diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4729 @@
+{
+  "best_metric": 0.09024298936128616,
+  "best_model_checkpoint": "./modeversion1_m7_e4/checkpoint-6800",
+  "epoch": 4.0,
+  "global_step": 6824,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00019970691676436108,
+      "loss": 4.9518,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00019941383352872216,
+      "loss": 4.7468,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00019912075029308323,
+      "loss": 4.6634,
+      "step": 30
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00019882766705744433,
+      "loss": 4.5601,
+      "step": 40
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0001985345838218054,
+      "loss": 4.4809,
+      "step": 50
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00019824150058616647,
+      "loss": 4.3918,
+      "step": 60
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00019794841735052755,
+      "loss": 4.1907,
+      "step": 70
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00019765533411488865,
+      "loss": 4.1004,
+      "step": 80
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00019736225087924972,
+      "loss": 4.07,
+      "step": 90
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0001970691676436108,
+      "loss": 4.073,
+      "step": 100
+    },
+    {
+      "epoch": 0.06,
+      "eval_accuracy": 0.1768493346687652,
+      "eval_loss": 3.93697452545166,
+      "eval_runtime": 144.6591,
+      "eval_samples_per_second": 48.314,
+      "eval_steps_per_second": 6.042,
+      "step": 100
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0001967760844079719,
+      "loss": 3.8669,
+      "step": 110
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00019648300117233296,
+      "loss": 3.7828,
+      "step": 120
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00019618991793669404,
+      "loss": 3.7687,
+      "step": 130
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0001958968347010551,
+      "loss": 3.693,
+      "step": 140
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00019560375146541618,
+      "loss": 3.5779,
+      "step": 150
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00019531066822977726,
+      "loss": 3.5634,
+      "step": 160
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00019501758499413833,
+      "loss": 3.4888,
+      "step": 170
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00019472450175849943,
+      "loss": 3.3963,
+      "step": 180
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0001944314185228605,
+      "loss": 3.2674,
+      "step": 190
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00019413833528722157,
+      "loss": 3.4186,
+      "step": 200
+    },
+    {
+      "epoch": 0.12,
+      "eval_accuracy": 0.25897839462011735,
+      "eval_loss": 3.272094964981079,
+      "eval_runtime": 144.7772,
+      "eval_samples_per_second": 48.274,
+      "eval_steps_per_second": 6.037,
+      "step": 200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00019384525205158265,
+      "loss": 3.2797,
+      "step": 210
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00019355216881594375,
+      "loss": 3.2347,
+      "step": 220
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00019325908558030482,
+      "loss": 3.1258,
+      "step": 230
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0001929660023446659,
+      "loss": 2.9637,
+      "step": 240
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.000192672919109027,
+      "loss": 2.9957,
+      "step": 250
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00019237983587338807,
+      "loss": 3.0592,
+      "step": 260
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00019208675263774914,
+      "loss": 2.778,
+      "step": 270
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0001917936694021102,
+      "loss": 2.7552,
+      "step": 280
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00019150058616647128,
+      "loss": 2.684,
+      "step": 290
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00019120750293083236,
+      "loss": 2.6745,
+      "step": 300
+    },
+    {
+      "epoch": 0.18,
+      "eval_accuracy": 0.38560595221061666,
+      "eval_loss": 2.6465089321136475,
+      "eval_runtime": 146.4706,
+      "eval_samples_per_second": 47.716,
+      "eval_steps_per_second": 5.967,
+      "step": 300
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00019091441969519343,
+      "loss": 2.764,
+      "step": 310
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00019062133645955453,
+      "loss": 2.6034,
+      "step": 320
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0001903282532239156,
+      "loss": 2.539,
+      "step": 330
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00019003516998827668,
+      "loss": 2.5864,
+      "step": 340
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00018974208675263775,
+      "loss": 2.3876,
+      "step": 350
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00018944900351699885,
+      "loss": 2.3781,
+      "step": 360
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00018915592028135992,
+      "loss": 2.2723,
+      "step": 370
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.000188862837045721,
+      "loss": 2.4228,
+      "step": 380
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00018856975381008207,
+      "loss": 2.1609,
+      "step": 390
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00018827667057444317,
+      "loss": 2.2806,
+      "step": 400
+    },
+    {
+      "epoch": 0.23,
+      "eval_accuracy": 0.45228215767634855,
+      "eval_loss": 2.260012626647949,
+      "eval_runtime": 144.5966,
+      "eval_samples_per_second": 48.334,
+      "eval_steps_per_second": 6.044,
+      "step": 400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00018798358733880424,
+      "loss": 2.2351,
+      "step": 410
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0001876905041031653,
+      "loss": 2.1572,
+      "step": 420
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00018739742086752638,
+      "loss": 2.3053,
+      "step": 430
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00018710433763188746,
+      "loss": 2.0842,
+      "step": 440
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00018681125439624853,
+      "loss": 2.1774,
+      "step": 450
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0001865181711606096,
+      "loss": 2.0003,
+      "step": 460
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0001862250879249707,
+      "loss": 2.1276,
+      "step": 470
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00018593200468933178,
+      "loss": 2.0091,
+      "step": 480
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00018563892145369285,
+      "loss": 1.9637,
+      "step": 490
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00018534583821805395,
+      "loss": 1.9275,
+      "step": 500
+    },
+    {
+      "epoch": 0.29,
+      "eval_accuracy": 0.5109457719273144,
+      "eval_loss": 1.965333104133606,
+      "eval_runtime": 142.8058,
+      "eval_samples_per_second": 48.941,
+      "eval_steps_per_second": 6.12,
+      "step": 500
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00018505275498241502,
+      "loss": 2.0284,
+      "step": 510
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0001847596717467761,
+      "loss": 2.0057,
+      "step": 520
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00018446658851113717,
+      "loss": 1.8957,
+      "step": 530
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00018417350527549827,
+      "loss": 1.8548,
+      "step": 540
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00018388042203985934,
+      "loss": 1.7549,
+      "step": 550
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0001835873388042204,
+      "loss": 1.6703,
+      "step": 560
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00018329425556858149,
+      "loss": 1.7037,
+      "step": 570
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00018300117233294256,
+      "loss": 1.5982,
+      "step": 580
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00018270808909730363,
+      "loss": 1.8877,
+      "step": 590
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0001824150058616647,
+      "loss": 1.6958,
+      "step": 600
+    },
+    {
+      "epoch": 0.35,
+      "eval_accuracy": 0.607812276434397,
+      "eval_loss": 1.6815019845962524,
+      "eval_runtime": 141.9874,
+      "eval_samples_per_second": 49.223,
+      "eval_steps_per_second": 6.155,
+      "step": 600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0001821219226260258,
+      "loss": 1.6292,
+      "step": 610
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00018182883939038688,
+      "loss": 1.6704,
+      "step": 620
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00018153575615474795,
+      "loss": 1.4707,
+      "step": 630
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00018124267291910902,
+      "loss": 1.5722,
+      "step": 640
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00018094958968347012,
+      "loss": 1.477,
+      "step": 650
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0001806565064478312,
+      "loss": 1.5452,
+      "step": 660
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00018036342321219227,
+      "loss": 1.6506,
+      "step": 670
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00018007033997655337,
+      "loss": 1.4681,
+      "step": 680
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00017977725674091444,
+      "loss": 1.4487,
+      "step": 690
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00017948417350527551,
+      "loss": 1.2797,
+      "step": 700
+    },
+    {
+      "epoch": 0.41,
+      "eval_accuracy": 0.6418657890971526,
+      "eval_loss": 1.4514340162277222,
+      "eval_runtime": 142.9219,
+      "eval_samples_per_second": 48.901,
+      "eval_steps_per_second": 6.115,
+      "step": 700
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00017919109026963659,
+      "loss": 1.4092,
+      "step": 710
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00017889800703399766,
+      "loss": 1.6711,
+      "step": 720
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00017860492379835873,
+      "loss": 1.5477,
+      "step": 730
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0001783118405627198,
+      "loss": 1.4402,
+      "step": 740
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0001780187573270809,
+      "loss": 1.5563,
+      "step": 750
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00017772567409144198,
+      "loss": 1.4736,
+      "step": 760
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00017743259085580305,
+      "loss": 1.5159,
+      "step": 770
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00017713950762016412,
+      "loss": 1.329,
+      "step": 780
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00017684642438452522,
+      "loss": 1.2187,
+      "step": 790
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0001765533411488863,
+      "loss": 1.3772,
+      "step": 800
+    },
+    {
+      "epoch": 0.47,
+      "eval_accuracy": 0.6762054657318644,
+      "eval_loss": 1.3212426900863647,
+      "eval_runtime": 143.9906,
+      "eval_samples_per_second": 48.538,
+      "eval_steps_per_second": 6.07,
+      "step": 800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00017626025791324737,
+      "loss": 1.1954,
+      "step": 810
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00017596717467760847,
+      "loss": 1.2532,
+      "step": 820
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00017567409144196954,
+      "loss": 1.2134,
+      "step": 830
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00017538100820633061,
+      "loss": 1.1357,
+      "step": 840
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0001750879249706917,
+      "loss": 1.2893,
+      "step": 850
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00017479484173505276,
+      "loss": 1.2528,
+      "step": 860
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00017450175849941383,
+      "loss": 1.3934,
+      "step": 870
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0001742086752637749,
