diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,596 +1,6702 @@
 {
-  "best_metric": 2.7241549491882324,
-  "best_model_checkpoint": "./results/models/checkpoint-26406",
-  "epoch": 27.0,
+  "best_metric": 3.3484203815460205,
+  "best_model_checkpoint": "./results/models/checkpoint-469080",
+  "epoch": 15.0,
   "eval_steps": 500,
-  "global_step": 26406,
+  "global_step": 469080,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.5112474437627812,
-      "grad_norm": 0.1318359375,
-      "learning_rate": 0.0009897750511247444,
-      "loss": 3.8641,
+      "epoch": 0.015988743924277307,
+      "grad_norm": 0.1171875,
+      "learning_rate": 0.0009996802251215145,
+      "loss": 1.9242,
       "step": 500
     },
+    {
+      "epoch": 0.031977487848554614,
+      "grad_norm": 0.09814453125,
+      "learning_rate": 0.000999360450243029,
+      "loss": 1.9017,
+      "step": 1000
+    },
+    {
+      "epoch": 0.04796623177283193,
+      "grad_norm": 0.111328125,
+      "learning_rate": 0.0009990406753645435,
+      "loss": 1.8959,
+      "step": 1500
+    },
+    {
+      "epoch": 0.06395497569710923,
+      "grad_norm": 0.0947265625,
+      "learning_rate": 0.0009987209004860577,
+      "loss": 1.8898,
+      "step": 2000
+    },
+    {
+      "epoch": 0.07994371962138655,
+      "grad_norm": 0.1005859375,
+      "learning_rate": 0.0009984011256075722,
+      "loss": 1.8867,
+      "step": 2500
+    },
+    {
+      "epoch": 0.09593246354566386,
+      "grad_norm": 0.09521484375,
+      "learning_rate": 0.0009980813507290869,
+      "loss": 1.8834,
+      "step": 3000
+    },
+    {
+      "epoch": 0.11192120746994116,
+      "grad_norm": 0.09326171875,
+      "learning_rate": 0.0009977615758506011,
+      "loss": 1.8813,
+      "step": 3500
+    },
+    {
+      "epoch": 0.12790995139421846,
+      "grad_norm": 0.09765625,
+      "learning_rate": 0.0009974418009721156,
+      "loss": 1.8789,
+      "step": 4000
+    },
+    {
+      "epoch": 0.14389869531849578,
+      "grad_norm": 0.08935546875,
+      "learning_rate": 0.00099712202609363,
+      "loss": 1.8767,
+      "step": 4500
+    },
+    {
+      "epoch": 0.1598874392427731,
+      "grad_norm": 0.0869140625,
+      "learning_rate": 0.0009968022512151446,
+      "loss": 1.8757,
+      "step": 5000
+    },
+    {
+      "epoch": 0.1758761831670504,
+      "grad_norm": 0.087890625,
+      "learning_rate": 0.000996482476336659,
+      "loss": 1.8734,
+      "step": 5500
+    },
+    {
+      "epoch": 0.1918649270913277,
+      "grad_norm": 0.095703125,
+      "learning_rate": 0.0009961627014581735,
+      "loss": 1.871,
+      "step": 6000
+    },
+    {
+      "epoch": 0.207853671015605,
+      "grad_norm": 0.10400390625,
+      "learning_rate": 0.0009958429265796878,
+      "loss": 1.8699,
+      "step": 6500
+    },
+    {
+      "epoch": 0.22384241493988233,
+      "grad_norm": 0.099609375,
+      "learning_rate": 0.0009955231517012023,
+      "loss": 1.8694,
+      "step": 7000
+    },
+    {
+      "epoch": 0.23983115886415962,
+      "grad_norm": 0.07763671875,
+      "learning_rate": 0.0009952033768227168,
+      "loss": 1.8677,
+      "step": 7500
+    },
+    {
+      "epoch": 0.2558199027884369,
+      "grad_norm": 0.083984375,
+      "learning_rate": 0.0009948836019442312,
+      "loss": 1.8643,
+      "step": 8000
+    },
+    {
+      "epoch": 0.27180864671271426,
+      "grad_norm": 0.09521484375,
+      "learning_rate": 0.0009945638270657457,
+      "loss": 1.8653,
+      "step": 8500
+    },
+    {
+      "epoch": 0.28779739063699156,
+      "grad_norm": 0.11328125,
+      "learning_rate": 0.0009942440521872602,
+      "loss": 1.863,
+      "step": 9000
+    },
+    {
+      "epoch": 0.30378613456126885,
+      "grad_norm": 0.08984375,
+      "learning_rate": 0.0009939242773087747,
+      "loss": 1.8638,
+      "step": 9500
+    },
+    {
+      "epoch": 0.3197748784855462,
+      "grad_norm": 0.09375,
+      "learning_rate": 0.0009936045024302891,
+      "loss": 1.8621,
+      "step": 10000
+    },
+    {
+      "epoch": 0.3357636224098235,
+      "grad_norm": 0.0927734375,
+      "learning_rate": 0.0009932847275518036,
+      "loss": 1.8617,
+      "step": 10500
+    },
+    {
+      "epoch": 0.3517523663341008,
+      "grad_norm": 0.08642578125,
+      "learning_rate": 0.000992964952673318,
+      "loss": 1.8592,
+      "step": 11000
+    },
+    {
+      "epoch": 0.3677411102583781,
+      "grad_norm": 0.10595703125,
+      "learning_rate": 0.0009926451777948324,
+      "loss": 1.8571,
+      "step": 11500
+    },
+    {
+      "epoch": 0.3837298541826554,
+      "grad_norm": 0.09619140625,
+      "learning_rate": 0.0009923254029163468,
+      "loss": 1.8564,
+      "step": 12000
+    },
+    {
+      "epoch": 0.3997185981069327,
+      "grad_norm": 0.0966796875,
+      "learning_rate": 0.0009920056280378613,
+      "loss": 1.8544,
+      "step": 12500
+    },
+    {
+      "epoch": 0.41570734203121,
+      "grad_norm": 0.10302734375,
+      "learning_rate": 0.0009916858531593758,
+      "loss": 1.8552,
+      "step": 13000
+    },
+    {
+      "epoch": 0.43169608595548736,
+      "grad_norm": 0.1005859375,
+      "learning_rate": 0.0009913660782808903,
+      "loss": 1.853,
+      "step": 13500
+    },
+    {
+      "epoch": 0.44768482987976466,
+      "grad_norm": 0.107421875,
+      "learning_rate": 0.0009910463034024047,
+      "loss": 1.8525,
+      "step": 14000
+    },
+    {
+      "epoch": 0.46367357380404195,
+      "grad_norm": 0.10400390625,
+      "learning_rate": 0.0009907265285239192,
+      "loss": 1.8503,
+      "step": 14500
+    },
+    {
+      "epoch": 0.47966231772831924,
+      "grad_norm": 0.1015625,
+      "learning_rate": 0.0009904067536454337,
+      "loss": 1.8501,
+      "step": 15000
+    },
+    {
+      "epoch": 0.4956510616525966,
+      "grad_norm": 0.11181640625,
+      "learning_rate": 0.0009900869787669482,
+      "loss": 1.8494,
+      "step": 15500
+    },
+    {
+      "epoch": 0.5116398055768738,
+      "grad_norm": 0.10546875,
+      "learning_rate": 0.0009897672038884624,
+      "loss": 1.8465,
+      "step": 16000
+    },
+    {
+      "epoch": 0.5276285495011512,
+      "grad_norm": 0.09326171875,
+      "learning_rate": 0.000989447429009977,
+      "loss": 1.8475,
+      "step": 16500
+    },
+    {
+      "epoch": 0.5436172934254285,
+      "grad_norm": 0.09375,
+      "learning_rate": 0.0009891276541314914,
+      "loss": 1.8443,
+      "step": 17000
+    },
+    {
+      "epoch": 0.5596060373497058,
+      "grad_norm": 0.0927734375,
+      "learning_rate": 0.0009888078792530059,
+      "loss": 1.8413,
+      "step": 17500
+    },
+    {
+      "epoch": 0.5755947812739831,
+      "grad_norm": 0.10302734375,
+      "learning_rate": 0.0009884881043745204,
+      "loss": 1.8418,
+      "step": 18000
+    },
+    {
+      "epoch": 0.5915835251982604,
+      "grad_norm": 0.09326171875,
+      "learning_rate": 0.0009881683294960348,
+      "loss": 1.8385,
+      "step": 18500
+    },
+    {
+      "epoch": 0.6075722691225377,
+      "grad_norm": 0.1064453125,
+      "learning_rate": 0.0009878485546175493,
+      "loss": 1.8376,
+      "step": 19000
+    },
+    {
+      "epoch": 0.623561013046815,
+      "grad_norm": 0.130859375,
+      "learning_rate": 0.0009875287797390638,
+      "loss": 1.8368,
+      "step": 19500
+    },
+    {
+      "epoch": 0.6395497569710924,
+      "grad_norm": 0.11279296875,
+      "learning_rate": 0.0009872090048605783,
+      "loss": 1.8357,
+      "step": 20000
+    },
+    {
+      "epoch": 0.6555385008953697,
+      "grad_norm": 0.1083984375,
+      "learning_rate": 0.0009868892299820925,
+      "loss": 1.8322,
+      "step": 20500
+    },
+    {
+      "epoch": 0.671527244819647,
+      "grad_norm": 0.1513671875,
+      "learning_rate": 0.000986569455103607,
+      "loss": 1.8316,
+      "step": 21000
+    },
+    {
+      "epoch": 0.6875159887439243,
+      "grad_norm": 0.1015625,
+      "learning_rate": 0.0009862496802251215,
+      "loss": 1.8315,
+      "step": 21500
+    },
+    {
+      "epoch": 0.7035047326682016,
+      "grad_norm": 0.1396484375,
+      "learning_rate": 0.000985929905346636,
+      "loss": 1.8303,
+      "step": 22000
+    },
+    {
+      "epoch": 0.7194934765924789,
+      "grad_norm": 0.09814453125,
+      "learning_rate": 0.0009856101304681504,
+      "loss": 1.829,
+      "step": 22500
+    },
+    {
+      "epoch": 0.7354822205167562,
+      "grad_norm": 0.1259765625,
+      "learning_rate": 0.000985290355589665,
+      "loss": 1.8293,
+      "step": 23000
+    },
+    {
+      "epoch": 0.7514709644410336,
+      "grad_norm": 0.130859375,
+      "learning_rate": 0.0009849705807111794,
+      "loss": 1.8242,
+      "step": 23500
+    },
+    {
+      "epoch": 0.7674597083653109,
+      "grad_norm": 0.103515625,
+      "learning_rate": 0.0009846508058326939,
+      "loss": 1.8208,
+      "step": 24000
+    },
+    {
+      "epoch": 0.7834484522895881,
+      "grad_norm": 0.1123046875,
+      "learning_rate": 0.0009843310309542083,
+      "loss": 1.8202,
+      "step": 24500
+    },
+    {
+      "epoch": 0.7994371962138654,
+      "grad_norm": 0.11376953125,
+      "learning_rate": 0.0009840112560757228,
+      "loss": 1.8182,
+      "step": 25000
+    },
+    {
+      "epoch": 0.8154259401381427,
+      "grad_norm": 0.1103515625,
+      "learning_rate": 0.000983691481197237,
+      "loss": 1.8167,
+      "step": 25500
+    },
+    {
+      "epoch": 0.83141468406242,
+      "grad_norm": 0.10888671875,
+      "learning_rate": 0.0009833717063187516,
+      "loss": 1.8184,
+      "step": 26000
+    },
+    {
+      "epoch": 0.8474034279866973,
+      "grad_norm": 0.2431640625,
+      "learning_rate": 0.000983051931440266,
+      "loss": 1.8157,
+      "step": 26500
+    },
+    {
+      "epoch": 0.8633921719109747,
+      "grad_norm": 0.1376953125,
+      "learning_rate": 0.0009827321565617805,
+      "loss": 1.8114,
+      "step": 27000
+    },
+    {
+      "epoch": 0.879380915835252,
+      "grad_norm": 0.12890625,
+      "learning_rate": 0.000982412381683295,
+      "loss": 1.8111,
+      "step": 27500
+    },
+    {
+      "epoch": 0.8953696597595293,
+      "grad_norm": 0.11181640625,
+      "learning_rate": 0.0009820926068048095,
+      "loss": 1.8065,
+      "step": 28000
+    },
+    {
+      "epoch": 0.9113584036838066,
+      "grad_norm": 0.119140625,
+      "learning_rate": 0.000981772831926324,
+      "loss": 1.8085,
+      "step": 28500
+    },
+    {
+      "epoch": 0.9273471476080839,
+      "grad_norm": 0.130859375,
+      "learning_rate": 0.0009814530570478384,
+      "loss": 1.8091,
+      "step": 29000
+    },
+    {
+      "epoch": 0.9433358915323612,
+      "grad_norm": 0.138671875,
+      "learning_rate": 0.000981133282169353,
+      "loss": 1.8083,
+      "step": 29500
+    },
+    {
+      "epoch": 0.9593246354566385,
+      "grad_norm": 0.10888671875,
+      "learning_rate": 0.0009808135072908672,
+      "loss": 1.8044,
+      "step": 30000
+    },
+    {
+      "epoch": 0.9753133793809159,
+      "grad_norm": 0.1015625,
+      "learning_rate": 0.0009804937324123816,
+      "loss": 1.8004,
+      "step": 30500
+    },
+    {
+      "epoch": 0.9913021233051932,
+      "grad_norm": 0.14453125,
+      "learning_rate": 0.0009801739575338961,
+      "loss": 1.8007,
+      "step": 31000
+    },
     {
       "epoch": 1.0,
-      "eval_loss": 3.6347575187683105,
-      "eval_runtime": 0.3183,
-      "eval_samples_per_second": 1571.027,
-      "eval_steps_per_second": 3.142,
-      "step": 978
+      "eval_loss": 3.5783417224884033,
+      "eval_runtime": 3.2334,
+      "eval_samples_per_second": 154.634,
+      "eval_steps_per_second": 4.948,
+      "step": 31272
+    },
+    {
+      "epoch": 1.0072908672294705,
+      "grad_norm": 0.1298828125,
+      "learning_rate": 0.0009798541826554106,
+      "loss": 1.7974,
+      "step": 31500
+    },
+    {
+      "epoch": 1.0232796111537477,
+      "grad_norm": 0.1357421875,
+      "learning_rate": 0.000979534407776925,
+      "loss": 1.7944,
+      "step": 32000
+    },
+    {
+      "epoch": 1.039268355078025,
+      "grad_norm": 0.1708984375,
+      "learning_rate": 0.0009792146328984396,
+      "loss": 1.7956,
+      "step": 32500
+    },
+    {
+      "epoch": 1.0552570990023025,
+      "grad_norm": 0.1240234375,
+      "learning_rate": 0.000978894858019954,
+      "loss": 1.7942,
+      "step": 33000
+    },
+    {
+      "epoch": 1.0712458429265797,
+      "grad_norm": 0.1181640625,
+      "learning_rate": 0.0009785750831414685,
+      "loss": 1.7909,
+      "step": 33500
+    },
+    {
+      "epoch": 1.087234586850857,
+      "grad_norm": 0.11962890625,
+      "learning_rate": 0.000978255308262983,
+      "loss": 1.7884,
+      "step": 34000
+    },
+    {
+      "epoch": 1.1032233307751342,
+      "grad_norm": 0.125,
+      "learning_rate": 0.0009779355333844975,
+      "loss": 1.7888,
+      "step": 34500
+    },
+    {
+      "epoch": 1.1192120746994116,
+      "grad_norm": 0.1396484375,
+      "learning_rate": 0.0009776157585060117,
+      "loss": 1.7884,
+      "step": 35000
+    },
+    {
+      "epoch": 1.135200818623689,
+      "grad_norm": 0.126953125,
+      "learning_rate": 0.0009772959836275262,
+      "loss": 1.7873,
+      "step": 35500
+    },
+    {
+      "epoch": 1.1511895625479662,
+      "grad_norm": 0.1064453125,
+      "learning_rate": 0.0009769762087490407,
+      "loss": 1.7872,
+      "step": 36000
+    },
+    {
+      "epoch": 1.1671783064722436,
+      "grad_norm": 0.1494140625,
+      "learning_rate": 0.0009766564338705552,
+      "loss": 1.7826,
+      "step": 36500
+    },
+    {
+      "epoch": 1.1831670503965208,
+      "grad_norm": 0.138671875,
+      "learning_rate": 0.0009763366589920695,
+      "loss": 1.7838,
+      "step": 37000
+    },
+    {
+      "epoch": 1.1991557943207982,
+      "grad_norm": 0.1728515625,
+      "learning_rate": 0.000976016884113584,
+      "loss": 1.7781,
+      "step": 37500
+    },
+    {
+      "epoch": 1.2151445382450754,
+      "grad_norm": 0.212890625,
+      "learning_rate": 0.0009756971092350985,
+      "loss": 1.7798,
+      "step": 38000
+    },
+    {
+      "epoch": 1.2311332821693528,
+      "grad_norm": 0.12060546875,
+      "learning_rate": 0.000975377334356613,
+      "loss": 1.7782,
+      "step": 38500
+    },
+    {
+      "epoch": 1.24712202609363,
+      "grad_norm": 0.1572265625,
+      "learning_rate": 0.0009750575594781274,
+      "loss": 1.7797,
+      "step": 39000
+    },
+    {
+      "epoch": 1.2631107700179074,
+      "grad_norm": 0.1455078125,
+      "learning_rate": 0.0009747377845996419,
+      "loss": 1.7788,
+      "step": 39500
+    },
+    {
+      "epoch": 1.2790995139421848,
+      "grad_norm": 0.15234375,
+      "learning_rate": 0.0009744180097211564,
+      "loss": 1.778,
+      "step": 40000
+    },
+    {
+      "epoch": 1.295088257866462,
+      "grad_norm": 0.12890625,
+      "learning_rate": 0.0009740982348426708,
+      "loss": 1.7779,
+      "step": 40500
+    },
+    {
+      "epoch": 1.3110770017907394,
+      "grad_norm": 0.13671875,
+      "learning_rate": 0.0009737784599641852,
+      "loss": 1.7741,
+      "step": 41000
+    },
+    {
+      "epoch": 1.3270657457150166,
+      "grad_norm": 0.16796875,
+      "learning_rate": 0.0009734586850856997,
+      "loss": 1.7724,
+      "step": 41500
+    },
+    {
+      "epoch": 1.343054489639294,
+      "grad_norm": 0.13671875,
+      "learning_rate": 0.0009731389102072141,
+      "loss": 1.7749,
+      "step": 42000
+    },
+    {
+      "epoch": 1.3590432335635714,
+      "grad_norm": 0.1767578125,
+      "learning_rate": 0.0009728191353287286,
+      "loss": 1.7737,
+      "step": 42500
+    },
+    {
+      "epoch": 1.3750319774878486,
+      "grad_norm": 0.134765625,
+      "learning_rate": 0.000972499360450243,
+      "loss": 1.7686,
+      "step": 43000
+    },
+    {
+      "epoch": 1.3910207214121257,
+      "grad_norm": 0.1337890625,
+      "learning_rate": 0.0009721795855717574,
+      "loss": 1.7698,
+      "step": 43500
+    },
+    {
+      "epoch": 1.4070094653364031,
+      "grad_norm": 0.1201171875,
+      "learning_rate": 0.000971859810693272,
+      "loss": 1.7679,
+      "step": 44000
+    },
+    {
+      "epoch": 1.4229982092606805,
+      "grad_norm": 0.1416015625,
+      "learning_rate": 0.0009715400358147865,
+      "loss": 1.7667,
+      "step": 44500
+    },
+    {
+      "epoch": 1.4389869531849577,
+      "grad_norm": 0.16796875,
+      "learning_rate": 0.0009712202609363008,
+      "loss": 1.7688,
+      "step": 45000
+    },
+    {
+      "epoch": 1.4549756971092351,
+      "grad_norm": 0.1845703125,
+      "learning_rate": 0.0009709004860578153,
+      "loss": 1.7665,
+      "step": 45500
+    },
+    {
+      "epoch": 1.4709644410335123,
+      "grad_norm": 0.1455078125,
+      "learning_rate": 0.0009705807111793298,
+      "loss": 1.7676,
+      "step": 46000
+    },
+    {
+      "epoch": 1.4869531849577897,
+      "grad_norm": 0.1572265625,
+      "learning_rate": 0.0009702609363008442,
+      "loss": 1.7637,
+      "step": 46500
+    },
+    {
+      "epoch": 1.5029419288820671,
+      "grad_norm": 0.173828125,
+      "learning_rate": 0.0009699411614223587,
+      "loss": 1.7648,
+      "step": 47000
+    },
+    {
+      "epoch": 1.5189306728063443,
+      "grad_norm": 0.1650390625,
+      "learning_rate": 0.0009696213865438731,
+      "loss": 1.7639,
+      "step": 47500
+    },
+    {
+      "epoch": 1.5349194167306215,
+      "grad_norm": 0.2021484375,
+      "learning_rate": 0.0009693016116653875,
+      "loss": 1.7585,
+      "step": 48000
+    },
+    {
+      "epoch": 1.550908160654899,
+      "grad_norm": 0.138671875,
+      "learning_rate": 0.000968981836786902,
+      "loss": 1.7589,
+      "step": 48500
+    },
+    {
+      "epoch": 1.5668969045791763,
+      "grad_norm": 0.1650390625,
+      "learning_rate": 0.0009686620619084166,
+      "loss": 1.7552,
+      "step": 49000
+    },
+    {
+      "epoch": 1.5828856485034537,
+      "grad_norm": 0.1572265625,
+      "learning_rate": 0.000968342287029931,
+      "loss": 1.7584,
+      "step": 49500
+    },
+    {
+      "epoch": 1.5988743924277309,
+      "grad_norm": 0.1689453125,
+      "learning_rate": 0.0009680225121514454,
+      "loss": 1.7563,
+      "step": 50000
+    },
+    {
+      "epoch": 1.614863136352008,
+      "grad_norm": 0.1689453125,
+      "learning_rate": 0.0009677027372729599,
+      "loss": 1.7574,
+      "step": 50500
+    },
+    {
+      "epoch": 1.6308518802762855,
+      "grad_norm": 0.1904296875,
+      "learning_rate": 0.0009673829623944744,
+      "loss": 1.7579,
+      "step": 51000
+    },
+    {
+      "epoch": 1.6468406242005629,
+      "grad_norm": 0.1865234375,
+      "learning_rate": 0.0009670631875159887,
+      "loss": 1.7566,
+      "step": 51500
+    },
+    {
+      "epoch": 1.66282936812484,
+      "grad_norm": 0.197265625,
+      "learning_rate": 0.0009667434126375032,
+      "loss": 1.7545,
+      "step": 52000
+    },
+    {
+      "epoch": 1.6788181120491175,
+      "grad_norm": 0.126953125,
+      "learning_rate": 0.0009664236377590177,
+      "loss": 1.7498,
+      "step": 52500
+    },
+    {
+      "epoch": 1.6948068559733946,
+      "grad_norm": 0.19921875,
+      "learning_rate": 0.0009661038628805321,
+      "loss": 1.755,
+      "step": 53000
+    },
+    {
+      "epoch": 1.710795599897672,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.0009657840880020465,
+      "loss": 1.752,
+      "step": 53500
+    },
+    {
+      "epoch": 1.7267843438219495,
+      "grad_norm": 0.146484375,
+      "learning_rate": 0.000965464313123561,
+      "loss": 1.7525,
+      "step": 54000
+    },
+    {
+      "epoch": 1.7427730877462266,
+      "grad_norm": 0.166015625,
+      "learning_rate": 0.0009651445382450755,
+      "loss": 1.7506,
+      "step": 54500
+    },
+    {
+      "epoch": 1.7587618316705038,
+      "grad_norm": 0.1953125,
+      "learning_rate": 0.00096482476336659,
+      "loss": 1.7528,
+      "step": 55000
+    },
+    {
+      "epoch": 1.7747505755947812,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.0009645049884881044,
+      "loss": 1.7493,
+      "step": 55500
+    },
+    {
+      "epoch": 1.7907393195190586,
+      "grad_norm": 0.134765625,
+      "learning_rate": 0.0009641852136096188,
+      "loss": 1.7478,
+      "step": 56000
+    },
+    {
+      "epoch": 1.806728063443336,
+      "grad_norm": 0.22265625,
+      "learning_rate": 0.0009638654387311333,
+      "loss": 1.7522,
+      "step": 56500
+    },
+    {
+      "epoch": 1.8227168073676132,
+      "grad_norm": 0.2158203125,
+      "learning_rate": 0.0009635456638526478,
+      "loss": 1.7467,
+      "step": 57000
+    },
+    {
+      "epoch": 1.8387055512918904,
+      "grad_norm": 0.1865234375,
+      "learning_rate": 0.0009632258889741621,
+      "loss": 1.748,
+      "step": 57500
+    },
+    {
+      "epoch": 1.8546942952161678,
+      "grad_norm": 0.2392578125,
+      "learning_rate": 0.0009629061140956766,
+      "loss": 1.7473,
+      "step": 58000
+    },
+    {
+      "epoch": 1.8706830391404452,
+      "grad_norm": 0.1904296875,
+      "learning_rate": 0.0009625863392171911,
+      "loss": 1.7478,
+      "step": 58500
+    },
+    {
+      "epoch": 1.8866717830647224,
+      "grad_norm": 0.18359375,
+      "learning_rate": 0.0009622665643387055,
+      "loss": 1.7446,
+      "step": 59000
+    },
+    {
+      "epoch": 1.9026605269889998,
+      "grad_norm": 0.1669921875,
+      "learning_rate": 0.00096194678946022,
+      "loss": 1.7483,
+      "step": 59500
+    },
+    {
+      "epoch": 1.918649270913277,
+      "grad_norm": 0.166015625,
+      "learning_rate": 0.0009616270145817345,
+      "loss": 1.7482,
+      "step": 60000
+    },
+    {
+      "epoch": 1.9346380148375544,
+      "grad_norm": 0.2109375,
+      "learning_rate": 0.000961307239703249,
+      "loss": 1.7448,
+      "step": 60500
