "loss": 0.3456,
+      "step": 568
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.03533458709716797,
+      "learning_rate": 5.836975187959223e-05,
+      "loss": 0.3283,
+      "step": 569
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.03490634635090828,
+      "learning_rate": 5.804468669036369e-05,
+      "loss": 0.3259,
+      "step": 570
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.033531077206134796,
+      "learning_rate": 5.772015865316205e-05,
+      "loss": 0.3236,
+      "step": 571
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.03854377195239067,
+      "learning_rate": 5.739617192290545e-05,
+      "loss": 0.3196,
+      "step": 572
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.03259904310107231,
+      "learning_rate": 5.707273064758178e-05,
+      "loss": 0.322,
+      "step": 573
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.034173525869846344,
+      "learning_rate": 5.6749838968195326e-05,
+      "loss": 0.3316,
+      "step": 574
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.03369203209877014,
+      "learning_rate": 5.642750101871408e-05,
+      "loss": 0.316,
+      "step": 575
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.034140925854444504,
+      "learning_rate": 5.610572092601659e-05,
+      "loss": 0.3129,
+      "step": 576
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.034850213676691055,
+      "learning_rate": 5.578450280983924e-05,
+      "loss": 0.3429,
+      "step": 577
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.036979470402002335,
+      "learning_rate": 5.5463850782723346e-05,
+      "loss": 0.316,
+      "step": 578
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.03544802591204643,
+      "learning_rate": 5.51437689499627e-05,
+      "loss": 0.3234,
+      "step": 579
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.03452309966087341,
+      "learning_rate": 5.48242614095509e-05,
+      "loss": 0.3299,
+      "step": 580
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.032840970903635025,
+      "learning_rate": 5.450533225212896e-05,
+      "loss": 0.3274,
+      "step": 581
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.034303292632102966,
+      "learning_rate": 5.418698556093271e-05,
+      "loss": 0.3326,
+      "step": 582
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.03548911213874817,
+      "learning_rate": 5.3869225411740884e-05,
+      "loss": 0.3369,
+      "step": 583
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.033440180122852325,
+      "learning_rate": 5.3552055872822636e-05,
+      "loss": 0.3335,
+      "step": 584
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.0349792055785656,
+      "learning_rate": 5.323548100488565e-05,
+      "loss": 0.3079,
+      "step": 585
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.03338850289583206,
+      "learning_rate": 5.2919504861023903e-05,
+      "loss": 0.3105,
+      "step": 586
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.03373012691736221,
+      "learning_rate": 5.260413148666614e-05,
+      "loss": 0.3267,
+      "step": 587
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.034098975360393524,
+      "learning_rate": 5.228936491952363e-05,
+      "loss": 0.3167,
+      "step": 588
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.0328991673886776,
+      "learning_rate": 5.197520918953904e-05,
+      "loss": 0.3325,
+      "step": 589
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.033464428037405014,
+      "learning_rate": 5.16616683188342e-05,
+      "loss": 0.3318,
+      "step": 590
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 0.03465236350893974,
+      "learning_rate": 5.134874632165913e-05,
+      "loss": 0.3275,
+      "step": 591
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.03109387867152691,
+      "learning_rate": 5.103644720434027e-05,
+      "loss": 0.3232,
+      "step": 592
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.035864707082509995,
+      "learning_rate": 5.072477496522958e-05,
+      "loss": 0.3095,
+      "step": 593
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.03278312832117081,
+      "learning_rate": 5.041373359465289e-05,
+      "loss": 0.3202,
+      "step": 594
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.03266187012195587,
+      "learning_rate": 5.010332707485925e-05,
+      "loss": 0.3188,
+      "step": 595
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.036194685846567154,
+      "learning_rate": 4.9793559379969566e-05,
+      "loss": 0.3143,
+      "step": 596
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.0343778058886528,
+      "learning_rate": 4.9484434475925976e-05,
+      "loss": 0.3204,
+      "step": 597
