diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4640 @@
+{
+  "best_metric": 0.6829969882965088,
+  "best_model_checkpoint": "./renovation/checkpoint-2600",
+  "epoch": 4.0,
+  "eval_steps": 100,
+  "global_step": 5844,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.7498056888580322,
+      "learning_rate": 0.0001996577686516085,
+      "loss": 1.6691,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.621170997619629,
+      "learning_rate": 0.00019931553730321699,
+      "loss": 1.5728,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 2.670748233795166,
+      "learning_rate": 0.00019897330595482546,
+      "loss": 1.3843,
+      "step": 30
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.2553956508636475,
+      "learning_rate": 0.00019863107460643396,
+      "loss": 1.3548,
+      "step": 40
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.815242290496826,
+      "learning_rate": 0.00019828884325804244,
+      "loss": 1.2678,
+      "step": 50
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.8017427921295166,
+      "learning_rate": 0.00019794661190965094,
+      "loss": 1.1998,
+      "step": 60
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.7234277725219727,
+      "learning_rate": 0.00019760438056125944,
+      "loss": 1.1949,
+      "step": 70
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.4754185676574707,
+      "learning_rate": 0.0001972621492128679,
+      "loss": 1.0971,
+      "step": 80
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.840914249420166,
+      "learning_rate": 0.0001969199178644764,
+      "loss": 1.2097,
+      "step": 90
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.4401698112487793,
+      "learning_rate": 0.0001965776865160849,
+      "loss": 1.0475,
+      "step": 100
+    },
+    {
+      "epoch": 0.07,
+      "eval_accuracy": 0.582421340629275,
+      "eval_loss": 1.0331681966781616,
+      "eval_runtime": 118.3782,
+      "eval_samples_per_second": 24.701,
+      "eval_steps_per_second": 3.092,
+      "step": 100
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.5946625471115112,
+      "learning_rate": 0.00019623545516769337,
+      "loss": 1.0628,
+      "step": 110
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.1281728744506836,
+      "learning_rate": 0.00019589322381930187,
+      "loss": 1.0603,
+      "step": 120
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.139054298400879,
+      "learning_rate": 0.00019555099247091034,
+      "loss": 1.0245,
+      "step": 130
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.660050392150879,
+      "learning_rate": 0.00019520876112251882,
+      "loss": 0.97,
+      "step": 140
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.315675735473633,
+      "learning_rate": 0.00019486652977412732,
+      "loss": 0.97,
+      "step": 150
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.5216097831726074,
+      "learning_rate": 0.00019452429842573582,
+      "loss": 0.979,
+      "step": 160
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 1.6023298501968384,
+      "learning_rate": 0.0001941820670773443,
+      "loss": 1.0199,
+      "step": 170
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 3.1774656772613525,
+      "learning_rate": 0.0001938398357289528,
+      "loss": 0.9622,
+      "step": 180
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 4.399940490722656,
+      "learning_rate": 0.00019349760438056127,
+      "loss": 0.9323,
+      "step": 190
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3.0077695846557617,
+      "learning_rate": 0.00019315537303216975,
+      "loss": 0.8651,
+      "step": 200
+    },
+    {
+      "epoch": 0.14,
+      "eval_accuracy": 0.6203830369357045,
+      "eval_loss": 0.9321824908256531,
+      "eval_runtime": 99.3808,
+      "eval_samples_per_second": 29.422,
+      "eval_steps_per_second": 3.683,
+      "step": 200
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.711514472961426,
+      "learning_rate": 0.00019281314168377825,
+      "loss": 0.9065,
+      "step": 210
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 3.137427568435669,
+      "learning_rate": 0.00019247091033538672,
+      "loss": 0.8798,
+      "step": 220
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.13639235496521,
+      "learning_rate": 0.00019212867898699522,
+      "loss": 0.9009,
+      "step": 230
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 3.441429615020752,
+      "learning_rate": 0.0001917864476386037,
+      "loss": 0.895,
+      "step": 240
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.388650894165039,
+      "learning_rate": 0.00019144421629021217,
+      "loss": 0.9888,
+      "step": 250
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.304922103881836,
+      "learning_rate": 0.00019110198494182067,
+      "loss": 0.7772,
+      "step": 260
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 3.080766439437866,
+      "learning_rate": 0.00019075975359342918,
+      "loss": 0.9646,
+      "step": 270
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.9031128883361816,
+      "learning_rate": 0.00019041752224503765,
+      "loss": 0.817,
+      "step": 280
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 3.220935106277466,
+      "learning_rate": 0.00019007529089664615,
+      "loss": 0.7329,
+      "step": 290
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.5123424530029297,
+      "learning_rate": 0.00018973305954825463,
+      "loss": 1.0022,
+      "step": 300
+    },
+    {
+      "epoch": 0.21,
+      "eval_accuracy": 0.5147058823529411,
+      "eval_loss": 1.2150331735610962,
+      "eval_runtime": 99.731,
+      "eval_samples_per_second": 29.319,
+      "eval_steps_per_second": 3.67,
+      "step": 300
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 3.335115671157837,
+      "learning_rate": 0.0001893908281998631,
+      "loss": 1.0487,
+      "step": 310
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 3.9753947257995605,
+      "learning_rate": 0.0001890485968514716,
+      "loss": 0.9879,
+      "step": 320
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 3.008366346359253,
+      "learning_rate": 0.0001887063655030801,
+      "loss": 1.063,
+      "step": 330
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.9694970846176147,
+      "learning_rate": 0.00018836413415468858,
+      "loss": 0.9867,
+      "step": 340
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 3.496166944503784,
+      "learning_rate": 0.00018802190280629705,
+      "loss": 0.9286,
+      "step": 350
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 4.208122730255127,
+      "learning_rate": 0.00018767967145790556,
+      "loss": 1.2562,
+      "step": 360
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 3.086482048034668,
+      "learning_rate": 0.00018733744010951403,
+      "loss": 0.8493,
+      "step": 370
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.9667807817459106,
+      "learning_rate": 0.00018699520876112253,
+      "loss": 0.8258,
+      "step": 380
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 5.695191860198975,
+      "learning_rate": 0.00018665297741273103,
+      "loss": 0.8873,
+      "step": 390
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.56025230884552,
+      "learning_rate": 0.00018631074606433948,
+      "loss": 1.0636,
+      "step": 400
+    },
+    {
+      "epoch": 0.27,
+      "eval_accuracy": 0.625170998632011,
+      "eval_loss": 0.9522940516471863,
+      "eval_runtime": 97.3502,
+      "eval_samples_per_second": 30.036,
+      "eval_steps_per_second": 3.76,
+      "step": 400
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 2.2179884910583496,
+      "learning_rate": 0.00018596851471594798,
+      "loss": 0.9573,
+      "step": 410
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.3255908489227295,
+      "learning_rate": 0.00018562628336755649,
+      "loss": 0.9596,
+      "step": 420
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.8122656345367432,
+      "learning_rate": 0.00018528405201916496,
+      "loss": 0.9204,
+      "step": 430
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.713568687438965,
+      "learning_rate": 0.00018494182067077346,
+      "loss": 1.005,
+      "step": 440
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 2.269219160079956,
+      "learning_rate": 0.00018459958932238194,
+      "loss": 0.8583,
+      "step": 450
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 2.7371773719787598,
+      "learning_rate": 0.0001842573579739904,
+      "loss": 0.8506,
+      "step": 460
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 3.109008550643921,
+      "learning_rate": 0.0001839151266255989,
+      "loss": 0.9721,
+      "step": 470
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 3.8040051460266113,
+      "learning_rate": 0.00018357289527720741,
+      "loss": 0.8799,
+      "step": 480
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 2.8032755851745605,
+      "learning_rate": 0.0001832306639288159,
+      "loss": 0.8998,
+      "step": 490
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 5.312169551849365,
+      "learning_rate": 0.0001828884325804244,
+      "loss": 0.8311,
+      "step": 500
+    },
+    {
+      "epoch": 0.34,
+      "eval_accuracy": 0.655608755129959,
+      "eval_loss": 0.8439633846282959,
+      "eval_runtime": 97.8414,
+      "eval_samples_per_second": 29.885,
+      "eval_steps_per_second": 3.741,
+      "step": 500
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 3.493340253829956,
+      "learning_rate": 0.00018254620123203286,
+      "loss": 0.8658,
+      "step": 510
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.711038827896118,
+      "learning_rate": 0.00018220396988364134,
+      "loss": 0.8064,
+      "step": 520
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.2599730491638184,
+      "learning_rate": 0.00018186173853524984,
+      "loss": 0.8902,
+      "step": 530
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 1.6086162328720093,
+      "learning_rate": 0.00018151950718685832,
+      "loss": 0.7395,
+      "step": 540
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.698824405670166,
+      "learning_rate": 0.00018117727583846682,
+      "loss": 0.8971,
+      "step": 550
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 3.232140064239502,
+      "learning_rate": 0.0001808350444900753,
+      "loss": 0.8815,
+      "step": 560
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.3350541591644287,
+      "learning_rate": 0.0001804928131416838,
+      "loss": 0.7075,
+      "step": 570
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 2.1570167541503906,
+      "learning_rate": 0.00018015058179329227,
+      "loss": 0.873,
+      "step": 580
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 3.0529322624206543,
+      "learning_rate": 0.00017980835044490077,
+      "loss": 0.8208,
+      "step": 590
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 2.9495370388031006,
+      "learning_rate": 0.00017946611909650924,
+      "loss": 0.88,
+      "step": 600
+    },
+    {
+      "epoch": 0.41,
+      "eval_accuracy": 0.649452804377565,
+      "eval_loss": 0.8707343935966492,
+      "eval_runtime": 97.0815,
+      "eval_samples_per_second": 30.119,
+      "eval_steps_per_second": 3.77,
+      "step": 600
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 4.997549533843994,
+      "learning_rate": 0.00017912388774811775,
+      "loss": 0.915,
+      "step": 610
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 3.1065518856048584,
+      "learning_rate": 0.00017878165639972622,
+      "loss": 0.8154,
+      "step": 620
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 2.30731201171875,
+      "learning_rate": 0.0001784394250513347,
+      "loss": 0.8252,
+      "step": 630
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.0879149436950684,
+      "learning_rate": 0.0001780971937029432,
+      "loss": 0.8409,
+      "step": 640
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 3.711212158203125,
+      "learning_rate": 0.0001777549623545517,
+      "loss": 0.7977,
+      "step": 650
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 2.27301287651062,
+      "learning_rate": 0.00017741273100616017,
+      "loss": 0.8718,
+      "step": 660
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 3.153876781463623,
+      "learning_rate": 0.00017707049965776865,
+      "loss": 0.8712,
+      "step": 670
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.7592906951904297,
+      "learning_rate": 0.00017672826830937715,
+      "loss": 0.9281,
+      "step": 680
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 6.171240329742432,
+      "learning_rate": 0.00017638603696098562,
+      "loss": 1.0087,
+      "step": 690
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 3.861326217651367,
+      "learning_rate": 0.00017604380561259413,
+      "loss": 0.8881,
+      "step": 700
+    },
+    {
+      "epoch": 0.48,
+      "eval_accuracy": 0.6333789329685362,
+      "eval_loss": 0.8902959227561951,
+      "eval_runtime": 98.1212,
+      "eval_samples_per_second": 29.8,
+      "eval_steps_per_second": 3.73,
+      "step": 700
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 5.067281246185303,
+      "learning_rate": 0.00017570157426420263,
+      "loss": 0.8387,
+      "step": 710
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 3.4782888889312744,
+      "learning_rate": 0.00017535934291581108,
+      "loss": 0.8257,
+      "step": 720
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 3.553341865539551,
+      "learning_rate": 0.00017501711156741958,
+      "loss": 0.7631,
+      "step": 730
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 2.3364970684051514,