+    },
+    {
+      "epoch": 1.9506267587618318,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.0009609874648247634,
+      "loss": 1.7418,
+      "step": 61000
+    },
+    {
+      "epoch": 1.966615502686109,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.0009606676899462779,
+      "loss": 1.7456,
+      "step": 61500
+    },
+    {
+      "epoch": 1.9826042466103861,
+      "grad_norm": 0.2216796875,
+      "learning_rate": 0.0009603479150677923,
+      "loss": 1.7471,
+      "step": 62000
+    },
+    {
+      "epoch": 1.9985929905346635,
+      "grad_norm": 0.203125,
+      "learning_rate": 0.0009600281401893067,
+      "loss": 1.7422,
+      "step": 62500
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 3.4795773029327393,
+      "eval_runtime": 3.1281,
+      "eval_samples_per_second": 159.84,
+      "eval_steps_per_second": 5.115,
+      "step": 62544
+    },
+    {
+      "epoch": 2.014581734458941,
+      "grad_norm": 0.1591796875,
+      "learning_rate": 0.0009597083653108212,
+      "loss": 1.738,
+      "step": 63000
+    },
+    {
+      "epoch": 2.0305704783832184,
+      "grad_norm": 0.1845703125,
+      "learning_rate": 0.0009593885904323357,
+      "loss": 1.7378,
+      "step": 63500
+    },
+    {
+      "epoch": 2.0465592223074953,
+      "grad_norm": 0.1806640625,
+      "learning_rate": 0.00095906881555385,
+      "loss": 1.7379,
+      "step": 64000
+    },
+    {
+      "epoch": 2.0625479662317727,
+      "grad_norm": 0.1767578125,
+      "learning_rate": 0.0009587490406753646,
+      "loss": 1.7346,
+      "step": 64500
+    },
+    {
+      "epoch": 2.07853671015605,
+      "grad_norm": 0.13671875,
+      "learning_rate": 0.0009584292657968791,
+      "loss": 1.7322,
+      "step": 65000
+    },
+    {
+      "epoch": 2.0945254540803275,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.0009581094909183935,
+      "loss": 1.7344,
+      "step": 65500
+    },
+    {
+      "epoch": 2.110514198004605,
+      "grad_norm": 0.1748046875,
+      "learning_rate": 0.0009577897160399079,
+      "loss": 1.7338,
+      "step": 66000
+    },
+    {
+      "epoch": 2.126502941928882,
+      "grad_norm": 0.18359375,
+      "learning_rate": 0.0009574699411614224,
+      "loss": 1.7377,
+      "step": 66500
+    },
+    {
+      "epoch": 2.1424916858531593,
+      "grad_norm": 0.1640625,
+      "learning_rate": 0.0009571501662829368,
+      "loss": 1.7388,
+      "step": 67000
+    },
+    {
+      "epoch": 2.1584804297774367,
+      "grad_norm": 0.1787109375,
+      "learning_rate": 0.0009568303914044513,
+      "loss": 1.74,
+      "step": 67500
+    },
+    {
+      "epoch": 2.174469173701714,
+      "grad_norm": 0.1640625,
+      "learning_rate": 0.0009565106165259657,
+      "loss": 1.7374,
+      "step": 68000
+    },
+    {
+      "epoch": 2.1904579176259915,
+      "grad_norm": 0.2431640625,
+      "learning_rate": 0.0009561908416474801,
+      "loss": 1.7339,
+      "step": 68500
+    },
+    {
+      "epoch": 2.2064466615502685,
+      "grad_norm": 0.205078125,
+      "learning_rate": 0.0009558710667689946,
+      "loss": 1.7372,
+      "step": 69000
+    },
+    {
+      "epoch": 2.222435405474546,
+      "grad_norm": 0.236328125,
+      "learning_rate": 0.0009555512918905092,
+      "loss": 1.7345,
+      "step": 69500
+    },
+    {
+      "epoch": 2.2384241493988233,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.0009552315170120236,
+      "loss": 1.7305,
+      "step": 70000
+    },
+    {
+      "epoch": 2.2544128933231007,
+      "grad_norm": 0.2216796875,
+      "learning_rate": 0.000954911742133538,
+      "loss": 1.736,
+      "step": 70500
+    },
+    {
+      "epoch": 2.270401637247378,
+      "grad_norm": 0.16796875,
+      "learning_rate": 0.0009545919672550525,
+      "loss": 1.7309,
+      "step": 71000
+    },
+    {
+      "epoch": 2.286390381171655,
+      "grad_norm": 0.1337890625,
+      "learning_rate": 0.000954272192376567,
+      "loss": 1.7349,
+      "step": 71500
+    },
+    {
+      "epoch": 2.3023791250959325,
+      "grad_norm": 0.177734375,
+      "learning_rate": 0.0009539524174980813,
+      "loss": 1.7295,
+      "step": 72000
+    },
+    {
+      "epoch": 2.31836786902021,
+      "grad_norm": 0.193359375,
+      "learning_rate": 0.0009536326426195958,
+      "loss": 1.7335,
+      "step": 72500
+    },
+    {
+      "epoch": 2.3343566129444873,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.0009533128677411103,
+      "loss": 1.7295,
+      "step": 73000
+    },
+    {
+      "epoch": 2.3503453568687642,
+      "grad_norm": 0.17578125,
+      "learning_rate": 0.0009529930928626247,
+      "loss": 1.7299,
+      "step": 73500
+    },
+    {
+      "epoch": 2.3663341007930416,
+      "grad_norm": 0.158203125,
+      "learning_rate": 0.0009526733179841391,
+      "loss": 1.7264,
+      "step": 74000
+    },
+    {
+      "epoch": 2.382322844717319,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.0009523535431056536,
+      "loss": 1.7264,
+      "step": 74500
+    },
+    {
+      "epoch": 2.3983115886415964,
+      "grad_norm": 0.1982421875,
+      "learning_rate": 0.0009520337682271681,
+      "loss": 1.7355,
+      "step": 75000
+    },
+    {
+      "epoch": 2.4143003325658734,
+      "grad_norm": 0.1943359375,
+      "learning_rate": 0.0009517139933486826,
+      "loss": 1.7242,
+      "step": 75500
+    },
+    {
+      "epoch": 2.430289076490151,
+      "grad_norm": 0.2080078125,
+      "learning_rate": 0.0009513942184701971,
+      "loss": 1.7274,
+      "step": 76000
+    },
+    {
+      "epoch": 2.446277820414428,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.0009510744435917114,
+      "loss": 1.7257,
+      "step": 76500
+    },
+    {
+      "epoch": 2.4622665643387056,
+      "grad_norm": 0.197265625,
+      "learning_rate": 0.0009507546687132259,
+      "loss": 1.7271,
+      "step": 77000
+    },
+    {
+      "epoch": 2.478255308262983,
+      "grad_norm": 0.1875,
+      "learning_rate": 0.0009504348938347404,
+      "loss": 1.7292,
+      "step": 77500
+    },
+    {
+      "epoch": 2.49424405218726,
+      "grad_norm": 0.2177734375,
+      "learning_rate": 0.0009501151189562548,
+      "loss": 1.7306,
+      "step": 78000
+    },
+    {
+      "epoch": 2.5102327961115374,
+      "grad_norm": 0.18359375,
+      "learning_rate": 0.0009497953440777692,
+      "loss": 1.73,
+      "step": 78500
+    },
+    {
+      "epoch": 2.526221540035815,
+      "grad_norm": 0.171875,
+      "learning_rate": 0.0009494755691992837,
+      "loss": 1.7257,
+      "step": 79000
+    },
+    {
+      "epoch": 2.542210283960092,
+      "grad_norm": 0.236328125,
+      "learning_rate": 0.0009491557943207981,
+      "loss": 1.728,
+      "step": 79500
+    },
+    {
+      "epoch": 2.5581990278843696,
+      "grad_norm": 0.1572265625,
+      "learning_rate": 0.0009488360194423127,
+      "loss": 1.7326,
+      "step": 80000
+    },
+    {
+      "epoch": 2.5741877718086466,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.0009485162445638271,
+      "loss": 1.725,
+      "step": 80500
+    },
+    {
+      "epoch": 2.590176515732924,
+      "grad_norm": 0.30078125,
+      "learning_rate": 0.0009481964696853416,
+      "loss": 1.7211,
+      "step": 81000
+    },
+    {
+      "epoch": 2.6061652596572014,
+      "grad_norm": 0.1728515625,
+      "learning_rate": 0.000947876694806856,
+      "loss": 1.7236,
+      "step": 81500
+    },
+    {
+      "epoch": 2.6221540035814788,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.0009475569199283705,
+      "loss": 1.7244,
+      "step": 82000
+    },
+    {
+      "epoch": 2.638142747505756,
+      "grad_norm": 0.1845703125,
+      "learning_rate": 0.0009472371450498849,
+      "loss": 1.7251,
+      "step": 82500
+    },
+    {
+      "epoch": 2.654131491430033,
+      "grad_norm": 0.20703125,
+      "learning_rate": 0.0009469173701713993,
+      "loss": 1.7223,
+      "step": 83000
+    },
+    {
+      "epoch": 2.6701202353543105,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.0009465975952929138,
+      "loss": 1.7214,
+      "step": 83500
+    },
+    {
+      "epoch": 2.686108979278588,
+      "grad_norm": 0.2431640625,
+      "learning_rate": 0.0009462778204144283,
+      "loss": 1.7228,
+      "step": 84000
+    },
+    {
+      "epoch": 2.7020977232028653,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0009459580455359426,
+      "loss": 1.7225,
+      "step": 84500
+    },
+    {
+      "epoch": 2.7180864671271427,
+      "grad_norm": 0.208984375,
+      "learning_rate": 0.0009456382706574572,
+      "loss": 1.7228,
+      "step": 85000
+    },
+    {
+      "epoch": 2.7340752110514197,
+      "grad_norm": 0.205078125,
+      "learning_rate": 0.0009453184957789717,
+      "loss": 1.7211,
+      "step": 85500
+    },
+    {
+      "epoch": 2.750063954975697,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0009449987209004861,
+      "loss": 1.7267,
+      "step": 86000
+    },
+    {
+      "epoch": 2.7660526988999745,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.0009446789460220005,
+      "loss": 1.7248,
+      "step": 86500
+    },
+    {
+      "epoch": 2.7820414428242515,
+      "grad_norm": 0.1611328125,
+      "learning_rate": 0.000944359171143515,
+      "loss": 1.725,
+      "step": 87000
+    },
+    {
+      "epoch": 2.798030186748529,
+      "grad_norm": 0.2041015625,
+      "learning_rate": 0.0009440393962650294,
+      "loss": 1.7281,
+      "step": 87500
+    },
+    {
+      "epoch": 2.8140189306728063,
+      "grad_norm": 0.2080078125,
+      "learning_rate": 0.0009437196213865439,
+      "loss": 1.7134,
+      "step": 88000
+    },
+    {
+      "epoch": 2.8300076745970837,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.0009433998465080583,
+      "loss": 1.7201,
+      "step": 88500
+    },
+    {
+      "epoch": 2.845996418521361,
+      "grad_norm": 0.2236328125,
+      "learning_rate": 0.0009430800716295727,
+      "loss": 1.7189,
+      "step": 89000
+    },
+    {
+      "epoch": 2.861985162445638,
+      "grad_norm": 0.34375,
+      "learning_rate": 0.0009427602967510872,
+      "loss": 1.7205,
+      "step": 89500
+    },
+    {
+      "epoch": 2.8779739063699155,
+      "grad_norm": 0.2158203125,
+      "learning_rate": 0.0009424405218726017,
+      "loss": 1.7175,
+      "step": 90000
+    },
+    {
+      "epoch": 2.893962650294193,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.0009421207469941163,
+      "loss": 1.7161,
+      "step": 90500
+    },
+    {
+      "epoch": 2.9099513942184703,
+      "grad_norm": 0.236328125,
+      "learning_rate": 0.0009418009721156306,
+      "loss": 1.7143,
+      "step": 91000
+    },
+    {
+      "epoch": 2.9259401381427477,
+      "grad_norm": 0.193359375,
+      "learning_rate": 0.0009414811972371451,
+      "loss": 1.7165,
+      "step": 91500
+    },
+    {
+      "epoch": 2.9419288820670246,
+      "grad_norm": 0.169921875,
+      "learning_rate": 0.0009411614223586596,
+      "loss": 1.7166,
+      "step": 92000
+    },
+    {
+      "epoch": 2.957917625991302,
+      "grad_norm": 0.2216796875,
+      "learning_rate": 0.000940841647480174,
+      "loss": 1.7178,
+      "step": 92500
+    },
+    {
+      "epoch": 2.9739063699155794,
+      "grad_norm": 0.1806640625,
+      "learning_rate": 0.0009405218726016884,
+      "loss": 1.7195,
+      "step": 93000
+    },
+    {
+      "epoch": 2.989895113839857,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.0009402020977232029,
+      "loss": 1.7231,
+      "step": 93500
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 3.432668685913086,
+      "eval_runtime": 2.8257,
+      "eval_samples_per_second": 176.95,
+      "eval_steps_per_second": 5.662,
+      "step": 93816
+    },
+    {
+      "epoch": 3.0058838577641342,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.0009398823228447173,
+      "loss": 1.716,
+      "step": 94000
+    },
+    {
+      "epoch": 3.021872601688411,
+      "grad_norm": 0.197265625,
+      "learning_rate": 0.0009395625479662318,
+      "loss": 1.7147,
+      "step": 94500
+    },
+    {
+      "epoch": 3.0378613456126886,
+      "grad_norm": 0.1689453125,
+      "learning_rate": 0.0009392427730877462,
+      "loss": 1.7154,
+      "step": 95000
+    },
+    {
+      "epoch": 3.053850089536966,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.0009389229982092607,
+      "loss": 1.7139,
+      "step": 95500
+    },
+    {
+      "epoch": 3.0698388334612434,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0009386032233307752,
+      "loss": 1.7133,
+      "step": 96000
+    },
+    {
+      "epoch": 3.0858275773855204,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.0009382834484522897,
+      "loss": 1.7221,
+      "step": 96500
+    },
+    {
+      "epoch": 3.101816321309798,
+      "grad_norm": 0.1884765625,
+      "learning_rate": 0.000937963673573804,
+      "loss": 1.7148,
+      "step": 97000
+    },
+    {
+      "epoch": 3.117805065234075,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.0009376438986953185,
+      "loss": 1.7106,
+      "step": 97500
+    },
+    {
+      "epoch": 3.1337938091583526,
+      "grad_norm": 0.20703125,
+      "learning_rate": 0.000937324123816833,
+      "loss": 1.7127,
+      "step": 98000
+    },
+    {
+      "epoch": 3.14978255308263,
+      "grad_norm": 0.216796875,
+      "learning_rate": 0.0009370043489383474,
+      "loss": 1.7107,
+      "step": 98500
+    },
+    {
+      "epoch": 3.165771297006907,
+      "grad_norm": 0.2314453125,
+      "learning_rate": 0.0009366845740598618,
+      "loss": 1.7121,
+      "step": 99000
+    },
+    {
+      "epoch": 3.1817600409311844,
+      "grad_norm": 0.193359375,
+      "learning_rate": 0.0009363647991813763,
+      "loss": 1.7148,
+      "step": 99500
+    },
+    {
+      "epoch": 3.1977487848554618,
+      "grad_norm": 0.2333984375,
+      "learning_rate": 0.0009360450243028907,
+      "loss": 1.7125,
+      "step": 100000
+    },
+    {
+      "epoch": 3.213737528779739,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.0009357252494244053,
+      "loss": 1.7157,
+      "step": 100500
+    },
+    {
+      "epoch": 3.2297262727040166,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.0009354054745459198,
+      "loss": 1.7121,
+      "step": 101000
+    },
+    {
+      "epoch": 3.2457150166282935,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.0009350856996674342,
+      "loss": 1.7096,
+      "step": 101500
+    },
+    {
+      "epoch": 3.261703760552571,
+      "grad_norm": 0.203125,
+      "learning_rate": 0.0009347659247889486,
+      "loss": 1.7097,
+      "step": 102000
+    },
+    {
+      "epoch": 3.2776925044768483,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.0009344461499104631,
+      "loss": 1.7169,
+      "step": 102500
+    },
+    {
+      "epoch": 3.2936812484011257,
+      "grad_norm": 0.201171875,
+      "learning_rate": 0.0009341263750319776,
+      "loss": 1.7145,
+      "step": 103000
+    },
+    {
+      "epoch": 3.3096699923254027,
+      "grad_norm": 0.158203125,
+      "learning_rate": 0.0009338066001534919,
+      "loss": 1.71,
+      "step": 103500
+    },
+    {
+      "epoch": 3.32565873624968,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.0009334868252750064,
+      "loss": 1.7087,
+      "step": 104000
+    },
+    {
+      "epoch": 3.3416474801739575,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.0009331670503965209,
+      "loss": 1.7159,
+      "step": 104500
+    },
+    {
+      "epoch": 3.357636224098235,
+      "grad_norm": 0.1904296875,
+      "learning_rate": 0.0009328472755180352,
+      "loss": 1.711,
+      "step": 105000
+    },
+    {
+      "epoch": 3.3736249680225123,
+      "grad_norm": 0.19921875,
+      "learning_rate": 0.0009325275006395498,
+      "loss": 1.7137,
+      "step": 105500
+    },
+    {
+      "epoch": 3.3896137119467893,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0009322077257610643,
+      "loss": 1.7118,
+      "step": 106000
+    },
+    {
+      "epoch": 3.4056024558710667,
+      "grad_norm": 0.1748046875,
+      "learning_rate": 0.0009318879508825787,
+      "loss": 1.7159,
+      "step": 106500
+    },
+    {
+      "epoch": 3.421591199795344,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0009315681760040932,
+      "loss": 1.7131,
+      "step": 107000
+    },
+    {
+      "epoch": 3.4375799437196215,
+      "grad_norm": 0.1884765625,
+      "learning_rate": 0.0009312484011256076,
+      "loss": 1.7108,
+      "step": 107500
+    },
+    {
+      "epoch": 3.453568687643899,
+      "grad_norm": 0.16796875,
+      "learning_rate": 0.000930928626247122,
+      "loss": 1.7101,
+      "step": 108000
+    },
+    {
+      "epoch": 3.469557431568176,
+      "grad_norm": 0.1962890625,
+      "learning_rate": 0.0009306088513686365,
+      "loss": 1.7096,
+      "step": 108500
+    },
+    {
+      "epoch": 3.4855461754924533,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.000930289076490151,
+      "loss": 1.7134,
+      "step": 109000
+    },
+    {
+      "epoch": 3.5015349194167307,
+      "grad_norm": 0.2236328125,
+      "learning_rate": 0.0009299693016116653,
+      "loss": 1.7105,
+      "step": 109500
+    },
+    {
+      "epoch": 3.517523663341008,
+      "grad_norm": 0.19140625,
+      "learning_rate": 0.0009296495267331798,
+      "loss": 1.7102,
+      "step": 110000
+    },
+    {
+      "epoch": 3.5335124072652855,
+      "grad_norm": 0.1845703125,
+      "learning_rate": 0.0009293297518546943,
+      "loss": 1.7067,
+      "step": 110500
+    },
+    {
+      "epoch": 3.5495011511895624,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0009290099769762088,
+      "loss": 1.7099,
+      "step": 111000
+    },
+    {
+      "epoch": 3.56548989511384,
+      "grad_norm": 0.2314453125,
+      "learning_rate": 0.0009286902020977232,
+      "loss": 1.7154,
+      "step": 111500
+    },
+    {
+      "epoch": 3.5814786390381173,
+      "grad_norm": 0.212890625,
+      "learning_rate": 0.0009283704272192377,
+      "loss": 1.7074,
+      "step": 112000
+    },
+    {
+      "epoch": 3.597467382962394,
+      "grad_norm": 0.140625,
+      "learning_rate": 0.0009280506523407522,
+      "loss": 1.7074,
+      "step": 112500
+    },
+    {
+      "epoch": 3.613456126886672,
+      "grad_norm": 0.2236328125,
+      "learning_rate": 0.0009277308774622666,
+      "loss": 1.7097,
+      "step": 113000
+    },
+    {
+      "epoch": 3.629444870810949,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.000927411102583781,
+      "loss": 1.7141,
+      "step": 113500
+    },
+    {
+      "epoch": 3.6454336147352264,
+      "grad_norm": 0.2177734375,
+      "learning_rate": 0.0009270913277052955,
+      "loss": 1.7029,
+      "step": 114000
+    },
+    {
+      "epoch": 3.661422358659504,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.0009267715528268099,
+      "loss": 1.7105,
+      "step": 114500
+    },
+    {
+      "epoch": 3.677411102583781,
+      "grad_norm": 0.205078125,
+      "learning_rate": 0.0009264517779483244,
+      "loss": 1.7101,
+      "step": 115000
+    },
+    {
+      "epoch": 3.693399846508058,
+      "grad_norm": 0.158203125,
+      "learning_rate": 0.0009261320030698388,
+      "loss": 1.7088,
+      "step": 115500
+    },
+    {
+      "epoch": 3.7093885904323356,
+      "grad_norm": 0.1923828125,
+      "learning_rate": 0.0009258122281913533,
+      "loss": 1.711,
+      "step": 116000
+    },
+    {
+      "epoch": 3.725377334356613,
+      "grad_norm": 0.2041015625,
+      "learning_rate": 0.0009254924533128678,
+      "loss": 1.7119,
+      "step": 116500
+    },
+    {
+      "epoch": 3.7413660782808904,
+      "grad_norm": 0.224609375,
+      "learning_rate": 0.0009251726784343823,
+      "loss": 1.7079,
+      "step": 117000
+    },
+    {
+      "epoch": 3.7573548222051674,
+      "grad_norm": 0.1875,
+      "learning_rate": 0.0009248529035558966,
+      "loss": 1.7072,
+      "step": 117500
+    },
+    {
+      "epoch": 3.7733435661294448,
+      "grad_norm": 0.169921875,
+      "learning_rate": 0.0009245331286774111,
+      "loss": 1.7062,
+      "step": 118000
+    },
+    {
+      "epoch": 3.789332310053722,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0009242133537989256,
+      "loss": 1.7027,
+      "step": 118500
+    },
+    {
+      "epoch": 3.8053210539779996,
+      "grad_norm": 0.2001953125,
+      "learning_rate": 0.00092389357892044,
+      "loss": 1.7069,
+      "step": 119000
+    },
+    {
+      "epoch": 3.821309797902277,
+      "grad_norm": 0.330078125,
+      "learning_rate": 0.0009235738040419545,
+      "loss": 1.7102,
+      "step": 119500
+    },
+    {
+      "epoch": 3.837298541826554,
+      "grad_norm": 0.205078125,
+      "learning_rate": 0.0009232540291634689,
+      "loss": 1.7099,
+      "step": 120000
+    },
+    {
+      "epoch": 3.8532872857508313,
+      "grad_norm": 0.328125,
+      "learning_rate": 0.0009229342542849833,
+      "loss": 1.7024,
+      "step": 120500
+    },
+    {
+      "epoch": 3.8692760296751088,
+      "grad_norm": 0.23046875,
+      "learning_rate": 0.0009226144794064979,
+      "loss": 1.711,
+      "step": 121000
+    },
+    {
+      "epoch": 3.885264773599386,
+      "grad_norm": 0.181640625,
+      "learning_rate": 0.0009222947045280124,
+      "loss": 1.7046,
+      "step": 121500
+    },
+    {
+      "epoch": 3.9012535175236636,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.0009219749296495267,
+      "loss": 1.7062,
+      "step": 122000
+    },
+    {
+      "epoch": 3.9172422614479405,
+      "grad_norm": 0.2021484375,
+      "learning_rate": 0.0009216551547710412,
+      "loss": 1.7027,
+      "step": 122500
+    },
+    {
+      "epoch": 3.933231005372218,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.0009213353798925557,
+      "loss": 1.7027,
+      "step": 123000
+    },
+    {