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.03244636580348015,
+      "learning_rate": 4.917595632044113e-05,
+      "loss": 0.3252,
+      "step": 598
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 0.036536213010549545,
+      "learning_rate": 4.8868128862947115e-05,
+      "loss": 0.313,
+      "step": 599
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.0331181101500988,
+      "learning_rate": 4.856095604454539e-05,
+      "loss": 0.328,
+      "step": 600
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.03526446595788002,
+      "learning_rate": 4.825444179795589e-05,
+      "loss": 0.3218,
+      "step": 601
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.03498705476522446,
+      "learning_rate": 4.7948590047467153e-05,
+      "loss": 0.3243,
+      "step": 602
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.03270057961344719,
+      "learning_rate": 4.764340470888556e-05,
+      "loss": 0.3184,
+      "step": 603
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.033852964639663696,
+      "learning_rate": 4.7338889689485624e-05,
+      "loss": 0.3219,
+      "step": 604
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.03443654999136925,
+      "learning_rate": 4.703504888795958e-05,
+      "loss": 0.3187,
+      "step": 605
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.03217703476548195,
+      "learning_rate": 4.673188619436798e-05,
+      "loss": 0.3208,
+      "step": 606
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.033705055713653564,
+      "learning_rate": 4.6429405490089306e-05,
+      "loss": 0.323,
+      "step": 607
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.033967047929763794,
+      "learning_rate": 4.6127610647770767e-05,
+      "loss": 0.3258,
+      "step": 608
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.031915146857500076,
+      "learning_rate": 4.582650553127835e-05,
+      "loss": 0.3117,
+      "step": 609
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.033060990273952484,
+      "learning_rate": 4.552609399564762e-05,
+      "loss": 0.3203,
+      "step": 610
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.032012149691581726,
+      "learning_rate": 4.522637988703424e-05,
+      "loss": 0.333,
+      "step": 611
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.033706024289131165,
+      "learning_rate": 4.492736704266475e-05,
+      "loss": 0.3233,
+      "step": 612
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.032826974987983704,
+      "learning_rate": 4.462905929078734e-05,
+      "loss": 0.3208,
+      "step": 613
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.03313816711306572,
+      "learning_rate": 4.4331460450623064e-05,
+      "loss": 0.3067,
+      "step": 614
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.032175060361623764,
+      "learning_rate": 4.403457433231674e-05,
+      "loss": 0.3259,
+      "step": 615
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.03576289862394333,
+      "learning_rate": 4.373840473688829e-05,
+      "loss": 0.321,
+      "step": 616
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 0.03244362026453018,
+      "learning_rate": 4.344295545618392e-05,
+      "loss": 0.3074,
+      "step": 617
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.03456572815775871,
+      "learning_rate": 4.3148230272827784e-05,
+      "loss": 0.3212,
+      "step": 618
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.037079472094774246,
+      "learning_rate": 4.285423296017341e-05,
+      "loss": 0.3389,
+      "step": 619
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.03216618299484253,
+      "learning_rate": 4.256096728225548e-05,
+      "loss": 0.314,
+      "step": 620
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.035881418734788895,
+      "learning_rate": 4.2268436993741454e-05,
+      "loss": 0.3327,
+      "step": 621
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.03452802449464798,
+      "learning_rate": 4.197664583988376e-05,
+      "loss": 0.3225,
+      "step": 622
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.03389676287770271,
+      "learning_rate": 4.168559755647169e-05,
+      "loss": 0.3234,
+      "step": 623
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.033218082040548325,
+      "learning_rate": 4.1395295869783615e-05,
+      "loss": 0.319,
+      "step": 624
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.03429538384079933,
+      "learning_rate": 4.110574449653916e-05,
+      "loss": 0.3201,
+      "step": 625
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 0.03393711894750595,
+      "learning_rate": 4.0816947143851816e-05,
+      "loss": 0.3249,
+      "step": 626