+      "learning_rate": 0.00017467488021902808,
+      "loss": 0.8531,
+      "step": 740
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 2.0276718139648438,
+      "learning_rate": 0.00017433264887063655,
+      "loss": 0.7493,
+      "step": 750
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 5.019596576690674,
+      "learning_rate": 0.00017399041752224506,
+      "loss": 0.7348,
+      "step": 760
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 3.7119674682617188,
+      "learning_rate": 0.00017364818617385356,
+      "loss": 0.8455,
+      "step": 770
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 2.028977870941162,
+      "learning_rate": 0.000173305954825462,
+      "loss": 0.8198,
+      "step": 780
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 1.9546798467636108,
+      "learning_rate": 0.0001729637234770705,
+      "loss": 0.754,
+      "step": 790
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.9083542823791504,
+      "learning_rate": 0.000172621492128679,
+      "loss": 0.7522,
+      "step": 800
+    },
+    {
+      "epoch": 0.55,
+      "eval_accuracy": 0.6576607387140903,
+      "eval_loss": 0.8479247093200684,
+      "eval_runtime": 98.2024,
+      "eval_samples_per_second": 29.775,
+      "eval_steps_per_second": 3.727,
+      "step": 800
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 2.516714572906494,
+      "learning_rate": 0.00017227926078028748,
+      "loss": 0.7789,
+      "step": 810
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 2.8401355743408203,
+      "learning_rate": 0.00017193702943189598,
+      "loss": 0.7046,
+      "step": 820
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 2.477896213531494,
+      "learning_rate": 0.00017159479808350446,
+      "loss": 0.879,
+      "step": 830
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 3.136587381362915,
+      "learning_rate": 0.00017125256673511293,
+      "loss": 0.8485,
+      "step": 840
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.215061068534851,
+      "learning_rate": 0.00017091033538672144,
+      "loss": 0.8007,
+      "step": 850
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 2.920921564102173,
+      "learning_rate": 0.00017056810403832994,
+      "loss": 0.8547,
+      "step": 860
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 2.390735626220703,
+      "learning_rate": 0.0001702258726899384,
+      "loss": 0.7409,
+      "step": 870
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 2.2901852130889893,
+      "learning_rate": 0.00016988364134154689,
+      "loss": 0.9101,
+      "step": 880
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 2.337519884109497,
+      "learning_rate": 0.0001695414099931554,
+      "loss": 0.7682,
+      "step": 890
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 2.016847848892212,
+      "learning_rate": 0.00016919917864476386,
+      "loss": 0.798,
+      "step": 900
+    },
+    {
+      "epoch": 0.62,
+      "eval_accuracy": 0.6843365253077975,
+      "eval_loss": 0.7738928198814392,
+      "eval_runtime": 98.3162,
+      "eval_samples_per_second": 29.741,
+      "eval_steps_per_second": 3.723,
+      "step": 900
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 2.343900203704834,
+      "learning_rate": 0.00016885694729637236,
+      "loss": 0.7555,
+      "step": 910
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.9535636901855469,
+      "learning_rate": 0.00016851471594798084,
+      "loss": 0.8414,
+      "step": 920
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 2.9519262313842773,
+      "learning_rate": 0.00016820670773442847,
+      "loss": 0.6694,
+      "step": 930
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 2.4800071716308594,
+      "learning_rate": 0.00016786447638603697,
+      "loss": 0.6643,
+      "step": 940
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 2.3085594177246094,
+      "learning_rate": 0.00016752224503764547,
+      "loss": 0.7348,
+      "step": 950
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 4.01909065246582,
+      "learning_rate": 0.00016718001368925392,
+      "loss": 0.8339,
+      "step": 960
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.9890522956848145,
+      "learning_rate": 0.00016683778234086242,
+      "loss": 0.7035,
+      "step": 970
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 2.7115402221679688,
+      "learning_rate": 0.00016649555099247093,
+      "loss": 0.8243,
+      "step": 980
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 3.829646348953247,
+      "learning_rate": 0.0001661533196440794,
+      "loss": 0.9374,
+      "step": 990
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 2.90616774559021,
+      "learning_rate": 0.0001658110882956879,
+      "loss": 0.7317,
+      "step": 1000
+    },
+    {
+      "epoch": 0.68,
+      "eval_accuracy": 0.6795485636114911,
+      "eval_loss": 0.7855507731437683,
+      "eval_runtime": 98.3649,
+      "eval_samples_per_second": 29.726,
+      "eval_steps_per_second": 3.721,
+      "step": 1000
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 2.388803482055664,
+      "learning_rate": 0.0001654688569472964,
+      "loss": 0.8345,
+      "step": 1010
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.2373769283294678,
+      "learning_rate": 0.00016512662559890485,
+      "loss": 0.712,
+      "step": 1020
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.6666200160980225,
+      "learning_rate": 0.00016478439425051335,
+      "loss": 0.8304,
+      "step": 1030
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 2.198072910308838,
+      "learning_rate": 0.00016444216290212185,
+      "loss": 0.6709,
+      "step": 1040
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 2.4663429260253906,
+      "learning_rate": 0.00016409993155373033,
+      "loss": 0.8756,
+      "step": 1050
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 3.6292316913604736,
+      "learning_rate": 0.00016375770020533883,
+      "loss": 0.8577,
+      "step": 1060
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 2.9756243228912354,
+      "learning_rate": 0.0001634154688569473,
+      "loss": 0.6306,
+      "step": 1070
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 3.7847914695739746,
+      "learning_rate": 0.00016307323750855578,
+      "loss": 0.7334,
+      "step": 1080
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 2.1593246459960938,
+      "learning_rate": 0.00016273100616016428,
+      "loss": 0.7105,
+      "step": 1090
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 3.798877239227295,
+      "learning_rate": 0.00016238877481177278,
+      "loss": 0.8372,
+      "step": 1100
+    },
+    {
+      "epoch": 0.75,
+      "eval_accuracy": 0.6354309165526676,
+      "eval_loss": 0.888414204120636,
+      "eval_runtime": 98.4587,
+      "eval_samples_per_second": 29.698,
+      "eval_steps_per_second": 3.717,
+      "step": 1100
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 3.8040831089019775,
+      "learning_rate": 0.00016204654346338126,
+      "loss": 0.8219,
+      "step": 1110
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 1.5006623268127441,
+      "learning_rate": 0.0001617385352498289,
+      "loss": 0.683,
+      "step": 1120
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 2.44973087310791,
+      "learning_rate": 0.0001613963039014374,
+      "loss": 0.7756,
+      "step": 1130
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 3.5681519508361816,
+      "learning_rate": 0.00016105407255304587,
+      "loss": 0.9079,
+      "step": 1140
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 2.513470411300659,
+      "learning_rate": 0.00016071184120465434,
+      "loss": 0.7569,
+      "step": 1150
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 5.322854518890381,
+      "learning_rate": 0.00016036960985626284,
+      "loss": 0.8551,
+      "step": 1160
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 2.9448812007904053,
+      "learning_rate": 0.00016002737850787132,
+      "loss": 0.8706,
+      "step": 1170
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 3.101590394973755,
+      "learning_rate": 0.00015968514715947982,
+      "loss": 0.8553,
+      "step": 1180
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 5.033396244049072,
+      "learning_rate": 0.00015934291581108832,
+      "loss": 0.7715,
+      "step": 1190
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.8060014247894287,
+      "learning_rate": 0.00015900068446269677,
+      "loss": 0.6629,
+      "step": 1200
+    },
+    {
+      "epoch": 0.82,
+      "eval_accuracy": 0.6870725034199726,
+      "eval_loss": 0.7572759389877319,
+      "eval_runtime": 97.9392,
+      "eval_samples_per_second": 29.855,
+      "eval_steps_per_second": 3.737,
+      "step": 1200
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 3.305678129196167,
+      "learning_rate": 0.00015865845311430527,
+      "loss": 0.6609,
+      "step": 1210
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.3885236978530884,
+      "learning_rate": 0.00015831622176591377,
+      "loss": 0.8702,
+      "step": 1220
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.796901822090149,
+      "learning_rate": 0.00015797399041752225,
+      "loss": 0.6958,
+      "step": 1230
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 2.182668685913086,
+      "learning_rate": 0.00015763175906913075,
+      "loss": 0.7975,
+      "step": 1240
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 2.096924066543579,
+      "learning_rate": 0.00015728952772073925,
+      "loss": 0.5707,
+      "step": 1250
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 2.922907590866089,
+      "learning_rate": 0.0001569472963723477,
+      "loss": 0.6806,
+      "step": 1260
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.7715729475021362,
+      "learning_rate": 0.0001566050650239562,
+      "loss": 0.7296,
+      "step": 1270
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 6.666139602661133,
+      "learning_rate": 0.0001562628336755647,
+      "loss": 0.6637,
+      "step": 1280
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 2.5472826957702637,
+      "learning_rate": 0.00015592060232717318,
+      "loss": 0.7683,
+      "step": 1290
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 3.5583953857421875,
+      "learning_rate": 0.00015557837097878168,
+      "loss": 0.7767,
+      "step": 1300
+    },
+    {
+      "epoch": 0.89,
+      "eval_accuracy": 0.686046511627907,
+      "eval_loss": 0.7543095946311951,
+      "eval_runtime": 97.3791,
+      "eval_samples_per_second": 30.027,
+      "eval_steps_per_second": 3.759,
+      "step": 1300
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 3.853813886642456,
+      "learning_rate": 0.00015523613963039015,
+      "loss": 0.8368,
+      "step": 1310
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 2.1980948448181152,
+      "learning_rate": 0.00015489390828199863,
+      "loss": 0.8148,
+      "step": 1320
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 2.6982269287109375,
+      "learning_rate": 0.00015455167693360713,
+      "loss": 0.7574,
+      "step": 1330
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.415897011756897,
+      "learning_rate": 0.00015420944558521563,
+      "loss": 0.7289,
+      "step": 1340
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.5153343677520752,
+      "learning_rate": 0.0001538672142368241,
+      "loss": 0.7146,
+      "step": 1350
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 4.069477558135986,
+      "learning_rate": 0.00015352498288843258,
+      "loss": 0.7833,
+      "step": 1360
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 2.2534329891204834,
+      "learning_rate": 0.00015318275154004108,
+      "loss": 0.7169,
+      "step": 1370
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 2.8250768184661865,
+      "learning_rate": 0.00015284052019164956,
+      "loss": 0.7735,
+      "step": 1380
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 4.815442085266113,
+      "learning_rate": 0.00015249828884325806,
+      "loss": 0.8695,
+      "step": 1390
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.844257116317749,
+      "learning_rate": 0.00015215605749486653,
+      "loss": 0.9246,
+      "step": 1400
+    },
+    {
+      "epoch": 0.96,
+      "eval_accuracy": 0.6634746922024624,
+      "eval_loss": 0.7895568013191223,
+      "eval_runtime": 97.6845,
+      "eval_samples_per_second": 29.933,
+      "eval_steps_per_second": 3.747,
+      "step": 1400
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 3.4265382289886475,
+      "learning_rate": 0.000151813826146475,
+      "loss": 0.8821,
+      "step": 1410
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.8195841312408447,
+      "learning_rate": 0.0001514715947980835,
+      "loss": 0.7777,
+      "step": 1420
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 2.0240731239318848,
+      "learning_rate": 0.000151129363449692,
+      "loss": 0.8862,
+      "step": 1430
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 2.748037338256836,
+      "learning_rate": 0.00015078713210130048,
+      "loss": 0.7216,
+      "step": 1440
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 2.5891735553741455,
+      "learning_rate": 0.00015044490075290899,
+      "loss": 0.8374,
+      "step": 1450
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 1.7044845819473267,
+      "learning_rate": 0.00015010266940451746,
+      "loss": 0.739,
+      "step": 1460
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 1.914048194885254,
+      "learning_rate": 0.00014976043805612594,
+      "loss": 0.6031,
+      "step": 1470
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 1.34175443649292,