+      "loss": 1.1394,
+      "step": 880
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.000173915592028136,
+      "loss": 1.224,
+      "step": 890
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00017362250879249708,
+      "loss": 1.1765,
+      "step": 900
+    },
+    {
+      "epoch": 0.53,
+      "eval_accuracy": 0.7028187151237659,
+      "eval_loss": 1.1476110219955444,
+      "eval_runtime": 142.6845,
+      "eval_samples_per_second": 48.982,
+      "eval_steps_per_second": 6.125,
+      "step": 900
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00017332942555685815,
+      "loss": 1.1505,
+      "step": 910
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00017303634232121922,
+      "loss": 1.0058,
+      "step": 920
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00017274325908558032,
+      "loss": 1.1991,
+      "step": 930
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0001724501758499414,
+      "loss": 1.1221,
+      "step": 940
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00017215709261430247,
+      "loss": 1.0279,
+      "step": 950
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00017186400937866357,
+      "loss": 1.0806,
+      "step": 960
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00017157092614302464,
+      "loss": 1.0342,
+      "step": 970
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00017127784290738572,
+      "loss": 1.095,
+      "step": 980
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0001709847596717468,
+      "loss": 1.0627,
+      "step": 990
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00017069167643610786,
+      "loss": 1.0152,
+      "step": 1000
+    },
+    {
+      "epoch": 0.59,
+      "eval_accuracy": 0.7312920303333811,
+      "eval_loss": 1.0357469320297241,
+      "eval_runtime": 141.502,
+      "eval_samples_per_second": 49.392,
+      "eval_steps_per_second": 6.177,
+      "step": 1000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00017039859320046893,
+      "loss": 1.1845,
+      "step": 1010
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00017010550996483,
+      "loss": 1.0731,
+      "step": 1020
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00016981242672919108,
+      "loss": 1.0936,
+      "step": 1030
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00016951934349355218,
+      "loss": 1.1276,
+      "step": 1040
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00016922626025791325,
+      "loss": 1.0028,
+      "step": 1050
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00016893317702227432,
+      "loss": 1.0649,
+      "step": 1060
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00016864009378663542,
+      "loss": 1.0313,
+      "step": 1070
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0001683470105509965,
+      "loss": 0.9503,
+      "step": 1080
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00016805392731535757,
+      "loss": 1.0665,
+      "step": 1090
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00016776084407971864,
+      "loss": 0.7861,
+      "step": 1100
+    },
+    {
+      "epoch": 0.64,
+      "eval_accuracy": 0.7184146515953641,
+      "eval_loss": 1.023007869720459,
+      "eval_runtime": 141.0512,
+      "eval_samples_per_second": 49.549,
+      "eval_steps_per_second": 6.196,
+      "step": 1100
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00016746776084407974,
+      "loss": 1.0444,
+      "step": 1110
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00016717467760844082,
+      "loss": 0.9562,
+      "step": 1120
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0001668815943728019,
+      "loss": 1.057,
+      "step": 1130
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016658851113716296,
+      "loss": 0.7354,
+      "step": 1140
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00016629542790152403,
+      "loss": 0.9462,
+      "step": 1150
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0001660023446658851,
+      "loss": 0.838,
+      "step": 1160
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00016570926143024618,
+      "loss": 1.0383,
+      "step": 1170
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00016541617819460728,
+      "loss": 0.9091,
+      "step": 1180
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00016512309495896835,
+      "loss": 0.9672,
+      "step": 1190
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00016483001172332943,
+      "loss": 1.0262,
+      "step": 1200
+    },
+    {
+      "epoch": 0.7,
+      "eval_accuracy": 0.7385892116182573,
+      "eval_loss": 0.9469121694564819,
+      "eval_runtime": 143.0373,
+      "eval_samples_per_second": 48.861,
+      "eval_steps_per_second": 6.11,
+      "step": 1200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001645369284876905,
+      "loss": 0.8081,
+      "step": 1210
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0001642438452520516,
+      "loss": 0.8826,
+      "step": 1220
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00016395076201641267,
+      "loss": 0.8504,
+      "step": 1230
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00016365767878077374,
+      "loss": 0.9641,
+      "step": 1240
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00016336459554513484,
+      "loss": 0.9718,
+      "step": 1250
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00016307151230949592,
+      "loss": 0.9118,
+      "step": 1260
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.000162778429073857,
+      "loss": 0.8592,
+      "step": 1270
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00016248534583821806,
+      "loss": 0.9236,
+      "step": 1280
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00016219226260257914,
+      "loss": 0.9054,
+      "step": 1290
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001618991793669402,
+      "loss": 0.8905,
+      "step": 1300
+    },
+    {
+      "epoch": 0.76,
+      "eval_accuracy": 0.7756474459865503,
+      "eval_loss": 0.818402886390686,
+      "eval_runtime": 142.6625,
+      "eval_samples_per_second": 48.99,
+      "eval_steps_per_second": 6.126,
+      "step": 1300
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00016160609613130128,
+      "loss": 0.9189,
+      "step": 1310
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00016131301289566238,
+      "loss": 0.7185,
+      "step": 1320
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00016101992966002345,
+      "loss": 0.7161,
+      "step": 1330
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00016072684642438453,
+      "loss": 0.78,
+      "step": 1340
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001604337631887456,
+      "loss": 0.8638,
+      "step": 1350
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001601406799531067,
+      "loss": 0.5591,
+      "step": 1360
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00015984759671746777,
+      "loss": 0.7001,
+      "step": 1370
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00015955451348182884,
+      "loss": 0.765,
+      "step": 1380
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00015926143024618994,
+      "loss": 0.7748,
+      "step": 1390
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00015896834701055102,
+      "loss": 0.6919,
+      "step": 1400
+    },
+    {
+      "epoch": 0.82,
+      "eval_accuracy": 0.7710688224352554,
+      "eval_loss": 0.8083020448684692,
+      "eval_runtime": 141.661,
+      "eval_samples_per_second": 49.336,
+      "eval_steps_per_second": 6.17,
+      "step": 1400
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0001586752637749121,
+      "loss": 0.736,
+      "step": 1410
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00015838218053927316,
+      "loss": 0.7675,
+      "step": 1420
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00015808909730363424,
+      "loss": 0.7939,
+      "step": 1430
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0001577960140679953,
+      "loss": 0.8095,
+      "step": 1440
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00015750293083235638,
+      "loss": 0.7648,
+      "step": 1450
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00015720984759671748,
+      "loss": 0.743,
+      "step": 1460
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00015691676436107855,
+      "loss": 0.7438,
+      "step": 1470
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00015662368112543963,
+      "loss": 0.7346,
+      "step": 1480
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0001563305978898007,
+      "loss": 0.8393,
+      "step": 1490
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0001560375146541618,
+      "loss": 0.7494,
+      "step": 1500
+    },
+    {
+      "epoch": 0.88,
+      "eval_accuracy": 0.7825153813134926,
+      "eval_loss": 0.7600758075714111,
+      "eval_runtime": 143.5803,
+      "eval_samples_per_second": 48.677,
+      "eval_steps_per_second": 6.087,
+      "step": 1500
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00015574443141852287,
+      "loss": 0.6779,
+      "step": 1510
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00015545134818288395,
+      "loss": 0.6841,
+      "step": 1520
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00015515826494724505,
+      "loss": 0.7694,
+      "step": 1530
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00015486518171160612,
+      "loss": 0.6697,
+      "step": 1540
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0001545720984759672,
+      "loss": 0.71,
+      "step": 1550
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00015427901524032826,
+      "loss": 0.6402,
+      "step": 1560
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00015398593200468934,
+      "loss": 0.6851,
+      "step": 1570
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0001536928487690504,
+      "loss": 0.6974,
+      "step": 1580
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00015339976553341148,
+      "loss": 0.7773,
+      "step": 1590
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00015310668229777258,
+      "loss": 0.5078,
+      "step": 1600
+    },
+    {
+      "epoch": 0.94,
+      "eval_accuracy": 0.805551581055945,
+      "eval_loss": 0.6883856058120728,
+      "eval_runtime": 143.3782,
+      "eval_samples_per_second": 48.745,
+      "eval_steps_per_second": 6.096,
+      "step": 1600
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00015281359906213366,
+      "loss": 0.8773,
+      "step": 1610
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00015252051582649473,
+      "loss": 0.728,
+      "step": 1620
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0001522274325908558,
+      "loss": 0.7335,
+      "step": 1630
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0001519343493552169,
+      "loss": 0.6446,
+      "step": 1640
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015164126611957797,
+      "loss": 0.6338,
+      "step": 1650
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015134818288393905,
+      "loss": 0.6337,
+      "step": 1660
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00015105509964830012,
+      "loss": 0.6159,
+      "step": 1670
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00015076201641266122,
+      "loss": 0.6605,
+      "step": 1680
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0001504689331770223,
+      "loss": 0.7442,
+      "step": 1690
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00015017584994138336,
+      "loss": 0.7134,
+      "step": 1700
+    },
+    {
+      "epoch": 1.0,
+      "eval_accuracy": 0.8159965660323365,
+      "eval_loss": 0.6311172842979431,
+      "eval_runtime": 142.9359,
+      "eval_samples_per_second": 48.896,
+      "eval_steps_per_second": 6.115,
+      "step": 1700
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00014988276670574444,
+      "loss": 0.4797,
+      "step": 1710
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.0001495896834701055,
+      "loss": 0.3648,
+      "step": 1720
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00014929660023446658,
+      "loss": 0.5075,
+      "step": 1730
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00014900351699882766,