+      "epoch": 3.9492197492964953,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.0009210156050140702,
+      "loss": 1.7021,
+      "step": 123500
+    },
+    {
+      "epoch": 3.9652084932207727,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.0009206958301355845,
+      "loss": 1.705,
+      "step": 124000
+    },
+    {
+      "epoch": 3.98119723714505,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.000920376055257099,
+      "loss": 1.7054,
+      "step": 124500
+    },
+    {
+      "epoch": 3.997185981069327,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.0009200562803786135,
+      "loss": 1.7036,
+      "step": 125000
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 3.41278076171875,
+      "eval_runtime": 3.0365,
+      "eval_samples_per_second": 164.665,
+      "eval_steps_per_second": 5.269,
+      "step": 125088
+    },
+    {
+      "epoch": 4.013174724993604,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0009197365055001279,
+      "loss": 1.7013,
+      "step": 125500
+    },
+    {
+      "epoch": 4.029163468917882,
+      "grad_norm": 0.197265625,
+      "learning_rate": 0.0009194167306216423,
+      "loss": 1.7043,
+      "step": 126000
+    },
+    {
+      "epoch": 4.045152212842159,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.0009190969557431569,
+      "loss": 1.7058,
+      "step": 126500
+    },
+    {
+      "epoch": 4.061140956766437,
+      "grad_norm": 0.19140625,
+      "learning_rate": 0.0009187771808646713,
+      "loss": 1.6991,
+      "step": 127000
+    },
+    {
+      "epoch": 4.077129700690714,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.0009184574059861858,
+      "loss": 1.703,
+      "step": 127500
+    },
+    {
+      "epoch": 4.093118444614991,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.0009181376311077002,
+      "loss": 1.7009,
+      "step": 128000
+    },
+    {
+      "epoch": 4.1091071885392685,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.0009178178562292146,
+      "loss": 1.7015,
+      "step": 128500
+    },
+    {
+      "epoch": 4.1250959324635454,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.0009174980813507291,
+      "loss": 1.7045,
+      "step": 129000
+    },
+    {
+      "epoch": 4.141084676387823,
+      "grad_norm": 0.1728515625,
+      "learning_rate": 0.0009171783064722436,
+      "loss": 1.7042,
+      "step": 129500
+    },
+    {
+      "epoch": 4.1570734203121,
+      "grad_norm": 0.1962890625,
+      "learning_rate": 0.0009168585315937579,
+      "loss": 1.7026,
+      "step": 130000
+    },
+    {
+      "epoch": 4.173062164236377,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.0009165387567152724,
+      "loss": 1.7042,
+      "step": 130500
+    },
+    {
+      "epoch": 4.189050908160655,
+      "grad_norm": 0.216796875,
+      "learning_rate": 0.0009162189818367869,
+      "loss": 1.702,
+      "step": 131000
+    },
+    {
+      "epoch": 4.205039652084932,
+      "grad_norm": 0.30078125,
+      "learning_rate": 0.0009158992069583014,
+      "loss": 1.7027,
+      "step": 131500
+    },
+    {
+      "epoch": 4.22102839600921,
+      "grad_norm": 0.2041015625,
+      "learning_rate": 0.0009155794320798159,
+      "loss": 1.7077,
+      "step": 132000
+    },
+    {
+      "epoch": 4.237017139933487,
+      "grad_norm": 0.2216796875,
+      "learning_rate": 0.0009152596572013303,
+      "loss": 1.7037,
+      "step": 132500
+    },
+    {
+      "epoch": 4.253005883857764,
+      "grad_norm": 0.345703125,
+      "learning_rate": 0.0009149398823228447,
+      "loss": 1.7015,
+      "step": 133000
+    },
+    {
+      "epoch": 4.268994627782042,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.0009146201074443592,
+      "loss": 1.7062,
+      "step": 133500
+    },
+    {
+      "epoch": 4.284983371706319,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0009143003325658737,
+      "loss": 1.7033,
+      "step": 134000
+    },
+    {
+      "epoch": 4.3009721156305964,
+      "grad_norm": 0.2119140625,
+      "learning_rate": 0.0009139805576873881,
+      "loss": 1.7051,
+      "step": 134500
+    },
+    {
+      "epoch": 4.316960859554873,
+      "grad_norm": 0.189453125,
+      "learning_rate": 0.0009136607828089025,
+      "loss": 1.7015,
+      "step": 135000
+    },
+    {
+      "epoch": 4.33294960347915,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.000913341007930417,
+      "loss": 1.7036,
+      "step": 135500
+    },
+    {
+      "epoch": 4.348938347403428,
+      "grad_norm": 0.2109375,
+      "learning_rate": 0.0009130212330519315,
+      "loss": 1.7003,
+      "step": 136000
+    },
+    {
+      "epoch": 4.364927091327705,
+      "grad_norm": 0.298828125,
+      "learning_rate": 0.0009127014581734459,
+      "loss": 1.6993,
+      "step": 136500
+    },
+    {
+      "epoch": 4.380915835251983,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.0009123816832949604,
+      "loss": 1.6958,
+      "step": 137000
+    },
+    {
+      "epoch": 4.39690457917626,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.0009120619084164749,
+      "loss": 1.7035,
+      "step": 137500
+    },
+    {
+      "epoch": 4.412893323100537,
+      "grad_norm": 0.197265625,
+      "learning_rate": 0.0009117421335379893,
+      "loss": 1.7053,
+      "step": 138000
+    },
+    {
+      "epoch": 4.428882067024815,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0009114223586595037,
+      "loss": 1.7002,
+      "step": 138500
+    },
+    {
+      "epoch": 4.444870810949092,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.0009111025837810182,
+      "loss": 1.7047,
+      "step": 139000
+    },
+    {
+      "epoch": 4.46085955487337,
+      "grad_norm": 0.2177734375,
+      "learning_rate": 0.0009107828089025326,
+      "loss": 1.7039,
+      "step": 139500
+    },
+    {
+      "epoch": 4.476848298797647,
+      "grad_norm": 0.298828125,
+      "learning_rate": 0.0009104630340240471,
+      "loss": 1.7002,
+      "step": 140000
+    },
+    {
+      "epoch": 4.4928370427219235,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0009101432591455615,
+      "loss": 1.7044,
+      "step": 140500
+    },
+    {
+      "epoch": 4.508825786646201,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.0009098234842670759,
+      "loss": 1.6948,
+      "step": 141000
+    },
+    {
+      "epoch": 4.524814530570478,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.0009095037093885905,
+      "loss": 1.7069,
+      "step": 141500
+    },
+    {
+      "epoch": 4.540803274494756,
+      "grad_norm": 0.326171875,
+      "learning_rate": 0.000909183934510105,
+      "loss": 1.6986,
+      "step": 142000
+    },
+    {
+      "epoch": 4.556792018419033,
+      "grad_norm": 0.25,
+      "learning_rate": 0.0009088641596316193,
+      "loss": 1.7004,
+      "step": 142500
+    },
+    {
+      "epoch": 4.57278076234331,
+      "grad_norm": 0.2734375,
+      "learning_rate": 0.0009085443847531338,
+      "loss": 1.6997,
+      "step": 143000
+    },
+    {
+      "epoch": 4.588769506267588,
+      "grad_norm": 0.2109375,
+      "learning_rate": 0.0009082246098746483,
+      "loss": 1.7034,
+      "step": 143500
+    },
+    {
+      "epoch": 4.604758250191865,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0009079048349961627,
+      "loss": 1.6996,
+      "step": 144000
+    },
+    {
+      "epoch": 4.620746994116142,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0009075850601176771,
+      "loss": 1.7026,
+      "step": 144500
+    },
+    {
+      "epoch": 4.63673573804042,
+      "grad_norm": 0.203125,
+      "learning_rate": 0.0009072652852391916,
+      "loss": 1.7015,
+      "step": 145000
+    },
+    {
+      "epoch": 4.652724481964697,
+      "grad_norm": 0.1982421875,
+      "learning_rate": 0.0009069455103607061,
+      "loss": 1.7014,
+      "step": 145500
+    },
+    {
+      "epoch": 4.6687132258889745,
+      "grad_norm": 0.2236328125,
+      "learning_rate": 0.0009066257354822205,
+      "loss": 1.698,
+      "step": 146000
+    },
+    {
+      "epoch": 4.6847019698132515,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.000906305960603735,
+      "loss": 1.7031,
+      "step": 146500
+    },
+    {
+      "epoch": 4.7006907137375284,
+      "grad_norm": 0.380859375,
+      "learning_rate": 0.0009059861857252495,
+      "loss": 1.6971,
+      "step": 147000
+    },
+    {
+      "epoch": 4.716679457661806,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0009056664108467639,
+      "loss": 1.6995,
+      "step": 147500
+    },
+    {
+      "epoch": 4.732668201586083,
+      "grad_norm": 0.2470703125,
+      "learning_rate": 0.0009053466359682784,
+      "loss": 1.6955,
+      "step": 148000
+    },
+    {
+      "epoch": 4.74865694551036,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.0009050268610897929,
+      "loss": 1.699,
+      "step": 148500
+    },
+    {
+      "epoch": 4.764645689434638,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0009047070862113072,
+      "loss": 1.7024,
+      "step": 149000
+    },
+    {
+      "epoch": 4.780634433358915,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.0009043873113328217,
+      "loss": 1.6941,
+      "step": 149500
+    },
+    {
+      "epoch": 4.796623177283193,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0009040675364543362,
+      "loss": 1.6972,
+      "step": 150000
+    },
+    {
+      "epoch": 4.81261192120747,
+      "grad_norm": 0.22265625,
+      "learning_rate": 0.0009037477615758506,
+      "loss": 1.7038,
+      "step": 150500
+    },
+    {
+      "epoch": 4.828600665131747,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.000903427986697365,
+      "loss": 1.7035,
+      "step": 151000
+    },
+    {
+      "epoch": 4.844589409056025,
+      "grad_norm": 0.1923828125,
+      "learning_rate": 0.0009031082118188795,
+      "loss": 1.7016,
+      "step": 151500
+    },
+    {
+      "epoch": 4.860578152980302,
+      "grad_norm": 0.185546875,
+      "learning_rate": 0.000902788436940394,
+      "loss": 1.7024,
+      "step": 152000
+    },
+    {
+      "epoch": 4.8765668969045795,
+      "grad_norm": 0.21875,
+      "learning_rate": 0.0009024686620619085,
+      "loss": 1.698,
+      "step": 152500
+    },
+    {
+      "epoch": 4.892555640828856,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.0009021488871834229,
+      "loss": 1.7005,
+      "step": 153000
+    },
+    {
+      "epoch": 4.908544384753133,
+      "grad_norm": 0.1923828125,
+      "learning_rate": 0.0009018291123049373,
+      "loss": 1.6972,
+      "step": 153500
+    },
+    {
+      "epoch": 4.924533128677411,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.0009015093374264518,
+      "loss": 1.6973,
+      "step": 154000
+    },
+    {
+      "epoch": 4.940521872601688,
+      "grad_norm": 0.1669921875,
+      "learning_rate": 0.0009011895625479663,
+      "loss": 1.6952,
+      "step": 154500
+    },
+    {
+      "epoch": 4.956510616525966,
+      "grad_norm": 0.22265625,
+      "learning_rate": 0.0009008697876694807,
+      "loss": 1.6995,
+      "step": 155000
+    },
+    {
+      "epoch": 4.972499360450243,
+      "grad_norm": 0.18359375,
+      "learning_rate": 0.0009005500127909951,
+      "loss": 1.6963,
+      "step": 155500
+    },
+    {
+      "epoch": 4.98848810437452,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0009002302379125096,
+      "loss": 1.7035,
+      "step": 156000
+    },
+    {
+      "epoch": 5.0,
+      "eval_loss": 3.3757665157318115,
+      "eval_runtime": 3.9479,
+      "eval_samples_per_second": 126.649,
+      "eval_steps_per_second": 4.053,
+      "step": 156360
+    },
+    {
+      "epoch": 5.004476848298798,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.0008999104630340241,
+      "loss": 1.699,
+      "step": 156500
+    },
+    {
+      "epoch": 5.020465592223075,
+      "grad_norm": 0.2021484375,
+      "learning_rate": 0.0008995906881555385,
+      "loss": 1.6938,
+      "step": 157000
+    },
+    {
+      "epoch": 5.036454336147353,
+      "grad_norm": 0.2333984375,
+      "learning_rate": 0.000899270913277053,
+      "loss": 1.6942,
+      "step": 157500
+    },
+    {
+      "epoch": 5.05244308007163,
+      "grad_norm": 0.201171875,
+      "learning_rate": 0.0008989511383985675,
+      "loss": 1.6958,
+      "step": 158000
+    },
+    {
+      "epoch": 5.0684318239959065,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.0008986313635200819,
+      "loss": 1.697,
+      "step": 158500
+    },
+    {
+      "epoch": 5.084420567920184,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0008983115886415963,
+      "loss": 1.6959,
+      "step": 159000
+    },
+    {
+      "epoch": 5.100409311844461,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.0008979918137631108,
+      "loss": 1.7003,
+      "step": 159500
+    },
+    {
+      "epoch": 5.116398055768739,
+      "grad_norm": 0.2041015625,
+      "learning_rate": 0.0008976720388846252,
+      "loss": 1.6963,
+      "step": 160000
+    },
+    {
+      "epoch": 5.132386799693016,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.0008973522640061397,
+      "loss": 1.6994,
+      "step": 160500
+    },
+    {
+      "epoch": 5.148375543617293,
+      "grad_norm": 0.203125,
+      "learning_rate": 0.0008970324891276542,
+      "loss": 1.6977,
+      "step": 161000
+    },
+    {
+      "epoch": 5.164364287541571,
+      "grad_norm": 0.1953125,
+      "learning_rate": 0.0008967127142491685,
+      "loss": 1.6937,
+      "step": 161500
+    },
+    {
+      "epoch": 5.180353031465848,
+      "grad_norm": 0.306640625,
+      "learning_rate": 0.000896392939370683,
+      "loss": 1.6994,
+      "step": 162000
+    },
+    {
+      "epoch": 5.196341775390126,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.0008960731644921976,
+      "loss": 1.6954,
+      "step": 162500
+    },
+    {
+      "epoch": 5.212330519314403,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.000895753389613712,
+      "loss": 1.6987,
+      "step": 163000
+    },
+    {
+      "epoch": 5.22831926323868,
+      "grad_norm": 0.220703125,
+      "learning_rate": 0.0008954336147352264,
+      "loss": 1.6908,
+      "step": 163500
+    },
+    {
+      "epoch": 5.2443080071629575,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.0008951138398567409,
+      "loss": 1.6921,
+      "step": 164000
+    },
+    {
+      "epoch": 5.2602967510872345,
+      "grad_norm": 0.224609375,
+      "learning_rate": 0.0008947940649782553,
+      "loss": 1.6987,
+      "step": 164500
+    },
+    {
+      "epoch": 5.276285495011512,
+      "grad_norm": 0.318359375,
+      "learning_rate": 0.0008944742900997698,
+      "loss": 1.6943,
+      "step": 165000
+    },
+    {
+      "epoch": 5.292274238935789,
+      "grad_norm": 0.1630859375,
+      "learning_rate": 0.0008941545152212842,
+      "loss": 1.6983,
+      "step": 165500
+    },
+    {
+      "epoch": 5.308262982860066,
+      "grad_norm": 0.2177734375,
+      "learning_rate": 0.0008938347403427987,
+      "loss": 1.6964,
+      "step": 166000
+    },
+    {
+      "epoch": 5.324251726784344,
+      "grad_norm": 0.25,
+      "learning_rate": 0.0008935149654643131,
+      "loss": 1.6894,
+      "step": 166500
+    },
+    {
+      "epoch": 5.340240470708621,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0008931951905858276,
+      "loss": 1.6944,
+      "step": 167000
+    },
+    {
+      "epoch": 5.356229214632898,
+      "grad_norm": 0.2353515625,
+      "learning_rate": 0.0008928754157073421,
+      "loss": 1.7001,
+      "step": 167500
+    },
+    {
+      "epoch": 5.372217958557176,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.0008925556408288565,
+      "loss": 1.6938,
+      "step": 168000
+    },
+    {
+      "epoch": 5.388206702481453,
+      "grad_norm": 0.228515625,
+      "learning_rate": 0.000892235865950371,
+      "loss": 1.6945,
+      "step": 168500
+    },
+    {
+      "epoch": 5.404195446405731,
+      "grad_norm": 0.203125,
+      "learning_rate": 0.0008919160910718855,
+      "loss": 1.6997,
+      "step": 169000
+    },
+    {
+      "epoch": 5.420184190330008,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.0008915963161933998,
+      "loss": 1.6947,
+      "step": 169500
+    },
+    {
+      "epoch": 5.436172934254285,
+      "grad_norm": 0.2470703125,
+      "learning_rate": 0.0008912765413149143,
+      "loss": 1.7007,
+      "step": 170000
+    },
+    {
+      "epoch": 5.4521616781785625,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.0008909567664364288,
+      "loss": 1.6932,
+      "step": 170500
+    },
+    {
+      "epoch": 5.468150422102839,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.0008906369915579432,
+      "loss": 1.6998,
+      "step": 171000
+    },
+    {
+      "epoch": 5.484139166027117,
+      "grad_norm": 0.205078125,
+      "learning_rate": 0.0008903172166794576,
+      "loss": 1.6901,
+      "step": 171500
+    },
+    {
+      "epoch": 5.500127909951394,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.0008899974418009721,
+      "loss": 1.7001,
+      "step": 172000
+    },
+    {
+      "epoch": 5.516116653875671,
+      "grad_norm": 0.2001953125,
+      "learning_rate": 0.0008896776669224866,
+      "loss": 1.7043,
+      "step": 172500
+    },
+    {
+      "epoch": 5.532105397799949,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.0008893578920440011,
+      "loss": 1.6994,
+      "step": 173000
+    },
+    {
+      "epoch": 5.548094141724226,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0008890381171655156,
+      "loss": 1.6985,
+      "step": 173500
+    },
+    {
+      "epoch": 5.564082885648504,
+      "grad_norm": 0.228515625,
+      "learning_rate": 0.0008887183422870299,
+      "loss": 1.6931,
+      "step": 174000
+    },
+    {
+      "epoch": 5.580071629572781,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.0008883985674085444,
+      "loss": 1.6996,
+      "step": 174500
+    },
+    {
+      "epoch": 5.596060373497058,
+      "grad_norm": 0.310546875,
+      "learning_rate": 0.0008880787925300589,
+      "loss": 1.6906,
+      "step": 175000
+    },
+    {
+      "epoch": 5.612049117421336,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.0008877590176515732,
+      "loss": 1.6937,
+      "step": 175500
+    },
+    {
+      "epoch": 5.628037861345613,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.0008874392427730877,
+      "loss": 1.6966,
+      "step": 176000
+    },
+    {
+      "epoch": 5.6440266052698895,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.0008871194678946022,
+      "loss": 1.6912,
+      "step": 176500
+    },
+    {
+      "epoch": 5.660015349194167,
+      "grad_norm": 0.322265625,
+      "learning_rate": 0.0008867996930161167,
+      "loss": 1.689,
+      "step": 177000
+    },
+    {
+      "epoch": 5.676004093118444,
+      "grad_norm": 0.1767578125,
+      "learning_rate": 0.0008864799181376312,
+      "loss": 1.6926,
+      "step": 177500
+    },
+    {
+      "epoch": 5.691992837042722,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.0008861601432591456,
+      "loss": 1.6968,
+      "step": 178000
+    },
+    {
+      "epoch": 5.707981580966999,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0008858403683806601,
+      "loss": 1.6962,
+      "step": 178500
+    },
+    {
+      "epoch": 5.723970324891276,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.0008855205935021745,
+      "loss": 1.6936,
+      "step": 179000
+    },
+    {
+      "epoch": 5.739959068815554,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.000885200818623689,
+      "loss": 1.6942,
+      "step": 179500
+    },
+    {
+      "epoch": 5.755947812739831,
+      "grad_norm": 0.298828125,
+      "learning_rate": 0.0008848810437452034,
+      "loss": 1.6932,
+      "step": 180000
+    },
+    {
+      "epoch": 5.771936556664109,
+      "grad_norm": 0.185546875,
+      "learning_rate": 0.0008845612688667178,
+      "loss": 1.6962,
+      "step": 180500
+    },
+    {
+      "epoch": 5.787925300588386,
+      "grad_norm": 0.1669921875,
+      "learning_rate": 0.0008842414939882323,
+      "loss": 1.6998,
+      "step": 181000
+    },
+    {
+      "epoch": 5.803914044512663,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.0008839217191097468,
+      "loss": 1.6932,
+      "step": 181500
+    },
+    {
+      "epoch": 5.8199027884369405,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0008836019442312611,
+      "loss": 1.6921,
+      "step": 182000
+    },
+    {
+      "epoch": 5.8358915323612175,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.0008832821693527756,
+      "loss": 1.691,
+      "step": 182500
+    },
+    {
+      "epoch": 5.851880276285495,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.0008829623944742902,
+      "loss": 1.6942,
+      "step": 183000
+    },
+    {
+      "epoch": 5.867869020209772,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0008826426195958046,
+      "loss": 1.6965,
+      "step": 183500
+    },
+    {
+      "epoch": 5.883857764134049,
+      "grad_norm": 0.2177734375,
+      "learning_rate": 0.000882322844717319,
+      "loss": 1.6946,
+      "step": 184000
+    },
+    {
+      "epoch": 5.899846508058327,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.0008820030698388335,
+      "loss": 1.6987,
+      "step": 184500
+    },
+    {
+      "epoch": 5.915835251982604,
+      "grad_norm": 0.224609375,
+      "learning_rate": 0.0008816832949603479,
+      "loss": 1.6927,
+      "step": 185000
+    },
+    {
+      "epoch": 5.931823995906882,
+      "grad_norm": 0.1806640625,
+      "learning_rate": 0.0008813635200818624,
+      "loss": 1.6956,
+      "step": 185500
+    },
+    {
+      "epoch": 5.947812739831159,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.0008810437452033768,
+      "loss": 1.6983,
+      "step": 186000
+    },
+    {
+      "epoch": 5.963801483755436,
+      "grad_norm": 0.2734375,
+      "learning_rate": 0.0008807239703248912,