+      "learning_rate": 0.00014941820670773444,
+      "loss": 0.5003,
+      "step": 1480
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 2.1494643688201904,
+      "learning_rate": 0.0001490759753593429,
+      "loss": 0.5636,
+      "step": 1490
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 1.9443460702896118,
+      "learning_rate": 0.0001487337440109514,
+      "loss": 0.5026,
+      "step": 1500
+    },
+    {
+      "epoch": 1.03,
+      "eval_accuracy": 0.6812585499316005,
+      "eval_loss": 0.787219226360321,
+      "eval_runtime": 98.8516,
+      "eval_samples_per_second": 29.58,
+      "eval_steps_per_second": 3.703,
+      "step": 1500
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 5.107473373413086,
+      "learning_rate": 0.00014839151266255992,
+      "loss": 0.6356,
+      "step": 1510
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 1.6247891187667847,
+      "learning_rate": 0.00014804928131416836,
+      "loss": 0.54,
+      "step": 1520
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 3.002816677093506,
+      "learning_rate": 0.00014770704996577686,
+      "loss": 0.5619,
+      "step": 1530
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 3.7361197471618652,
+      "learning_rate": 0.00014736481861738537,
+      "loss": 0.6403,
+      "step": 1540
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 2.038827896118164,
+      "learning_rate": 0.00014702258726899384,
+      "loss": 0.5458,
+      "step": 1550
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 2.4605889320373535,
+      "learning_rate": 0.00014668035592060234,
+      "loss": 0.4914,
+      "step": 1560
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 3.549699544906616,
+      "learning_rate": 0.00014633812457221082,
+      "loss": 0.5927,
+      "step": 1570
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 2.2531206607818604,
+      "learning_rate": 0.0001459958932238193,
+      "loss": 0.5065,
+      "step": 1580
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 2.869708299636841,
+      "learning_rate": 0.0001456536618754278,
+      "loss": 0.6245,
+      "step": 1590
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 1.7271431684494019,
+      "learning_rate": 0.0001453114305270363,
+      "loss": 0.7599,
+      "step": 1600
+    },
+    {
+      "epoch": 1.1,
+      "eval_accuracy": 0.6757865937072504,
+      "eval_loss": 0.7860671281814575,
+      "eval_runtime": 98.0642,
+      "eval_samples_per_second": 29.817,
+      "eval_steps_per_second": 3.732,
+      "step": 1600
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 3.388765573501587,
+      "learning_rate": 0.00014496919917864477,
+      "loss": 0.6075,
+      "step": 1610
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 1.6970986127853394,
+      "learning_rate": 0.00014462696783025327,
+      "loss": 0.506,
+      "step": 1620
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 3.88580584526062,
+      "learning_rate": 0.00014428473648186175,
+      "loss": 0.6463,
+      "step": 1630
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 4.629645347595215,
+      "learning_rate": 0.00014394250513347022,
+      "loss": 0.5743,
+      "step": 1640
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 3.3085250854492188,
+      "learning_rate": 0.00014360027378507872,
+      "loss": 0.5702,
+      "step": 1650
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 1.6527128219604492,
+      "learning_rate": 0.00014325804243668722,
+      "loss": 0.3941,
+      "step": 1660
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 4.697700500488281,
+      "learning_rate": 0.0001429158110882957,
+      "loss": 0.6121,
+      "step": 1670
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 2.6221749782562256,
+      "learning_rate": 0.00014257357973990417,
+      "loss": 0.5243,
+      "step": 1680
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 2.580265522003174,
+      "learning_rate": 0.00014223134839151267,
+      "loss": 0.5563,
+      "step": 1690
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 2.5967156887054443,
+      "learning_rate": 0.00014188911704312115,
+      "loss": 0.5764,
+      "step": 1700
+    },
+    {
+      "epoch": 1.16,
+      "eval_accuracy": 0.6802325581395349,
+      "eval_loss": 0.808827817440033,
+      "eval_runtime": 97.9955,
+      "eval_samples_per_second": 29.838,
+      "eval_steps_per_second": 3.735,
+      "step": 1700
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 2.171257257461548,
+      "learning_rate": 0.00014154688569472965,
+      "loss": 0.6382,
+      "step": 1710
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 1.1206802129745483,
+      "learning_rate": 0.00014120465434633813,
+      "loss": 0.5243,
+      "step": 1720
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 1.483469009399414,
+      "learning_rate": 0.0001408624229979466,
+      "loss": 0.7305,
+      "step": 1730
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 2.7231271266937256,
+      "learning_rate": 0.0001405201916495551,
+      "loss": 0.4636,
+      "step": 1740
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 3.154658794403076,
+      "learning_rate": 0.0001401779603011636,
+      "loss": 0.5783,
+      "step": 1750
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 3.6419148445129395,
+      "learning_rate": 0.00013983572895277208,
+      "loss": 0.5313,
+      "step": 1760
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 2.149718999862671,
+      "learning_rate": 0.00013949349760438058,
+      "loss": 0.5444,
+      "step": 1770
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 3.819136142730713,
+      "learning_rate": 0.00013915126625598905,
+      "loss": 0.7068,
+      "step": 1780
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 1.5383284091949463,
+      "learning_rate": 0.00013880903490759753,
+      "loss": 0.5585,
+      "step": 1790
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 2.2872893810272217,
+      "learning_rate": 0.00013846680355920603,
+      "loss": 0.4329,
+      "step": 1800
+    },
+    {
+      "epoch": 1.23,
+      "eval_accuracy": 0.7058823529411765,
+      "eval_loss": 0.7281420826911926,
+      "eval_runtime": 98.0691,
+      "eval_samples_per_second": 29.816,
+      "eval_steps_per_second": 3.732,
+      "step": 1800
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 4.656331539154053,
+      "learning_rate": 0.0001381245722108145,
+      "loss": 0.4951,
+      "step": 1810
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 3.128675937652588,
+      "learning_rate": 0.000137782340862423,
+      "loss": 0.6682,
+      "step": 1820
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 2.384826898574829,
+      "learning_rate": 0.0001374401095140315,
+      "loss": 0.5825,
+      "step": 1830
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 2.245954990386963,
+      "learning_rate": 0.00013709787816563998,
+      "loss": 0.48,
+      "step": 1840
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 2.726475715637207,
+      "learning_rate": 0.00013675564681724846,
+      "loss": 0.4232,
+      "step": 1850
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 2.0738930702209473,
+      "learning_rate": 0.00013641341546885696,
+      "loss": 0.5,
+      "step": 1860
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 1.9418065547943115,
+      "learning_rate": 0.00013607118412046543,
+      "loss": 0.4846,
+      "step": 1870
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 3.351672410964966,
+      "learning_rate": 0.00013572895277207394,
+      "loss": 0.4581,
+      "step": 1880
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 3.308870792388916,
+      "learning_rate": 0.0001353867214236824,
+      "loss": 0.5721,
+      "step": 1890
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 3.86739182472229,
+      "learning_rate": 0.00013504449007529089,
+      "loss": 0.6271,
+      "step": 1900
+    },
+    {
+      "epoch": 1.3,
+      "eval_accuracy": 0.7116963064295485,
+      "eval_loss": 0.7290834784507751,
+      "eval_runtime": 98.3785,
+      "eval_samples_per_second": 29.722,
+      "eval_steps_per_second": 3.72,
+      "step": 1900
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 2.4816877841949463,
+      "learning_rate": 0.0001347022587268994,
+      "loss": 0.4315,
+      "step": 1910
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 2.6154468059539795,
+      "learning_rate": 0.0001343600273785079,
+      "loss": 0.5494,
+      "step": 1920
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 4.586129188537598,
+      "learning_rate": 0.00013401779603011636,
+      "loss": 0.5922,
+      "step": 1930
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 3.945017099380493,
+      "learning_rate": 0.00013367556468172487,
+      "loss": 0.7348,
+      "step": 1940
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 9.25570297241211,
+      "learning_rate": 0.00013333333333333334,
+      "loss": 0.4907,
+      "step": 1950
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 2.919928789138794,
+      "learning_rate": 0.00013299110198494181,
+      "loss": 0.5759,
+      "step": 1960
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 3.0628156661987305,
+      "learning_rate": 0.00013264887063655032,
+      "loss": 0.5712,
+      "step": 1970
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 3.5648550987243652,
+      "learning_rate": 0.00013230663928815882,
+      "loss": 0.6428,
+      "step": 1980
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 3.348015546798706,
+      "learning_rate": 0.0001319644079397673,
+      "loss": 0.5054,
+      "step": 1990
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 3.1563289165496826,
+      "learning_rate": 0.00013162217659137577,
+      "loss": 0.5498,
+      "step": 2000
+    },
+    {
+      "epoch": 1.37,
+      "eval_accuracy": 0.7058823529411765,
+      "eval_loss": 0.7744925618171692,
+      "eval_runtime": 98.5948,
+      "eval_samples_per_second": 29.657,
+      "eval_steps_per_second": 3.712,
+      "step": 2000
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 2.594055414199829,
+      "learning_rate": 0.00013127994524298427,
+      "loss": 0.5086,
+      "step": 2010
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 5.243160724639893,
+      "learning_rate": 0.00013093771389459274,
+      "loss": 0.6349,
+      "step": 2020
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 3.0656027793884277,
+      "learning_rate": 0.00013059548254620124,
+      "loss": 0.5671,
+      "step": 2030
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 2.7779109477996826,
+      "learning_rate": 0.00013025325119780975,
+      "loss": 0.5813,
+      "step": 2040
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 3.042433023452759,
+      "learning_rate": 0.0001299110198494182,
+      "loss": 0.6665,
+      "step": 2050
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 2.6497788429260254,
+      "learning_rate": 0.0001295687885010267,
+      "loss": 0.517,
+      "step": 2060
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 2.5200939178466797,
+      "learning_rate": 0.0001292265571526352,
+      "loss": 0.6524,
+      "step": 2070
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 3.1486799716949463,
+      "learning_rate": 0.00012888432580424367,
+      "loss": 0.4255,
+      "step": 2080
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 3.680635929107666,
+      "learning_rate": 0.00012854209445585217,
+      "loss": 0.499,
+      "step": 2090
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 4.004785060882568,
+      "learning_rate": 0.00012819986310746065,
+      "loss": 0.5247,
+      "step": 2100
+    },
+    {
+      "epoch": 1.44,
+      "eval_accuracy": 0.689124487004104,
+      "eval_loss": 0.8001527190208435,
+      "eval_runtime": 98.0128,
+      "eval_samples_per_second": 29.833,
+      "eval_steps_per_second": 3.734,
+      "step": 2100
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 3.6504130363464355,
+      "learning_rate": 0.00012785763175906912,
+      "loss": 0.575,
+      "step": 2110
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 2.601299285888672,
+      "learning_rate": 0.00012751540041067762,
+      "loss": 0.4903,
+      "step": 2120
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 3.950484275817871,
+      "learning_rate": 0.0001271731690622861,
+      "loss": 0.5872,
+      "step": 2130
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 4.166860103607178,
+      "learning_rate": 0.0001268309377138946,
+      "loss": 0.5385,
+      "step": 2140
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 1.910089373588562,
+      "learning_rate": 0.0001264887063655031,
+      "loss": 0.6209,
+      "step": 2150
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 14.525965690612793,
+      "learning_rate": 0.00012614647501711158,
+      "loss": 0.6604,
+      "step": 2160
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 6.641749858856201,
+      "learning_rate": 0.00012580424366872005,
+      "loss": 0.5131,
+      "step": 2170
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 3.671076536178589,
+      "learning_rate": 0.00012546201232032855,
+      "loss": 0.5391,
+      "step": 2180
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 3.4304141998291016,
+      "learning_rate": 0.00012511978097193703,
+      "loss": 0.4564,
+      "step": 2190
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 2.004720687866211,
+      "learning_rate": 0.00012477754962354553,
+      "loss": 0.4891,
+      "step": 2200
+    },
+    {
+      "epoch": 1.51,
+      "eval_accuracy": 0.7099863201094391,
+      "eval_loss": 0.7013882994651794,
+      "eval_runtime": 98.2469,
+      "eval_samples_per_second": 29.762,