+      "loss": 0.4528,
+      "step": 1740
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00014871043376318876,
+      "loss": 0.3223,
+      "step": 1750
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00014841735052754983,
+      "loss": 0.5243,
+      "step": 1760
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.0001481242672919109,
+      "loss": 0.5827,
+      "step": 1770
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00014783118405627197,
+      "loss": 0.5099,
+      "step": 1780
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00014753810082063307,
+      "loss": 0.2818,
+      "step": 1790
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00014724501758499415,
+      "loss": 0.4328,
+      "step": 1800
+    },
+    {
+      "epoch": 1.06,
+      "eval_accuracy": 0.8251538131349263,
+      "eval_loss": 0.5739801526069641,
+      "eval_runtime": 142.1993,
+      "eval_samples_per_second": 49.149,
+      "eval_steps_per_second": 6.146,
+      "step": 1800
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00014695193434935522,
+      "loss": 0.4137,
+      "step": 1810
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00014665885111371632,
+      "loss": 0.3978,
+      "step": 1820
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.0001463657678780774,
+      "loss": 0.4585,
+      "step": 1830
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00014607268464243847,
+      "loss": 0.3887,
+      "step": 1840
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00014577960140679954,
+      "loss": 0.4743,
+      "step": 1850
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.0001454865181711606,
+      "loss": 0.4315,
+      "step": 1860
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00014519343493552168,
+      "loss": 0.4949,
+      "step": 1870
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00014490035169988276,
+      "loss": 0.5806,
+      "step": 1880
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00014460726846424386,
+      "loss": 0.4127,
+      "step": 1890
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00014431418522860493,
+      "loss": 0.4971,
+      "step": 1900
+    },
+    {
+      "epoch": 1.11,
+      "eval_accuracy": 0.8290170267563314,
+      "eval_loss": 0.5856359601020813,
+      "eval_runtime": 141.3649,
+      "eval_samples_per_second": 49.439,
+      "eval_steps_per_second": 6.183,
+      "step": 1900
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.000144021101992966,
+      "loss": 0.4027,
+      "step": 1910
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00014372801875732708,
+      "loss": 0.3814,
+      "step": 1920
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00014343493552168818,
+      "loss": 0.4191,
+      "step": 1930
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00014314185228604925,
+      "loss": 0.4999,
+      "step": 1940
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00014284876905041032,
+      "loss": 0.3142,
+      "step": 1950
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00014255568581477142,
+      "loss": 0.383,
+      "step": 1960
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.0001422626025791325,
+      "loss": 0.4117,
+      "step": 1970
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00014196951934349357,
+      "loss": 0.3326,
+      "step": 1980
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00014167643610785464,
+      "loss": 0.332,
+      "step": 1990
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.0001413833528722157,
+      "loss": 0.5207,
+      "step": 2000
+    },
+    {
+      "epoch": 1.17,
+      "eval_accuracy": 0.8167119759622263,
+      "eval_loss": 0.6218913197517395,
+      "eval_runtime": 142.1541,
+      "eval_samples_per_second": 49.165,
+      "eval_steps_per_second": 6.148,
+      "step": 2000
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00014109026963657678,
+      "loss": 0.3766,
+      "step": 2010
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00014079718640093786,
+      "loss": 0.3261,
+      "step": 2020
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.00014050410316529896,
+      "loss": 0.5744,
+      "step": 2030
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00014024032825322392,
+      "loss": 0.4988,
+      "step": 2040
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00013994724501758502,
+      "loss": 0.3837,
+      "step": 2050
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.0001396541617819461,
+      "loss": 0.4694,
+      "step": 2060
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.00013936107854630716,
+      "loss": 0.4364,
+      "step": 2070
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.00013906799531066824,
+      "loss": 0.5789,
+      "step": 2080
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.0001387749120750293,
+      "loss": 0.3874,
+      "step": 2090
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.0001384818288393904,
+      "loss": 0.4027,
+      "step": 2100
+    },
+    {
+      "epoch": 1.23,
+      "eval_accuracy": 0.826584632994706,
+      "eval_loss": 0.5703310966491699,
+      "eval_runtime": 144.3732,
+      "eval_samples_per_second": 48.409,
+      "eval_steps_per_second": 6.054,
+      "step": 2100
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.00013818874560375148,
+      "loss": 0.4893,
+      "step": 2110
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.00013789566236811255,
+      "loss": 0.5338,
+      "step": 2120
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.00013760257913247363,
+      "loss": 0.4958,
+      "step": 2130
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.0001373094958968347,
+      "loss": 0.3926,
+      "step": 2140
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.00013701641266119577,
+      "loss": 0.3796,
+      "step": 2150
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.00013672332942555685,
+      "loss": 0.4433,
+      "step": 2160
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.00013643024618991795,
+      "loss": 0.3509,
+      "step": 2170
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.00013613716295427902,
+      "loss": 0.2775,
+      "step": 2180
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.0001358440797186401,
+      "loss": 0.3138,
+      "step": 2190
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.0001355509964830012,
+      "loss": 0.5605,
+      "step": 2200
+    },
+    {
+      "epoch": 1.29,
+      "eval_accuracy": 0.8371726999570754,
+      "eval_loss": 0.5217174291610718,
+      "eval_runtime": 142.7973,
+      "eval_samples_per_second": 48.944,
+      "eval_steps_per_second": 6.121,
+      "step": 2200
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.00013525791324736226,
+      "loss": 0.3436,
+      "step": 2210
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.00013496483001172334,
+      "loss": 0.3696,
+      "step": 2220
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.0001346717467760844,
+      "loss": 0.4021,
+      "step": 2230
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.00013437866354044548,
+      "loss": 0.4185,
+      "step": 2240
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.00013408558030480658,
+      "loss": 0.2992,
+      "step": 2250
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.00013379249706916766,
+      "loss": 0.4494,
+      "step": 2260
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.00013349941383352873,
+      "loss": 0.4397,
+      "step": 2270
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.0001332063305978898,
+      "loss": 0.4446,
+      "step": 2280
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.00013291324736225087,
+      "loss": 0.4627,
+      "step": 2290
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00013262016412661195,
+      "loss": 0.2723,
+      "step": 2300
+    },
+    {
+      "epoch": 1.35,
+      "eval_accuracy": 0.8564887680641007,
+      "eval_loss": 0.48052945733070374,
+      "eval_runtime": 141.8523,
+      "eval_samples_per_second": 49.27,
+      "eval_steps_per_second": 6.161,
+      "step": 2300
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00013232708089097302,
+      "loss": 0.3139,
+      "step": 2310
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.00013203399765533412,
+      "loss": 0.3448,
+      "step": 2320
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.0001317409144196952,
+      "loss": 0.4392,
+      "step": 2330
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 0.0001314478311840563,
+      "loss": 0.417,
+      "step": 2340
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.00013115474794841736,
+      "loss": 0.4455,
+      "step": 2350
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.00013086166471277844,
+      "loss": 0.304,
+      "step": 2360
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.0001305685814771395,
+      "loss": 0.2896,
+      "step": 2370
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.00013027549824150058,
+      "loss": 0.3634,
+      "step": 2380
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.00012998241500586168,
+      "loss": 0.3162,
+      "step": 2390
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.00012968933177022276,
+      "loss": 0.401,
+      "step": 2400
+    },
+    {
+      "epoch": 1.41,
+      "eval_accuracy": 0.8490485047932466,
+      "eval_loss": 0.48114609718322754,
+      "eval_runtime": 141.1595,
+      "eval_samples_per_second": 49.511,
+      "eval_steps_per_second": 6.192,
+      "step": 2400
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 0.00012939624853458383,
+      "loss": 0.4162,
+      "step": 2410
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.0001291031652989449,
+      "loss": 0.3563,
+      "step": 2420
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.00012881008206330597,
+      "loss": 0.2935,
+      "step": 2430
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.00012851699882766705,
+      "loss": 0.2772,
+      "step": 2440
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.00012822391559202812,
+      "loss": 0.1815,
+      "step": 2450
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 0.00012793083235638922,
+      "loss": 0.2421,
+      "step": 2460
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.0001276377491207503,
+      "loss": 0.3135,
+      "step": 2470
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.0001273446658851114,
+      "loss": 0.2401,
+      "step": 2480
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.00012705158264947247,
+      "loss": 0.2437,
+      "step": 2490
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.00012675849941383354,
+      "loss": 0.3419,
+      "step": 2500
+    },
+    {
+      "epoch": 1.47,
+      "eval_accuracy": 0.8607812276434397,
+      "eval_loss": 0.46192315220832825,
+      "eval_runtime": 141.0585,
+      "eval_samples_per_second": 49.547,
+      "eval_steps_per_second": 6.196,
+      "step": 2500
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.0001264654161781946,
+      "loss": 0.3481,
+      "step": 2510
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 0.00012617233294255568,
+      "loss": 0.3368,
+      "step": 2520
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 0.00012587924970691678,
+      "loss": 0.4863,
+      "step": 2530
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.00012558616647127786,
+      "loss": 0.3587,
+      "step": 2540
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.00012529308323563893,
+      "loss": 0.3619,
+      "step": 2550
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.000125,
+      "loss": 0.32,
+      "step": 2560
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.00012470691676436108,
+      "loss": 0.4175,
+      "step": 2570
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.00012441383352872215,
+      "loss": 0.2654,
+      "step": 2580
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.00012412075029308322,
+      "loss": 0.3706,
+      "step": 2590
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 0.00012382766705744432,
+      "loss": 0.301,
+      "step": 2600
+    },
+    {
+      "epoch": 1.52,
+      "eval_accuracy": 0.8712262126198311,