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.03543519601225853,
+      "learning_rate": 4.052890750918138e-05,
+      "loss": 0.3082,
+      "step": 627
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.03144745156168938,
+      "learning_rate": 4.024162928028663e-05,
+      "loss": 0.3232,
+      "step": 628
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.03410906717181206,
+      "learning_rate": 3.995511613517799e-05,
+      "loss": 0.3176,
+      "step": 629
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.03316747397184372,
+      "learning_rate": 3.966937174207066e-05,
+      "loss": 0.3325,
+      "step": 630
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.032455362379550934,
+      "learning_rate": 3.938439975933752e-05,
+      "loss": 0.3175,
+      "step": 631
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.03558775410056114,
+      "learning_rate": 3.910020383546233e-05,
+      "loss": 0.322,
+      "step": 632
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.032109521329402924,
+      "learning_rate": 3.881678760899291e-05,
+      "loss": 0.3261,
+      "step": 633
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.03333992138504982,
+      "learning_rate": 3.853415470849479e-05,
+      "loss": 0.3173,
+      "step": 634
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 0.03266488015651703,
+      "learning_rate": 3.8252308752504404e-05,
+      "loss": 0.3252,
+      "step": 635
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.03426700085401535,
+      "learning_rate": 3.7971253349483285e-05,
+      "loss": 0.319,
+      "step": 636
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.03221927583217621,
+      "learning_rate": 3.7690992097771224e-05,
+      "loss": 0.3276,
+      "step": 637
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.031010834500193596,
+      "learning_rate": 3.741152858554077e-05,
+      "loss": 0.3198,
+      "step": 638
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.03406721353530884,
+      "learning_rate": 3.713286639075083e-05,
+      "loss": 0.3323,
+      "step": 639
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.03283034637570381,
+      "learning_rate": 3.6855009081101355e-05,
+      "loss": 0.3177,
+      "step": 640
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.03155268728733063,
+      "learning_rate": 3.6577960213987104e-05,
+      "loss": 0.3235,
+      "step": 641
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.034217044711112976,
+      "learning_rate": 3.630172333645261e-05,
+      "loss": 0.3229,
+      "step": 642
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.03232642635703087,
+      "learning_rate": 3.602630198514632e-05,
+      "loss": 0.3214,
+      "step": 643
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.03322529047727585,
+      "learning_rate": 3.5751699686275786e-05,
+      "loss": 0.3269,
+      "step": 644
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.03211558982729912,
+      "learning_rate": 3.547791995556203e-05,
+      "loss": 0.3173,
+      "step": 645
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.03321954980492592,
+      "learning_rate": 3.520496629819494e-05,
+      "loss": 0.309,
+      "step": 646
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.03272011876106262,
+      "learning_rate": 3.493284220878817e-05,
+      "loss": 0.3191,
+      "step": 647
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.03565934672951698,
+      "learning_rate": 3.466155117133433e-05,
+      "loss": 0.3137,
+      "step": 648
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.03242676332592964,
+      "learning_rate": 3.439109665916076e-05,
+      "loss": 0.3269,
+      "step": 649
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.03284203261137009,
+      "learning_rate": 3.4121482134884575e-05,
+      "loss": 0.3254,
+      "step": 650
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.035054489970207214,
+      "learning_rate": 3.385271105036868e-05,
+      "loss": 0.3263,
+      "step": 651
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.03570910543203354,
+      "learning_rate": 3.358478684667734e-05,
+      "loss": 0.3322,
+      "step": 652
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 0.030848857015371323,
+      "learning_rate": 3.3317712954032454e-05,
+      "loss": 0.321,
+      "step": 653
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.03351164981722832,
+      "learning_rate": 3.305149279176921e-05,
+      "loss": 0.3192,
+      "step": 654
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.03559879586100578,
+      "learning_rate": 3.278612976829265e-05,
+      "loss": 0.3202,
+      "step": 655