+      "eval_steps_per_second": 3.725,
+      "step": 2200
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 4.185796737670898,
+      "learning_rate": 0.000124435318275154,
+      "loss": 0.5397,
+      "step": 2210
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 4.123441696166992,
+      "learning_rate": 0.00012409308692676248,
+      "loss": 0.5452,
+      "step": 2220
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 4.169721603393555,
+      "learning_rate": 0.00012375085557837098,
+      "loss": 0.658,
+      "step": 2230
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 1.8797214031219482,
+      "learning_rate": 0.00012340862422997948,
+      "loss": 0.5156,
+      "step": 2240
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 2.8339829444885254,
+      "learning_rate": 0.00012306639288158796,
+      "loss": 0.5107,
+      "step": 2250
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 3.297180414199829,
+      "learning_rate": 0.00012272416153319643,
+      "loss": 0.7244,
+      "step": 2260
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 2.3934600353240967,
+      "learning_rate": 0.00012238193018480493,
+      "loss": 0.6198,
+      "step": 2270
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 3.875546932220459,
+      "learning_rate": 0.00012203969883641342,
+      "loss": 0.5995,
+      "step": 2280
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 4.619594097137451,
+      "learning_rate": 0.00012169746748802191,
+      "loss": 0.6293,
+      "step": 2290
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 2.215853452682495,
+      "learning_rate": 0.0001213552361396304,
+      "loss": 0.5211,
+      "step": 2300
+    },
+    {
+      "epoch": 1.57,
+      "eval_accuracy": 0.6863885088919288,
+      "eval_loss": 0.7724740505218506,
+      "eval_runtime": 99.41,
+      "eval_samples_per_second": 29.414,
+      "eval_steps_per_second": 3.682,
+      "step": 2300
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 1.7644644975662231,
+      "learning_rate": 0.0001210130047912389,
+      "loss": 0.5552,
+      "step": 2310
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 2.6976282596588135,
+      "learning_rate": 0.00012067077344284736,
+      "loss": 0.4784,
+      "step": 2320
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 2.179863691329956,
+      "learning_rate": 0.00012032854209445585,
+      "loss": 0.5625,
+      "step": 2330
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 1.2474085092544556,
+      "learning_rate": 0.00011998631074606435,
+      "loss": 0.5838,
+      "step": 2340
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 3.6004438400268555,
+      "learning_rate": 0.00011964407939767284,
+      "loss": 0.5386,
+      "step": 2350
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 2.642122268676758,
+      "learning_rate": 0.00011930184804928133,
+      "loss": 0.5415,
+      "step": 2360
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 6.505972862243652,
+      "learning_rate": 0.0001189596167008898,
+      "loss": 0.5824,
+      "step": 2370
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 4.343991279602051,
+      "learning_rate": 0.00011861738535249829,
+      "loss": 0.484,
+      "step": 2380
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 3.2172842025756836,
+      "learning_rate": 0.00011827515400410678,
+      "loss": 0.5003,
+      "step": 2390
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 3.127413272857666,
+      "learning_rate": 0.00011793292265571527,
+      "loss": 0.659,
+      "step": 2400
+    },
+    {
+      "epoch": 1.64,
+      "eval_accuracy": 0.7086183310533516,
+      "eval_loss": 0.7477269172668457,
+      "eval_runtime": 98.4799,
+      "eval_samples_per_second": 29.691,
+      "eval_steps_per_second": 3.716,
+      "step": 2400
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 2.628113269805908,
+      "learning_rate": 0.00011759069130732377,
+      "loss": 0.5786,
+      "step": 2410
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 1.421924114227295,
+      "learning_rate": 0.00011724845995893223,
+      "loss": 0.5878,
+      "step": 2420
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 3.8365235328674316,
+      "learning_rate": 0.00011690622861054073,
+      "loss": 0.4624,
+      "step": 2430
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 3.5438849925994873,
+      "learning_rate": 0.00011656399726214922,
+      "loss": 0.5347,
+      "step": 2440
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 2.891608238220215,
+      "learning_rate": 0.00011622176591375771,
+      "loss": 0.4244,
+      "step": 2450
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 3.9522762298583984,
+      "learning_rate": 0.0001158795345653662,
+      "loss": 0.5261,
+      "step": 2460
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 3.749552011489868,
+      "learning_rate": 0.0001155373032169747,
+      "loss": 0.5521,
+      "step": 2470
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 3.110929250717163,
+      "learning_rate": 0.00011519507186858316,
+      "loss": 0.5003,
+      "step": 2480
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 2.78469181060791,
+      "learning_rate": 0.00011485284052019165,
+      "loss": 0.5088,
+      "step": 2490
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 4.185746669769287,
+      "learning_rate": 0.00011451060917180015,
+      "loss": 0.4878,
+      "step": 2500
+    },
+    {
+      "epoch": 1.71,
+      "eval_accuracy": 0.7051983584131327,
+      "eval_loss": 0.7128526568412781,
+      "eval_runtime": 98.1754,
+      "eval_samples_per_second": 29.783,
+      "eval_steps_per_second": 3.728,
+      "step": 2500
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 2.7198736667633057,
+      "learning_rate": 0.00011416837782340864,
+      "loss": 0.4998,
+      "step": 2510
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 1.5983054637908936,
+      "learning_rate": 0.00011382614647501712,
+      "loss": 0.509,
+      "step": 2520
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 3.75059175491333,
+      "learning_rate": 0.0001134839151266256,
+      "loss": 0.6075,
+      "step": 2530
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 3.323063611984253,
+      "learning_rate": 0.00011314168377823409,
+      "loss": 0.5783,
+      "step": 2540
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 5.036363124847412,
+      "learning_rate": 0.00011279945242984257,
+      "loss": 0.647,
+      "step": 2550
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 2.75390887260437,
+      "learning_rate": 0.00011245722108145106,
+      "loss": 0.5227,
+      "step": 2560
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 2.0095183849334717,
+      "learning_rate": 0.00011211498973305956,
+      "loss": 0.4561,
+      "step": 2570
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 3.252497911453247,
+      "learning_rate": 0.00011177275838466803,
+      "loss": 0.5389,
+      "step": 2580
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 2.183528184890747,
+      "learning_rate": 0.00011143052703627653,
+      "loss": 0.4228,
+      "step": 2590
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 2.2591006755828857,
+      "learning_rate": 0.00011108829568788502,
+      "loss": 0.4941,
+      "step": 2600
+    },
+    {
+      "epoch": 1.78,
+      "eval_accuracy": 0.7219562243502052,
+      "eval_loss": 0.6829969882965088,
+      "eval_runtime": 99.2532,
+      "eval_samples_per_second": 29.46,
+      "eval_steps_per_second": 3.688,
+      "step": 2600
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 2.982792377471924,
+      "learning_rate": 0.0001107460643394935,
+      "loss": 0.4714,
+      "step": 2610
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 2.341184616088867,
+      "learning_rate": 0.00011040383299110199,
+      "loss": 0.4921,
+      "step": 2620
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 4.106105327606201,
+      "learning_rate": 0.0001100616016427105,
+      "loss": 0.5576,
+      "step": 2630
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 2.408310651779175,
+      "learning_rate": 0.00010971937029431895,
+      "loss": 0.5945,
+      "step": 2640
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 2.786745071411133,
+      "learning_rate": 0.00010937713894592744,
+      "loss": 0.411,
+      "step": 2650
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 2.012185573577881,
+      "learning_rate": 0.00010903490759753594,
+      "loss": 0.5151,
+      "step": 2660
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 4.303542137145996,
+      "learning_rate": 0.00010869267624914443,
+      "loss": 0.5431,
+      "step": 2670
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 3.5029568672180176,
+      "learning_rate": 0.00010835044490075292,
+      "loss": 0.62,
+      "step": 2680
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 2.5191617012023926,
+      "learning_rate": 0.0001080082135523614,
+      "loss": 0.5678,
+      "step": 2690
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 2.9402916431427,
+      "learning_rate": 0.00010766598220396988,
+      "loss": 0.4648,
+      "step": 2700
+    },
+    {
+      "epoch": 1.85,
+      "eval_accuracy": 0.7028043775649795,
+      "eval_loss": 0.7181953191757202,
+      "eval_runtime": 97.5335,
+      "eval_samples_per_second": 29.979,
+      "eval_steps_per_second": 3.753,
+      "step": 2700
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 2.9497809410095215,
+      "learning_rate": 0.00010732375085557837,
+      "loss": 0.4891,
+      "step": 2710
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 4.476220607757568,
+      "learning_rate": 0.00010698151950718687,
+      "loss": 0.4989,
+      "step": 2720
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 3.7788009643554688,
+      "learning_rate": 0.00010663928815879536,
+      "loss": 0.649,
+      "step": 2730
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 3.8084425926208496,
+      "learning_rate": 0.00010629705681040382,
+      "loss": 0.6365,
+      "step": 2740
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 2.09145188331604,
+      "learning_rate": 0.00010595482546201232,
+      "loss": 0.621,
+      "step": 2750
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 1.95127272605896,
+      "learning_rate": 0.00010561259411362081,
+      "loss": 0.5456,
+      "step": 2760
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 3.267646074295044,
+      "learning_rate": 0.0001052703627652293,
+      "loss": 0.529,
+      "step": 2770
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 3.5166428089141846,
+      "learning_rate": 0.00010492813141683779,
+      "loss": 0.442,
+      "step": 2780
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 2.4923620223999023,
+      "learning_rate": 0.00010458590006844629,
+      "loss": 0.4446,
+      "step": 2790
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 4.28298282623291,
+      "learning_rate": 0.00010424366872005475,
+      "loss": 0.5501,
+      "step": 2800
+    },
+    {
+      "epoch": 1.92,
+      "eval_accuracy": 0.7144322845417237,
+      "eval_loss": 0.7190529108047485,
+      "eval_runtime": 97.7263,
+      "eval_samples_per_second": 29.92,
+      "eval_steps_per_second": 3.745,
+      "step": 2800
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 3.6538245677948,
+      "learning_rate": 0.00010390143737166324,
+      "loss": 0.484,
+      "step": 2810
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 4.382143974304199,
+      "learning_rate": 0.00010355920602327174,
+      "loss": 0.4907,
+      "step": 2820
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 4.952005386352539,
+      "learning_rate": 0.00010321697467488023,
+      "loss": 0.6108,
+      "step": 2830
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 5.33458948135376,
+      "learning_rate": 0.00010287474332648872,
+      "loss": 0.5257,
+      "step": 2840
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 1.6532644033432007,
+      "learning_rate": 0.00010253251197809719,
+      "loss": 0.5796,
+      "step": 2850
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 3.1267786026000977,
+      "learning_rate": 0.00010219028062970568,
+      "loss": 0.5385,
+      "step": 2860
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 2.2853283882141113,
+      "learning_rate": 0.00010184804928131417,
+      "loss": 0.4369,
+      "step": 2870
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 4.428330421447754,
+      "learning_rate": 0.00010150581793292267,
+      "loss": 0.5144,
+      "step": 2880
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 2.0124640464782715,
+      "learning_rate": 0.00010116358658453116,
+      "loss": 0.427,
+      "step": 2890
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 5.213257789611816,
+      "learning_rate": 0.00010082135523613962,
+      "loss": 0.5491,
+      "step": 2900
+    },
+    {
+      "epoch": 1.98,
+      "eval_accuracy": 0.7154582763337893,
+      "eval_loss": 0.7131847143173218,
+      "eval_runtime": 97.9351,
+      "eval_samples_per_second": 29.857,
+      "eval_steps_per_second": 3.737,
+      "step": 2900
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 4.522421836853027,
+      "learning_rate": 0.00010047912388774812,
+      "loss": 0.4969,
+      "step": 2910
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 2.8199572563171387,
+      "learning_rate": 0.00010013689253935661,
+      "loss": 0.535,
+      "step": 2920
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 3.0851550102233887,
+      "learning_rate": 9.97946611909651e-05,
+      "loss": 0.3725,
+      "step": 2930
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 2.9317426681518555,