+      "eval_loss": 0.43177559971809387,
+      "eval_runtime": 142.4669,
+      "eval_samples_per_second": 49.057,
+      "eval_steps_per_second": 6.135,
+      "step": 2600
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.0001235345838218054,
+      "loss": 0.3649,
+      "step": 2610
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.0001232415005861665,
+      "loss": 0.2837,
+      "step": 2620
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.00012294841735052757,
+      "loss": 0.2603,
+      "step": 2630
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.00012265533411488864,
+      "loss": 0.3673,
+      "step": 2640
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 0.0001223622508792497,
+      "loss": 0.2663,
+      "step": 2650
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.00012206916764361079,
+      "loss": 0.2428,
+      "step": 2660
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.00012177608440797188,
+      "loss": 0.3494,
+      "step": 2670
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.00012148300117233296,
+      "loss": 0.4883,
+      "step": 2680
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.00012118991793669403,
+      "loss": 0.4268,
+      "step": 2690
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.0001208968347010551,
+      "loss": 0.2872,
+      "step": 2700
+    },
+    {
+      "epoch": 1.58,
+      "eval_accuracy": 0.8573472599799685,
+      "eval_loss": 0.46980831027030945,
+      "eval_runtime": 142.0482,
+      "eval_samples_per_second": 49.202,
+      "eval_steps_per_second": 6.153,
+      "step": 2700
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.00012060375146541618,
+      "loss": 0.356,
+      "step": 2710
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 0.00012031066822977726,
+      "loss": 0.4152,
+      "step": 2720
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.00012001758499413834,
+      "loss": 0.3628,
+      "step": 2730
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.00011972450175849944,
+      "loss": 0.3722,
+      "step": 2740
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.00011943141852286051,
+      "loss": 0.3239,
+      "step": 2750
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.00011913833528722158,
+      "loss": 0.2286,
+      "step": 2760
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.00011884525205158265,
+      "loss": 0.2773,
+      "step": 2770
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 0.00011855216881594373,
+      "loss": 0.3792,
+      "step": 2780
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.00011825908558030481,
+      "loss": 0.3792,
+      "step": 2790
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.00011796600234466589,
+      "loss": 0.2451,
+      "step": 2800
+    },
+    {
+      "epoch": 1.64,
+      "eval_accuracy": 0.8729431964515667,
+      "eval_loss": 0.42098021507263184,
+      "eval_runtime": 141.8474,
+      "eval_samples_per_second": 49.271,
+      "eval_steps_per_second": 6.162,
+      "step": 2800
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.00011767291910902699,
+      "loss": 0.3349,
+      "step": 2810
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.00011737983587338806,
+      "loss": 0.2566,
+      "step": 2820
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.00011708675263774913,
+      "loss": 0.2207,
+      "step": 2830
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 0.0001167936694021102,
+      "loss": 0.2553,
+      "step": 2840
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.00011650058616647128,
+      "loss": 0.3184,
+      "step": 2850
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.00011620750293083236,
+      "loss": 0.2927,
+      "step": 2860
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.00011591441969519344,
+      "loss": 0.3839,
+      "step": 2870
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.00011562133645955454,
+      "loss": 0.3236,
+      "step": 2880
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.00011532825322391561,
+      "loss": 0.3058,
+      "step": 2890
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 0.00011503516998827668,
+      "loss": 0.2211,
+      "step": 2900
+    },
+    {
+      "epoch": 1.7,
+      "eval_accuracy": 0.8851051652596938,
+      "eval_loss": 0.36445751786231995,
+      "eval_runtime": 142.4316,
+      "eval_samples_per_second": 49.069,
+      "eval_steps_per_second": 6.136,
+      "step": 2900
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.00011474208675263775,
+      "loss": 0.4079,
+      "step": 2910
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.00011444900351699883,
+      "loss": 0.285,
+      "step": 2920
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.00011415592028135991,
+      "loss": 0.268,
+      "step": 2930
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.00011386283704572099,
+      "loss": 0.3587,
+      "step": 2940
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.00011356975381008206,
+      "loss": 0.2891,
+      "step": 2950
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.00011327667057444316,
+      "loss": 0.4267,
+      "step": 2960
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.00011298358733880423,
+      "loss": 0.3827,
+      "step": 2970
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.0001126905041031653,
+      "loss": 0.2797,
+      "step": 2980
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.00011239742086752638,
+      "loss": 0.2858,
+      "step": 2990
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.00011210433763188745,
+      "loss": 0.3145,
+      "step": 3000
+    },
+    {
+      "epoch": 1.76,
+      "eval_accuracy": 0.871512376591787,
+      "eval_loss": 0.41394171118736267,
+      "eval_runtime": 143.8279,
+      "eval_samples_per_second": 48.593,
+      "eval_steps_per_second": 6.077,
+      "step": 3000
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.00011181125439624854,
+      "loss": 0.2412,
+      "step": 3010
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 0.00011151817116060961,
+      "loss": 0.4263,
+      "step": 3020
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.00011122508792497071,
+      "loss": 0.2569,
+      "step": 3030
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.00011093200468933178,
+      "loss": 0.3211,
+      "step": 3040
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.00011063892145369286,
+      "loss": 0.2235,
+      "step": 3050
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.00011034583821805393,
+      "loss": 0.2982,
+      "step": 3060
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.000110052754982415,
+      "loss": 0.2236,
+      "step": 3070
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.00010975967174677609,
+      "loss": 0.2339,
+      "step": 3080
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 0.00010946658851113716,
+      "loss": 0.2568,
+      "step": 3090
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.00010917350527549826,
+      "loss": 0.2001,
+      "step": 3100
+    },
+    {
+      "epoch": 1.82,
+      "eval_accuracy": 0.8863929031334955,
+      "eval_loss": 0.3604518175125122,
+      "eval_runtime": 143.3768,
+      "eval_samples_per_second": 48.746,
+      "eval_steps_per_second": 6.096,
+      "step": 3100
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.00010888042203985933,
+      "loss": 0.2798,
+      "step": 3110
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.0001085873388042204,
+      "loss": 0.2949,
+      "step": 3120
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.00010829425556858148,
+      "loss": 0.2569,
+      "step": 3130
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.00010800117233294255,
+      "loss": 0.2033,
+      "step": 3140
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.00010770808909730364,
+      "loss": 0.2469,
+      "step": 3150
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 0.00010741500586166471,
+      "loss": 0.3182,
+      "step": 3160
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.00010712192262602581,
+      "loss": 0.2325,
+      "step": 3170
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.00010682883939038688,
+      "loss": 0.3356,
+      "step": 3180
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.00010653575615474796,
+      "loss": 0.2707,
+      "step": 3190
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.00010624267291910903,
+      "loss": 0.3095,
+      "step": 3200
+    },
+    {
+      "epoch": 1.88,
+      "eval_accuracy": 0.8675060809844041,
+      "eval_loss": 0.4273549020290375,
+      "eval_runtime": 143.5124,
+      "eval_samples_per_second": 48.7,
+      "eval_steps_per_second": 6.09,
+      "step": 3200
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 0.0001059495896834701,
+      "loss": 0.3672,
+      "step": 3210
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.00010565650644783119,
+      "loss": 0.3408,
+      "step": 3220
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.00010536342321219226,
+      "loss": 0.277,
+      "step": 3230
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.00010507033997655336,
+      "loss": 0.2123,
+      "step": 3240
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.00010477725674091443,
+      "loss": 0.1926,
+      "step": 3250
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0001044841735052755,
+      "loss": 0.2049,
+      "step": 3260
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.00010419109026963658,
+      "loss": 0.3891,
+      "step": 3270
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 0.00010389800703399765,
+      "loss": 0.3387,
+      "step": 3280
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.00010360492379835874,
+      "loss": 0.2356,
+      "step": 3290
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.00010331184056271981,
+      "loss": 0.1915,
+      "step": 3300
+    },
+    {
+      "epoch": 1.93,
+      "eval_accuracy": 0.9101445128058377,
+      "eval_loss": 0.2909921109676361,
+      "eval_runtime": 142.0389,
+      "eval_samples_per_second": 49.205,
+      "eval_steps_per_second": 6.153,
+      "step": 3300
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.00010301875732708091,
+      "loss": 0.2416,
+      "step": 3310
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.00010272567409144198,
+      "loss": 0.2543,
+      "step": 3320
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 0.00010243259085580306,
+      "loss": 0.2175,
+      "step": 3330
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.00010213950762016413,
+      "loss": 0.335,
+      "step": 3340
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.0001018464243845252,
+      "loss": 0.3157,
+      "step": 3350
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.00010155334114888629,
+      "loss": 0.2386,
+      "step": 3360
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.00010126025791324736,
+      "loss": 0.289,
+      "step": 3370
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.00010096717467760846,
+      "loss": 0.2696,
+      "step": 3380
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.00010067409144196953,
+      "loss": 0.504,
+      "step": 3390
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 0.00010038100820633061,
+      "loss": 0.2465,
+      "step": 3400
+    },
+    {
+      "epoch": 1.99,
+      "eval_accuracy": 0.9102875947918158,
+      "eval_loss": 0.27264514565467834,
+      "eval_runtime": 141.7632,
+      "eval_samples_per_second": 49.301,
+      "eval_steps_per_second": 6.165,
+      "step": 3400
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.00010008792497069168,
+      "loss": 0.1975,
+      "step": 3410
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 9.979484173505275e-05,
+      "loss": 0.1501,
+      "step": 3420
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 9.950175849941384e-05,
+      "loss": 0.1063,
+      "step": 3430
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 9.920867526377493e-05,
+      "loss": 0.125,
+      "step": 3440
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 9.8915592028136e-05,
+      "loss": 0.0868,
+      "step": 3450