+      "loss": 1.6971,
+      "step": 186500
+    },
+    {
+      "epoch": 5.979790227679714,
+      "grad_norm": 0.27734375,
+      "learning_rate": 0.0008804041954464057,
+      "loss": 1.6983,
+      "step": 187000
+    },
+    {
+      "epoch": 5.995778971603991,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0008800844205679202,
+      "loss": 1.6948,
+      "step": 187500
+    },
+    {
+      "epoch": 6.0,
+      "eval_loss": 3.3661954402923584,
+      "eval_runtime": 3.4589,
+      "eval_samples_per_second": 144.553,
+      "eval_steps_per_second": 4.626,
+      "step": 187632
+    },
+    {
+      "epoch": 6.0117677155282685,
+      "grad_norm": 0.1865234375,
+      "learning_rate": 0.0008797646456894348,
+      "loss": 1.6969,
+      "step": 188000
+    },
+    {
+      "epoch": 6.0277564594525455,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.0008794448708109491,
+      "loss": 1.6895,
+      "step": 188500
+    },
+    {
+      "epoch": 6.043745203376822,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.0008791250959324636,
+      "loss": 1.69,
+      "step": 189000
+    },
+    {
+      "epoch": 6.0597339473011,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.0008788053210539781,
+      "loss": 1.6933,
+      "step": 189500
+    },
+    {
+      "epoch": 6.075722691225377,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0008784855461754924,
+      "loss": 1.6929,
+      "step": 190000
+    },
+    {
+      "epoch": 6.091711435149655,
+      "grad_norm": 0.1953125,
+      "learning_rate": 0.0008781657712970069,
+      "loss": 1.6975,
+      "step": 190500
+    },
+    {
+      "epoch": 6.107700179073932,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0008778459964185214,
+      "loss": 1.6918,
+      "step": 191000
+    },
+    {
+      "epoch": 6.123688922998209,
+      "grad_norm": 0.205078125,
+      "learning_rate": 0.0008775262215400358,
+      "loss": 1.6926,
+      "step": 191500
+    },
+    {
+      "epoch": 6.139677666922487,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.0008772064466615503,
+      "loss": 1.7001,
+      "step": 192000
+    },
+    {
+      "epoch": 6.155666410846764,
+      "grad_norm": 0.21875,
+      "learning_rate": 0.0008768866717830647,
+      "loss": 1.6856,
+      "step": 192500
+    },
+    {
+      "epoch": 6.171655154771041,
+      "grad_norm": 0.220703125,
+      "learning_rate": 0.0008765668969045792,
+      "loss": 1.6922,
+      "step": 193000
+    },
+    {
+      "epoch": 6.187643898695319,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.0008762471220260937,
+      "loss": 1.6965,
+      "step": 193500
+    },
+    {
+      "epoch": 6.203632642619596,
+      "grad_norm": 0.162109375,
+      "learning_rate": 0.0008759273471476082,
+      "loss": 1.6933,
+      "step": 194000
+    },
+    {
+      "epoch": 6.219621386543873,
+      "grad_norm": 0.2314453125,
+      "learning_rate": 0.0008756075722691225,
+      "loss": 1.694,
+      "step": 194500
+    },
+    {
+      "epoch": 6.23561013046815,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.000875287797390637,
+      "loss": 1.702,
+      "step": 195000
+    },
+    {
+      "epoch": 6.251598874392427,
+      "grad_norm": 0.33203125,
+      "learning_rate": 0.0008749680225121515,
+      "loss": 1.6911,
+      "step": 195500
+    },
+    {
+      "epoch": 6.267587618316705,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.0008746482476336659,
+      "loss": 1.6878,
+      "step": 196000
+    },
+    {
+      "epoch": 6.283576362240982,
+      "grad_norm": 0.302734375,
+      "learning_rate": 0.0008743284727551803,
+      "loss": 1.69,
+      "step": 196500
+    },
+    {
+      "epoch": 6.29956510616526,
+      "grad_norm": 0.220703125,
+      "learning_rate": 0.0008740086978766948,
+      "loss": 1.6958,
+      "step": 197000
+    },
+    {
+      "epoch": 6.315553850089537,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.0008736889229982092,
+      "loss": 1.6949,
+      "step": 197500
+    },
+    {
+      "epoch": 6.331542594013814,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.0008733691481197237,
+      "loss": 1.6944,
+      "step": 198000
+    },
+    {
+      "epoch": 6.347531337938092,
+      "grad_norm": 0.1953125,
+      "learning_rate": 0.0008730493732412382,
+      "loss": 1.6915,
+      "step": 198500
+    },
+    {
+      "epoch": 6.363520081862369,
+      "grad_norm": 0.2236328125,
+      "learning_rate": 0.0008727295983627527,
+      "loss": 1.6979,
+      "step": 199000
+    },
+    {
+      "epoch": 6.379508825786647,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0008724098234842671,
+      "loss": 1.6891,
+      "step": 199500
+    },
+    {
+      "epoch": 6.3954975697109235,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.0008720900486057816,
+      "loss": 1.6916,
+      "step": 200000
+    },
+    {
+      "epoch": 6.4114863136352005,
+      "grad_norm": 0.236328125,
+      "learning_rate": 0.000871770273727296,
+      "loss": 1.6906,
+      "step": 200500
+    },
+    {
+      "epoch": 6.427475057559478,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.0008714504988488104,
+      "loss": 1.694,
+      "step": 201000
+    },
+    {
+      "epoch": 6.443463801483755,
+      "grad_norm": 0.216796875,
+      "learning_rate": 0.0008711307239703249,
+      "loss": 1.6908,
+      "step": 201500
+    },
+    {
+      "epoch": 6.459452545408033,
+      "grad_norm": 0.310546875,
+      "learning_rate": 0.0008708109490918394,
+      "loss": 1.6955,
+      "step": 202000
+    },
+    {
+      "epoch": 6.47544128933231,
+      "grad_norm": 0.224609375,
+      "learning_rate": 0.0008704911742133537,
+      "loss": 1.6885,
+      "step": 202500
+    },
+    {
+      "epoch": 6.491430033256587,
+      "grad_norm": 0.2333984375,
+      "learning_rate": 0.0008701713993348682,
+      "loss": 1.6932,
+      "step": 203000
+    },
+    {
+      "epoch": 6.507418777180865,
+      "grad_norm": 0.224609375,
+      "learning_rate": 0.0008698516244563828,
+      "loss": 1.6929,
+      "step": 203500
+    },
+    {
+      "epoch": 6.523407521105142,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0008695318495778972,
+      "loss": 1.6877,
+      "step": 204000
+    },
+    {
+      "epoch": 6.539396265029419,
+      "grad_norm": 0.158203125,
+      "learning_rate": 0.0008692120746994117,
+      "loss": 1.6912,
+      "step": 204500
+    },
+    {
+      "epoch": 6.555385008953697,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.0008688922998209261,
+      "loss": 1.6915,
+      "step": 205000
+    },
+    {
+      "epoch": 6.571373752877974,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.0008685725249424405,
+      "loss": 1.6913,
+      "step": 205500
+    },
+    {
+      "epoch": 6.5873624968022515,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.000868252750063955,
+      "loss": 1.6872,
+      "step": 206000
+    },
+    {
+      "epoch": 6.6033512407265285,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.0008679329751854695,
+      "loss": 1.6942,
+      "step": 206500
+    },
+    {
+      "epoch": 6.619339984650805,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0008676132003069838,
+      "loss": 1.6867,
+      "step": 207000
+    },
+    {
+      "epoch": 6.635328728575083,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0008672934254284983,
+      "loss": 1.6921,
+      "step": 207500
+    },
+    {
+      "epoch": 6.65131747249936,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.0008669736505500128,
+      "loss": 1.6921,
+      "step": 208000
+    },
+    {
+      "epoch": 6.667306216423638,
+      "grad_norm": 0.154296875,
+      "learning_rate": 0.0008666538756715274,
+      "loss": 1.69,
+      "step": 208500
+    },
+    {
+      "epoch": 6.683294960347915,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.0008663341007930417,
+      "loss": 1.6934,
+      "step": 209000
+    },
+    {
+      "epoch": 6.699283704272192,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.0008660143259145562,
+      "loss": 1.6928,
+      "step": 209500
+    },
+    {
+      "epoch": 6.71527244819647,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.0008656945510360707,
+      "loss": 1.6917,
+      "step": 210000
+    },
+    {
+      "epoch": 6.731261192120747,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0008653747761575851,
+      "loss": 1.6912,
+      "step": 210500
+    },
+    {
+      "epoch": 6.747249936045025,
+      "grad_norm": 0.2236328125,
+      "learning_rate": 0.0008650550012790995,
+      "loss": 1.6974,
+      "step": 211000
+    },
+    {
+      "epoch": 6.763238679969302,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.000864735226400614,
+      "loss": 1.6929,
+      "step": 211500
+    },
+    {
+      "epoch": 6.779227423893579,
+      "grad_norm": 0.302734375,
+      "learning_rate": 0.0008644154515221284,
+      "loss": 1.6888,
+      "step": 212000
+    },
+    {
+      "epoch": 6.795216167817856,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.0008640956766436429,
+      "loss": 1.691,
+      "step": 212500
+    },
+    {
+      "epoch": 6.811204911742133,
+      "grad_norm": 0.2734375,
+      "learning_rate": 0.0008637759017651573,
+      "loss": 1.6958,
+      "step": 213000
+    },
+    {
+      "epoch": 6.827193655666411,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0008634561268866718,
+      "loss": 1.6948,
+      "step": 213500
+    },
+    {
+      "epoch": 6.843182399590688,
+      "grad_norm": 0.35546875,
+      "learning_rate": 0.0008631363520081863,
+      "loss": 1.6926,
+      "step": 214000
+    },
+    {
+      "epoch": 6.859171143514965,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0008628165771297008,
+      "loss": 1.6872,
+      "step": 214500
+    },
+    {
+      "epoch": 6.875159887439243,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.0008624968022512151,
+      "loss": 1.6921,
+      "step": 215000
+    },
+    {
+      "epoch": 6.89114863136352,
+      "grad_norm": 0.23828125,
+      "learning_rate": 0.0008621770273727296,
+      "loss": 1.6952,
+      "step": 215500
+    },
+    {
+      "epoch": 6.907137375287798,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.0008618572524942441,
+      "loss": 1.6911,
+      "step": 216000
+    },
+    {
+      "epoch": 6.923126119212075,
+      "grad_norm": 0.2734375,
+      "learning_rate": 0.0008615374776157585,
+      "loss": 1.6904,
+      "step": 216500
+    },
+    {
+      "epoch": 6.939114863136352,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.0008612177027372729,
+      "loss": 1.6902,
+      "step": 217000
+    },
+    {
+      "epoch": 6.95510360706063,
+      "grad_norm": 0.2109375,
+      "learning_rate": 0.0008608979278587874,
+      "loss": 1.6922,
+      "step": 217500
+    },
+    {
+      "epoch": 6.9710923509849065,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.0008605781529803018,
+      "loss": 1.6936,
+      "step": 218000
+    },
+    {
+      "epoch": 6.987081094909184,
+      "grad_norm": 0.171875,
+      "learning_rate": 0.0008602583781018163,
+      "loss": 1.6896,
+      "step": 218500
+    },
+    {
+      "epoch": 7.0,
+      "eval_loss": 3.3631489276885986,
+      "eval_runtime": 3.5018,
+      "eval_samples_per_second": 142.783,
+      "eval_steps_per_second": 4.569,
+      "step": 218904
+    },
+    {
+      "epoch": 7.003069838833461,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.0008599386032233309,
+      "loss": 1.6899,
+      "step": 219000
+    },
+    {
+      "epoch": 7.019058582757738,
+      "grad_norm": 0.2890625,
+      "learning_rate": 0.0008596188283448453,
+      "loss": 1.6891,
+      "step": 219500
+    },
+    {
+      "epoch": 7.035047326682016,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.0008592990534663597,
+      "loss": 1.6907,
+      "step": 220000
+    },
+    {
+      "epoch": 7.051036070606293,
+      "grad_norm": 0.333984375,
+      "learning_rate": 0.0008589792785878742,
+      "loss": 1.6892,
+      "step": 220500
+    },
+    {
+      "epoch": 7.06702481453057,
+      "grad_norm": 0.224609375,
+      "learning_rate": 0.0008586595037093887,
+      "loss": 1.6916,
+      "step": 221000
+    },
+    {
+      "epoch": 7.083013558454848,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.000858339728830903,
+      "loss": 1.6861,
+      "step": 221500
+    },
+    {
+      "epoch": 7.099002302379125,
+      "grad_norm": 0.318359375,
+      "learning_rate": 0.0008580199539524175,
+      "loss": 1.6886,
+      "step": 222000
+    },
+    {
+      "epoch": 7.114991046303403,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.000857700179073932,
+      "loss": 1.6841,
+      "step": 222500
+    },
+    {
+      "epoch": 7.13097979022768,
+      "grad_norm": 0.197265625,
+      "learning_rate": 0.0008573804041954464,
+      "loss": 1.6898,
+      "step": 223000
+    },
+    {
+      "epoch": 7.146968534151957,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.0008570606293169608,
+      "loss": 1.6865,
+      "step": 223500
+    },
+    {
+      "epoch": 7.1629572780762345,
+      "grad_norm": 0.302734375,
+      "learning_rate": 0.0008567408544384754,
+      "loss": 1.6986,
+      "step": 224000
+    },
+    {
+      "epoch": 7.1789460220005115,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.0008564210795599898,
+      "loss": 1.6921,
+      "step": 224500
+    },
+    {
+      "epoch": 7.194934765924789,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.0008561013046815043,
+      "loss": 1.6844,
+      "step": 225000
+    },
+    {
+      "epoch": 7.210923509849066,
+      "grad_norm": 0.34765625,
+      "learning_rate": 0.0008557815298030187,
+      "loss": 1.6929,
+      "step": 225500
+    },
+    {
+      "epoch": 7.226912253773343,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.0008554617549245331,
+      "loss": 1.6958,
+      "step": 226000
+    },
+    {
+      "epoch": 7.242900997697621,
+      "grad_norm": 0.2470703125,
+      "learning_rate": 0.0008551419800460476,
+      "loss": 1.6929,
+      "step": 226500
+    },
+    {
+      "epoch": 7.258889741621898,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.0008548222051675621,
+      "loss": 1.6913,
+      "step": 227000
+    },
+    {
+      "epoch": 7.274878485546176,
+      "grad_norm": 0.189453125,
+      "learning_rate": 0.0008545024302890764,
+      "loss": 1.689,
+      "step": 227500
+    },
+    {
+      "epoch": 7.290867229470453,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.0008541826554105909,
+      "loss": 1.689,
+      "step": 228000
+    },
+    {
+      "epoch": 7.30685597339473,
+      "grad_norm": 0.2392578125,
+      "learning_rate": 0.0008538628805321054,
+      "loss": 1.6941,
+      "step": 228500
+    },
+    {
+      "epoch": 7.322844717319008,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0008535431056536199,
+      "loss": 1.6894,
+      "step": 229000
+    },
+    {
+      "epoch": 7.338833461243285,
+      "grad_norm": 0.1962890625,
+      "learning_rate": 0.0008532233307751343,
+      "loss": 1.6902,
+      "step": 229500
+    },
+    {
+      "epoch": 7.354822205167562,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.0008529035558966488,
+      "loss": 1.6888,
+      "step": 230000
+    },
+    {
+      "epoch": 7.370810949091839,
+      "grad_norm": 0.19921875,
+      "learning_rate": 0.0008525837810181633,
+      "loss": 1.6937,
+      "step": 230500
+    },
+    {
+      "epoch": 7.386799693016116,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.0008522640061396777,
+      "loss": 1.6962,
+      "step": 231000
+    },
+    {
+      "epoch": 7.402788436940394,
+      "grad_norm": 0.2353515625,
+      "learning_rate": 0.0008519442312611921,
+      "loss": 1.6887,
+      "step": 231500
+    },
+    {
+      "epoch": 7.418777180864671,
+      "grad_norm": 0.1826171875,
+      "learning_rate": 0.0008516244563827066,
+      "loss": 1.6882,
+      "step": 232000
+    },
+    {
+      "epoch": 7.434765924788948,
+      "grad_norm": 0.203125,
+      "learning_rate": 0.000851304681504221,
+      "loss": 1.6916,
+      "step": 232500
+    },
+    {
+      "epoch": 7.450754668713226,
+      "grad_norm": 0.330078125,
+      "learning_rate": 0.0008509849066257355,
+      "loss": 1.6898,
+      "step": 233000
+    },
+    {
+      "epoch": 7.466743412637503,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.00085066513174725,
+      "loss": 1.6925,
+      "step": 233500
+    },
+    {
+      "epoch": 7.482732156561781,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.0008503453568687643,
+      "loss": 1.6863,
+      "step": 234000
+    },
+    {
+      "epoch": 7.498720900486058,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0008500255819902789,
+      "loss": 1.6845,
+      "step": 234500
+    },
+    {
+      "epoch": 7.514709644410335,
+      "grad_norm": 0.2490234375,
+      "learning_rate": 0.0008497058071117934,
+      "loss": 1.688,
+      "step": 235000
+    },
+    {
+      "epoch": 7.530698388334613,
+      "grad_norm": 0.3125,
+      "learning_rate": 0.0008493860322333078,
+      "loss": 1.6919,
+      "step": 235500
+    },
+    {
+      "epoch": 7.5466871322588895,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.0008490662573548222,
+      "loss": 1.6952,
+      "step": 236000
+    },
+    {
+      "epoch": 7.562675876183167,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.0008487464824763367,
+      "loss": 1.6883,
+      "step": 236500
+    },
+    {
+      "epoch": 7.578664620107444,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.0008484267075978511,
+      "loss": 1.6893,
+      "step": 237000
+    },
+    {
+      "epoch": 7.594653364031721,
+      "grad_norm": 0.2392578125,
+      "learning_rate": 0.0008481069327193656,
+      "loss": 1.6895,
+      "step": 237500
+    },
+    {
+      "epoch": 7.610642107955999,
+      "grad_norm": 0.1689453125,
+      "learning_rate": 0.00084778715784088,
+      "loss": 1.6836,
+      "step": 238000
+    },
+    {
+      "epoch": 7.626630851880276,
+      "grad_norm": 0.23046875,
+      "learning_rate": 0.0008474673829623944,
+      "loss": 1.6826,
+      "step": 238500
+    },
+    {
+      "epoch": 7.642619595804554,
+      "grad_norm": 0.193359375,
+      "learning_rate": 0.0008471476080839089,
+      "loss": 1.6943,
+      "step": 239000
+    },
+    {
+      "epoch": 7.658608339728831,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.0008468278332054235,
+      "loss": 1.6861,
+      "step": 239500
+    },
+    {
+      "epoch": 7.674597083653108,
+      "grad_norm": 0.205078125,
+      "learning_rate": 0.0008465080583269378,
+      "loss": 1.6908,
+      "step": 240000
+    },
+    {
+      "epoch": 7.690585827577386,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.0008461882834484523,
+      "loss": 1.6873,
+      "step": 240500
+    },
+    {
+      "epoch": 7.706574571501663,
+      "grad_norm": 0.2177734375,
+      "learning_rate": 0.0008458685085699668,
+      "loss": 1.6903,
+      "step": 241000
+    },
+    {
+      "epoch": 7.7225633154259405,
+      "grad_norm": 0.24609375,
+      "learning_rate": 0.0008455487336914813,
+      "loss": 1.6906,
+      "step": 241500
+    },
+    {
+      "epoch": 7.7385520593502175,
+      "grad_norm": 0.2236328125,
+      "learning_rate": 0.0008452289588129956,
+      "loss": 1.6899,
+      "step": 242000
+    },
+    {
+      "epoch": 7.7545408032744945,
+      "grad_norm": 0.27734375,
+      "learning_rate": 0.0008449091839345101,
+      "loss": 1.6898,
+      "step": 242500
+    },
+    {
+      "epoch": 7.770529547198772,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.0008445894090560246,
+      "loss": 1.6934,
+      "step": 243000
+    },
+    {
+      "epoch": 7.786518291123049,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.000844269634177539,
+      "loss": 1.6887,
+      "step": 243500
+    },
+    {
+      "epoch": 7.802507035047327,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.0008439498592990534,
+      "loss": 1.6937,
+      "step": 244000
+    },
+    {
+      "epoch": 7.818495778971604,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.000843630084420568,
+      "loss": 1.6895,
+      "step": 244500
+    },
+    {
+      "epoch": 7.834484522895881,
+      "grad_norm": 0.1767578125,
+      "learning_rate": 0.0008433103095420824,
+      "loss": 1.6888,
+      "step": 245000
+    },
+    {
+      "epoch": 7.850473266820159,
+      "grad_norm": 0.18359375,
+      "learning_rate": 0.0008429905346635969,
+      "loss": 1.6893,
+      "step": 245500
+    },
+    {
+      "epoch": 7.866462010744436,
+      "grad_norm": 0.22265625,
+      "learning_rate": 0.0008426707597851114,
+      "loss": 1.6833,
+      "step": 246000
+    },
+    {
+      "epoch": 7.882450754668714,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.0008423509849066257,
+      "loss": 1.6946,
+      "step": 246500
+    },
+    {
+      "epoch": 7.898439498592991,
+      "grad_norm": 0.16015625,
+      "learning_rate": 0.0008420312100281402,
+      "loss": 1.6885,
+      "step": 247000
+    },
+    {
+      "epoch": 7.914428242517268,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.0008417114351496547,
+      "loss": 1.6848,
+      "step": 247500
+    },
+    {
+      "epoch": 7.9304169864415455,
+      "grad_norm": 0.1865234375,
+      "learning_rate": 0.000841391660271169,
+      "loss": 1.6921,
+      "step": 248000
+    },
+    {
+      "epoch": 7.946405730365822,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.0008410718853926835,
+      "loss": 1.6884,
+      "step": 248500
+    },
+    {
+      "epoch": 7.962394474290099,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.000840752110514198,
+      "loss": 1.6921,
+      "step": 249000
+    },
+    {
+      "epoch": 7.978383218214377,
+      "grad_norm": 0.1875,
+      "learning_rate": 0.0008404323356357125,
+      "loss": 1.6854,
+      "step": 249500