+      "learning_rate": 9.945242984257359e-05,
+      "loss": 0.2253,
+      "step": 2940
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 2.0250513553619385,
+      "learning_rate": 9.911019849418207e-05,
+      "loss": 0.226,
+      "step": 2950
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 2.064840316772461,
+      "learning_rate": 9.876796714579056e-05,
+      "loss": 0.2686,
+      "step": 2960
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 2.1055028438568115,
+      "learning_rate": 9.842573579739904e-05,
+      "loss": 0.26,
+      "step": 2970
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 1.3024044036865234,
+      "learning_rate": 9.808350444900754e-05,
+      "loss": 0.2297,
+      "step": 2980
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 5.22949743270874,
+      "learning_rate": 9.774127310061603e-05,
+      "loss": 0.2884,
+      "step": 2990
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 3.9687552452087402,
+      "learning_rate": 9.73990417522245e-05,
+      "loss": 0.2373,
+      "step": 3000
+    },
+    {
+      "epoch": 2.05,
+      "eval_accuracy": 0.7096443228454172,
+      "eval_loss": 0.7831349968910217,
+      "eval_runtime": 98.1821,
+      "eval_samples_per_second": 29.781,
+      "eval_steps_per_second": 3.728,
+      "step": 3000
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 2.5891599655151367,
+      "learning_rate": 9.7056810403833e-05,
+      "loss": 0.1881,
+      "step": 3010
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 1.736016035079956,
+      "learning_rate": 9.671457905544148e-05,
+      "loss": 0.2452,
+      "step": 3020
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 2.477384567260742,
+      "learning_rate": 9.637234770704997e-05,
+      "loss": 0.3086,
+      "step": 3030
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 2.031851053237915,
+      "learning_rate": 9.603011635865847e-05,
+      "loss": 0.2753,
+      "step": 3040
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 2.2902073860168457,
+      "learning_rate": 9.568788501026694e-05,
+      "loss": 0.2279,
+      "step": 3050
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 3.447719097137451,
+      "learning_rate": 9.534565366187543e-05,
+      "loss": 0.3219,
+      "step": 3060
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 3.44710373878479,
+      "learning_rate": 9.500342231348393e-05,
+      "loss": 0.2313,
+      "step": 3070
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 2.4824321269989014,
+      "learning_rate": 9.46611909650924e-05,
+      "loss": 0.2762,
+      "step": 3080
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 2.154989719390869,
+      "learning_rate": 9.43189596167009e-05,
+      "loss": 0.2523,
+      "step": 3090
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 6.111990928649902,
+      "learning_rate": 9.397672826830938e-05,
+      "loss": 0.2756,
+      "step": 3100
+    },
+    {
+      "epoch": 2.12,
+      "eval_accuracy": 0.7246922024623803,
+      "eval_loss": 0.7965042591094971,
+      "eval_runtime": 97.9768,
+      "eval_samples_per_second": 29.844,
+      "eval_steps_per_second": 3.736,
+      "step": 3100
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 3.067180633544922,
+      "learning_rate": 9.363449691991787e-05,
+      "loss": 0.3076,
+      "step": 3110
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 1.2516257762908936,
+      "learning_rate": 9.329226557152636e-05,
+      "loss": 0.308,
+      "step": 3120
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 0.4879588484764099,
+      "learning_rate": 9.295003422313485e-05,
+      "loss": 0.2395,
+      "step": 3130
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 1.9442752599716187,
+      "learning_rate": 9.260780287474334e-05,
+      "loss": 0.1651,
+      "step": 3140
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 1.563724160194397,
+      "learning_rate": 9.226557152635181e-05,
+      "loss": 0.1976,
+      "step": 3150
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 2.747708320617676,
+      "learning_rate": 9.19233401779603e-05,
+      "loss": 0.1894,
+      "step": 3160
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 1.413163185119629,
+      "learning_rate": 9.15811088295688e-05,
+      "loss": 0.257,
+      "step": 3170
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 5.1449875831604,
+      "learning_rate": 9.123887748117727e-05,
+      "loss": 0.2783,
+      "step": 3180
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 8.056723594665527,
+      "learning_rate": 9.089664613278576e-05,
+      "loss": 0.4055,
+      "step": 3190
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 3.5667428970336914,
+      "learning_rate": 9.058863791923341e-05,
+      "loss": 0.2299,
+      "step": 3200
+    },
+    {
+      "epoch": 2.19,
+      "eval_accuracy": 0.7219562243502052,
+      "eval_loss": 0.8241410851478577,
+      "eval_runtime": 97.6979,
+      "eval_samples_per_second": 29.929,
+      "eval_steps_per_second": 3.746,
+      "step": 3200
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 0.5471103191375732,
+      "learning_rate": 9.02464065708419e-05,
+      "loss": 0.2466,
+      "step": 3210
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 3.4794716835021973,
+      "learning_rate": 8.990417522245039e-05,
+      "loss": 0.3227,
+      "step": 3220
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 4.952136993408203,
+      "learning_rate": 8.956194387405887e-05,
+      "loss": 0.2804,
+      "step": 3230
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 3.02123761177063,
+      "learning_rate": 8.921971252566735e-05,
+      "loss": 0.3655,
+      "step": 3240
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 3.419013500213623,
+      "learning_rate": 8.887748117727585e-05,
+      "loss": 0.2,
+      "step": 3250
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 2.6587626934051514,
+      "learning_rate": 8.853524982888432e-05,
+      "loss": 0.1915,
+      "step": 3260
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 6.789358615875244,
+      "learning_rate": 8.819301848049281e-05,
+      "loss": 0.2843,
+      "step": 3270
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 4.29807710647583,
+      "learning_rate": 8.785078713210131e-05,
+      "loss": 0.314,
+      "step": 3280
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 3.908632755279541,
+      "learning_rate": 8.750855578370979e-05,
+      "loss": 0.1721,
+      "step": 3290
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 2.6210243701934814,
+      "learning_rate": 8.716632443531828e-05,
+      "loss": 0.2323,
+      "step": 3300
+    },
+    {
+      "epoch": 2.26,
+      "eval_accuracy": 0.7110123119015048,
+      "eval_loss": 0.8286489248275757,
+      "eval_runtime": 97.6813,
+      "eval_samples_per_second": 29.934,
+      "eval_steps_per_second": 3.747,
+      "step": 3300
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 2.0647199153900146,
+      "learning_rate": 8.682409308692678e-05,
+      "loss": 0.2073,
+      "step": 3310
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 3.1741795539855957,
+      "learning_rate": 8.648186173853525e-05,
+      "loss": 0.3861,
+      "step": 3320
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 2.711451530456543,
+      "learning_rate": 8.613963039014374e-05,
+      "loss": 0.2986,
+      "step": 3330
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 3.5253520011901855,
+      "learning_rate": 8.579739904175223e-05,
+      "loss": 0.1992,
+      "step": 3340
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 2.7021403312683105,
+      "learning_rate": 8.545516769336072e-05,
+      "loss": 0.2105,
+      "step": 3350
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 1.193359136581421,
+      "learning_rate": 8.51129363449692e-05,
+      "loss": 0.3453,
+      "step": 3360
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 4.054200649261475,
+      "learning_rate": 8.47707049965777e-05,
+      "loss": 0.2351,
+      "step": 3370
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 5.733345985412598,
+      "learning_rate": 8.442847364818618e-05,
+      "loss": 0.3263,
+      "step": 3380
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 0.9603079557418823,
+      "learning_rate": 8.408624229979466e-05,
+      "loss": 0.2073,
+      "step": 3390
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 3.1017444133758545,
+      "learning_rate": 8.374401095140314e-05,
+      "loss": 0.1979,
+      "step": 3400
+    },
+    {
+      "epoch": 2.33,
+      "eval_accuracy": 0.7301641586867305,
+      "eval_loss": 0.7993032336235046,
+      "eval_runtime": 98.5314,
+      "eval_samples_per_second": 29.676,
+      "eval_steps_per_second": 3.715,
+      "step": 3400
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 4.0206146240234375,
+      "learning_rate": 8.340177960301165e-05,
+      "loss": 0.277,
+      "step": 3410
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 5.1184797286987305,
+      "learning_rate": 8.305954825462012e-05,
+      "loss": 0.2841,
+      "step": 3420
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 1.4796267747879028,
+      "learning_rate": 8.271731690622861e-05,
+      "loss": 0.1861,
+      "step": 3430
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 3.755347967147827,
+      "learning_rate": 8.237508555783711e-05,
+      "loss": 0.2651,
+      "step": 3440
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 1.0390390157699585,
+      "learning_rate": 8.203285420944559e-05,
+      "loss": 0.2076,
+      "step": 3450
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 3.164752244949341,
+      "learning_rate": 8.169062286105407e-05,
+      "loss": 0.2688,
+      "step": 3460
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 1.338983416557312,
+      "learning_rate": 8.134839151266256e-05,
+      "loss": 0.1923,
+      "step": 3470
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 3.2453787326812744,
+      "learning_rate": 8.100616016427105e-05,
+      "loss": 0.3519,
+      "step": 3480
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 1.3863258361816406,
+      "learning_rate": 8.066392881587954e-05,
+      "loss": 0.3351,
+      "step": 3490
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 3.6055376529693604,
+      "learning_rate": 8.032169746748803e-05,
+      "loss": 0.2507,
+      "step": 3500
+    },
+    {
+      "epoch": 2.4,
+      "eval_accuracy": 0.7188782489740082,
+      "eval_loss": 0.8477036952972412,
+      "eval_runtime": 98.8022,
+      "eval_samples_per_second": 29.594,
+      "eval_steps_per_second": 3.704,
+      "step": 3500
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 3.500161647796631,
+      "learning_rate": 7.997946611909651e-05,
+      "loss": 0.2426,
+      "step": 3510
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 2.6914896965026855,
+      "learning_rate": 7.9637234770705e-05,
+      "loss": 0.2767,
+      "step": 3520
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 3.296560287475586,
+      "learning_rate": 7.929500342231349e-05,
+      "loss": 0.3303,
+      "step": 3530
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 3.575965642929077,
+      "learning_rate": 7.895277207392198e-05,
+      "loss": 0.2548,
+      "step": 3540
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 3.445319175720215,
+      "learning_rate": 7.861054072553045e-05,
+      "loss": 0.2209,
+      "step": 3550
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 3.6920342445373535,
+      "learning_rate": 7.826830937713896e-05,
+      "loss": 0.3507,
+      "step": 3560
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 4.31784200668335,
+      "learning_rate": 7.792607802874744e-05,
+      "loss": 0.339,
+      "step": 3570
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 3.469020128250122,
+      "learning_rate": 7.758384668035592e-05,
+      "loss": 0.297,
+      "step": 3580
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 2.89150333404541,
+      "learning_rate": 7.72416153319644e-05,
+      "loss": 0.2303,
+      "step": 3590
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 2.1436691284179688,
+      "learning_rate": 7.689938398357291e-05,
+      "loss": 0.205,
+      "step": 3600
+    },
+    {
+      "epoch": 2.46,
+      "eval_accuracy": 0.7123803009575923,
+      "eval_loss": 0.8196815252304077,
+      "eval_runtime": 98.7766,
+      "eval_samples_per_second": 29.602,
+      "eval_steps_per_second": 3.705,
+      "step": 3600
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 3.9782419204711914,
+      "learning_rate": 7.655715263518138e-05,
+      "loss": 0.2987,
+      "step": 3610
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 2.255920648574829,
+      "learning_rate": 7.621492128678987e-05,
+      "loss": 0.2195,
+      "step": 3620
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 1.3451207876205444,
+      "learning_rate": 7.587268993839836e-05,
+      "loss": 0.1647,
+      "step": 3630
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 3.4238128662109375,
+      "learning_rate": 7.553045859000685e-05,
+      "loss": 0.2629,
+      "step": 3640
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 4.091916084289551,
+      "learning_rate": 7.518822724161534e-05,
+      "loss": 0.2447,
+      "step": 3650
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 5.296712398529053,
+      "learning_rate": 7.484599589322382e-05,
+      "loss": 0.2414,
+      "step": 3660
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 1.0360019207000732,
+      "learning_rate": 7.450376454483231e-05,
+      "loss": 0.1963,
+      "step": 3670
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 3.331775665283203,