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.030793210491538048,
+      "learning_rate": 3.252162728103382e-05,
+      "loss": 0.3151,
+      "step": 656
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.032545629888772964,
+      "learning_rate": 3.225798871640644e-05,
+      "loss": 0.3214,
+      "step": 657
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.03530002385377884,
+      "learning_rate": 3.199521744976342e-05,
+      "loss": 0.3224,
+      "step": 658
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.0325043648481369,
+      "learning_rate": 3.1733316845353735e-05,
+      "loss": 0.3265,
+      "step": 659
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.033760733902454376,
+      "learning_rate": 3.147229025627922e-05,
+      "loss": 0.3304,
+      "step": 660
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.0321899875998497,
+      "learning_rate": 3.1212141024451826e-05,
+      "loss": 0.3176,
+      "step": 661
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 0.03273585066199303,
+      "learning_rate": 3.095287248055069e-05,
+      "loss": 0.3286,
+      "step": 662
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.03182963281869888,
+      "learning_rate": 3.069448794397959e-05,
+      "loss": 0.3201,
+      "step": 663
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.03337734937667847,
+      "learning_rate": 3.043699072282429e-05,
+      "loss": 0.3269,
+      "step": 664
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.03036559373140335,
+      "learning_rate": 3.01803841138104e-05,
+      "loss": 0.3203,
+      "step": 665
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.03221311792731285,
+      "learning_rate": 2.9924671402261018e-05,
+      "loss": 0.3354,
+      "step": 666
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.031059028580784798,
+      "learning_rate": 2.966985586205475e-05,
+      "loss": 0.3171,
+      "step": 667
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.03237179294228554,
+      "learning_rate": 2.941594075558366e-05,
+      "loss": 0.3173,
+      "step": 668
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.03317666053771973,
+      "learning_rate": 2.9162929333711697e-05,
+      "loss": 0.3063,
+      "step": 669
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.034094829112291336,
+      "learning_rate": 2.8910824835732952e-05,
+      "loss": 0.3012,
+      "step": 670
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 0.031877171248197556,
+      "learning_rate": 2.8659630489330213e-05,
+      "loss": 0.3268,
+      "step": 671
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.03302241861820221,
+      "learning_rate": 2.8409349510533578e-05,
+      "loss": 0.3093,
+      "step": 672
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.03291558101773262,
+      "learning_rate": 2.8159985103679408e-05,
+      "loss": 0.3179,
+      "step": 673
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.031883399933576584,
+      "learning_rate": 2.7911540461369222e-05,
+      "loss": 0.3204,
+      "step": 674
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.03172933682799339,
+      "learning_rate": 2.7664018764428835e-05,
+      "loss": 0.3192,
+      "step": 675
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.03233201429247856,
+      "learning_rate": 2.7417423181867585e-05,
+      "loss": 0.3118,
+      "step": 676
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.032947055995464325,
+      "learning_rate": 2.717175687083785e-05,
+      "loss": 0.318,
+      "step": 677
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.03231576085090637,
+      "learning_rate": 2.6927022976594607e-05,
+      "loss": 0.3064,
+      "step": 678
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.032052453607320786,
+      "learning_rate": 2.668322463245513e-05,
+      "loss": 0.3202,
+      "step": 679
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.03498750552535057,
+      "learning_rate": 2.6440364959758813e-05,
+      "loss": 0.3167,
+      "step": 680
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.033017512410879135,
+      "learning_rate": 2.619844706782737e-05,
+      "loss": 0.3166,
+      "step": 681
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.033758796751499176,
+      "learning_rate": 2.595747405392491e-05,
+      "loss": 0.3112,
+      "step": 682
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.03454572334885597,
+      "learning_rate": 2.5717449003218342e-05,
+      "loss": 0.317,
+      "step": 683
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.03280545398592949,
+      "learning_rate": 2.5478374988737753e-05,
+      "loss": 0.3161,
+      "step": 684
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.03240388259291649,