+    },
+    {
+      "epoch": 7.994371962138654,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.000840112560757227,
+      "loss": 1.6831,
+      "step": 250000
+    },
+    {
+      "epoch": 8.0,
+      "eval_loss": 3.3572514057159424,
+      "eval_runtime": 4.2182,
+      "eval_samples_per_second": 118.533,
+      "eval_steps_per_second": 3.793,
+      "step": 250176
+    },
+    {
+      "epoch": 8.010360706062931,
+      "grad_norm": 0.185546875,
+      "learning_rate": 0.0008397927858787414,
+      "loss": 1.6869,
+      "step": 250500
+    },
+    {
+      "epoch": 8.026349449987208,
+      "grad_norm": 0.2353515625,
+      "learning_rate": 0.0008394730110002559,
+      "loss": 1.6838,
+      "step": 251000
+    },
+    {
+      "epoch": 8.042338193911487,
+      "grad_norm": 0.2216796875,
+      "learning_rate": 0.0008391532361217703,
+      "loss": 1.6856,
+      "step": 251500
+    },
+    {
+      "epoch": 8.058326937835764,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0008388334612432848,
+      "loss": 1.6877,
+      "step": 252000
+    },
+    {
+      "epoch": 8.07431568176004,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.0008385136863647992,
+      "loss": 1.6853,
+      "step": 252500
+    },
+    {
+      "epoch": 8.090304425684318,
+      "grad_norm": 0.1708984375,
+      "learning_rate": 0.0008381939114863136,
+      "loss": 1.6906,
+      "step": 253000
+    },
+    {
+      "epoch": 8.106293169608595,
+      "grad_norm": 0.19921875,
+      "learning_rate": 0.0008378741366078281,
+      "loss": 1.683,
+      "step": 253500
+    },
+    {
+      "epoch": 8.122281913532873,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.0008375543617293426,
+      "loss": 1.6885,
+      "step": 254000
+    },
+    {
+      "epoch": 8.13827065745715,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.0008372345868508569,
+      "loss": 1.6849,
+      "step": 254500
+    },
+    {
+      "epoch": 8.154259401381427,
+      "grad_norm": 0.25,
+      "learning_rate": 0.0008369148119723715,
+      "loss": 1.6855,
+      "step": 255000
+    },
+    {
+      "epoch": 8.170248145305704,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.000836595037093886,
+      "loss": 1.6853,
+      "step": 255500
+    },
+    {
+      "epoch": 8.186236889229981,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.0008362752622154004,
+      "loss": 1.6847,
+      "step": 256000
+    },
+    {
+      "epoch": 8.20222563315426,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.0008359554873369148,
+      "loss": 1.6861,
+      "step": 256500
+    },
+    {
+      "epoch": 8.218214377078537,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0008356357124584293,
+      "loss": 1.685,
+      "step": 257000
+    },
+    {
+      "epoch": 8.234203121002814,
+      "grad_norm": 0.220703125,
+      "learning_rate": 0.0008353159375799437,
+      "loss": 1.6873,
+      "step": 257500
+    },
+    {
+      "epoch": 8.250191864927091,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.0008349961627014582,
+      "loss": 1.6904,
+      "step": 258000
+    },
+    {
+      "epoch": 8.266180608851368,
+      "grad_norm": 0.17578125,
+      "learning_rate": 0.0008346763878229726,
+      "loss": 1.6884,
+      "step": 258500
+    },
+    {
+      "epoch": 8.282169352775647,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.000834356612944487,
+      "loss": 1.6912,
+      "step": 259000
+    },
+    {
+      "epoch": 8.298158096699924,
+      "grad_norm": 0.22265625,
+      "learning_rate": 0.0008340368380660015,
+      "loss": 1.6891,
+      "step": 259500
+    },
+    {
+      "epoch": 8.3141468406242,
+      "grad_norm": 0.2001953125,
+      "learning_rate": 0.0008337170631875161,
+      "loss": 1.6916,
+      "step": 260000
+    },
+    {
+      "epoch": 8.330135584548477,
+      "grad_norm": 0.330078125,
+      "learning_rate": 0.0008333972883090304,
+      "loss": 1.6828,
+      "step": 260500
+    },
+    {
+      "epoch": 8.346124328472754,
+      "grad_norm": 0.2431640625,
+      "learning_rate": 0.0008330775134305449,
+      "loss": 1.6849,
+      "step": 261000
+    },
+    {
+      "epoch": 8.362113072397033,
+      "grad_norm": 0.33203125,
+      "learning_rate": 0.0008327577385520594,
+      "loss": 1.6919,
+      "step": 261500
+    },
+    {
+      "epoch": 8.37810181632131,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.0008324379636735739,
+      "loss": 1.6842,
+      "step": 262000
+    },
+    {
+      "epoch": 8.394090560245587,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.0008321181887950883,
+      "loss": 1.6893,
+      "step": 262500
+    },
+    {
+      "epoch": 8.410079304169864,
+      "grad_norm": 0.20703125,
+      "learning_rate": 0.0008317984139166027,
+      "loss": 1.6855,
+      "step": 263000
+    },
+    {
+      "epoch": 8.426068048094141,
+      "grad_norm": 0.197265625,
+      "learning_rate": 0.0008314786390381172,
+      "loss": 1.6931,
+      "step": 263500
+    },
+    {
+      "epoch": 8.44205679201842,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.0008311588641596316,
+      "loss": 1.6883,
+      "step": 264000
+    },
+    {
+      "epoch": 8.458045535942697,
+      "grad_norm": 0.21875,
+      "learning_rate": 0.000830839089281146,
+      "loss": 1.6838,
+      "step": 264500
+    },
+    {
+      "epoch": 8.474034279866974,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.0008305193144026606,
+      "loss": 1.6812,
+      "step": 265000
+    },
+    {
+      "epoch": 8.49002302379125,
+      "grad_norm": 0.322265625,
+      "learning_rate": 0.000830199539524175,
+      "loss": 1.6897,
+      "step": 265500
+    },
+    {
+      "epoch": 8.506011767715528,
+      "grad_norm": 0.2216796875,
+      "learning_rate": 0.0008298797646456895,
+      "loss": 1.6856,
+      "step": 266000
+    },
+    {
+      "epoch": 8.522000511639806,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.000829559989767204,
+      "loss": 1.692,
+      "step": 266500
+    },
+    {
+      "epoch": 8.537989255564083,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.0008292402148887183,
+      "loss": 1.6901,
+      "step": 267000
+    },
+    {
+      "epoch": 8.55397799948836,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0008289204400102328,
+      "loss": 1.6841,
+      "step": 267500
+    },
+    {
+      "epoch": 8.569966743412637,
+      "grad_norm": 0.2431640625,
+      "learning_rate": 0.0008286006651317473,
+      "loss": 1.6864,
+      "step": 268000
+    },
+    {
+      "epoch": 8.585955487336914,
+      "grad_norm": 0.318359375,
+      "learning_rate": 0.0008282808902532617,
+      "loss": 1.6856,
+      "step": 268500
+    },
+    {
+      "epoch": 8.601944231261193,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.0008279611153747761,
+      "loss": 1.6866,
+      "step": 269000
+    },
+    {
+      "epoch": 8.61793297518547,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0008276413404962906,
+      "loss": 1.6881,
+      "step": 269500
+    },
+    {
+      "epoch": 8.633921719109747,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.000827321565617805,
+      "loss": 1.6845,
+      "step": 270000
+    },
+    {
+      "epoch": 8.649910463034024,
+      "grad_norm": 0.166015625,
+      "learning_rate": 0.0008270017907393196,
+      "loss": 1.6862,
+      "step": 270500
+    },
+    {
+      "epoch": 8.6658992069583,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.000826682015860834,
+      "loss": 1.69,
+      "step": 271000
+    },
+    {
+      "epoch": 8.68188795088258,
+      "grad_norm": 0.3359375,
+      "learning_rate": 0.0008263622409823484,
+      "loss": 1.691,
+      "step": 271500
+    },
+    {
+      "epoch": 8.697876694806856,
+      "grad_norm": 0.2314453125,
+      "learning_rate": 0.0008260424661038629,
+      "loss": 1.6865,
+      "step": 272000
+    },
+    {
+      "epoch": 8.713865438731133,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.0008257226912253774,
+      "loss": 1.6866,
+      "step": 272500
+    },
+    {
+      "epoch": 8.72985418265541,
+      "grad_norm": 0.375,
+      "learning_rate": 0.0008254029163468918,
+      "loss": 1.6886,
+      "step": 273000
+    },
+    {
+      "epoch": 8.745842926579687,
+      "grad_norm": 0.236328125,
+      "learning_rate": 0.0008250831414684062,
+      "loss": 1.6843,
+      "step": 273500
+    },
+    {
+      "epoch": 8.761831670503966,
+      "grad_norm": 0.27734375,
+      "learning_rate": 0.0008247633665899207,
+      "loss": 1.6901,
+      "step": 274000
+    },
+    {
+      "epoch": 8.777820414428243,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.0008244435917114352,
+      "loss": 1.6808,
+      "step": 274500
+    },
+    {
+      "epoch": 8.79380915835252,
+      "grad_norm": 0.177734375,
+      "learning_rate": 0.0008241238168329495,
+      "loss": 1.6811,
+      "step": 275000
+    },
+    {
+      "epoch": 8.809797902276797,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.0008238040419544641,
+      "loss": 1.6865,
+      "step": 275500
+    },
+    {
+      "epoch": 8.825786646201074,
+      "grad_norm": 0.1953125,
+      "learning_rate": 0.0008234842670759786,
+      "loss": 1.6928,
+      "step": 276000
+    },
+    {
+      "epoch": 8.841775390125353,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.000823164492197493,
+      "loss": 1.6853,
+      "step": 276500
+    },
+    {
+      "epoch": 8.85776413404963,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.0008228447173190075,
+      "loss": 1.687,
+      "step": 277000
+    },
+    {
+      "epoch": 8.873752877973907,
+      "grad_norm": 0.2490234375,
+      "learning_rate": 0.0008225249424405219,
+      "loss": 1.6845,
+      "step": 277500
+    },
+    {
+      "epoch": 8.889741621898184,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0008222051675620363,
+      "loss": 1.6883,
+      "step": 278000
+    },
+    {
+      "epoch": 8.90573036582246,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0008218853926835508,
+      "loss": 1.6892,
+      "step": 278500
+    },
+    {
+      "epoch": 8.92171910974674,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.0008215656178050653,
+      "loss": 1.6883,
+      "step": 279000
+    },
+    {
+      "epoch": 8.937707853671016,
+      "grad_norm": 0.2490234375,
+      "learning_rate": 0.0008212458429265796,
+      "loss": 1.689,
+      "step": 279500
+    },
+    {
+      "epoch": 8.953696597595293,
+      "grad_norm": 0.318359375,
+      "learning_rate": 0.0008209260680480941,
+      "loss": 1.6885,
+      "step": 280000
+    },
+    {
+      "epoch": 8.96968534151957,
+      "grad_norm": 0.34765625,
+      "learning_rate": 0.0008206062931696087,
+      "loss": 1.6934,
+      "step": 280500
+    },
+    {
+      "epoch": 8.985674085443847,
+      "grad_norm": 0.220703125,
+      "learning_rate": 0.0008202865182911231,
+      "loss": 1.6851,
+      "step": 281000
+    },
+    {
+      "epoch": 9.0,
+      "eval_loss": 3.354891061782837,
+      "eval_runtime": 3.5512,
+      "eval_samples_per_second": 140.797,
+      "eval_steps_per_second": 4.505,
+      "step": 281448
+    },
+    {
+      "epoch": 9.001662829368124,
+      "grad_norm": 0.1650390625,
+      "learning_rate": 0.0008199667434126375,
+      "loss": 1.6857,
+      "step": 281500
+    },
+    {
+      "epoch": 9.017651573292403,
+      "grad_norm": 0.162109375,
+      "learning_rate": 0.000819646968534152,
+      "loss": 1.6831,
+      "step": 282000
+    },
+    {
+      "epoch": 9.03364031721668,
+      "grad_norm": 0.1806640625,
+      "learning_rate": 0.0008193271936556664,
+      "loss": 1.6917,
+      "step": 282500
+    },
+    {
+      "epoch": 9.049629061140957,
+      "grad_norm": 0.2158203125,
+      "learning_rate": 0.0008190074187771809,
+      "loss": 1.6805,
+      "step": 283000
+    },
+    {
+      "epoch": 9.065617805065234,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0008186876438986953,
+      "loss": 1.6876,
+      "step": 283500
+    },
+    {
+      "epoch": 9.08160654898951,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.0008183678690202098,
+      "loss": 1.6868,
+      "step": 284000
+    },
+    {
+      "epoch": 9.09759529291379,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.0008180480941417242,
+      "loss": 1.683,
+      "step": 284500
+    },
+    {
+      "epoch": 9.113584036838066,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.0008177283192632387,
+      "loss": 1.6879,
+      "step": 285000
+    },
+    {
+      "epoch": 9.129572780762343,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.0008174085443847533,
+      "loss": 1.6901,
+      "step": 285500
+    },
+    {
+      "epoch": 9.14556152468662,
+      "grad_norm": 0.330078125,
+      "learning_rate": 0.0008170887695062676,
+      "loss": 1.6911,
+      "step": 286000
+    },
+    {
+      "epoch": 9.161550268610897,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.0008167689946277821,
+      "loss": 1.6931,
+      "step": 286500
+    },
+    {
+      "epoch": 9.177539012535176,
+      "grad_norm": 0.2119140625,
+      "learning_rate": 0.0008164492197492966,
+      "loss": 1.684,
+      "step": 287000
+    },
+    {
+      "epoch": 9.193527756459453,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.0008161294448708109,
+      "loss": 1.6892,
+      "step": 287500
+    },
+    {
+      "epoch": 9.20951650038373,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0008158096699923254,
+      "loss": 1.6841,
+      "step": 288000
+    },
+    {
+      "epoch": 9.225505244308007,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.0008154898951138399,
+      "loss": 1.6802,
+      "step": 288500
+    },
+    {
+      "epoch": 9.241493988232284,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.0008151701202353543,
+      "loss": 1.6821,
+      "step": 289000
+    },
+    {
+      "epoch": 9.257482732156562,
+      "grad_norm": 0.212890625,
+      "learning_rate": 0.0008148503453568687,
+      "loss": 1.6878,
+      "step": 289500
+    },
+    {
+      "epoch": 9.27347147608084,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.0008145305704783832,
+      "loss": 1.6812,
+      "step": 290000
+    },
+    {
+      "epoch": 9.289460220005116,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0008142107955998976,
+      "loss": 1.6925,
+      "step": 290500
+    },
+    {
+      "epoch": 9.305448963929393,
+      "grad_norm": 0.22265625,
+      "learning_rate": 0.0008138910207214122,
+      "loss": 1.6889,
+      "step": 291000
+    },
+    {
+      "epoch": 9.32143770785367,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.0008135712458429267,
+      "loss": 1.6845,
+      "step": 291500
+    },
+    {
+      "epoch": 9.337426451777949,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.000813251470964441,
+      "loss": 1.6888,
+      "step": 292000
+    },
+    {
+      "epoch": 9.353415195702226,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.0008129316960859555,
+      "loss": 1.6906,
+      "step": 292500
+    },
+    {
+      "epoch": 9.369403939626503,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.00081261192120747,
+      "loss": 1.6885,
+      "step": 293000
+    },
+    {
+      "epoch": 9.38539268355078,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.0008122921463289844,
+      "loss": 1.683,
+      "step": 293500
+    },
+    {
+      "epoch": 9.401381427475057,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0008119723714504988,
+      "loss": 1.6914,
+      "step": 294000
+    },
+    {
+      "epoch": 9.417370171399336,
+      "grad_norm": 0.3203125,
+      "learning_rate": 0.0008116525965720133,
+      "loss": 1.6831,
+      "step": 294500
+    },
+    {
+      "epoch": 9.433358915323613,
+      "grad_norm": 0.2177734375,
+      "learning_rate": 0.0008113328216935278,
+      "loss": 1.6899,
+      "step": 295000
+    },
+    {
+      "epoch": 9.44934765924789,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.0008110130468150422,
+      "loss": 1.6902,
+      "step": 295500
+    },
+    {
+      "epoch": 9.465336403172167,
+      "grad_norm": 0.1962890625,
+      "learning_rate": 0.0008106932719365567,
+      "loss": 1.6851,
+      "step": 296000
+    },
+    {
+      "epoch": 9.481325147096443,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.0008103734970580712,
+      "loss": 1.6909,
+      "step": 296500
+    },
+    {
+      "epoch": 9.497313891020722,
+      "grad_norm": 0.224609375,
+      "learning_rate": 0.0008100537221795856,
+      "loss": 1.683,
+      "step": 297000
+    },
+    {
+      "epoch": 9.513302634945,
+      "grad_norm": 0.3125,
+      "learning_rate": 0.0008097339473011001,
+      "loss": 1.6889,
+      "step": 297500
+    },
+    {
+      "epoch": 9.529291378869276,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0008094141724226145,
+      "loss": 1.6843,
+      "step": 298000
+    },
+    {
+      "epoch": 9.545280122793553,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.0008090943975441289,
+      "loss": 1.689,
+      "step": 298500
+    },
+    {
+      "epoch": 9.56126886671783,
+      "grad_norm": 0.2353515625,
+      "learning_rate": 0.0008087746226656434,
+      "loss": 1.6915,
+      "step": 299000
+    },
+    {
+      "epoch": 9.577257610642109,
+      "grad_norm": 0.19140625,
+      "learning_rate": 0.0008084548477871579,
+      "loss": 1.6844,
+      "step": 299500
+    },
+    {
+      "epoch": 9.593246354566386,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0008081350729086722,
+      "loss": 1.6855,
+      "step": 300000
+    },
+    {
+      "epoch": 9.609235098490663,
+      "grad_norm": 0.3125,
+      "learning_rate": 0.0008078152980301867,
+      "loss": 1.6843,
+      "step": 300500
+    },
+    {
+      "epoch": 9.62522384241494,
+      "grad_norm": 0.349609375,
+      "learning_rate": 0.0008074955231517013,
+      "loss": 1.6825,
+      "step": 301000
+    },
+    {
+      "epoch": 9.641212586339217,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.0008071757482732157,
+      "loss": 1.6854,
+      "step": 301500
+    },
+    {
+      "epoch": 9.657201330263494,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.0008068559733947301,
+      "loss": 1.6852,
+      "step": 302000
+    },
+    {
+      "epoch": 9.673190074187772,
+      "grad_norm": 0.203125,
+      "learning_rate": 0.0008065361985162446,
+      "loss": 1.6841,
+      "step": 302500
+    },
+    {
+      "epoch": 9.68917881811205,
+      "grad_norm": 0.3125,
+      "learning_rate": 0.000806216423637759,
+      "loss": 1.6875,
+      "step": 303000
+    },
+    {
+      "epoch": 9.705167562036326,
+      "grad_norm": 0.2470703125,
+      "learning_rate": 0.0008058966487592735,
+      "loss": 1.6846,
+      "step": 303500
+    },
+    {
+      "epoch": 9.721156305960603,
+      "grad_norm": 0.236328125,
+      "learning_rate": 0.000805576873880788,
+      "loss": 1.6844,
+      "step": 304000
+    },
+    {
+      "epoch": 9.73714504988488,
+      "grad_norm": 0.306640625,
+      "learning_rate": 0.0008052570990023023,
+      "loss": 1.6858,
+      "step": 304500
+    },
+    {
+      "epoch": 9.753133793809159,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0008049373241238168,
+      "loss": 1.689,
+      "step": 305000
+    },
+    {
+      "epoch": 9.769122537733436,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0008046175492453313,
+      "loss": 1.6919,
+      "step": 305500
+    },
+    {
+      "epoch": 9.785111281657713,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.0008042977743668458,
+      "loss": 1.6813,
+      "step": 306000
+    },
+    {
+      "epoch": 9.80110002558199,
+      "grad_norm": 0.1865234375,
+      "learning_rate": 0.0008039779994883602,
+      "loss": 1.6887,
+      "step": 306500
+    },
+    {
+      "epoch": 9.817088769506267,
+      "grad_norm": 0.2236328125,
+      "learning_rate": 0.0008036582246098747,
+      "loss": 1.6877,
+      "step": 307000
+    },
+    {
+      "epoch": 9.833077513430545,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0008033384497313892,
+      "loss": 1.684,
+      "step": 307500
+    },
+    {
+      "epoch": 9.849066257354822,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0008030186748529036,
+      "loss": 1.6893,
+      "step": 308000
+    },
+    {
+      "epoch": 9.8650550012791,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.000802698899974418,
+      "loss": 1.6853,
+      "step": 308500
+    },
+    {
+      "epoch": 9.881043745203376,
+      "grad_norm": 0.23046875,
+      "learning_rate": 0.0008023791250959325,
+      "loss": 1.6852,
+      "step": 309000
+    },
+    {
+      "epoch": 9.897032489127653,
+      "grad_norm": 0.1806640625,
+      "learning_rate": 0.0008020593502174469,
+      "loss": 1.6788,
+      "step": 309500
+    },
+    {
+      "epoch": 9.913021233051932,
+      "grad_norm": 0.31640625,
+      "learning_rate": 0.0008017395753389614,
+      "loss": 1.6856,
+      "step": 310000
+    },
+    {
+      "epoch": 9.929009976976209,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.0008014198004604758,
+      "loss": 1.6889,
+      "step": 310500
+    },
+    {
+      "epoch": 9.944998720900486,
+      "grad_norm": 0.2197265625,
+      "learning_rate": 0.0008011000255819902,
+      "loss": 1.682,
+      "step": 311000
+    },
+    {
+      "epoch": 9.960987464824763,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.0008007802507035048,
+      "loss": 1.6871,
+      "step": 311500
+    },
+    {
+      "epoch": 9.97697620874904,
+      "grad_norm": 0.3359375,
+      "learning_rate": 0.0008004604758250193,
+      "loss": 1.6846,
+      "step": 312000
+    },
+    {
+      "epoch": 9.992964952673319,
+      "grad_norm": 0.318359375,
+      "learning_rate": 0.0008001407009465336,
+      "loss": 1.6835,
+      "step": 312500
+    },
+    {
+      "epoch": 10.0,
+      "eval_loss": 3.3549692630767822,
+      "eval_runtime": 3.7493,