+      "learning_rate": 7.41615331964408e-05,
+      "loss": 0.275,
+      "step": 3680
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 1.4381409883499146,
+      "learning_rate": 7.381930184804929e-05,
+      "loss": 0.2396,
+      "step": 3690
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 2.7507870197296143,
+      "learning_rate": 7.347707049965778e-05,
+      "loss": 0.35,
+      "step": 3700
+    },
+    {
+      "epoch": 2.53,
+      "eval_accuracy": 0.7127222982216143,
+      "eval_loss": 0.8347957730293274,
+      "eval_runtime": 98.1633,
+      "eval_samples_per_second": 29.787,
+      "eval_steps_per_second": 3.728,
+      "step": 3700
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 5.777230262756348,
+      "learning_rate": 7.313483915126625e-05,
+      "loss": 0.2073,
+      "step": 3710
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 2.3496434688568115,
+      "learning_rate": 7.279260780287475e-05,
+      "loss": 0.1783,
+      "step": 3720
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 0.5175051689147949,
+      "learning_rate": 7.245037645448324e-05,
+      "loss": 0.3318,
+      "step": 3730
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 3.0439980030059814,
+      "learning_rate": 7.210814510609171e-05,
+      "loss": 0.1888,
+      "step": 3740
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 3.7360622882843018,
+      "learning_rate": 7.17659137577002e-05,
+      "loss": 0.1365,
+      "step": 3750
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 2.4673309326171875,
+      "learning_rate": 7.14236824093087e-05,
+      "loss": 0.3447,
+      "step": 3760
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 5.723246097564697,
+      "learning_rate": 7.108145106091718e-05,
+      "loss": 0.256,
+      "step": 3770
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 1.58307683467865,
+      "learning_rate": 7.073921971252567e-05,
+      "loss": 0.2825,
+      "step": 3780
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 5.220044136047363,
+      "learning_rate": 7.039698836413416e-05,
+      "loss": 0.2681,
+      "step": 3790
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 4.61745548248291,
+      "learning_rate": 7.005475701574264e-05,
+      "loss": 0.3372,
+      "step": 3800
+    },
+    {
+      "epoch": 2.6,
+      "eval_accuracy": 0.7199042407660738,
+      "eval_loss": 0.8999379277229309,
+      "eval_runtime": 97.7794,
+      "eval_samples_per_second": 29.904,
+      "eval_steps_per_second": 3.743,
+      "step": 3800
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 3.353637218475342,
+      "learning_rate": 6.971252566735113e-05,
+      "loss": 0.3247,
+      "step": 3810
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 3.8344199657440186,
+      "learning_rate": 6.937029431895962e-05,
+      "loss": 0.1697,
+      "step": 3820
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 6.490875244140625,
+      "learning_rate": 6.902806297056811e-05,
+      "loss": 0.1713,
+      "step": 3830
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 5.620298862457275,
+      "learning_rate": 6.86858316221766e-05,
+      "loss": 0.204,
+      "step": 3840
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 2.180828094482422,
+      "learning_rate": 6.834360027378508e-05,
+      "loss": 0.2048,
+      "step": 3850
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 4.349495887756348,
+      "learning_rate": 6.800136892539357e-05,
+      "loss": 0.189,
+      "step": 3860
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 4.728153705596924,
+      "learning_rate": 6.765913757700205e-05,
+      "loss": 0.2495,
+      "step": 3870
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 0.8773788809776306,
+      "learning_rate": 6.731690622861055e-05,
+      "loss": 0.2749,
+      "step": 3880
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 5.594605922698975,
+      "learning_rate": 6.697467488021904e-05,
+      "loss": 0.2288,
+      "step": 3890
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 2.9209389686584473,
+      "learning_rate": 6.663244353182751e-05,
+      "loss": 0.1968,
+      "step": 3900
+    },
+    {
+      "epoch": 2.67,
+      "eval_accuracy": 0.7274281805745554,
+      "eval_loss": 0.8263109922409058,
+      "eval_runtime": 97.795,
+      "eval_samples_per_second": 29.899,
+      "eval_steps_per_second": 3.743,
+      "step": 3900
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 5.26107120513916,
+      "learning_rate": 6.6290212183436e-05,
+      "loss": 0.2062,
+      "step": 3910
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 3.8810832500457764,
+      "learning_rate": 6.59479808350445e-05,
+      "loss": 0.2644,
+      "step": 3920
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 6.908627510070801,
+      "learning_rate": 6.560574948665298e-05,
+      "loss": 0.234,
+      "step": 3930
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 3.4059367179870605,
+      "learning_rate": 6.526351813826146e-05,
+      "loss": 0.2296,
+      "step": 3940
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 4.1873698234558105,
+      "learning_rate": 6.492128678986995e-05,
+      "loss": 0.2907,
+      "step": 3950
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 3.698305606842041,
+      "learning_rate": 6.457905544147844e-05,
+      "loss": 0.2561,
+      "step": 3960
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 6.283819198608398,
+      "learning_rate": 6.423682409308693e-05,
+      "loss": 0.1497,
+      "step": 3970
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 3.8906710147857666,
+      "learning_rate": 6.389459274469542e-05,
+      "loss": 0.2858,
+      "step": 3980
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 3.4375922679901123,
+      "learning_rate": 6.35523613963039e-05,
+      "loss": 0.1774,
+      "step": 3990
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 0.2142799347639084,
+      "learning_rate": 6.32101300479124e-05,
+      "loss": 0.1443,
+      "step": 4000
+    },
+    {
+      "epoch": 2.74,
+      "eval_accuracy": 0.7243502051983585,
+      "eval_loss": 0.8703867793083191,
+      "eval_runtime": 98.045,
+      "eval_samples_per_second": 29.823,
+      "eval_steps_per_second": 3.733,
+      "step": 4000
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 3.87772798538208,
+      "learning_rate": 6.286789869952088e-05,
+      "loss": 0.2021,
+      "step": 4010
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 0.43029889464378357,
+      "learning_rate": 6.252566735112937e-05,
+      "loss": 0.2097,
+      "step": 4020
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 7.595574855804443,
+      "learning_rate": 6.2217659137577e-05,
+      "loss": 0.2467,
+      "step": 4030
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 4.542736053466797,
+      "learning_rate": 6.187542778918549e-05,
+      "loss": 0.2685,
+      "step": 4040
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 3.0954976081848145,
+      "learning_rate": 6.153319644079398e-05,
+      "loss": 0.2651,
+      "step": 4050
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 0.6948625445365906,
+      "learning_rate": 6.119096509240247e-05,
+      "loss": 0.3066,
+      "step": 4060
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 2.493579387664795,
+      "learning_rate": 6.0848733744010955e-05,
+      "loss": 0.286,
+      "step": 4070
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 4.651486873626709,
+      "learning_rate": 6.050650239561945e-05,
+      "loss": 0.3193,
+      "step": 4080
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 2.718686103820801,
+      "learning_rate": 6.0164271047227924e-05,
+      "loss": 0.2494,
+      "step": 4090
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 3.578540563583374,
+      "learning_rate": 5.982203969883642e-05,
+      "loss": 0.1933,
+      "step": 4100
+    },
+    {
+      "epoch": 2.81,
+      "eval_accuracy": 0.7243502051983585,
+      "eval_loss": 0.8270108103752136,
+      "eval_runtime": 97.3019,
+      "eval_samples_per_second": 30.051,
+      "eval_steps_per_second": 3.761,
+      "step": 4100
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 5.739526271820068,
+      "learning_rate": 5.94798083504449e-05,
+      "loss": 0.2929,
+      "step": 4110
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 5.148767471313477,
+      "learning_rate": 5.913757700205339e-05,
+      "loss": 0.245,
+      "step": 4120
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 3.4480843544006348,
+      "learning_rate": 5.8795345653661884e-05,
+      "loss": 0.1627,
+      "step": 4130
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 8.554824829101562,
+      "learning_rate": 5.8453114305270365e-05,
+      "loss": 0.2578,
+      "step": 4140
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 4.859767436981201,
+      "learning_rate": 5.8110882956878853e-05,
+      "loss": 0.1952,
+      "step": 4150
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 3.225494146347046,
+      "learning_rate": 5.776865160848735e-05,
+      "loss": 0.2609,
+      "step": 4160
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 6.27465295791626,
+      "learning_rate": 5.742642026009582e-05,
+      "loss": 0.3345,
+      "step": 4170
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 3.6918678283691406,
+      "learning_rate": 5.708418891170432e-05,
+      "loss": 0.2109,
+      "step": 4180
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 5.811365604400635,
+      "learning_rate": 5.67419575633128e-05,
+      "loss": 0.1586,
+      "step": 4190
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 10.013077735900879,
+      "learning_rate": 5.639972621492129e-05,
+      "loss": 0.2044,
+      "step": 4200
+    },
+    {
+      "epoch": 2.87,
+      "eval_accuracy": 0.7274281805745554,
+      "eval_loss": 0.8322764039039612,
+      "eval_runtime": 97.5717,
+      "eval_samples_per_second": 29.968,
+      "eval_steps_per_second": 3.751,
+      "step": 4200
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 0.11126231402158737,
+      "learning_rate": 5.605749486652978e-05,
+      "loss": 0.2304,
+      "step": 4210
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 8.398926734924316,
+      "learning_rate": 5.5715263518138264e-05,
+      "loss": 0.259,
+      "step": 4220
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 4.514709949493408,
+      "learning_rate": 5.537303216974675e-05,
+      "loss": 0.1818,
+      "step": 4230
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 5.3074235916137695,
+      "learning_rate": 5.503080082135525e-05,
+      "loss": 0.2162,
+      "step": 4240
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 6.365301132202148,
+      "learning_rate": 5.468856947296372e-05,
+      "loss": 0.3011,
+      "step": 4250
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 1.199822187423706,
+      "learning_rate": 5.4346338124572216e-05,
+      "loss": 0.2197,
+      "step": 4260
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 5.970396995544434,
+      "learning_rate": 5.40041067761807e-05,
+      "loss": 0.2257,
+      "step": 4270
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 4.547550678253174,
+      "learning_rate": 5.3661875427789186e-05,
+      "loss": 0.1782,
+      "step": 4280
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 1.4566155672073364,
+      "learning_rate": 5.331964407939768e-05,
+      "loss": 0.1697,
+      "step": 4290
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 0.7294021248817444,
+      "learning_rate": 5.297741273100616e-05,
+      "loss": 0.2709,
+      "step": 4300
+    },
+    {
+      "epoch": 2.94,
+      "eval_accuracy": 0.7294801641586868,
+      "eval_loss": 0.8493737578392029,
+      "eval_runtime": 98.5093,
+      "eval_samples_per_second": 29.682,
+      "eval_steps_per_second": 3.715,
+      "step": 4300
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 1.6976951360702515,
+      "learning_rate": 5.263518138261465e-05,
+      "loss": 0.1856,
+      "step": 4310
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 2.718456268310547,
+      "learning_rate": 5.2292950034223145e-05,
+      "loss": 0.148,
+      "step": 4320
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 2.8870649337768555,
+      "learning_rate": 5.195071868583162e-05,
+      "loss": 0.2673,
+      "step": 4330
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 8.342497825622559,
+      "learning_rate": 5.1608487337440115e-05,
+      "loss": 0.1964,
+      "step": 4340
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 4.078546524047852,
+      "learning_rate": 5.1266255989048596e-05,
+      "loss": 0.2781,
+      "step": 4350
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 4.1737799644470215,
+      "learning_rate": 5.0924024640657084e-05,
+      "loss": 0.2913,
+      "step": 4360
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 4.929452896118164,
+      "learning_rate": 5.058179329226558e-05,
+      "loss": 0.1411,
+      "step": 4370
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 3.994762420654297,
+      "learning_rate": 5.023956194387406e-05,
+      "loss": 0.2904,
+      "step": 4380
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 0.9369414448738098,
+      "learning_rate": 4.989733059548255e-05,
+      "loss": 0.1364,
+      "step": 4390
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.4886672496795654,
+      "learning_rate": 4.955509924709104e-05,
+      "loss": 0.1021,
+      "step": 4400
+    },
+    {
+      "epoch": 3.01,
+      "eval_accuracy": 0.7335841313269493,
+      "eval_loss": 0.857344388961792,
+      "eval_runtime": 97.6576,
+      "eval_samples_per_second": 29.941,
+      "eval_steps_per_second": 3.748,