+      "learning_rate": 2.524025507133726e-05,
+      "loss": 0.3259,
+      "step": 685
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.03514571860432625,
+      "learning_rate": 2.5003092299655584e-05,
+      "loss": 0.3247,
+      "step": 686
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.03480707108974457,
+      "learning_rate": 2.4766889710077357e-05,
+      "loss": 0.323,
+      "step": 687
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.03250759467482567,
+      "learning_rate": 2.4531650326693822e-05,
+      "loss": 0.313,
+      "step": 688
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.03347860276699066,
+      "learning_rate": 2.429737716126451e-05,
+      "loss": 0.3262,
+      "step": 689
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.035569798201322556,
+      "learning_rate": 2.406407321317835e-05,
+      "loss": 0.3147,
+      "step": 690
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.03313907980918884,
+      "learning_rate": 2.3831741469415593e-05,
+      "loss": 0.3252,
+      "step": 691
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.03403162583708763,
+      "learning_rate": 2.3600384904509254e-05,
+      "loss": 0.2971,
+      "step": 692
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.03517422825098038,
+      "learning_rate": 2.3370006480507244e-05,
+      "loss": 0.3191,
+      "step": 693
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.03399303928017616,
+      "learning_rate": 2.31406091469344e-05,
+      "loss": 0.3182,
+      "step": 694
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.03280871361494064,
+      "learning_rate": 2.2912195840754604e-05,
+      "loss": 0.3165,
+      "step": 695
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.03217533603310585,
+      "learning_rate": 2.2684769486333445e-05,
+      "loss": 0.3258,
+      "step": 696
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 0.03353995084762573,
+      "learning_rate": 2.2458332995400433e-05,
+      "loss": 0.3231,
+      "step": 697
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.03184211999177933,
+      "learning_rate": 2.2232889267012038e-05,
+      "loss": 0.3027,
+      "step": 698
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.03190856799483299,
+      "learning_rate": 2.2008441187514293e-05,
+      "loss": 0.3238,
+      "step": 699
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.03220997750759125,
+      "learning_rate": 2.178499163050617e-05,
+      "loss": 0.3197,
+      "step": 700
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.03379112109541893,
+      "learning_rate": 2.156254345680242e-05,
+      "loss": 0.3115,
+      "step": 701
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.031570915132761,
+      "learning_rate": 2.1341099514397266e-05,
+      "loss": 0.3131,
+      "step": 702
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.032580941915512085,
+      "learning_rate": 2.1120662638427647e-05,
+      "loss": 0.3098,
+      "step": 703
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.03251396119594574,
+      "learning_rate": 2.0901235651137284e-05,
+      "loss": 0.3124,
+      "step": 704
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.03203850984573364,
+      "learning_rate": 2.0682821361840154e-05,
+      "loss": 0.3219,
+      "step": 705
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 0.03141346946358681,
+      "learning_rate": 2.0465422566884805e-05,
+      "loss": 0.3144,
+      "step": 706
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.031967923045158386,
+      "learning_rate": 2.0249042049618384e-05,
+      "loss": 0.3129,
+      "step": 707
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.031204957515001297,
+      "learning_rate": 2.0033682580351144e-05,
+      "loss": 0.315,
+      "step": 708
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.03257647901773453,
+      "learning_rate": 1.981934691632088e-05,
+      "loss": 0.3158,
+      "step": 709
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.03063308075070381,
+      "learning_rate": 1.9606037801657673e-05,
+      "loss": 0.3272,
+      "step": 710
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.0314013808965683,
+      "learning_rate": 1.9393757967348658e-05,
+      "loss": 0.324,
+      "step": 711
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.03184258192777634,
+      "learning_rate": 1.9182510131203224e-05,
+      "loss": 0.322,
+      "step": 712
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 888,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 89,
+  "total_flos": 1.4202538154367386e+19,
+  "train_batch_size": 14,
+  "trial_name": null,
+  "trial_params": null
+}