+      "eval_samples_per_second": 133.359,
+      "eval_steps_per_second": 4.267,
+      "step": 312720
+    },
+    {
+      "epoch": 10.008953696597596,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.0007998209260680481,
+      "loss": 1.6791,
+      "step": 313000
+    },
+    {
+      "epoch": 10.024942440521873,
+      "grad_norm": 0.240234375,
+      "learning_rate": 0.0007995011511895626,
+      "loss": 1.6872,
+      "step": 313500
+    },
+    {
+      "epoch": 10.04093118444615,
+      "grad_norm": 0.1884765625,
+      "learning_rate": 0.000799181376311077,
+      "loss": 1.6793,
+      "step": 314000
+    },
+    {
+      "epoch": 10.056919928370426,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.0007988616014325914,
+      "loss": 1.6862,
+      "step": 314500
+    },
+    {
+      "epoch": 10.072908672294705,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.0007985418265541059,
+      "loss": 1.6831,
+      "step": 315000
+    },
+    {
+      "epoch": 10.088897416218982,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.0007982220516756204,
+      "loss": 1.6909,
+      "step": 315500
+    },
+    {
+      "epoch": 10.10488616014326,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0007979022767971348,
+      "loss": 1.6881,
+      "step": 316000
+    },
+    {
+      "epoch": 10.120874904067536,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.0007975825019186494,
+      "loss": 1.6844,
+      "step": 316500
+    },
+    {
+      "epoch": 10.136863647991813,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0007972627270401638,
+      "loss": 1.6934,
+      "step": 317000
+    },
+    {
+      "epoch": 10.152852391916092,
+      "grad_norm": 0.337890625,
+      "learning_rate": 0.0007969429521616782,
+      "loss": 1.6819,
+      "step": 317500
+    },
+    {
+      "epoch": 10.168841135840369,
+      "grad_norm": 0.2021484375,
+      "learning_rate": 0.0007966231772831927,
+      "loss": 1.6797,
+      "step": 318000
+    },
+    {
+      "epoch": 10.184829879764646,
+      "grad_norm": 0.21875,
+      "learning_rate": 0.0007963034024047072,
+      "loss": 1.6824,
+      "step": 318500
+    },
+    {
+      "epoch": 10.200818623688923,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.0007959836275262215,
+      "loss": 1.6842,
+      "step": 319000
+    },
+    {
+      "epoch": 10.2168073676132,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.000795663852647736,
+      "loss": 1.6912,
+      "step": 319500
+    },
+    {
+      "epoch": 10.232796111537478,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.0007953440777692505,
+      "loss": 1.6856,
+      "step": 320000
+    },
+    {
+      "epoch": 10.248784855461755,
+      "grad_norm": 0.248046875,
+      "learning_rate": 0.0007950243028907648,
+      "loss": 1.6885,
+      "step": 320500
+    },
+    {
+      "epoch": 10.264773599386032,
+      "grad_norm": 0.2236328125,
+      "learning_rate": 0.0007947045280122793,
+      "loss": 1.6897,
+      "step": 321000
+    },
+    {
+      "epoch": 10.28076234331031,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.0007943847531337939,
+      "loss": 1.685,
+      "step": 321500
+    },
+    {
+      "epoch": 10.296751087234586,
+      "grad_norm": 0.25,
+      "learning_rate": 0.0007940649782553083,
+      "loss": 1.6896,
+      "step": 322000
+    },
+    {
+      "epoch": 10.312739831158865,
+      "grad_norm": 0.3359375,
+      "learning_rate": 0.0007937452033768228,
+      "loss": 1.6934,
+      "step": 322500
+    },
+    {
+      "epoch": 10.328728575083142,
+      "grad_norm": 0.306640625,
+      "learning_rate": 0.0007934254284983372,
+      "loss": 1.6864,
+      "step": 323000
+    },
+    {
+      "epoch": 10.344717319007419,
+      "grad_norm": 0.212890625,
+      "learning_rate": 0.0007931056536198516,
+      "loss": 1.6819,
+      "step": 323500
+    },
+    {
+      "epoch": 10.360706062931696,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0007927858787413661,
+      "loss": 1.6884,
+      "step": 324000
+    },
+    {
+      "epoch": 10.376694806855973,
+      "grad_norm": 0.2353515625,
+      "learning_rate": 0.0007924661038628806,
+      "loss": 1.6848,
+      "step": 324500
+    },
+    {
+      "epoch": 10.392683550780252,
+      "grad_norm": 0.1865234375,
+      "learning_rate": 0.0007921463289843949,
+      "loss": 1.6846,
+      "step": 325000
+    },
+    {
+      "epoch": 10.408672294704528,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.0007918265541059094,
+      "loss": 1.6831,
+      "step": 325500
+    },
+    {
+      "epoch": 10.424661038628805,
+      "grad_norm": 0.1884765625,
+      "learning_rate": 0.0007915067792274239,
+      "loss": 1.6902,
+      "step": 326000
+    },
+    {
+      "epoch": 10.440649782553082,
+      "grad_norm": 0.326171875,
+      "learning_rate": 0.0007911870043489384,
+      "loss": 1.6766,
+      "step": 326500
+    },
+    {
+      "epoch": 10.45663852647736,
+      "grad_norm": 0.2080078125,
+      "learning_rate": 0.0007908672294704528,
+      "loss": 1.6825,
+      "step": 327000
+    },
+    {
+      "epoch": 10.472627270401638,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.0007905474545919673,
+      "loss": 1.6905,
+      "step": 327500
+    },
+    {
+      "epoch": 10.488616014325915,
+      "grad_norm": 0.2734375,
+      "learning_rate": 0.0007902276797134818,
+      "loss": 1.6839,
+      "step": 328000
+    },
+    {
+      "epoch": 10.504604758250192,
+      "grad_norm": 0.330078125,
+      "learning_rate": 0.0007899079048349962,
+      "loss": 1.691,
+      "step": 328500
+    },
+    {
+      "epoch": 10.520593502174469,
+      "grad_norm": 0.2041015625,
+      "learning_rate": 0.0007895881299565106,
+      "loss": 1.6893,
+      "step": 329000
+    },
+    {
+      "epoch": 10.536582246098746,
+      "grad_norm": 0.3203125,
+      "learning_rate": 0.0007892683550780251,
+      "loss": 1.681,
+      "step": 329500
+    },
+    {
+      "epoch": 10.552570990023025,
+      "grad_norm": 0.2353515625,
+      "learning_rate": 0.0007889485801995395,
+      "loss": 1.6864,
+      "step": 330000
+    },
+    {
+      "epoch": 10.568559733947302,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.000788628805321054,
+      "loss": 1.6912,
+      "step": 330500
+    },
+    {
+      "epoch": 10.584548477871579,
+      "grad_norm": 0.298828125,
+      "learning_rate": 0.0007883090304425684,
+      "loss": 1.6874,
+      "step": 331000
+    },
+    {
+      "epoch": 10.600537221795856,
+      "grad_norm": 0.1904296875,
+      "learning_rate": 0.0007879892555640828,
+      "loss": 1.6895,
+      "step": 331500
+    },
+    {
+      "epoch": 10.616525965720133,
+      "grad_norm": 0.1943359375,
+      "learning_rate": 0.0007876694806855974,
+      "loss": 1.6866,
+      "step": 332000
+    },
+    {
+      "epoch": 10.632514709644411,
+      "grad_norm": 0.302734375,
+      "learning_rate": 0.0007873497058071119,
+      "loss": 1.6891,
+      "step": 332500
+    },
+    {
+      "epoch": 10.648503453568688,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0007870299309286262,
+      "loss": 1.687,
+      "step": 333000
+    },
+    {
+      "epoch": 10.664492197492965,
+      "grad_norm": 0.2470703125,
+      "learning_rate": 0.0007867101560501407,
+      "loss": 1.6833,
+      "step": 333500
+    },
+    {
+      "epoch": 10.680480941417242,
+      "grad_norm": 0.212890625,
+      "learning_rate": 0.0007863903811716552,
+      "loss": 1.6839,
+      "step": 334000
+    },
+    {
+      "epoch": 10.696469685341519,
+      "grad_norm": 0.2734375,
+      "learning_rate": 0.0007860706062931696,
+      "loss": 1.6827,
+      "step": 334500
+    },
+    {
+      "epoch": 10.712458429265796,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.000785750831414684,
+      "loss": 1.6825,
+      "step": 335000
+    },
+    {
+      "epoch": 10.728447173190075,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.0007854310565361985,
+      "loss": 1.6876,
+      "step": 335500
+    },
+    {
+      "epoch": 10.744435917114352,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0007851112816577129,
+      "loss": 1.6868,
+      "step": 336000
+    },
+    {
+      "epoch": 10.760424661038629,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.0007847915067792274,
+      "loss": 1.688,
+      "step": 336500
+    },
+    {
+      "epoch": 10.776413404962906,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.000784471731900742,
+      "loss": 1.6914,
+      "step": 337000
+    },
+    {
+      "epoch": 10.792402148887183,
+      "grad_norm": 0.3671875,
+      "learning_rate": 0.0007841519570222564,
+      "loss": 1.6881,
+      "step": 337500
+    },
+    {
+      "epoch": 10.808390892811461,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.0007838321821437708,
+      "loss": 1.683,
+      "step": 338000
+    },
+    {
+      "epoch": 10.824379636735738,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.0007835124072652853,
+      "loss": 1.6845,
+      "step": 338500
+    },
+    {
+      "epoch": 10.840368380660015,
+      "grad_norm": 0.314453125,
+      "learning_rate": 0.0007831926323867998,
+      "loss": 1.6832,
+      "step": 339000
+    },
+    {
+      "epoch": 10.856357124584292,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.0007828728575083141,
+      "loss": 1.6859,
+      "step": 339500
+    },
+    {
+      "epoch": 10.87234586850857,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.0007825530826298286,
+      "loss": 1.6859,
+      "step": 340000
+    },
+    {
+      "epoch": 10.888334612432848,
+      "grad_norm": 0.34765625,
+      "learning_rate": 0.0007822333077513431,
+      "loss": 1.6909,
+      "step": 340500
+    },
+    {
+      "epoch": 10.904323356357125,
+      "grad_norm": 0.181640625,
+      "learning_rate": 0.0007819135328728575,
+      "loss": 1.6818,
+      "step": 341000
+    },
+    {
+      "epoch": 10.920312100281402,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0007815937579943719,
+      "loss": 1.6803,
+      "step": 341500
+    },
+    {
+      "epoch": 10.936300844205679,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0007812739831158864,
+      "loss": 1.6845,
+      "step": 342000
+    },
+    {
+      "epoch": 10.952289588129956,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.0007809542082374009,
+      "loss": 1.6819,
+      "step": 342500
+    },
+    {
+      "epoch": 10.968278332054235,
+      "grad_norm": 0.2392578125,
+      "learning_rate": 0.0007806344333589154,
+      "loss": 1.6857,
+      "step": 343000
+    },
+    {
+      "epoch": 10.984267075978511,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.0007803146584804298,
+      "loss": 1.6826,
+      "step": 343500
+    },
+    {
+      "epoch": 11.0,
+      "eval_loss": 3.353982448577881,
+      "eval_runtime": 4.5101,
+      "eval_samples_per_second": 110.863,
+      "eval_steps_per_second": 3.548,
+      "step": 343992
+    },
+    {
+      "epoch": 11.000255819902788,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.0007799948836019442,
+      "loss": 1.6828,
+      "step": 344000
+    },
+    {
+      "epoch": 11.016244563827065,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.0007796751087234587,
+      "loss": 1.678,
+      "step": 344500
+    },
+    {
+      "epoch": 11.032233307751342,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0007793553338449732,
+      "loss": 1.6804,
+      "step": 345000
+    },
+    {
+      "epoch": 11.048222051675621,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.0007790355589664875,
+      "loss": 1.6832,
+      "step": 345500
+    },
+    {
+      "epoch": 11.064210795599898,
+      "grad_norm": 0.25,
+      "learning_rate": 0.000778715784088002,
+      "loss": 1.6883,
+      "step": 346000
+    },
+    {
+      "epoch": 11.080199539524175,
+      "grad_norm": 0.333984375,
+      "learning_rate": 0.0007783960092095165,
+      "loss": 1.682,
+      "step": 346500
+    },
+    {
+      "epoch": 11.096188283448452,
+      "grad_norm": 0.197265625,
+      "learning_rate": 0.0007780762343310309,
+      "loss": 1.6849,
+      "step": 347000
+    },
+    {
+      "epoch": 11.112177027372729,
+      "grad_norm": 0.1943359375,
+      "learning_rate": 0.0007777564594525455,
+      "loss": 1.6775,
+      "step": 347500
+    },
+    {
+      "epoch": 11.128165771297008,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0007774366845740599,
+      "loss": 1.6823,
+      "step": 348000
+    },
+    {
+      "epoch": 11.144154515221285,
+      "grad_norm": 0.27734375,
+      "learning_rate": 0.0007771169096955744,
+      "loss": 1.6815,
+      "step": 348500
+    },
+    {
+      "epoch": 11.160143259145562,
+      "grad_norm": 0.2080078125,
+      "learning_rate": 0.0007767971348170888,
+      "loss": 1.6813,
+      "step": 349000
+    },
+    {
+      "epoch": 11.176132003069839,
+      "grad_norm": 0.345703125,
+      "learning_rate": 0.0007764773599386033,
+      "loss": 1.6849,
+      "step": 349500
+    },
+    {
+      "epoch": 11.192120746994116,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0007761575850601177,
+      "loss": 1.6865,
+      "step": 350000
+    },
+    {
+      "epoch": 11.208109490918394,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.0007758378101816321,
+      "loss": 1.6851,
+      "step": 350500
+    },
+    {
+      "epoch": 11.224098234842671,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.0007755180353031466,
+      "loss": 1.6813,
+      "step": 351000
+    },
+    {
+      "epoch": 11.240086978766948,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0007751982604246611,
+      "loss": 1.6835,
+      "step": 351500
+    },
+    {
+      "epoch": 11.256075722691225,
+      "grad_norm": 0.3203125,
+      "learning_rate": 0.0007748784855461754,
+      "loss": 1.6899,
+      "step": 352000
+    },
+    {
+      "epoch": 11.272064466615502,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.00077455871066769,
+      "loss": 1.6819,
+      "step": 352500
+    },
+    {
+      "epoch": 11.28805321053978,
+      "grad_norm": 0.3359375,
+      "learning_rate": 0.0007742389357892045,
+      "loss": 1.6862,
+      "step": 353000
+    },
+    {
+      "epoch": 11.304041954464058,
+      "grad_norm": 0.1875,
+      "learning_rate": 0.0007739191609107189,
+      "loss": 1.6824,
+      "step": 353500
+    },
+    {
+      "epoch": 11.320030698388335,
+      "grad_norm": 0.2080078125,
+      "learning_rate": 0.0007735993860322333,
+      "loss": 1.6812,
+      "step": 354000
+    },
+    {
+      "epoch": 11.336019442312612,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.0007732796111537478,
+      "loss": 1.6827,
+      "step": 354500
+    },
+    {
+      "epoch": 11.352008186236889,
+      "grad_norm": 0.23046875,
+      "learning_rate": 0.0007729598362752622,
+      "loss": 1.6921,
+      "step": 355000
+    },
+    {
+      "epoch": 11.367996930161166,
+      "grad_norm": 0.298828125,
+      "learning_rate": 0.0007726400613967767,
+      "loss": 1.6838,
+      "step": 355500
+    },
+    {
+      "epoch": 11.383985674085444,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0007723202865182911,
+      "loss": 1.686,
+      "step": 356000
+    },
+    {
+      "epoch": 11.399974418009721,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0007720005116398055,
+      "loss": 1.6838,
+      "step": 356500
+    },
+    {
+      "epoch": 11.415963161933998,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.00077168073676132,
+      "loss": 1.6815,
+      "step": 357000
+    },
+    {
+      "epoch": 11.431951905858275,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.0007713609618828346,
+      "loss": 1.6894,
+      "step": 357500
+    },
+    {
+      "epoch": 11.447940649782552,
+      "grad_norm": 0.30078125,
+      "learning_rate": 0.000771041187004349,
+      "loss": 1.6866,
+      "step": 358000
+    },
+    {
+      "epoch": 11.463929393706831,
+      "grad_norm": 0.2890625,
+      "learning_rate": 0.0007707214121258634,
+      "loss": 1.6903,
+      "step": 358500
+    },
+    {
+      "epoch": 11.479918137631108,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.0007704016372473779,
+      "loss": 1.6827,
+      "step": 359000
+    },
+    {
+      "epoch": 11.495906881555385,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.0007700818623688924,
+      "loss": 1.6824,
+      "step": 359500
+    },
+    {
+      "epoch": 11.511895625479662,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.0007697620874904067,
+      "loss": 1.6799,
+      "step": 360000
+    },
+    {
+      "epoch": 11.527884369403939,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0007694423126119212,
+      "loss": 1.6863,
+      "step": 360500
+    },
+    {
+      "epoch": 11.543873113328218,
+      "grad_norm": 0.18359375,
+      "learning_rate": 0.0007691225377334357,
+      "loss": 1.6885,
+      "step": 361000
+    },
+    {
+      "epoch": 11.559861857252494,
+      "grad_norm": 0.19140625,
+      "learning_rate": 0.0007688027628549501,
+      "loss": 1.6851,
+      "step": 361500
+    },
+    {
+      "epoch": 11.575850601176771,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.0007684829879764645,
+      "loss": 1.6768,
+      "step": 362000
+    },
+    {
+      "epoch": 11.591839345101048,
+      "grad_norm": 0.349609375,
+      "learning_rate": 0.000768163213097979,
+      "loss": 1.6832,
+      "step": 362500
+    },
+    {
+      "epoch": 11.607828089025325,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.0007678434382194935,
+      "loss": 1.6874,
+      "step": 363000
+    },
+    {
+      "epoch": 11.623816832949604,
+      "grad_norm": 0.2734375,
+      "learning_rate": 0.000767523663341008,
+      "loss": 1.6839,
+      "step": 363500
+    },
+    {
+      "epoch": 11.639805576873881,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.0007672038884625225,
+      "loss": 1.6875,
+      "step": 364000
+    },
+    {
+      "epoch": 11.655794320798158,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0007668841135840368,
+      "loss": 1.6896,
+      "step": 364500
+    },
+    {
+      "epoch": 11.671783064722435,
+      "grad_norm": 0.2021484375,
+      "learning_rate": 0.0007665643387055513,
+      "loss": 1.6829,
+      "step": 365000
+    },
+    {
+      "epoch": 11.687771808646712,
+      "grad_norm": 0.1484375,
+      "learning_rate": 0.0007662445638270658,
+      "loss": 1.6871,
+      "step": 365500
+    },
+    {
+      "epoch": 11.70376055257099,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.0007659247889485802,
+      "loss": 1.6842,
+      "step": 366000
+    },
+    {
+      "epoch": 11.719749296495268,
+      "grad_norm": 0.2890625,
+      "learning_rate": 0.0007656050140700946,
+      "loss": 1.6819,
+      "step": 366500
+    },
+    {
+      "epoch": 11.735738040419545,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0007652852391916091,
+      "loss": 1.6854,
+      "step": 367000
+    },
+    {
+      "epoch": 11.751726784343822,
+      "grad_norm": 0.2431640625,
+      "learning_rate": 0.0007649654643131235,
+      "loss": 1.6823,
+      "step": 367500
+    },
+    {
+      "epoch": 11.767715528268099,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.0007646456894346381,
+      "loss": 1.6865,
+      "step": 368000
+    },
+    {
+      "epoch": 11.783704272192377,
+      "grad_norm": 0.33984375,
+      "learning_rate": 0.0007643259145561525,
+      "loss": 1.685,
+      "step": 368500
+    },
+    {
+      "epoch": 11.799693016116654,
+      "grad_norm": 0.2158203125,
+      "learning_rate": 0.000764006139677667,
+      "loss": 1.6824,
+      "step": 369000
+    },
+    {
+      "epoch": 11.815681760040931,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.0007636863647991814,
+      "loss": 1.6835,
+      "step": 369500
+    },
+    {
+      "epoch": 11.831670503965208,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0007633665899206959,
+      "loss": 1.6885,
+      "step": 370000
+    },
+    {
+      "epoch": 11.847659247889485,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.0007630468150422103,
+      "loss": 1.6825,
+      "step": 370500
+    },
+    {
+      "epoch": 11.863647991813764,
+      "grad_norm": 0.2490234375,
+      "learning_rate": 0.0007627270401637247,
+      "loss": 1.6861,
+      "step": 371000
+    },
+    {
+      "epoch": 11.87963673573804,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.0007624072652852392,
+      "loss": 1.6862,
+      "step": 371500
+    },
+    {
+      "epoch": 11.895625479662318,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0007620874904067537,
+      "loss": 1.6842,
+      "step": 372000
+    },
+    {
+      "epoch": 11.911614223586595,
+      "grad_norm": 0.22265625,
+      "learning_rate": 0.000761767715528268,
+      "loss": 1.6844,
+      "step": 372500
+    },
+    {
+      "epoch": 11.927602967510872,
+      "grad_norm": 0.21875,
+      "learning_rate": 0.0007614479406497826,
+      "loss": 1.6846,
+      "step": 373000
+    },
+    {
+      "epoch": 11.94359171143515,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0007611281657712971,
+      "loss": 1.678,
+      "step": 373500
+    },
+    {
+      "epoch": 11.959580455359427,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.0007608083908928115,
+      "loss": 1.6817,
+      "step": 374000
+    },
+    {
+      "epoch": 11.975569199283704,
+      "grad_norm": 0.1923828125,
+      "learning_rate": 0.000760488616014326,
+      "loss": 1.6818,
+      "step": 374500
+    },
+    {
+      "epoch": 11.991557943207981,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.0007601688411358404,
+      "loss": 1.685,
+      "step": 375000
+    },
+    {
+      "epoch": 12.0,
+      "eval_loss": 3.349221706390381,
+      "eval_runtime": 3.4957,
+      "eval_samples_per_second": 143.035,
+      "eval_steps_per_second": 4.577,
+      "step": 375264
+    },
+    {