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 9.862250879249707e-05,
+      "loss": 0.1477,
+      "step": 3460
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 9.832942555685816e-05,
+      "loss": 0.1333,
+      "step": 3470
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 9.803634232121923e-05,
+      "loss": 0.0903,
+      "step": 3480
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 9.77432590855803e-05,
+      "loss": 0.0753,
+      "step": 3490
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 9.745017584994138e-05,
+      "loss": 0.1218,
+      "step": 3500
+    },
+    {
+      "epoch": 2.05,
+      "eval_accuracy": 0.9128630705394191,
+      "eval_loss": 0.2742130756378174,
+      "eval_runtime": 141.9071,
+      "eval_samples_per_second": 49.251,
+      "eval_steps_per_second": 6.159,
+      "step": 3500
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 9.715709261430248e-05,
+      "loss": 0.1116,
+      "step": 3510
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 9.686400937866355e-05,
+      "loss": 0.1161,
+      "step": 3520
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 9.657092614302462e-05,
+      "loss": 0.1803,
+      "step": 3530
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 9.627784290738571e-05,
+      "loss": 0.0855,
+      "step": 3540
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 9.598475967174678e-05,
+      "loss": 0.1219,
+      "step": 3550
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 9.569167643610785e-05,
+      "loss": 0.128,
+      "step": 3560
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 9.539859320046893e-05,
+      "loss": 0.1373,
+      "step": 3570
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 9.510550996483001e-05,
+      "loss": 0.1646,
+      "step": 3580
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 9.48124267291911e-05,
+      "loss": 0.1103,
+      "step": 3590
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 9.451934349355217e-05,
+      "loss": 0.0752,
+      "step": 3600
+    },
+    {
+      "epoch": 2.11,
+      "eval_accuracy": 0.9183001860065818,
+      "eval_loss": 0.25723978877067566,
+      "eval_runtime": 141.3288,
+      "eval_samples_per_second": 49.452,
+      "eval_steps_per_second": 6.184,
+      "step": 3600
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 9.422626025791326e-05,
+      "loss": 0.1596,
+      "step": 3610
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 9.393317702227433e-05,
+      "loss": 0.1059,
+      "step": 3620
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 9.36400937866354e-05,
+      "loss": 0.186,
+      "step": 3630
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 9.334701055099648e-05,
+      "loss": 0.106,
+      "step": 3640
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 9.305392731535756e-05,
+      "loss": 0.1323,
+      "step": 3650
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 9.276084407971865e-05,
+      "loss": 0.1517,
+      "step": 3660
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 9.246776084407972e-05,
+      "loss": 0.1218,
+      "step": 3670
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 9.217467760844081e-05,
+      "loss": 0.1591,
+      "step": 3680
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 9.188159437280188e-05,
+      "loss": 0.129,
+      "step": 3690
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 9.158851113716295e-05,
+      "loss": 0.1067,
+      "step": 3700
+    },
+    {
+      "epoch": 2.17,
+      "eval_accuracy": 0.9203033338102733,
+      "eval_loss": 0.25835496187210083,
+      "eval_runtime": 155.4296,
+      "eval_samples_per_second": 44.966,
+      "eval_steps_per_second": 5.623,
+      "step": 3700
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 9.129542790152403e-05,
+      "loss": 0.0996,
+      "step": 3710
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 9.100234466588511e-05,
+      "loss": 0.1542,
+      "step": 3720
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 9.07092614302462e-05,
+      "loss": 0.1354,
+      "step": 3730
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 9.041617819460727e-05,
+      "loss": 0.1129,
+      "step": 3740
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 9.012309495896835e-05,
+      "loss": 0.188,
+      "step": 3750
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 8.983001172332943e-05,
+      "loss": 0.1329,
+      "step": 3760
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 8.95369284876905e-05,
+      "loss": 0.1951,
+      "step": 3770
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 8.924384525205158e-05,
+      "loss": 0.0462,
+      "step": 3780
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 8.895076201641266e-05,
+      "loss": 0.0564,
+      "step": 3790
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 8.865767878077375e-05,
+      "loss": 0.0838,
+      "step": 3800
+    },
+    {
+      "epoch": 2.23,
+      "eval_accuracy": 0.921161825726141,
+      "eval_loss": 0.2457675188779831,
+      "eval_runtime": 162.6766,
+      "eval_samples_per_second": 42.963,
+      "eval_steps_per_second": 5.373,
+      "step": 3800
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 8.836459554513482e-05,
+      "loss": 0.151,
+      "step": 3810
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 8.80715123094959e-05,
+      "loss": 0.1161,
+      "step": 3820
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 8.777842907385698e-05,
+      "loss": 0.1928,
+      "step": 3830
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 8.748534583821806e-05,
+      "loss": 0.0891,
+      "step": 3840
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 8.719226260257913e-05,
+      "loss": 0.1545,
+      "step": 3850
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 8.689917936694021e-05,
+      "loss": 0.1813,
+      "step": 3860
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 8.66060961313013e-05,
+      "loss": 0.1153,
+      "step": 3870
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 8.631301289566237e-05,
+      "loss": 0.0978,
+      "step": 3880
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 8.601992966002345e-05,
+      "loss": 0.1,
+      "step": 3890
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 8.572684642438453e-05,
+      "loss": 0.1106,
+      "step": 3900
+    },
+    {
+      "epoch": 2.29,
+      "eval_accuracy": 0.9237373014737444,
+      "eval_loss": 0.24122054874897003,
+      "eval_runtime": 164.1489,
+      "eval_samples_per_second": 42.577,
+      "eval_steps_per_second": 5.324,
+      "step": 3900
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 8.54337631887456e-05,
+      "loss": 0.1534,
+      "step": 3910
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 8.514067995310668e-05,
+      "loss": 0.1362,
+      "step": 3920
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 8.484759671746777e-05,
+      "loss": 0.1671,
+      "step": 3930
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 8.455451348182885e-05,
+      "loss": 0.1264,
+      "step": 3940
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 8.426143024618992e-05,
+      "loss": 0.0528,
+      "step": 3950
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 8.3968347010551e-05,
+      "loss": 0.1812,
+      "step": 3960
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 8.367526377491208e-05,
+      "loss": 0.0458,
+      "step": 3970
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 8.338218053927316e-05,
+      "loss": 0.1345,
+      "step": 3980
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 8.308909730363423e-05,
+      "loss": 0.0905,
+      "step": 3990
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 8.279601406799532e-05,
+      "loss": 0.092,
+      "step": 4000
+    },
+    {
+      "epoch": 2.34,
+      "eval_accuracy": 0.9277435970811275,
+      "eval_loss": 0.22324928641319275,
+      "eval_runtime": 164.4007,
+      "eval_samples_per_second": 42.512,
+      "eval_steps_per_second": 5.316,
+      "step": 4000
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 8.25029308323564e-05,
+      "loss": 0.1248,
+      "step": 4010
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 8.220984759671747e-05,
+      "loss": 0.1342,
+      "step": 4020
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 8.191676436107855e-05,
+      "loss": 0.1925,
+      "step": 4030
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 8.162368112543963e-05,
+      "loss": 0.1011,
+      "step": 4040
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 8.13305978898007e-05,
+      "loss": 0.0761,
+      "step": 4050
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 8.106682297772568e-05,
+      "loss": 0.1509,
+      "step": 4060
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 8.077373974208675e-05,
+      "loss": 0.0951,
+      "step": 4070
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 8.048065650644784e-05,
+      "loss": 0.2193,
+      "step": 4080
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 8.018757327080891e-05,
+      "loss": 0.0828,
+      "step": 4090
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 7.989449003516999e-05,
+      "loss": 0.1056,
+      "step": 4100
+    },
+    {
+      "epoch": 2.4,
+      "eval_accuracy": 0.9077121190442123,
+      "eval_loss": 0.2816599905490875,
+      "eval_runtime": 161.9883,
+      "eval_samples_per_second": 43.145,
+      "eval_steps_per_second": 5.395,
+      "step": 4100
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 7.963071512309497e-05,
+      "loss": 0.1311,
+      "step": 4110
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 7.933763188745605e-05,
+      "loss": 0.042,
+      "step": 4120
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 7.904454865181712e-05,
+      "loss": 0.0865,
+      "step": 4130
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 7.875146541617819e-05,
+      "loss": 0.0694,
+      "step": 4140
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 7.845838218053928e-05,
+      "loss": 0.085,
+      "step": 4150
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 7.816529894490035e-05,
+      "loss": 0.0738,
+      "step": 4160
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 7.787221570926144e-05,
+      "loss": 0.0573,
+      "step": 4170
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 7.757913247362252e-05,
+      "loss": 0.0758,
+      "step": 4180
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 7.72860492379836e-05,
+      "loss": 0.144,
+      "step": 4190
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 7.699296600234467e-05,
+      "loss": 0.0696,
+      "step": 4200
+    },
+    {
+      "epoch": 2.46,
+      "eval_accuracy": 0.9284590070110174,
+      "eval_loss": 0.23343555629253387,
+      "eval_runtime": 162.6927,
+      "eval_samples_per_second": 42.958,
+      "eval_steps_per_second": 5.372,
+      "step": 4200
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 7.669988276670574e-05,
+      "loss": 0.0767,
+      "step": 4210
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 7.640679953106683e-05,
+      "loss": 0.1279,
+      "step": 4220
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 7.61137162954279e-05,
+      "loss": 0.0458,
+      "step": 4230
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 7.582063305978899e-05,
+      "loss": 0.0958,
+      "step": 4240
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 7.552754982415006e-05,
+      "loss": 0.1067,
+      "step": 4250
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 7.523446658851115e-05,
+      "loss": 0.0582,
+      "step": 4260
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 7.494138335287222e-05,
+      "loss": 0.0322,
+      "step": 4270
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 7.464830011723329e-05,
+      "loss": 0.069,
+      "step": 4280
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 7.435521688159438e-05,
+      "loss": 0.0548,
+      "step": 4290
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 7.406213364595545e-05,
+      "loss": 0.0444,
+      "step": 4300
+    },
+    {
+      "epoch": 2.52,
+      "eval_accuracy": 0.9363285162398054,
+      "eval_loss": 0.21417850255966187,
+      "eval_runtime": 164.9686,
+      "eval_samples_per_second": 42.366,