+      "step": 4400
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 1.9758846759796143,
+      "learning_rate": 4.921286789869952e-05,
+      "loss": 0.0595,
+      "step": 4410
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 0.5239822268486023,
+      "learning_rate": 4.887063655030801e-05,
+      "loss": 0.0655,
+      "step": 4420
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 3.8521106243133545,
+      "learning_rate": 4.85284052019165e-05,
+      "loss": 0.0548,
+      "step": 4430
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 0.09303455054759979,
+      "learning_rate": 4.818617385352498e-05,
+      "loss": 0.0626,
+      "step": 4440
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 0.5870689749717712,
+      "learning_rate": 4.784394250513347e-05,
+      "loss": 0.0421,
+      "step": 4450
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 2.024603843688965,
+      "learning_rate": 4.7501711156741966e-05,
+      "loss": 0.0241,
+      "step": 4460
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 1.6237764358520508,
+      "learning_rate": 4.715947980835045e-05,
+      "loss": 0.092,
+      "step": 4470
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.9192093014717102,
+      "learning_rate": 4.6817248459958935e-05,
+      "loss": 0.0629,
+      "step": 4480
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 0.39932507276535034,
+      "learning_rate": 4.6475017111567424e-05,
+      "loss": 0.0191,
+      "step": 4490
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 0.11515524983406067,
+      "learning_rate": 4.6132785763175905e-05,
+      "loss": 0.0393,
+      "step": 4500
+    },
+    {
+      "epoch": 3.08,
+      "eval_accuracy": 0.737688098495212,
+      "eval_loss": 0.9333314299583435,
+      "eval_runtime": 100.3956,
+      "eval_samples_per_second": 29.125,
+      "eval_steps_per_second": 3.646,
+      "step": 4500
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 1.743781566619873,
+      "learning_rate": 4.57905544147844e-05,
+      "loss": 0.0314,
+      "step": 4510
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 0.39528560638427734,
+      "learning_rate": 4.544832306639288e-05,
+      "loss": 0.0695,
+      "step": 4520
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 6.3506340980529785,
+      "learning_rate": 4.510609171800137e-05,
+      "loss": 0.0587,
+      "step": 4530
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 4.544747352600098,
+      "learning_rate": 4.476386036960986e-05,
+      "loss": 0.0302,
+      "step": 4540
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 1.1574302911758423,
+      "learning_rate": 4.4421629021218346e-05,
+      "loss": 0.1469,
+      "step": 4550
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 9.069709777832031,
+      "learning_rate": 4.4079397672826834e-05,
+      "loss": 0.0678,
+      "step": 4560
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.6530300974845886,
+      "learning_rate": 4.373716632443532e-05,
+      "loss": 0.0596,
+      "step": 4570
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 0.2227853536605835,
+      "learning_rate": 4.3394934976043803e-05,
+      "loss": 0.0933,
+      "step": 4580
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 4.514962673187256,
+      "learning_rate": 4.30527036276523e-05,
+      "loss": 0.0967,
+      "step": 4590
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 3.422731399536133,
+      "learning_rate": 4.271047227926078e-05,
+      "loss": 0.0973,
+      "step": 4600
+    },
+    {
+      "epoch": 3.15,
+      "eval_accuracy": 0.7335841313269493,
+      "eval_loss": 0.964632511138916,
+      "eval_runtime": 98.6581,
+      "eval_samples_per_second": 29.638,
+      "eval_steps_per_second": 3.71,
+      "step": 4600
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 2.566582202911377,
+      "learning_rate": 4.236824093086927e-05,
+      "loss": 0.0826,
+      "step": 4610
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 0.1476365625858307,
+      "learning_rate": 4.2026009582477756e-05,
+      "loss": 0.0658,
+      "step": 4620
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 0.049972761422395706,
+      "learning_rate": 4.1683778234086244e-05,
+      "loss": 0.0615,
+      "step": 4630
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 0.12682102620601654,
+      "learning_rate": 4.134154688569473e-05,
+      "loss": 0.0509,
+      "step": 4640
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 5.897772789001465,
+      "learning_rate": 4.099931553730322e-05,
+      "loss": 0.0915,
+      "step": 4650
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 0.03461030498147011,
+      "learning_rate": 4.06570841889117e-05,
+      "loss": 0.0837,
+      "step": 4660
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 0.14550870656967163,
+      "learning_rate": 4.03148528405202e-05,
+      "loss": 0.0638,
+      "step": 4670
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 0.03923880308866501,
+      "learning_rate": 3.997262149212868e-05,
+      "loss": 0.0367,
+      "step": 4680
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 3.9474973678588867,
+      "learning_rate": 3.9630390143737166e-05,
+      "loss": 0.1589,
+      "step": 4690
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 0.4555126428604126,
+      "learning_rate": 3.9288158795345655e-05,
+      "loss": 0.0317,
+      "step": 4700
+    },
+    {
+      "epoch": 3.22,
+      "eval_accuracy": 0.7335841313269493,
+      "eval_loss": 0.9820448160171509,
+      "eval_runtime": 98.4684,
+      "eval_samples_per_second": 29.695,
+      "eval_steps_per_second": 3.717,
+      "step": 4700
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 0.07397227734327316,
+      "learning_rate": 3.894592744695414e-05,
+      "loss": 0.0509,
+      "step": 4710
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 0.12771746516227722,
+      "learning_rate": 3.860369609856263e-05,
+      "loss": 0.0432,
+      "step": 4720
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 0.5672203302383423,
+      "learning_rate": 3.826146475017112e-05,
+      "loss": 0.0284,
+      "step": 4730
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 0.2664686441421509,
+      "learning_rate": 3.79192334017796e-05,
+      "loss": 0.0858,
+      "step": 4740
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 6.073988437652588,
+      "learning_rate": 3.7577002053388095e-05,
+      "loss": 0.0706,
+      "step": 4750
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 0.13147962093353271,
+      "learning_rate": 3.723477070499658e-05,
+      "loss": 0.0774,
+      "step": 4760
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 6.215001106262207,
+      "learning_rate": 3.6892539356605065e-05,
+      "loss": 0.039,
+      "step": 4770
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 0.5598773956298828,
+      "learning_rate": 3.655030800821355e-05,
+      "loss": 0.0236,
+      "step": 4780
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 0.738316535949707,
+      "learning_rate": 3.620807665982204e-05,
+      "loss": 0.0366,
+      "step": 4790
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 1.859127402305603,
+      "learning_rate": 3.586584531143053e-05,
+      "loss": 0.0458,
+      "step": 4800
+    },
+    {
+      "epoch": 3.29,
+      "eval_accuracy": 0.7325581395348837,
+      "eval_loss": 1.0716344118118286,
+      "eval_runtime": 98.9214,
+      "eval_samples_per_second": 29.559,
+      "eval_steps_per_second": 3.7,
+      "step": 4800
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 3.993727684020996,
+      "learning_rate": 3.552361396303902e-05,
+      "loss": 0.0542,
+      "step": 4810
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 3.287825345993042,
+      "learning_rate": 3.51813826146475e-05,
+      "loss": 0.0408,
+      "step": 4820
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 5.9396071434021,
+      "learning_rate": 3.4839151266255994e-05,
+      "loss": 0.0919,
+      "step": 4830
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 0.04825394228100777,
+      "learning_rate": 3.449691991786448e-05,
+      "loss": 0.0754,
+      "step": 4840
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 1.768473744392395,
+      "learning_rate": 3.415468856947296e-05,
+      "loss": 0.0924,
+      "step": 4850
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 0.08893775939941406,
+      "learning_rate": 3.381245722108145e-05,
+      "loss": 0.0526,
+      "step": 4860
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 10.308042526245117,
+      "learning_rate": 3.347022587268994e-05,
+      "loss": 0.1114,
+      "step": 4870
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 7.975988388061523,
+      "learning_rate": 3.312799452429843e-05,
+      "loss": 0.0368,
+      "step": 4880
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 1.1892056465148926,
+      "learning_rate": 3.2785763175906916e-05,
+      "loss": 0.0976,
+      "step": 4890
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 4.882401466369629,
+      "learning_rate": 3.24435318275154e-05,
+      "loss": 0.164,
+      "step": 4900
+    },
+    {
+      "epoch": 3.35,
+      "eval_accuracy": 0.7311901504787962,
+      "eval_loss": 1.088935375213623,
+      "eval_runtime": 99.102,
+      "eval_samples_per_second": 29.505,
+      "eval_steps_per_second": 3.693,
+      "step": 4900
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 0.2973441481590271,
+      "learning_rate": 3.210130047912389e-05,
+      "loss": 0.0566,
+      "step": 4910
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 2.9655909538269043,
+      "learning_rate": 3.175906913073238e-05,
+      "loss": 0.0773,
+      "step": 4920
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 0.0812312662601471,
+      "learning_rate": 3.141683778234086e-05,
+      "loss": 0.0484,
+      "step": 4930
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 0.9421347975730896,
+      "learning_rate": 3.107460643394935e-05,
+      "loss": 0.044,
+      "step": 4940
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 0.39745262265205383,
+      "learning_rate": 3.073237508555784e-05,
+      "loss": 0.0686,
+      "step": 4950
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 5.0486159324646,
+      "learning_rate": 3.0390143737166326e-05,
+      "loss": 0.0724,
+      "step": 4960
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 0.15137074887752533,
+      "learning_rate": 3.004791238877481e-05,
+      "loss": 0.0659,
+      "step": 4970
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 0.8398496508598328,
+      "learning_rate": 2.97056810403833e-05,
+      "loss": 0.0367,
+      "step": 4980
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 0.19229641556739807,
+      "learning_rate": 2.936344969199179e-05,
+      "loss": 0.0558,
+      "step": 4990
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 1.1081570386886597,
+      "learning_rate": 2.9021218343600276e-05,
+      "loss": 0.0578,
+      "step": 5000
+    },
+    {
+      "epoch": 3.42,
+      "eval_accuracy": 0.7311901504787962,
+      "eval_loss": 1.1010822057724,
+      "eval_runtime": 98.8332,
+      "eval_samples_per_second": 29.585,
+      "eval_steps_per_second": 3.703,
+      "step": 5000
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 0.5777124166488647,
+      "learning_rate": 2.867898699520876e-05,
+      "loss": 0.0801,
+      "step": 5010
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 0.2859288156032562,
+      "learning_rate": 2.833675564681725e-05,
+      "loss": 0.0107,
+      "step": 5020
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 0.21665354073047638,
+      "learning_rate": 2.799452429842574e-05,
+      "loss": 0.0714,
+      "step": 5030
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 0.04660337045788765,
+      "learning_rate": 2.7652292950034225e-05,
+      "loss": 0.033,
+      "step": 5040
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 0.08590657263994217,
+      "learning_rate": 2.7310061601642713e-05,
+      "loss": 0.0928,
+      "step": 5050
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 4.4041032791137695,
+      "learning_rate": 2.6967830253251198e-05,
+      "loss": 0.0976,
+      "step": 5060
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 4.6001081466674805,
+      "learning_rate": 2.662559890485969e-05,
+      "loss": 0.0908,
+      "step": 5070
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 1.22696852684021,
+      "learning_rate": 2.6283367556468174e-05,
+      "loss": 0.0581,
+      "step": 5080
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 5.576196193695068,
+      "learning_rate": 2.5941136208076662e-05,
+      "loss": 0.0522,
+      "step": 5090
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 0.21262912452220917,
+      "learning_rate": 2.5598904859685147e-05,
+      "loss": 0.0563,
+      "step": 5100
+    },
+    {
+      "epoch": 3.49,
+      "eval_accuracy": 0.7356361149110807,
+      "eval_loss": 1.1010431051254272,
+      "eval_runtime": 98.8855,
+      "eval_samples_per_second": 29.57,
+      "eval_steps_per_second": 3.701,
+      "step": 5100
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 3.5604753494262695,
+      "learning_rate": 2.525667351129364e-05,
+      "loss": 0.0959,
+      "step": 5110
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 0.3624362349510193,
+      "learning_rate": 2.4914442162902123e-05,
+      "loss": 0.0614,
+      "step": 5120
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 0.05665013566613197,
+      "learning_rate": 2.457221081451061e-05,
+      "loss": 0.078,
+      "step": 5130
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 0.3010866641998291,