+      "epoch": 12.007546687132258,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0007598490662573548,
+      "loss": 1.6838,
+      "step": 375500
+    },
+    {
+      "epoch": 12.023535431056537,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0007595292913788693,
+      "loss": 1.6833,
+      "step": 376000
+    },
+    {
+      "epoch": 12.039524174980814,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.0007592095165003838,
+      "loss": 1.6824,
+      "step": 376500
+    },
+    {
+      "epoch": 12.055512918905091,
+      "grad_norm": 0.27734375,
+      "learning_rate": 0.0007588897416218981,
+      "loss": 1.686,
+      "step": 377000
+    },
+    {
+      "epoch": 12.071501662829368,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0007585699667434126,
+      "loss": 1.6866,
+      "step": 377500
+    },
+    {
+      "epoch": 12.087490406753645,
+      "grad_norm": 0.220703125,
+      "learning_rate": 0.0007582501918649271,
+      "loss": 1.682,
+      "step": 378000
+    },
+    {
+      "epoch": 12.103479150677924,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.0007579304169864416,
+      "loss": 1.6839,
+      "step": 378500
+    },
+    {
+      "epoch": 12.1194678946022,
+      "grad_norm": 0.361328125,
+      "learning_rate": 0.000757610642107956,
+      "loss": 1.6817,
+      "step": 379000
+    },
+    {
+      "epoch": 12.135456638526477,
+      "grad_norm": 0.16015625,
+      "learning_rate": 0.0007572908672294705,
+      "loss": 1.6838,
+      "step": 379500
+    },
+    {
+      "epoch": 12.151445382450754,
+      "grad_norm": 0.24609375,
+      "learning_rate": 0.000756971092350985,
+      "loss": 1.6865,
+      "step": 380000
+    },
+    {
+      "epoch": 12.167434126375031,
+      "grad_norm": 0.1669921875,
+      "learning_rate": 0.0007566513174724994,
+      "loss": 1.6826,
+      "step": 380500
+    },
+    {
+      "epoch": 12.18342287029931,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0007563315425940138,
+      "loss": 1.6847,
+      "step": 381000
+    },
+    {
+      "epoch": 12.199411614223587,
+      "grad_norm": 0.373046875,
+      "learning_rate": 0.0007560117677155283,
+      "loss": 1.6853,
+      "step": 381500
+    },
+    {
+      "epoch": 12.215400358147864,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.0007556919928370427,
+      "loss": 1.6827,
+      "step": 382000
+    },
+    {
+      "epoch": 12.231389102072141,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.0007553722179585572,
+      "loss": 1.6862,
+      "step": 382500
+    },
+    {
+      "epoch": 12.247377845996418,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.0007550524430800716,
+      "loss": 1.68,
+      "step": 383000
+    },
+    {
+      "epoch": 12.263366589920697,
+      "grad_norm": 0.27734375,
+      "learning_rate": 0.0007547326682015861,
+      "loss": 1.6812,
+      "step": 383500
+    },
+    {
+      "epoch": 12.279355333844974,
+      "grad_norm": 0.259765625,
+      "learning_rate": 0.0007544128933231006,
+      "loss": 1.6786,
+      "step": 384000
+    },
+    {
+      "epoch": 12.29534407776925,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.0007540931184446151,
+      "loss": 1.6809,
+      "step": 384500
+    },
+    {
+      "epoch": 12.311332821693528,
+      "grad_norm": 0.1826171875,
+      "learning_rate": 0.0007537733435661294,
+      "loss": 1.6819,
+      "step": 385000
+    },
+    {
+      "epoch": 12.327321565617805,
+      "grad_norm": 0.236328125,
+      "learning_rate": 0.0007534535686876439,
+      "loss": 1.6816,
+      "step": 385500
+    },
+    {
+      "epoch": 12.343310309542082,
+      "grad_norm": 0.2333984375,
+      "learning_rate": 0.0007531337938091584,
+      "loss": 1.6863,
+      "step": 386000
+    },
+    {
+      "epoch": 12.35929905346636,
+      "grad_norm": 0.2431640625,
+      "learning_rate": 0.0007528140189306728,
+      "loss": 1.6755,
+      "step": 386500
+    },
+    {
+      "epoch": 12.375287797390637,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0007524942440521872,
+      "loss": 1.6845,
+      "step": 387000
+    },
+    {
+      "epoch": 12.391276541314914,
+      "grad_norm": 0.4609375,
+      "learning_rate": 0.0007521744691737017,
+      "loss": 1.6853,
+      "step": 387500
+    },
+    {
+      "epoch": 12.407265285239191,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.0007518546942952161,
+      "loss": 1.692,
+      "step": 388000
+    },
+    {
+      "epoch": 12.423254029163468,
+      "grad_norm": 0.314453125,
+      "learning_rate": 0.0007515349194167307,
+      "loss": 1.6783,
+      "step": 388500
+    },
+    {
+      "epoch": 12.439242773087747,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0007512151445382452,
+      "loss": 1.6829,
+      "step": 389000
+    },
+    {
+      "epoch": 12.455231517012024,
+      "grad_norm": 0.361328125,
+      "learning_rate": 0.0007508953696597595,
+      "loss": 1.6763,
+      "step": 389500
+    },
+    {
+      "epoch": 12.4712202609363,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.000750575594781274,
+      "loss": 1.6793,
+      "step": 390000
+    },
+    {
+      "epoch": 12.487209004860578,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0007502558199027885,
+      "loss": 1.6837,
+      "step": 390500
+    },
+    {
+      "epoch": 12.503197748784855,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.000749936045024303,
+      "loss": 1.6839,
+      "step": 391000
+    },
+    {
+      "epoch": 12.519186492709133,
+      "grad_norm": 0.205078125,
+      "learning_rate": 0.0007496162701458173,
+      "loss": 1.6836,
+      "step": 391500
+    },
+    {
+      "epoch": 12.53517523663341,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.0007492964952673318,
+      "loss": 1.6856,
+      "step": 392000
+    },
+    {
+      "epoch": 12.551163980557687,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0007489767203888463,
+      "loss": 1.6828,
+      "step": 392500
+    },
+    {
+      "epoch": 12.567152724481964,
+      "grad_norm": 0.37109375,
+      "learning_rate": 0.0007486569455103606,
+      "loss": 1.6877,
+      "step": 393000
+    },
+    {
+      "epoch": 12.583141468406241,
+      "grad_norm": 0.1982421875,
+      "learning_rate": 0.0007483371706318752,
+      "loss": 1.6764,
+      "step": 393500
+    },
+    {
+      "epoch": 12.59913021233052,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.0007480173957533897,
+      "loss": 1.6825,
+      "step": 394000
+    },
+    {
+      "epoch": 12.615118956254797,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0007476976208749041,
+      "loss": 1.6918,
+      "step": 394500
+    },
+    {
+      "epoch": 12.631107700179074,
+      "grad_norm": 0.328125,
+      "learning_rate": 0.0007473778459964186,
+      "loss": 1.684,
+      "step": 395000
+    },
+    {
+      "epoch": 12.64709644410335,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.000747058071117933,
+      "loss": 1.6831,
+      "step": 395500
+    },
+    {
+      "epoch": 12.663085188027628,
+      "grad_norm": 0.2255859375,
+      "learning_rate": 0.0007467382962394474,
+      "loss": 1.6866,
+      "step": 396000
+    },
+    {
+      "epoch": 12.679073931951907,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.0007464185213609619,
+      "loss": 1.6822,
+      "step": 396500
+    },
+    {
+      "epoch": 12.695062675876184,
+      "grad_norm": 0.2021484375,
+      "learning_rate": 0.0007460987464824764,
+      "loss": 1.6853,
+      "step": 397000
+    },
+    {
+      "epoch": 12.71105141980046,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.0007457789716039907,
+      "loss": 1.6853,
+      "step": 397500
+    },
+    {
+      "epoch": 12.727040163724737,
+      "grad_norm": 0.208984375,
+      "learning_rate": 0.0007454591967255052,
+      "loss": 1.6828,
+      "step": 398000
+    },
+    {
+      "epoch": 12.743028907649014,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.0007451394218470197,
+      "loss": 1.6884,
+      "step": 398500
+    },
+    {
+      "epoch": 12.759017651573293,
+      "grad_norm": 0.2158203125,
+      "learning_rate": 0.0007448196469685342,
+      "loss": 1.681,
+      "step": 399000
+    },
+    {
+      "epoch": 12.77500639549757,
+      "grad_norm": 0.279296875,
+      "learning_rate": 0.0007444998720900486,
+      "loss": 1.6813,
+      "step": 399500
+    },
+    {
+      "epoch": 12.790995139421847,
+      "grad_norm": 0.263671875,
+      "learning_rate": 0.0007441800972115631,
+      "loss": 1.6818,
+      "step": 400000
+    },
+    {
+      "epoch": 12.806983883346124,
+      "grad_norm": 0.1806640625,
+      "learning_rate": 0.0007438603223330776,
+      "loss": 1.6825,
+      "step": 400500
+    },
+    {
+      "epoch": 12.822972627270401,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.000743540547454592,
+      "loss": 1.6822,
+      "step": 401000
+    },
+    {
+      "epoch": 12.83896137119468,
+      "grad_norm": 0.2119140625,
+      "learning_rate": 0.0007432207725761064,
+      "loss": 1.6798,
+      "step": 401500
+    },
+    {
+      "epoch": 12.854950115118957,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.0007429009976976209,
+      "loss": 1.6818,
+      "step": 402000
+    },
+    {
+      "epoch": 12.870938859043234,
+      "grad_norm": 0.23046875,
+      "learning_rate": 0.0007425812228191353,
+      "loss": 1.6838,
+      "step": 402500
+    },
+    {
+      "epoch": 12.88692760296751,
+      "grad_norm": 0.33203125,
+      "learning_rate": 0.0007422614479406498,
+      "loss": 1.6819,
+      "step": 403000
     },
     {
-      "epoch": 1.0224948875255624,
-      "grad_norm": 0.2255859375,
-      "learning_rate": 0.0009795501022494889,
-      "loss": 3.6938,
-      "step": 1000
+      "epoch": 12.902916346891788,
+      "grad_norm": 0.24609375,
+      "learning_rate": 0.0007419416730621642,
+      "loss": 1.6857,
+      "step": 403500
     },
     {
-      "epoch": 1.5337423312883436,
-      "grad_norm": 0.263671875,
-      "learning_rate": 0.0009693251533742331,
-      "loss": 3.488,
-      "step": 1500
+      "epoch": 12.918905090816066,
+      "grad_norm": 0.2275390625,
+      "learning_rate": 0.0007416218981836787,
+      "loss": 1.6831,
+      "step": 404000
     },
     {
-      "epoch": 2.0,
-      "eval_loss": 3.214155912399292,
-      "eval_runtime": 0.95,
-      "eval_samples_per_second": 526.301,
-      "eval_steps_per_second": 1.053,
-      "step": 1956
+      "epoch": 12.934893834740343,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.0007413021233051932,
+      "loss": 1.6899,
+      "step": 404500
     },
     {
-      "epoch": 2.044989775051125,
-      "grad_norm": 0.224609375,
-      "learning_rate": 0.0009591002044989775,
-      "loss": 3.2879,
-      "step": 2000
+      "epoch": 12.95088257866462,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.0007409823484267077,
+      "loss": 1.6884,
+      "step": 405000
     },
     {
-      "epoch": 2.556237218813906,
-      "grad_norm": 0.2373046875,
-      "learning_rate": 0.0009488752556237219,
-      "loss": 3.1641,
-      "step": 2500
+      "epoch": 12.966871322588897,
+      "grad_norm": 0.2734375,
+      "learning_rate": 0.000740662573548222,
+      "loss": 1.6842,
+      "step": 405500
     },
     {
-      "epoch": 3.0,
-      "eval_loss": 3.0714735984802246,
-      "eval_runtime": 0.9223,
-      "eval_samples_per_second": 542.135,
-      "eval_steps_per_second": 1.084,
-      "step": 2934
+      "epoch": 12.982860066513174,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0007403427986697365,
+      "loss": 1.6833,
+      "step": 406000
     },
     {
-      "epoch": 3.067484662576687,
-      "grad_norm": 0.29296875,
-      "learning_rate": 0.0009386503067484663,
-      "loss": 3.0912,
-      "step": 3000
+      "epoch": 12.998848810437451,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.000740023023791251,
+      "loss": 1.6797,
+      "step": 406500
     },
     {
-      "epoch": 3.5787321063394684,
-      "grad_norm": 0.255859375,
-      "learning_rate": 0.0009284253578732107,
-      "loss": 3.0323,
-      "step": 3500
+      "epoch": 13.0,
+      "eval_loss": 3.3510327339172363,
+      "eval_runtime": 4.0919,
+      "eval_samples_per_second": 122.192,
+      "eval_steps_per_second": 3.91,
+      "step": 406536
     },
     {
-      "epoch": 4.0,
-      "eval_loss": 2.990953207015991,
-      "eval_runtime": 1.0538,
-      "eval_samples_per_second": 474.475,
-      "eval_steps_per_second": 0.949,
-      "step": 3912
+      "epoch": 13.01483755436173,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0007397032489127654,
+      "loss": 1.6775,
+      "step": 407000
+    },
+    {
+      "epoch": 13.030826298286007,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.0007393834740342799,
+      "loss": 1.6844,
+      "step": 407500
+    },
+    {
+      "epoch": 13.046815042210284,
+      "grad_norm": 0.1953125,
+      "learning_rate": 0.0007390636991557943,
+      "loss": 1.6797,
+      "step": 408000
+    },
+    {
+      "epoch": 13.06280378613456,
+      "grad_norm": 0.2578125,
+      "learning_rate": 0.0007387439242773087,
+      "loss": 1.6858,
+      "step": 408500
+    },
+    {
+      "epoch": 13.078792530058838,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0007384241493988233,
+      "loss": 1.6788,
+      "step": 409000
+    },
+    {
+      "epoch": 13.094781273983116,
+      "grad_norm": 0.376953125,
+      "learning_rate": 0.0007381043745203378,
+      "loss": 1.6788,
+      "step": 409500
+    },
+    {
+      "epoch": 13.110770017907393,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0007377845996418521,
+      "loss": 1.6763,
+      "step": 410000
+    },
+    {
+      "epoch": 13.12675876183167,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.0007374648247633666,
+      "loss": 1.6797,
+      "step": 410500
+    },
+    {
+      "epoch": 13.142747505755947,
+      "grad_norm": 0.341796875,
+      "learning_rate": 0.0007371450498848811,
+      "loss": 1.6782,
+      "step": 411000
+    },
+    {
+      "epoch": 13.158736249680224,
+      "grad_norm": 0.2001953125,
+      "learning_rate": 0.0007368252750063956,
+      "loss": 1.6809,
+      "step": 411500
+    },
+    {
+      "epoch": 13.174724993604503,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0007365055001279099,
+      "loss": 1.6839,
+      "step": 412000
+    },
+    {
+      "epoch": 13.19071373752878,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0007361857252494244,
+      "loss": 1.6896,
+      "step": 412500
+    },
+    {
+      "epoch": 13.206702481453057,
+      "grad_norm": 0.234375,
+      "learning_rate": 0.0007358659503709389,
+      "loss": 1.6854,
+      "step": 413000
+    },
+    {
+      "epoch": 13.222691225377334,
+      "grad_norm": 0.2294921875,
+      "learning_rate": 0.0007355461754924533,
+      "loss": 1.6822,
+      "step": 413500
+    },
+    {
+      "epoch": 13.23867996930161,
+      "grad_norm": 0.216796875,
+      "learning_rate": 0.0007352264006139677,
+      "loss": 1.6894,
+      "step": 414000
+    },
+    {
+      "epoch": 13.25466871322589,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0007349066257354823,
+      "loss": 1.6834,
+      "step": 414500
+    },
+    {
+      "epoch": 13.270657457150167,
+      "grad_norm": 0.2216796875,
+      "learning_rate": 0.0007345868508569967,
+      "loss": 1.683,
+      "step": 415000
+    },
+    {
+      "epoch": 13.286646201074443,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0007342670759785112,
+      "loss": 1.6798,
+      "step": 415500
+    },
+    {
+      "epoch": 13.30263494499872,
+      "grad_norm": 0.1962890625,
+      "learning_rate": 0.0007339473011000256,
+      "loss": 1.6759,
+      "step": 416000
+    },
+    {
+      "epoch": 13.318623688922997,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.00073362752622154,
+      "loss": 1.6853,
+      "step": 416500
+    },
+    {
+      "epoch": 13.334612432847276,
+      "grad_norm": 0.328125,
+      "learning_rate": 0.0007333077513430545,
+      "loss": 1.6799,
+      "step": 417000
+    },
+    {
+      "epoch": 13.350601176771553,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.000732987976464569,
+      "loss": 1.6835,
+      "step": 417500
+    },
+    {
+      "epoch": 13.36658992069583,
+      "grad_norm": 0.322265625,
+      "learning_rate": 0.0007326682015860833,
+      "loss": 1.6812,
+      "step": 418000
+    },
+    {
+      "epoch": 13.382578664620107,
+      "grad_norm": 0.25,
+      "learning_rate": 0.0007323484267075978,
+      "loss": 1.6857,
+      "step": 418500
+    },
+    {
+      "epoch": 13.398567408544384,
+      "grad_norm": 0.322265625,
+      "learning_rate": 0.0007320286518291123,
+      "loss": 1.6803,
+      "step": 419000
+    },
+    {
+      "epoch": 13.414556152468663,
+      "grad_norm": 0.310546875,
+      "learning_rate": 0.0007317088769506268,
+      "loss": 1.6835,
+      "step": 419500
+    },
+    {
+      "epoch": 13.43054489639294,
+      "grad_norm": 0.2412109375,
+      "learning_rate": 0.0007313891020721413,
+      "loss": 1.6811,
+      "step": 420000
+    },
+    {
+      "epoch": 13.446533640317217,
+      "grad_norm": 0.16015625,
+      "learning_rate": 0.0007310693271936557,
+      "loss": 1.6863,
+      "step": 420500
+    },
+    {
+      "epoch": 13.462522384241494,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0007307495523151701,
+      "loss": 1.6766,
+      "step": 421000
+    },
+    {
+      "epoch": 13.47851112816577,
+      "grad_norm": 0.2021484375,
+      "learning_rate": 0.0007304297774366846,
+      "loss": 1.6843,
+      "step": 421500
+    },
+    {
+      "epoch": 13.49449987209005,
+      "grad_norm": 0.2470703125,
+      "learning_rate": 0.0007301100025581991,
+      "loss": 1.6821,
+      "step": 422000
+    },
+    {
+      "epoch": 13.510488616014326,
+      "grad_norm": 0.1845703125,
+      "learning_rate": 0.0007297902276797135,
+      "loss": 1.6884,
+      "step": 422500
+    },
+    {
+      "epoch": 13.526477359938603,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.0007294704528012279,
+      "loss": 1.6868,
+      "step": 423000
+    },
+    {
+      "epoch": 13.54246610386288,
+      "grad_norm": 0.3125,
+      "learning_rate": 0.0007291506779227424,
+      "loss": 1.6794,
+      "step": 423500
+    },
+    {
+      "epoch": 13.558454847787157,
+      "grad_norm": 0.23046875,
+      "learning_rate": 0.0007288309030442569,
+      "loss": 1.6877,
+      "step": 424000
     },
     {
-      "epoch": 4.08997955010225,
+      "epoch": 13.574443591711436,
       "grad_norm": 0.259765625,
-      "learning_rate": 0.000918200408997955,
-      "loss": 2.9912,
-      "step": 4000
+      "learning_rate": 0.0007285111281657713,
+      "loss": 1.6831,
+      "step": 424500
     },
     {
-      "epoch": 4.601226993865031,
-      "grad_norm": 0.337890625,
-      "learning_rate": 0.0009079754601226994,
-      "loss": 2.9553,
-      "step": 4500
+      "epoch": 13.590432335635713,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.0007281913532872858,
+      "loss": 1.6905,
+      "step": 425000
     },
     {
-      "epoch": 5.0,
-      "eval_loss": 2.9412269592285156,
-      "eval_runtime": 1.0599,
-      "eval_samples_per_second": 471.73,
-      "eval_steps_per_second": 0.943,
-      "step": 4890
+      "epoch": 13.60642107955999,
+      "grad_norm": 0.2001953125,
+      "learning_rate": 0.0007278715784088003,
+      "loss": 1.6816,
+      "step": 425500
     },
     {
-      "epoch": 5.112474437627812,
+      "epoch": 13.622409823484267,
       "grad_norm": 0.26953125,
-      "learning_rate": 0.0008977505112474438,
-      "loss": 2.9255,
-      "step": 5000
+      "learning_rate": 0.0007275518035303147,
+      "loss": 1.6781,
+      "step": 426000
     },
     {
-      "epoch": 5.623721881390593,
-      "grad_norm": 0.25390625,
-      "learning_rate": 0.0008875255623721882,
-      "loss": 2.8991,
-      "step": 5500
+      "epoch": 13.638398567408544,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0007272320286518291,
+      "loss": 1.6826,
+      "step": 426500
     },
     {
-      "epoch": 6.0,
-      "eval_loss": 2.905097723007202,
-      "eval_runtime": 1.0476,
-      "eval_samples_per_second": 477.298,
-      "eval_steps_per_second": 0.955,
-      "step": 5868
+      "epoch": 13.654387311332822,
+      "grad_norm": 0.30078125,
+      "learning_rate": 0.0007269122537733436,
+      "loss": 1.6788,
+      "step": 427000
     },
     {
-      "epoch": 6.134969325153374,
-      "grad_norm": 0.3203125,
-      "learning_rate": 0.0008773006134969325,
-      "loss": 2.8794,
-      "step": 6000
+      "epoch": 13.6703760552571,
+      "grad_norm": 0.30859375,
+      "learning_rate": 0.000726592478894858,
+      "loss": 1.6839,
+      "step": 427500
     },
     {
-      "epoch": 6.6462167689161555,
-      "grad_norm": 0.29296875,
-      "learning_rate": 0.0008670756646216768,
-      "loss": 2.8577,
-      "step": 6500
+      "epoch": 13.686364799181376,
+      "grad_norm": 0.185546875,
+      "learning_rate": 0.0007262727040163725,
+      "loss": 1.6901,
+      "step": 428000
     },
     {
-      "epoch": 7.0,
-      "eval_loss": 2.8703627586364746,
-      "eval_runtime": 1.0515,
-      "eval_samples_per_second": 475.508,
-      "eval_steps_per_second": 0.951,
-      "step": 6846
+      "epoch": 13.702353543105653,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.0007259529291378869,
+      "loss": 1.6824,
+      "step": 428500
+    },
+    {
+      "epoch": 13.71834228702993,
+      "grad_norm": 0.275390625,
+      "learning_rate": 0.0007256331542594013,
+      "loss": 1.6785,
+      "step": 429000
     },
     {
-      "epoch": 7.157464212678937,
+      "epoch": 13.734331030954209,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0007253133793809159,
+      "loss": 1.6809,
+      "step": 429500
+    },
+    {
+      "epoch": 13.750319774878486,
+      "grad_norm": 0.41796875,