+      "eval_steps_per_second": 5.298,
+      "step": 4300
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 7.376905041031654e-05,
+      "loss": 0.086,
+      "step": 4310
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 7.347596717467761e-05,
+      "loss": 0.0917,
+      "step": 4320
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 7.31828839390387e-05,
+      "loss": 0.0744,
+      "step": 4330
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 7.288980070339977e-05,
+      "loss": 0.0346,
+      "step": 4340
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 7.259671746776084e-05,
+      "loss": 0.0869,
+      "step": 4350
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 7.230363423212193e-05,
+      "loss": 0.0513,
+      "step": 4360
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 7.2010550996483e-05,
+      "loss": 0.1099,
+      "step": 4370
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 7.171746776084409e-05,
+      "loss": 0.102,
+      "step": 4380
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 7.142438452520516e-05,
+      "loss": 0.0844,
+      "step": 4390
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 7.113130128956625e-05,
+      "loss": 0.1046,
+      "step": 4400
+    },
+    {
+      "epoch": 2.58,
+      "eval_accuracy": 0.9351838603519816,
+      "eval_loss": 0.2036106288433075,
+      "eval_runtime": 164.8201,
+      "eval_samples_per_second": 42.404,
+      "eval_steps_per_second": 5.303,
+      "step": 4400
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 7.083821805392732e-05,
+      "loss": 0.0337,
+      "step": 4410
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 7.054513481828839e-05,
+      "loss": 0.0956,
+      "step": 4420
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 7.025205158264948e-05,
+      "loss": 0.0669,
+      "step": 4430
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 6.995896834701055e-05,
+      "loss": 0.0814,
+      "step": 4440
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 6.966588511137164e-05,
+      "loss": 0.0544,
+      "step": 4450
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 6.937280187573271e-05,
+      "loss": 0.087,
+      "step": 4460
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 6.90797186400938e-05,
+      "loss": 0.111,
+      "step": 4470
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 6.878663540445487e-05,
+      "loss": 0.0844,
+      "step": 4480
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 6.849355216881594e-05,
+      "loss": 0.0468,
+      "step": 4490
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 6.820046893317703e-05,
+      "loss": 0.066,
+      "step": 4500
+    },
+    {
+      "epoch": 2.64,
+      "eval_accuracy": 0.9364715982257834,
+      "eval_loss": 0.2114696204662323,
+      "eval_runtime": 164.4144,
+      "eval_samples_per_second": 42.508,
+      "eval_steps_per_second": 5.316,
+      "step": 4500
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 6.79073856975381e-05,
+      "loss": 0.0762,
+      "step": 4510
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 6.761430246189917e-05,
+      "loss": 0.077,
+      "step": 4520
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 6.732121922626026e-05,
+      "loss": 0.0801,
+      "step": 4530
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 6.702813599062135e-05,
+      "loss": 0.1392,
+      "step": 4540
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 6.673505275498242e-05,
+      "loss": 0.04,
+      "step": 4550
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 6.64419695193435e-05,
+      "loss": 0.0774,
+      "step": 4560
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 6.614888628370457e-05,
+      "loss": 0.098,
+      "step": 4570
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 6.585580304806565e-05,
+      "loss": 0.0882,
+      "step": 4580
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 6.556271981242673e-05,
+      "loss": 0.0846,
+      "step": 4590
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 6.526963657678781e-05,
+      "loss": 0.0649,
+      "step": 4600
+    },
+    {
+      "epoch": 2.7,
+      "eval_accuracy": 0.9447703534125054,
+      "eval_loss": 0.1729690283536911,
+      "eval_runtime": 164.155,
+      "eval_samples_per_second": 42.576,
+      "eval_steps_per_second": 5.324,
+      "step": 4600
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 6.49765533411489e-05,
+      "loss": 0.0805,
+      "step": 4610
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 6.468347010550997e-05,
+      "loss": 0.0386,
+      "step": 4620
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 6.439038686987104e-05,
+      "loss": 0.0576,
+      "step": 4630
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 6.409730363423212e-05,
+      "loss": 0.0721,
+      "step": 4640
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 6.38042203985932e-05,
+      "loss": 0.1281,
+      "step": 4650
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 6.351113716295428e-05,
+      "loss": 0.0783,
+      "step": 4660
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 6.321805392731536e-05,
+      "loss": 0.0638,
+      "step": 4670
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 6.292497069167645e-05,
+      "loss": 0.1168,
+      "step": 4680
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 6.263188745603752e-05,
+      "loss": 0.0586,
+      "step": 4690
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 6.23388042203986e-05,
+      "loss": 0.0513,
+      "step": 4700
+    },
+    {
+      "epoch": 2.75,
+      "eval_accuracy": 0.93389612247818,
+      "eval_loss": 0.21480464935302734,
+      "eval_runtime": 164.883,
+      "eval_samples_per_second": 42.388,
+      "eval_steps_per_second": 5.301,
+      "step": 4700
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 6.204572098475967e-05,
+      "loss": 0.0793,
+      "step": 4710
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 6.175263774912075e-05,
+      "loss": 0.0444,
+      "step": 4720
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 6.145955451348183e-05,
+      "loss": 0.0528,
+      "step": 4730
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 6.116647127784291e-05,
+      "loss": 0.05,
+      "step": 4740
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 6.087338804220399e-05,
+      "loss": 0.086,
+      "step": 4750
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 6.058030480656507e-05,
+      "loss": 0.0332,
+      "step": 4760
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 6.0287221570926144e-05,
+      "loss": 0.0645,
+      "step": 4770
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 5.999413833528722e-05,
+      "loss": 0.084,
+      "step": 4780
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 5.9701055099648304e-05,
+      "loss": 0.0662,
+      "step": 4790
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 5.940797186400938e-05,
+      "loss": 0.0917,
+      "step": 4800
+    },
+    {
+      "epoch": 2.81,
+      "eval_accuracy": 0.9437687795106596,
+      "eval_loss": 0.18099913001060486,
+      "eval_runtime": 163.6306,
+      "eval_samples_per_second": 42.712,
+      "eval_steps_per_second": 5.341,
+      "step": 4800
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 5.9114888628370456e-05,
+      "loss": 0.0719,
+      "step": 4810
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 5.882180539273154e-05,
+      "loss": 0.0301,
+      "step": 4820
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 5.852872215709262e-05,
+      "loss": 0.0396,
+      "step": 4830
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 5.8235638921453695e-05,
+      "loss": 0.0671,
+      "step": 4840
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 5.794255568581477e-05,
+      "loss": 0.0185,
+      "step": 4850
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 5.7649472450175854e-05,
+      "loss": 0.0473,
+      "step": 4860
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 5.7356389214536934e-05,
+      "loss": 0.079,
+      "step": 4870
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 5.7063305978898006e-05,
+      "loss": 0.0578,
+      "step": 4880
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 5.677022274325909e-05,
+      "loss": 0.0324,
+      "step": 4890
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 5.647713950762017e-05,
+      "loss": 0.0879,
+      "step": 4900
+    },
+    {
+      "epoch": 2.87,
+      "eval_accuracy": 0.9387609100014308,
+      "eval_loss": 0.19710442423820496,
+      "eval_runtime": 164.6231,
+      "eval_samples_per_second": 42.455,
+      "eval_steps_per_second": 5.309,
+      "step": 4900
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 5.6184056271981245e-05,
+      "loss": 0.0679,
+      "step": 4910
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 5.589097303634232e-05,
+      "loss": 0.0441,
+      "step": 4920
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 5.5597889800703404e-05,
+      "loss": 0.0352,
+      "step": 4930
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 5.5304806565064484e-05,
+      "loss": 0.0405,
+      "step": 4940
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 5.501172332942556e-05,
+      "loss": 0.0397,
+      "step": 4950
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 5.471864009378663e-05,
+      "loss": 0.0531,
+      "step": 4960
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 5.4425556858147716e-05,
+      "loss": 0.0564,
+      "step": 4970
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 5.4132473622508796e-05,
+      "loss": 0.048,
+      "step": 4980
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 5.383939038686987e-05,
+      "loss": 0.0597,
+      "step": 4990
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 5.3546307151230955e-05,
+      "loss": 0.1052,
+      "step": 5000
+    },
+    {
+      "epoch": 2.93,
+      "eval_accuracy": 0.9507797968235799,
+      "eval_loss": 0.16023729741573334,
+      "eval_runtime": 164.4511,
+      "eval_samples_per_second": 42.499,
+      "eval_steps_per_second": 5.315,
+      "step": 5000
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 5.3253223915592034e-05,
+      "loss": 0.0319,
+      "step": 5010
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 5.296014067995311e-05,
+      "loss": 0.0565,
+      "step": 5020
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 5.266705744431418e-05,
+      "loss": 0.0545,
+      "step": 5030
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 5.2373974208675266e-05,
+      "loss": 0.0658,
+      "step": 5040
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 5.2080890973036346e-05,
+      "loss": 0.0377,
+      "step": 5050
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 5.178780773739742e-05,
+      "loss": 0.0384,
+      "step": 5060
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 5.1494724501758505e-05,
+      "loss": 0.0243,
+      "step": 5070
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 5.1201641266119585e-05,
+      "loss": 0.0173,
+      "step": 5080
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 5.090855803048066e-05,
+      "loss": 0.0739,
+      "step": 5090
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 5.061547479484173e-05,
+      "loss": 0.0362,
+      "step": 5100
+    },
+    {
+      "epoch": 2.99,
+      "eval_accuracy": 0.9556445843468307,
+      "eval_loss": 0.1475149393081665,
+      "eval_runtime": 163.8639,
+      "eval_samples_per_second": 42.651,
+      "eval_steps_per_second": 5.334,
+      "step": 5100
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 5.032239155920282e-05,
+      "loss": 0.0649,
+      "step": 5110
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 5.0029308323563896e-05,
+      "loss": 0.0223,
+      "step": 5120
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 4.9736225087924976e-05,
+      "loss": 0.0067,
+      "step": 5130
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 4.944314185228605e-05,
+      "loss": 0.0183,
+      "step": 5140
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 4.9150058616647135e-05,
+      "loss": 0.0171,
+      "step": 5150
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 4.885697538100821e-05,