+      "learning_rate": 2.42299794661191e-05,
+      "loss": 0.0446,
+      "step": 5140
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 2.0640757083892822,
+      "learning_rate": 2.3887748117727584e-05,
+      "loss": 0.0422,
+      "step": 5150
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 0.3488443195819855,
+      "learning_rate": 2.3545516769336072e-05,
+      "loss": 0.0837,
+      "step": 5160
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 3.6810433864593506,
+      "learning_rate": 2.320328542094456e-05,
+      "loss": 0.1436,
+      "step": 5170
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 6.979311943054199,
+      "learning_rate": 2.286105407255305e-05,
+      "loss": 0.0283,
+      "step": 5180
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 0.41332653164863586,
+      "learning_rate": 2.2518822724161534e-05,
+      "loss": 0.0409,
+      "step": 5190
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 4.3381853103637695,
+      "learning_rate": 2.2176591375770022e-05,
+      "loss": 0.0318,
+      "step": 5200
+    },
+    {
+      "epoch": 3.56,
+      "eval_accuracy": 0.7342681258549931,
+      "eval_loss": 1.0923420190811157,
+      "eval_runtime": 98.234,
+      "eval_samples_per_second": 29.766,
+      "eval_steps_per_second": 3.726,
+      "step": 5200
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 5.9086761474609375,
+      "learning_rate": 2.183436002737851e-05,
+      "loss": 0.0669,
+      "step": 5210
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 0.3409196734428406,
+      "learning_rate": 2.1492128678986998e-05,
+      "loss": 0.0446,
+      "step": 5220
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 1.891818642616272,
+      "learning_rate": 2.1149897330595483e-05,
+      "loss": 0.0644,
+      "step": 5230
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 3.4832639694213867,
+      "learning_rate": 2.080766598220397e-05,
+      "loss": 0.0489,
+      "step": 5240
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 0.3685603141784668,
+      "learning_rate": 2.046543463381246e-05,
+      "loss": 0.064,
+      "step": 5250
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 0.036173015832901,
+      "learning_rate": 2.0123203285420947e-05,
+      "loss": 0.0618,
+      "step": 5260
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 0.8872077465057373,
+      "learning_rate": 1.9780971937029432e-05,
+      "loss": 0.088,
+      "step": 5270
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 15.102840423583984,
+      "learning_rate": 1.943874058863792e-05,
+      "loss": 0.0912,
+      "step": 5280
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 0.5182992815971375,
+      "learning_rate": 1.909650924024641e-05,
+      "loss": 0.0427,
+      "step": 5290
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 4.398647308349609,
+      "learning_rate": 1.8754277891854896e-05,
+      "loss": 0.0255,
+      "step": 5300
+    },
+    {
+      "epoch": 3.63,
+      "eval_accuracy": 0.7332421340629275,
+      "eval_loss": 1.1156063079833984,
+      "eval_runtime": 97.9604,
+      "eval_samples_per_second": 29.849,
+      "eval_steps_per_second": 3.736,
+      "step": 5300
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 0.02750484086573124,
+      "learning_rate": 1.841204654346338e-05,
+      "loss": 0.0359,
+      "step": 5310
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 0.08833522349596024,
+      "learning_rate": 1.806981519507187e-05,
+      "loss": 0.0361,
+      "step": 5320
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 5.376247882843018,
+      "learning_rate": 1.7727583846680358e-05,
+      "loss": 0.0454,
+      "step": 5330
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 2.7907521724700928,
+      "learning_rate": 1.7385352498288842e-05,
+      "loss": 0.0519,
+      "step": 5340
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 3.718421220779419,
+      "learning_rate": 1.704312114989733e-05,
+      "loss": 0.0496,
+      "step": 5350
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 4.242332458496094,
+      "learning_rate": 1.670088980150582e-05,
+      "loss": 0.0148,
+      "step": 5360
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.1361379623413086,
+      "learning_rate": 1.6358658453114307e-05,
+      "loss": 0.0713,
+      "step": 5370
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 1.4509526491165161,
+      "learning_rate": 1.601642710472279e-05,
+      "loss": 0.1173,
+      "step": 5380
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 0.025651389732956886,
+      "learning_rate": 1.567419575633128e-05,
+      "loss": 0.0497,
+      "step": 5390
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 0.9164614081382751,
+      "learning_rate": 1.5331964407939768e-05,
+      "loss": 0.0169,
+      "step": 5400
+    },
+    {
+      "epoch": 3.7,
+      "eval_accuracy": 0.7414500683994528,
+      "eval_loss": 1.1049845218658447,
+      "eval_runtime": 97.5294,
+      "eval_samples_per_second": 29.981,
+      "eval_steps_per_second": 3.753,
+      "step": 5400
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 0.16985657811164856,
+      "learning_rate": 1.4989733059548256e-05,
+      "loss": 0.0156,
+      "step": 5410
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 0.046321526169776917,
+      "learning_rate": 1.464750171115674e-05,
+      "loss": 0.0675,
+      "step": 5420
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 0.12195578962564468,
+      "learning_rate": 1.430527036276523e-05,
+      "loss": 0.0291,
+      "step": 5430
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 0.8293215036392212,
+      "learning_rate": 1.3963039014373715e-05,
+      "loss": 0.0309,
+      "step": 5440
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 0.5577486753463745,
+      "learning_rate": 1.3620807665982205e-05,
+      "loss": 0.0484,
+      "step": 5450
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 0.04991504177451134,
+      "learning_rate": 1.327857631759069e-05,
+      "loss": 0.0577,
+      "step": 5460
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 9.814260482788086,
+      "learning_rate": 1.293634496919918e-05,
+      "loss": 0.0549,
+      "step": 5470
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 2.222853183746338,
+      "learning_rate": 1.2594113620807665e-05,
+      "loss": 0.0377,
+      "step": 5480
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 0.03162392973899841,
+      "learning_rate": 1.2251882272416155e-05,
+      "loss": 0.0259,
+      "step": 5490
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 0.04703206941485405,
+      "learning_rate": 1.1909650924024641e-05,
+      "loss": 0.0629,
+      "step": 5500
+    },
+    {
+      "epoch": 3.76,
+      "eval_accuracy": 0.7373461012311902,
+      "eval_loss": 1.1132022142410278,
+      "eval_runtime": 97.6002,
+      "eval_samples_per_second": 29.959,
+      "eval_steps_per_second": 3.75,
+      "step": 5500
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 0.29484233260154724,
+      "learning_rate": 1.1567419575633129e-05,
+      "loss": 0.028,
+      "step": 5510
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 0.05921239033341408,
+      "learning_rate": 1.1225188227241616e-05,
+      "loss": 0.004,
+      "step": 5520
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 0.025789450854063034,
+      "learning_rate": 1.0882956878850104e-05,
+      "loss": 0.0729,
+      "step": 5530
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 2.618316411972046,
+      "learning_rate": 1.054072553045859e-05,
+      "loss": 0.133,
+      "step": 5540
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 0.7046251893043518,
+      "learning_rate": 1.023271731690623e-05,
+      "loss": 0.0963,
+      "step": 5550
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 0.7423825263977051,
+      "learning_rate": 9.890485968514716e-06,
+      "loss": 0.1037,
+      "step": 5560
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 0.02404310554265976,
+      "learning_rate": 9.548254620123204e-06,
+      "loss": 0.0922,
+      "step": 5570
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 0.8627146482467651,
+      "learning_rate": 9.20602327173169e-06,
+      "loss": 0.0687,
+      "step": 5580
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 0.3027166426181793,
+      "learning_rate": 8.863791923340179e-06,
+      "loss": 0.0347,
+      "step": 5590
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 5.652677059173584,
+      "learning_rate": 8.521560574948665e-06,
+      "loss": 0.0627,
+      "step": 5600
+    },
+    {
+      "epoch": 3.83,
+      "eval_accuracy": 0.738030095759234,
+      "eval_loss": 1.1109539270401,
+      "eval_runtime": 98.2791,
+      "eval_samples_per_second": 29.752,
+      "eval_steps_per_second": 3.724,
+      "step": 5600
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 0.16831336915493011,
+      "learning_rate": 8.179329226557153e-06,
+      "loss": 0.0305,
+      "step": 5610
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 0.02433227188885212,
+      "learning_rate": 7.83709787816564e-06,
+      "loss": 0.0803,
+      "step": 5620
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 0.018608298152685165,
+      "learning_rate": 7.494866529774128e-06,
+      "loss": 0.0106,
+      "step": 5630
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 0.16128766536712646,
+      "learning_rate": 7.152635181382615e-06,
+      "loss": 0.0339,
+      "step": 5640
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 5.512686729431152,
+      "learning_rate": 6.810403832991103e-06,
+      "loss": 0.1132,
+      "step": 5650
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 0.03263100981712341,
+      "learning_rate": 6.46817248459959e-06,
+      "loss": 0.1002,
+      "step": 5660
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 0.07603902369737625,
+      "learning_rate": 6.125941136208077e-06,
+      "loss": 0.0517,
+      "step": 5670
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 0.04678818956017494,
+      "learning_rate": 5.7837097878165646e-06,
+      "loss": 0.0745,
+      "step": 5680
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 0.03259843587875366,
+      "learning_rate": 5.441478439425052e-06,
+      "loss": 0.0306,
+      "step": 5690
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 0.06205374002456665,
+      "learning_rate": 5.099247091033539e-06,
+      "loss": 0.0078,
+      "step": 5700
+    },
+    {
+      "epoch": 3.9,
+      "eval_accuracy": 0.7349521203830369,
+      "eval_loss": 1.1116923093795776,
+      "eval_runtime": 98.2862,
+      "eval_samples_per_second": 29.75,
+      "eval_steps_per_second": 3.724,
+      "step": 5700
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 12.98844051361084,
+      "learning_rate": 4.7570157426420265e-06,
+      "loss": 0.0493,
+      "step": 5710
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 3.6260194778442383,
+      "learning_rate": 4.414784394250514e-06,
+      "loss": 0.1285,
+      "step": 5720
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 1.3537558317184448,
+      "learning_rate": 4.072553045859001e-06,
+      "loss": 0.055,
+      "step": 5730
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 6.886358737945557,
+      "learning_rate": 3.7303216974674884e-06,
+      "loss": 0.0996,
+      "step": 5740
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 6.763581275939941,
+      "learning_rate": 3.3880903490759757e-06,
+      "loss": 0.0322,
+      "step": 5750
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 0.09468955546617508,
+      "learning_rate": 3.0458590006844626e-06,
+      "loss": 0.0433,
+      "step": 5760
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 0.07523488253355026,
+      "learning_rate": 2.70362765229295e-06,
+      "loss": 0.0596,
+      "step": 5770
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 6.736556529998779,
+      "learning_rate": 2.3613963039014377e-06,
+      "loss": 0.0475,
+      "step": 5780
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 13.317132949829102,
+      "learning_rate": 2.019164955509925e-06,
+      "loss": 0.0395,
+      "step": 5790
+    },
+    {
+      "epoch": 3.97,
+      "grad_norm": 0.16380663216114044,
+      "learning_rate": 1.676933607118412e-06,
+      "loss": 0.027,
+      "step": 5800
+    },
+    {
+      "epoch": 3.97,
+      "eval_accuracy": 0.7342681258549931,
+      "eval_loss": 1.1201249361038208,
+      "eval_runtime": 97.9757,
+      "eval_samples_per_second": 29.844,
+      "eval_steps_per_second": 3.736,
+      "step": 5800
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 0.9811205267906189,
+      "learning_rate": 1.3347022587268994e-06,
+      "loss": 0.0991,
+      "step": 5810
+    },
+    {
+      "epoch": 3.98,
+      "grad_norm": 0.2990233898162842,
+      "learning_rate": 9.924709103353869e-07,
+      "loss": 0.0435,
+      "step": 5820
+    },
+    {
+      "epoch": 3.99,
+      "grad_norm": 0.09798883646726608,
+      "learning_rate": 6.502395619438741e-07,
+      "loss": 0.03,
+      "step": 5830
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 0.43798768520355225,
+      "learning_rate": 3.080082135523614e-07,
+      "loss": 0.0463,
+      "step": 5840
+    },
+    {
+      "epoch": 4.0,
+      "step": 5844,
+      "total_flos": 7.243900911073935e+18,
+      "train_loss": 0.43194858603623126,
+      "train_runtime": 10540.446,
+      "train_samples_per_second": 8.868,
+      "train_steps_per_second": 0.554
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 5844,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 100,
+  "total_flos": 7.243900911073935e+18,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}