+      "loss": 0.0287,
+      "step": 5160
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 4.856389214536929e-05,
+      "loss": 0.0121,
+      "step": 5170
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 4.827080890973037e-05,
+      "loss": 0.0064,
+      "step": 5180
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 4.797772567409145e-05,
+      "loss": 0.0179,
+      "step": 5190
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 4.768464243845252e-05,
+      "loss": 0.041,
+      "step": 5200
+    },
+    {
+      "epoch": 3.05,
+      "eval_accuracy": 0.9585062240663901,
+      "eval_loss": 0.13277386128902435,
+      "eval_runtime": 162.7191,
+      "eval_samples_per_second": 42.951,
+      "eval_steps_per_second": 5.371,
+      "step": 5200
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 4.73915592028136e-05,
+      "loss": 0.0101,
+      "step": 5210
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 4.709847596717468e-05,
+      "loss": 0.0035,
+      "step": 5220
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 4.680539273153576e-05,
+      "loss": 0.0073,
+      "step": 5230
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 4.651230949589684e-05,
+      "loss": 0.0123,
+      "step": 5240
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 4.621922626025791e-05,
+      "loss": 0.0272,
+      "step": 5250
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 4.5926143024619e-05,
+      "loss": 0.0129,
+      "step": 5260
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 4.563305978898007e-05,
+      "loss": 0.0255,
+      "step": 5270
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 4.533997655334115e-05,
+      "loss": 0.0218,
+      "step": 5280
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 4.504689331770223e-05,
+      "loss": 0.0068,
+      "step": 5290
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 4.475381008206331e-05,
+      "loss": 0.0156,
+      "step": 5300
+    },
+    {
+      "epoch": 3.11,
+      "eval_accuracy": 0.9570754042066104,
+      "eval_loss": 0.13893470168113708,
+      "eval_runtime": 163.8232,
+      "eval_samples_per_second": 42.662,
+      "eval_steps_per_second": 5.335,
+      "step": 5300
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 4.446072684642439e-05,
+      "loss": 0.0103,
+      "step": 5310
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 4.416764361078546e-05,
+      "loss": 0.04,
+      "step": 5320
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 4.387456037514655e-05,
+      "loss": 0.0212,
+      "step": 5330
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 4.358147713950762e-05,
+      "loss": 0.0047,
+      "step": 5340
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 4.32883939038687e-05,
+      "loss": 0.014,
+      "step": 5350
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 4.299531066822978e-05,
+      "loss": 0.005,
+      "step": 5360
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 4.270222743259086e-05,
+      "loss": 0.0296,
+      "step": 5370
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 4.240914419695194e-05,
+      "loss": 0.0037,
+      "step": 5380
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 4.211606096131301e-05,
+      "loss": 0.0559,
+      "step": 5390
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 4.18229777256741e-05,
+      "loss": 0.0047,
+      "step": 5400
+    },
+    {
+      "epoch": 3.17,
+      "eval_accuracy": 0.9638002575475748,
+      "eval_loss": 0.12242697924375534,
+      "eval_runtime": 164.1713,
+      "eval_samples_per_second": 42.571,
+      "eval_steps_per_second": 5.324,
+      "step": 5400
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 4.152989449003517e-05,
+      "loss": 0.0042,
+      "step": 5410
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 4.123681125439625e-05,
+      "loss": 0.0082,
+      "step": 5420
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 4.094372801875733e-05,
+      "loss": 0.0034,
+      "step": 5430
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 4.065064478311841e-05,
+      "loss": 0.0092,
+      "step": 5440
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 4.035756154747949e-05,
+      "loss": 0.0039,
+      "step": 5450
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 4.006447831184056e-05,
+      "loss": 0.0375,
+      "step": 5460
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 3.977139507620165e-05,
+      "loss": 0.0033,
+      "step": 5470
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 3.947831184056272e-05,
+      "loss": 0.0089,
+      "step": 5480
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 3.91852286049238e-05,
+      "loss": 0.0028,
+      "step": 5490
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 3.8892145369284874e-05,
+      "loss": 0.0174,
+      "step": 5500
+    },
+    {
+      "epoch": 3.22,
+      "eval_accuracy": 0.9650879954213765,
+      "eval_loss": 0.11929039657115936,
+      "eval_runtime": 163.6627,
+      "eval_samples_per_second": 42.704,
+      "eval_steps_per_second": 5.34,
+      "step": 5500
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 3.859906213364596e-05,
+      "loss": 0.004,
+      "step": 5510
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 3.830597889800703e-05,
+      "loss": 0.0091,
+      "step": 5520
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 3.801289566236811e-05,
+      "loss": 0.0044,
+      "step": 5530
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 3.771981242672919e-05,
+      "loss": 0.0061,
+      "step": 5540
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 3.742672919109027e-05,
+      "loss": 0.0065,
+      "step": 5550
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 3.713364595545135e-05,
+      "loss": 0.0106,
+      "step": 5560
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 3.6840562719812424e-05,
+      "loss": 0.008,
+      "step": 5570
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 3.654747948417351e-05,
+      "loss": 0.0162,
+      "step": 5580
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 3.625439624853458e-05,
+      "loss": 0.009,
+      "step": 5590
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 3.596131301289566e-05,
+      "loss": 0.0087,
+      "step": 5600
+    },
+    {
+      "epoch": 3.28,
+      "eval_accuracy": 0.9622263557018171,
+      "eval_loss": 0.12759189307689667,
+      "eval_runtime": 163.9026,
+      "eval_samples_per_second": 42.641,
+      "eval_steps_per_second": 5.332,
+      "step": 5600
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 3.566822977725674e-05,
+      "loss": 0.0094,
+      "step": 5610
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 3.537514654161782e-05,
+      "loss": 0.0125,
+      "step": 5620
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 3.50820633059789e-05,
+      "loss": 0.0086,
+      "step": 5630
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 3.4788980070339975e-05,
+      "loss": 0.027,
+      "step": 5640
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 3.449589683470106e-05,
+      "loss": 0.0098,
+      "step": 5650
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 3.4202813599062134e-05,
+      "loss": 0.0269,
+      "step": 5660
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 3.390973036342321e-05,
+      "loss": 0.0052,
+      "step": 5670
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 3.361664712778429e-05,
+      "loss": 0.007,
+      "step": 5680
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 3.332356389214537e-05,
+      "loss": 0.0048,
+      "step": 5690
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 3.303048065650645e-05,
+      "loss": 0.0084,
+      "step": 5700
+    },
+    {
+      "epoch": 3.34,
+      "eval_accuracy": 0.9662326513092002,
+      "eval_loss": 0.11341094970703125,
+      "eval_runtime": 163.1714,
+      "eval_samples_per_second": 42.832,
+      "eval_steps_per_second": 5.356,
+      "step": 5700
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 3.2737397420867525e-05,
+      "loss": 0.0069,
+      "step": 5710
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 3.244431418522861e-05,
+      "loss": 0.0221,
+      "step": 5720
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 3.2151230949589684e-05,
+      "loss": 0.0104,
+      "step": 5730
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 3.1858147713950764e-05,
+      "loss": 0.0036,
+      "step": 5740
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 3.1565064478311837e-05,
+      "loss": 0.0026,
+      "step": 5750
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 3.127198124267292e-05,
+      "loss": 0.0053,
+      "step": 5760
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 3.0978898007034e-05,
+      "loss": 0.0101,
+      "step": 5770
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 3.0685814771395075e-05,
+      "loss": 0.0083,
+      "step": 5780
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 3.0392731535756158e-05,
+      "loss": 0.0082,
+      "step": 5790
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 3.0099648300117235e-05,
+      "loss": 0.0141,
+      "step": 5800
+    },
+    {
+      "epoch": 3.4,
+      "eval_accuracy": 0.963084847617685,
+      "eval_loss": 0.12386169284582138,
+      "eval_runtime": 164.253,
+      "eval_samples_per_second": 42.55,
+      "eval_steps_per_second": 5.321,
+      "step": 5800
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 2.9806565064478314e-05,
+      "loss": 0.0041,
+      "step": 5810
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 2.951348182883939e-05,
+      "loss": 0.0075,
+      "step": 5820
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 2.922039859320047e-05,
+      "loss": 0.0064,
+      "step": 5830
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 2.8927315357561546e-05,
+      "loss": 0.018,
+      "step": 5840
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 2.8634232121922626e-05,
+      "loss": 0.0052,
+      "step": 5850
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 2.834114888628371e-05,
+      "loss": 0.0049,
+      "step": 5860
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 2.8048065650644785e-05,
+      "loss": 0.004,
+      "step": 5870
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 2.7754982415005865e-05,
+      "loss": 0.0325,
+      "step": 5880
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 2.746189917936694e-05,
+      "loss": 0.0048,
+      "step": 5890
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 2.716881594372802e-05,
+      "loss": 0.0291,
+      "step": 5900
+    },
+    {
+      "epoch": 3.46,
+      "eval_accuracy": 0.9645156674774645,
+      "eval_loss": 0.1198735386133194,
+      "eval_runtime": 162.4699,
+      "eval_samples_per_second": 43.017,
+      "eval_steps_per_second": 5.379,
+      "step": 5900
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 2.6875732708089097e-05,
+      "loss": 0.0153,
+      "step": 5910
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 2.6582649472450176e-05,
+      "loss": 0.0131,
+      "step": 5920
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 2.628956623681126e-05,
+      "loss": 0.0074,
+      "step": 5930
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 2.5996483001172335e-05,
+      "loss": 0.0028,
+      "step": 5940
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 2.5703399765533415e-05,
+      "loss": 0.0065,
+      "step": 5950
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 2.541031652989449e-05,
+      "loss": 0.0081,
+      "step": 5960
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 2.511723329425557e-05,
+      "loss": 0.0438,
+      "step": 5970
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 2.482415005861665e-05,
+      "loss": 0.0041,
+      "step": 5980
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 2.4531066822977727e-05,
+      "loss": 0.003,
+      "step": 5990
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 2.4237983587338806e-05,
+      "loss": 0.0049,
+      "step": 6000
+    },
+    {
+      "epoch": 3.52,
+      "eval_accuracy": 0.9679496351409358,
+      "eval_loss": 0.11025016009807587,
+      "eval_runtime": 163.4902,
+      "eval_samples_per_second": 42.749,
+      "eval_steps_per_second": 5.346,
+      "step": 6000
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 2.3944900351699882e-05,
+      "loss": 0.0028,
+      "step": 6010
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 2.3651817116060962e-05,
+      "loss": 0.0028,
+      "step": 6020
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 2.335873388042204e-05,
+      "loss": 0.0022,
+      "step": 6030
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 2.3065650644783118e-05,
+      "loss": 0.0028,
+      "step": 6040
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 2.27725674091442e-05,
+      "loss": 0.0139,
+      "step": 6050
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 2.2479484173505277e-05,
+      "loss": 0.0051,
+      "step": 6060
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 2.2186400937866357e-05,
+      "loss": 0.0038,
+      "step": 6070
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 2.1893317702227433e-05,
+      "loss": 0.0237,
+      "step": 6080
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 2.1600234466588512e-05,
+      "loss": 0.0108,
+      "step": 6090
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 2.130715123094959e-05,
+      "loss": 0.0055,
+      "step": 6100
+    },
+    {
+      "epoch": 3.58,
+      "eval_accuracy": 0.9662326513092002,
+      "eval_loss": 0.11199595779180527,
+      "eval_runtime": 163.683,
+      "eval_samples_per_second": 42.698,
+      "eval_steps_per_second": 5.34,
+      "step": 6100
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 2.1014067995310668e-05,
+      "loss": 0.0029,
+      "step": 6110
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 2.0720984759671748e-05,
+      "loss": 0.0096,
+      "step": 6120
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 2.0427901524032827e-05,
+      "loss": 0.0067,
+      "step": 6130
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 2.0134818288393907e-05,
+      "loss": 0.0029,
+      "step": 6140
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 1.9841735052754983e-05,
+      "loss": 0.0075,
+      "step": 6150
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 1.9548651817116063e-05,
+      "loss": 0.031,
+      "step": 6160
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 1.925556858147714e-05,
+      "loss": 0.0029,
+      "step": 6170
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 1.896248534583822e-05,
+      "loss": 0.0025,
+      "step": 6180
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 1.8669402110199298e-05,
+      "loss": 0.0026,
+      "step": 6190
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 1.8376318874560374e-05,
+      "loss": 0.0061,
+      "step": 6200
+    },
+    {
+      "epoch": 3.63,
+      "eval_accuracy": 0.966804979253112,
+      "eval_loss": 0.10705357789993286,
+      "eval_runtime": 164.8249,
+      "eval_samples_per_second": 42.403,
+      "eval_steps_per_second": 5.303,
+      "step": 6200
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 1.8083235638921457e-05,
+      "loss": 0.0059,
+      "step": 6210
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 1.7790152403282534e-05,
+      "loss": 0.011,
+      "step": 6220
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 1.7497069167643613e-05,
+      "loss": 0.0026,
+      "step": 6230
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 1.720398593200469e-05,
+      "loss": 0.009,
+      "step": 6240
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 1.691090269636577e-05,
+      "loss": 0.0024,
+      "step": 6250
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 1.6617819460726845e-05,
+      "loss": 0.0024,
+      "step": 6260
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 1.6324736225087925e-05,
+      "loss": 0.0027,
+      "step": 6270
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 1.6031652989449004e-05,
+      "loss": 0.0069,
+      "step": 6280
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 1.5738569753810084e-05,
+      "loss": 0.004,
+      "step": 6290
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 1.5445486518171164e-05,
+      "loss": 0.0054,
+      "step": 6300
+    },
+    {
+      "epoch": 3.69,
+      "eval_accuracy": 0.9696666189726714,
+      "eval_loss": 0.10320978611707687,
+      "eval_runtime": 163.4309,
+      "eval_samples_per_second": 42.764,
+      "eval_steps_per_second": 5.348,
+      "step": 6300
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 1.515240328253224e-05,
+      "loss": 0.0181,
+      "step": 6310
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 1.485932004689332e-05,
+      "loss": 0.0026,
+      "step": 6320
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 1.4566236811254397e-05,
+      "loss": 0.0032,
+      "step": 6330
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 1.4273153575615475e-05,
+      "loss": 0.0136,
+      "step": 6340
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 1.3980070339976553e-05,
+      "loss": 0.0058,
+      "step": 6350
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 1.3686987104337631e-05,
+      "loss": 0.0371,
+      "step": 6360
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 1.3393903868698712e-05,
+      "loss": 0.0061,
+      "step": 6370
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 1.310082063305979e-05,
+      "loss": 0.0019,
+      "step": 6380
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 1.2807737397420868e-05,
+      "loss": 0.0103,
+      "step": 6390
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 1.2514654161781948e-05,
+      "loss": 0.0041,
+      "step": 6400
+    },
+    {
+      "epoch": 3.75,
+      "eval_accuracy": 0.971097438832451,
+      "eval_loss": 0.09614234417676926,
+      "eval_runtime": 163.0518,
+      "eval_samples_per_second": 42.864,
+      "eval_steps_per_second": 5.36,
+      "step": 6400
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 1.2221570926143026e-05,
+      "loss": 0.0044,
+      "step": 6410
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 1.1928487690504103e-05,
+      "loss": 0.0048,
+      "step": 6420
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 1.1635404454865183e-05,
+      "loss": 0.0047,
+      "step": 6430
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 1.1342321219226261e-05,
+      "loss": 0.0027,
+      "step": 6440
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 1.1049237983587339e-05,
+      "loss": 0.0023,
+      "step": 6450
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 1.0756154747948417e-05,
+      "loss": 0.0025,
+      "step": 6460
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 1.0463071512309496e-05,
+      "loss": 0.0087,
+      "step": 6470
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 1.0169988276670574e-05,
+      "loss": 0.0271,
+      "step": 6480
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 9.876905041031654e-06,
+      "loss": 0.0036,
+      "step": 6490
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 9.583821805392732e-06,
+      "loss": 0.0018,
+      "step": 6500
+    },
+    {
+      "epoch": 3.81,
+      "eval_accuracy": 0.9718128487623409,
+      "eval_loss": 0.0929703488945961,
+      "eval_runtime": 164.4896,
+      "eval_samples_per_second": 42.489,
+      "eval_steps_per_second": 5.313,
+      "step": 6500
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 9.290738569753811e-06,
+      "loss": 0.0029,
+      "step": 6510
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 8.99765533411489e-06,
+      "loss": 0.0026,
+      "step": 6520
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 8.704572098475967e-06,
+      "loss": 0.0023,
+      "step": 6530
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 8.411488862837045e-06,
+      "loss": 0.002,
+      "step": 6540
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 8.118405627198125e-06,
+      "loss": 0.0039,
+      "step": 6550
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 7.825322391559203e-06,
+      "loss": 0.0107,
+      "step": 6560
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 7.532239155920281e-06,
+      "loss": 0.0023,
+      "step": 6570
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 7.23915592028136e-06,
+      "loss": 0.0057,
+      "step": 6580
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 6.94607268464244e-06,
+      "loss": 0.0031,
+      "step": 6590
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 6.6529894490035175e-06,
+      "loss": 0.0032,
+      "step": 6600
+    },
+    {
+      "epoch": 3.87,
+      "eval_accuracy": 0.9729575046501645,
+      "eval_loss": 0.09183436632156372,
+      "eval_runtime": 165.4991,
+      "eval_samples_per_second": 42.23,
+      "eval_steps_per_second": 5.281,
+      "step": 6600
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 6.3599062133645955e-06,
+      "loss": 0.0042,
+      "step": 6610
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 6.066822977725674e-06,
+      "loss": 0.0053,
+      "step": 6620
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 5.773739742086753e-06,
+      "loss": 0.0077,
+      "step": 6630
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 5.480656506447832e-06,
+      "loss": 0.0212,
+      "step": 6640
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 5.18757327080891e-06,
+      "loss": 0.0022,
+      "step": 6650
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 4.894490035169988e-06,
+      "loss": 0.0027,
+      "step": 6660
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 4.601406799531067e-06,
+      "loss": 0.0093,
+      "step": 6670
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 4.308323563892146e-06,
+      "loss": 0.0036,
+      "step": 6680
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 4.015240328253224e-06,
+      "loss": 0.0034,
+      "step": 6690
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 3.722157092614303e-06,
+      "loss": 0.0048,
+      "step": 6700
+    },
+    {
+      "epoch": 3.93,
+      "eval_accuracy": 0.9732436686221205,
+      "eval_loss": 0.09061435610055923,
+      "eval_runtime": 162.7794,
+      "eval_samples_per_second": 42.935,
+      "eval_steps_per_second": 5.369,
+      "step": 6700
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 3.429073856975381e-06,
+      "loss": 0.0028,
+      "step": 6710
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 3.13599062133646e-06,
+      "loss": 0.0023,
+      "step": 6720
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 2.8429073856975383e-06,
+      "loss": 0.0029,
+      "step": 6730
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 2.549824150058617e-06,
+      "loss": 0.0042,
+      "step": 6740
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 2.2567409144196954e-06,
+      "loss": 0.002,
+      "step": 6750
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 1.963657678780774e-06,
+      "loss": 0.0099,
+      "step": 6760
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 1.6705744431418525e-06,
+      "loss": 0.0017,
+      "step": 6770
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 1.3774912075029308e-06,
+      "loss": 0.0031,
+      "step": 6780
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 1.0844079718640093e-06,
+      "loss": 0.0018,
+      "step": 6790
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 7.913247362250879e-07,
+      "loss": 0.002,
+      "step": 6800
+    },
+    {
+      "epoch": 3.99,
+      "eval_accuracy": 0.9731005866361425,
+      "eval_loss": 0.09024298936128616,
+      "eval_runtime": 163.9652,
+      "eval_samples_per_second": 42.625,
+      "eval_steps_per_second": 5.33,
+      "step": 6800
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 4.982415005861665e-07,
+      "loss": 0.0037,
+      "step": 6810
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 2.0515826494724504e-07,
+      "loss": 0.0266,
+      "step": 6820
+    },
+    {
+      "epoch": 4.0,
+      "step": 6824,
+      "total_flos": 8.470480851570254e+18,
+      "train_loss": 0.5286654638341308,
+      "train_runtime": 15256.5796,
+      "train_samples_per_second": 7.155,
+      "train_steps_per_second": 0.447
+    }
+  ],
+  "max_steps": 6824,
+  "num_train_epochs": 4,
+  "total_flos": 8.470480851570254e+18,
+  "trial_name": null,
+  "trial_params": null
+}