+      "learning_rate": 0.0007249936045024304,
+      "loss": 1.6818,
+      "step": 430000
+    },
+    {
+      "epoch": 13.766308518802763,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0007246738296239447,
+      "loss": 1.6815,
+      "step": 430500
+    },
+    {
+      "epoch": 13.78229726272704,
+      "grad_norm": 0.193359375,
+      "learning_rate": 0.0007243540547454592,
+      "loss": 1.6835,
+      "step": 431000
+    },
+    {
+      "epoch": 13.798286006651317,
+      "grad_norm": 0.294921875,
+      "learning_rate": 0.0007240342798669737,
+      "loss": 1.6832,
+      "step": 431500
+    },
+    {
+      "epoch": 13.814274750575596,
+      "grad_norm": 0.2392578125,
+      "learning_rate": 0.0007237145049884881,
+      "loss": 1.6877,
+      "step": 432000
+    },
+    {
+      "epoch": 13.830263494499873,
+      "grad_norm": 0.1875,
+      "learning_rate": 0.0007233947301100025,
+      "loss": 1.6829,
+      "step": 432500
+    },
+    {
+      "epoch": 13.84625223842415,
       "grad_norm": 0.29296875,
-      "learning_rate": 0.0008568507157464213,
-      "loss": 2.8435,
-      "step": 7000
+      "learning_rate": 0.000723074955231517,
+      "loss": 1.6882,
+      "step": 433000
     },
     {
-      "epoch": 7.668711656441718,
-      "grad_norm": 0.30078125,
-      "learning_rate": 0.0008466257668711657,
-      "loss": 2.8289,
-      "step": 7500
+      "epoch": 13.862240982348426,
+      "grad_norm": 0.228515625,
+      "learning_rate": 0.0007227551803530315,
+      "loss": 1.685,
+      "step": 433500
     },
     {
-      "epoch": 8.0,
-      "eval_loss": 2.847052812576294,
-      "eval_runtime": 1.0464,
-      "eval_samples_per_second": 477.826,
-      "eval_steps_per_second": 0.956,
-      "step": 7824
+      "epoch": 13.878229726272703,
+      "grad_norm": 0.291015625,
+      "learning_rate": 0.0007224354054745459,
+      "loss": 1.682,
+      "step": 434000
     },
     {
-      "epoch": 8.1799591002045,
-      "grad_norm": 0.3046875,
-      "learning_rate": 0.0008364008179959101,
-      "loss": 2.8131,
-      "step": 8000
+      "epoch": 13.894218470196982,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0007221156305960603,
+      "loss": 1.6885,
+      "step": 434500
     },
     {
-      "epoch": 8.69120654396728,
-      "grad_norm": 0.359375,
-      "learning_rate": 0.0008261758691206544,
-      "loss": 2.8022,
-      "step": 8500
+      "epoch": 13.91020721412126,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0007217958557175749,
+      "loss": 1.6838,
+      "step": 435000
     },
     {
-      "epoch": 9.0,
-      "eval_loss": 2.8293135166168213,
-      "eval_runtime": 0.9764,
-      "eval_samples_per_second": 512.074,
-      "eval_steps_per_second": 1.024,
-      "step": 8802
+      "epoch": 13.926195958045536,
+      "grad_norm": 0.2177734375,
+      "learning_rate": 0.0007214760808390893,
+      "loss": 1.6861,
+      "step": 435500
+    },
+    {
+      "epoch": 13.942184701969813,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0007211563059606038,
+      "loss": 1.6829,
+      "step": 436000
+    },
+    {
+      "epoch": 13.95817344589409,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.0007208365310821183,
+      "loss": 1.6844,
+      "step": 436500
+    },
+    {
+      "epoch": 13.974162189818369,
+      "grad_norm": 0.25390625,
+      "learning_rate": 0.0007205167562036326,
+      "loss": 1.6825,
+      "step": 437000
+    },
+    {
+      "epoch": 13.990150933742646,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.0007201969813251471,
+      "loss": 1.6818,
+      "step": 437500
+    },
+    {
+      "epoch": 14.0,
+      "eval_loss": 3.35012149810791,
+      "eval_runtime": 3.6622,
+      "eval_samples_per_second": 136.532,
+      "eval_steps_per_second": 4.369,
+      "step": 437808
+    },
+    {
+      "epoch": 14.006139677666923,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0007198772064466616,
+      "loss": 1.6854,
+      "step": 438000
     },
     {
-      "epoch": 9.202453987730062,
+      "epoch": 14.0221284215912,
       "grad_norm": 0.310546875,
-      "learning_rate": 0.0008159509202453987,
-      "loss": 2.7914,
-      "step": 9000
+      "learning_rate": 0.000719557431568176,
+      "loss": 1.6792,
+      "step": 438500
     },
     {
-      "epoch": 9.713701431492842,
-      "grad_norm": 0.314453125,
-      "learning_rate": 0.0008057259713701431,
-      "loss": 2.7813,
-      "step": 9500
+      "epoch": 14.038117165515477,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0007192376566896904,
+      "loss": 1.6824,
+      "step": 439000
     },
     {
-      "epoch": 10.0,
-      "eval_loss": 2.8122880458831787,
-      "eval_runtime": 1.0064,
-      "eval_samples_per_second": 496.84,
-      "eval_steps_per_second": 0.994,
-      "step": 9780
+      "epoch": 14.054105909439754,
+      "grad_norm": 0.212890625,
+      "learning_rate": 0.0007189178818112049,
+      "loss": 1.6811,
+      "step": 439500
     },
     {
-      "epoch": 10.224948875255624,
-      "grad_norm": 0.296875,
-      "learning_rate": 0.0007955010224948876,
-      "loss": 2.7682,
-      "step": 10000
+      "epoch": 14.070094653364032,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.0007185981069327194,
+      "loss": 1.6773,
+      "step": 440000
+    },
+    {
+      "epoch": 14.08608339728831,
+      "grad_norm": 0.27734375,
+      "learning_rate": 0.0007182783320542339,
+      "loss": 1.6765,
+      "step": 440500
+    },
+    {
+      "epoch": 14.102072141212586,
+      "grad_norm": 0.2890625,
+      "learning_rate": 0.0007179585571757483,
+      "loss": 1.6827,
+      "step": 441000
+    },
+    {
+      "epoch": 14.118060885136863,
+      "grad_norm": 0.2060546875,
+      "learning_rate": 0.0007176387822972627,
+      "loss": 1.6818,
+      "step": 441500
     },
     {
-      "epoch": 10.736196319018404,
+      "epoch": 14.13404962906114,
       "grad_norm": 0.328125,
-      "learning_rate": 0.0007852760736196319,
-      "loss": 2.7607,
-      "step": 10500
+      "learning_rate": 0.0007173190074187772,
+      "loss": 1.685,
+      "step": 442000
     },
     {
-      "epoch": 11.0,
-      "eval_loss": 2.8023903369903564,
-      "eval_runtime": 1.0601,
-      "eval_samples_per_second": 471.642,
-      "eval_steps_per_second": 0.943,
-      "step": 10758
+      "epoch": 14.150038372985419,
+      "grad_norm": 0.333984375,
+      "learning_rate": 0.0007169992325402917,
+      "loss": 1.6818,
+      "step": 442500
     },
     {
-      "epoch": 11.247443762781186,
-      "grad_norm": 0.294921875,
-      "learning_rate": 0.0007750511247443763,
-      "loss": 2.7527,
-      "step": 11000
+      "epoch": 14.166027116909696,
+      "grad_norm": 0.23046875,
+      "learning_rate": 0.000716679457661806,
+      "loss": 1.6883,
+      "step": 443000
     },
     {
-      "epoch": 11.758691206543967,
-      "grad_norm": 0.3046875,
-      "learning_rate": 0.0007648261758691206,
-      "loss": 2.7443,
-      "step": 11500
+      "epoch": 14.182015860833973,
+      "grad_norm": 0.2890625,
+      "learning_rate": 0.0007163596827833205,
+      "loss": 1.6839,
+      "step": 443500
     },
     {
-      "epoch": 12.0,
-      "eval_loss": 2.7869932651519775,
-      "eval_runtime": 1.0222,
-      "eval_samples_per_second": 489.149,
-      "eval_steps_per_second": 0.978,
-      "step": 11736
+      "epoch": 14.19800460475825,
+      "grad_norm": 0.1982421875,
+      "learning_rate": 0.000716039907904835,
+      "loss": 1.6816,
+      "step": 444000
+    },
+    {
+      "epoch": 14.213993348682527,
+      "grad_norm": 0.2373046875,
+      "learning_rate": 0.0007157201330263495,
+      "loss": 1.6867,
+      "step": 444500
     },
     {
-      "epoch": 12.269938650306749,
+      "epoch": 14.229982092606805,
       "grad_norm": 0.298828125,
-      "learning_rate": 0.000754601226993865,
-      "loss": 2.7366,
-      "step": 12000
+      "learning_rate": 0.000715400358147864,
+      "loss": 1.6769,
+      "step": 445000
     },
     {
-      "epoch": 12.781186094069529,
-      "grad_norm": 0.34375,
-      "learning_rate": 0.0007443762781186095,
-      "loss": 2.7313,
-      "step": 12500
+      "epoch": 14.245970836531082,
+      "grad_norm": 0.271484375,
+      "learning_rate": 0.0007150805832693784,
+      "loss": 1.6847,
+      "step": 445500
     },
     {
-      "epoch": 13.0,
-      "eval_loss": 2.785067081451416,
-      "eval_runtime": 1.051,
-      "eval_samples_per_second": 475.716,
-      "eval_steps_per_second": 0.951,
-      "step": 12714
+      "epoch": 14.26195958045536,
+      "grad_norm": 0.27734375,
+      "learning_rate": 0.0007147608083908929,
+      "loss": 1.6789,
+      "step": 446000
     },
     {
-      "epoch": 13.292433537832311,
-      "grad_norm": 0.33203125,
-      "learning_rate": 0.0007341513292433538,
-      "loss": 2.7223,
-      "step": 13000
+      "epoch": 14.277948324379636,
+      "grad_norm": 0.296875,
+      "learning_rate": 0.0007144410335124073,
+      "loss": 1.6829,
+      "step": 446500
     },
     {
-      "epoch": 13.803680981595091,
-      "grad_norm": 0.33203125,
-      "learning_rate": 0.0007239263803680982,
-      "loss": 2.7179,
-      "step": 13500
+      "epoch": 14.293937068303913,
+      "grad_norm": 0.21484375,
+      "learning_rate": 0.0007141212586339217,
+      "loss": 1.6794,
+      "step": 447000
     },
     {
-      "epoch": 14.0,
-      "eval_loss": 2.772521495819092,
-      "eval_runtime": 1.0385,
-      "eval_samples_per_second": 481.485,
-      "eval_steps_per_second": 0.963,
-      "step": 13692
+      "epoch": 14.309925812228192,
+      "grad_norm": 0.212890625,
+      "learning_rate": 0.0007138014837554362,
+      "loss": 1.6834,
+      "step": 447500
     },
     {
-      "epoch": 14.314928425357873,
-      "grad_norm": 0.330078125,
-      "learning_rate": 0.0007137014314928425,
-      "loss": 2.7088,
-      "step": 14000
+      "epoch": 14.325914556152469,
+      "grad_norm": 0.359375,
+      "learning_rate": 0.0007134817088769506,
+      "loss": 1.6826,
+      "step": 448000
     },
     {
-      "epoch": 14.826175869120654,
-      "grad_norm": 0.34375,
-      "learning_rate": 0.0007034764826175869,
-      "loss": 2.7051,
-      "step": 14500
+      "epoch": 14.341903300076746,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0007131619339984651,
+      "loss": 1.6878,
+      "step": 448500
     },
     {
-      "epoch": 15.0,
-      "eval_loss": 2.7638583183288574,
-      "eval_runtime": 0.9397,
-      "eval_samples_per_second": 532.097,
-      "eval_steps_per_second": 1.064,
-      "step": 14670
+      "epoch": 14.357892044001023,
+      "grad_norm": 0.29296875,
+      "learning_rate": 0.0007128421591199796,
+      "loss": 1.6825,
+      "step": 449000
     },
     {
-      "epoch": 15.337423312883436,
-      "grad_norm": 0.3203125,
-      "learning_rate": 0.0006932515337423313,
-      "loss": 2.6999,
-      "step": 15000
+      "epoch": 14.3738807879253,
+      "grad_norm": 0.33203125,
+      "learning_rate": 0.0007125223842414939,
+      "loss": 1.6798,
+      "step": 449500
     },
     {
-      "epoch": 15.848670756646216,
-      "grad_norm": 0.31640625,
-      "learning_rate": 0.0006830265848670757,
-      "loss": 2.6961,
-      "step": 15500
+      "epoch": 14.389869531849579,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.0007122026093630084,
+      "loss": 1.6828,
+      "step": 450000
     },
     {
-      "epoch": 16.0,
-      "eval_loss": 2.7651634216308594,
-      "eval_runtime": 0.9625,
-      "eval_samples_per_second": 519.471,
-      "eval_steps_per_second": 1.039,
-      "step": 15648
+      "epoch": 14.405858275773856,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.000711882834484523,
+      "loss": 1.6844,
+      "step": 450500
     },
     {
-      "epoch": 16.359918200409,
-      "grad_norm": 0.388671875,
-      "learning_rate": 0.0006728016359918201,
-      "loss": 2.6903,
-      "step": 16000
+      "epoch": 14.421847019698133,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.0007115630596060374,
+      "loss": 1.6864,
+      "step": 451000
     },
     {
-      "epoch": 16.87116564417178,
-      "grad_norm": 0.345703125,
-      "learning_rate": 0.0006625766871165644,
-      "loss": 2.6905,
-      "step": 16500
+      "epoch": 14.43783576362241,
+      "grad_norm": 0.2392578125,
+      "learning_rate": 0.0007112432847275518,
+      "loss": 1.6858,
+      "step": 451500
     },
     {
-      "epoch": 17.0,
-      "eval_loss": 2.75215744972229,
-      "eval_runtime": 0.8724,
-      "eval_samples_per_second": 573.163,
-      "eval_steps_per_second": 1.146,
-      "step": 16626
+      "epoch": 14.453824507546686,
+      "grad_norm": 0.2451171875,
+      "learning_rate": 0.0007109235098490663,
+      "loss": 1.6816,
+      "step": 452000
     },
     {
-      "epoch": 17.38241308793456,
-      "grad_norm": 0.337890625,
-      "learning_rate": 0.0006523517382413088,
-      "loss": 2.6791,
-      "step": 17000
+      "epoch": 14.469813251470965,
+      "grad_norm": 0.1845703125,
+      "learning_rate": 0.0007106037349705807,
+      "loss": 1.6784,
+      "step": 452500
     },
     {
-      "epoch": 17.893660531697343,
-      "grad_norm": 0.328125,
-      "learning_rate": 0.0006421267893660531,
-      "loss": 2.6817,
-      "step": 17500
+      "epoch": 14.485801995395242,
+      "grad_norm": 0.2734375,
+      "learning_rate": 0.0007102839600920952,
+      "loss": 1.6817,
+      "step": 453000
     },
     {
-      "epoch": 18.0,
-      "eval_loss": 2.750427722930908,
-      "eval_runtime": 1.0601,
-      "eval_samples_per_second": 471.655,
-      "eval_steps_per_second": 0.943,
-      "step": 17604
+      "epoch": 14.50179073931952,
+      "grad_norm": 0.2890625,
+      "learning_rate": 0.0007099641852136096,
+      "loss": 1.6867,
+      "step": 453500
     },
     {
-      "epoch": 18.404907975460123,
-      "grad_norm": 0.33984375,
-      "learning_rate": 0.0006319018404907975,
-      "loss": 2.6712,
-      "step": 18000
+      "epoch": 14.517779483243796,
+      "grad_norm": 0.232421875,
+      "learning_rate": 0.000709644410335124,
+      "loss": 1.6884,
+      "step": 454000
     },
     {
-      "epoch": 18.916155419222903,
-      "grad_norm": 0.349609375,
-      "learning_rate": 0.000621676891615542,
-      "loss": 2.6721,
-      "step": 18500
+      "epoch": 14.533768227168073,
+      "grad_norm": 0.2138671875,
+      "learning_rate": 0.0007093246354566385,
+      "loss": 1.6854,
+      "step": 454500
     },
     {
-      "epoch": 19.0,
-      "eval_loss": 2.740670919418335,
-      "eval_runtime": 0.9797,
-      "eval_samples_per_second": 510.381,
-      "eval_steps_per_second": 1.021,
-      "step": 18582
+      "epoch": 14.549756971092352,
+      "grad_norm": 0.2099609375,
+      "learning_rate": 0.000709004860578153,
+      "loss": 1.6822,
+      "step": 455000
     },
     {
-      "epoch": 19.427402862985684,
-      "grad_norm": 0.328125,
-      "learning_rate": 0.0006114519427402863,
-      "loss": 2.6632,
-      "step": 19000
+      "epoch": 14.565745715016629,
+      "grad_norm": 0.26953125,
+      "learning_rate": 0.0007086850856996675,
+      "loss": 1.6795,
+      "step": 455500
     },
     {
-      "epoch": 19.938650306748468,
-      "grad_norm": 0.36328125,
-      "learning_rate": 0.0006012269938650307,
-      "loss": 2.6677,
-      "step": 19500
+      "epoch": 14.581734458940906,
+      "grad_norm": 0.298828125,
+      "learning_rate": 0.0007083653108211819,
+      "loss": 1.6802,
+      "step": 456000
     },
     {
-      "epoch": 20.0,
-      "eval_loss": 2.740766763687134,
-      "eval_runtime": 0.9358,
-      "eval_samples_per_second": 534.326,
-      "eval_steps_per_second": 1.069,
-      "step": 19560
+      "epoch": 14.597723202865183,
+      "grad_norm": 0.267578125,
+      "learning_rate": 0.0007080455359426964,
+      "loss": 1.6809,
+      "step": 456500
     },
     {
-      "epoch": 20.449897750511248,
-      "grad_norm": 0.345703125,
-      "learning_rate": 0.000591002044989775,
-      "loss": 2.6576,
-      "step": 20000
+      "epoch": 14.61371194678946,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0007077257610642109,
+      "loss": 1.6788,
+      "step": 457000
     },
     {
-      "epoch": 20.961145194274028,
-      "grad_norm": 0.349609375,
-      "learning_rate": 0.0005807770961145194,
-      "loss": 2.6624,
-      "step": 20500
+      "epoch": 14.629700690713738,
+      "grad_norm": 0.3203125,
+      "learning_rate": 0.0007074059861857252,
+      "loss": 1.6821,
+      "step": 457500
     },
     {
-      "epoch": 21.0,
-      "eval_loss": 2.7370216846466064,
-      "eval_runtime": 0.9579,
-      "eval_samples_per_second": 521.987,
-      "eval_steps_per_second": 1.044,
-      "step": 20538
+      "epoch": 14.645689434638015,
+      "grad_norm": 0.310546875,
+      "learning_rate": 0.0007070862113072397,
+      "loss": 1.6775,
+      "step": 458000
     },
     {
-      "epoch": 21.47239263803681,
-      "grad_norm": 0.314453125,
-      "learning_rate": 0.0005705521472392639,
-      "loss": 2.65,
-      "step": 21000
+      "epoch": 14.661678178562292,
+      "grad_norm": 0.2421875,
+      "learning_rate": 0.0007067664364287542,
+      "loss": 1.679,
+      "step": 458500
     },
     {
-      "epoch": 21.983640081799592,
-      "grad_norm": 0.341796875,
-      "learning_rate": 0.0005603271983640082,
-      "loss": 2.6575,
-      "step": 21500
+      "epoch": 14.67766692248657,
+      "grad_norm": 0.287109375,
+      "learning_rate": 0.0007064466615502686,
+      "loss": 1.6806,
+      "step": 459000
     },
     {
-      "epoch": 22.0,
-      "eval_loss": 2.734588384628296,
-      "eval_runtime": 0.8297,
-      "eval_samples_per_second": 602.609,
-      "eval_steps_per_second": 1.205,
-      "step": 21516
+      "epoch": 14.693655666410846,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.000706126886671783,
+      "loss": 1.6836,
+      "step": 459500
     },
     {
-      "epoch": 22.494887525562373,
-      "grad_norm": 0.349609375,
-      "learning_rate": 0.0005501022494887526,
-      "loss": 2.6453,
-      "step": 22000
+      "epoch": 14.709644410335123,
+      "grad_norm": 0.2265625,
+      "learning_rate": 0.0007058071117932975,
+      "loss": 1.6906,
+      "step": 460000
     },
     {
-      "epoch": 23.0,
-      "eval_loss": 2.73508882522583,
-      "eval_runtime": 1.0104,
-      "eval_samples_per_second": 494.868,
-      "eval_steps_per_second": 0.99,
-      "step": 22494
+      "epoch": 14.725633154259402,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.000705487336914812,
+      "loss": 1.6832,
+      "step": 460500
     },
     {
-      "epoch": 23.006134969325153,
-      "grad_norm": 0.345703125,
-      "learning_rate": 0.0005398773006134969,
-      "loss": 2.6533,
-      "step": 22500
+      "epoch": 14.741621898183679,
+      "grad_norm": 0.251953125,
+      "learning_rate": 0.0007051675620363265,
+      "loss": 1.6874,
+      "step": 461000
     },
     {
-      "epoch": 23.517382413087933,
-      "grad_norm": 0.33984375,
-      "learning_rate": 0.0005296523517382413,
-      "loss": 2.6441,
-      "step": 23000
+      "epoch": 14.757610642107956,
+      "grad_norm": 0.3125,
+      "learning_rate": 0.000704847787157841,
+      "loss": 1.6832,
+      "step": 461500
     },
     {
-      "epoch": 24.0,
-      "eval_loss": 2.731626033782959,
-      "eval_runtime": 0.674,
-      "eval_samples_per_second": 741.886,
-      "eval_steps_per_second": 1.484,
-      "step": 23472
+      "epoch": 14.773599386032233,
+      "grad_norm": 0.265625,
+      "learning_rate": 0.0007045280122793553,
+      "loss": 1.6804,
+      "step": 462000
     },
     {
-      "epoch": 24.028629856850717,
-      "grad_norm": 0.322265625,
-      "learning_rate": 0.0005194274028629857,
-      "loss": 2.6442,
-      "step": 23500
+      "epoch": 14.78958812995651,
+      "grad_norm": 0.17578125,
+      "learning_rate": 0.0007042082374008698,
+      "loss": 1.6808,
+      "step": 462500
     },
     {
-      "epoch": 24.539877300613497,
-      "grad_norm": 0.330078125,
-      "learning_rate": 0.00050920245398773,
-      "loss": 2.637,
-      "step": 24000
+      "epoch": 14.805576873880788,
+      "grad_norm": 0.162109375,
+      "learning_rate": 0.0007038884625223843,
+      "loss": 1.6811,
+      "step": 463000
+    },
+    {
+      "epoch": 14.821565617805065,
+      "grad_norm": 0.255859375,
+      "learning_rate": 0.0007035686876438986,
+      "loss": 1.6826,
+      "step": 463500
+    },
+    {
+      "epoch": 14.837554361729342,
+      "grad_norm": 0.3046875,
+      "learning_rate": 0.0007032489127654131,
+      "loss": 1.6856,
+      "step": 464000
+    },
+    {
+      "epoch": 14.85354310565362,
+      "grad_norm": 0.26171875,
+      "learning_rate": 0.0007029291378869276,
+      "loss": 1.687,
+      "step": 464500
+    },
+    {
+      "epoch": 14.869531849577896,
+      "grad_norm": 0.244140625,
+      "learning_rate": 0.0007026093630084421,
+      "loss": 1.6878,
+      "step": 465000
     },
     {
-      "epoch": 25.0,
-      "eval_loss": 2.731029510498047,
-      "eval_runtime": 0.7022,
-      "eval_samples_per_second": 712.014,
-      "eval_steps_per_second": 1.424,
-      "step": 24450
+      "epoch": 14.885520593502175,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0007022895881299566,
+      "loss": 1.6782,
+      "step": 465500
     },
     {
-      "epoch": 25.051124744376278,
+      "epoch": 14.901509337426452,
       "grad_norm": 0.326171875,
-      "learning_rate": 0.0004989775051124745,
-      "loss": 2.6437,
-      "step": 24500
+      "learning_rate": 0.000701969813251471,
+      "loss": 1.6841,
+      "step": 466000
     },
     {
-      "epoch": 25.562372188139058,
-      "grad_norm": 0.333984375,
-      "learning_rate": 0.0004887525562372188,
-      "loss": 2.6364,
-      "step": 25000
+      "epoch": 14.917498081350729,
+      "grad_norm": 0.28515625,
+      "learning_rate": 0.0007016500383729855,
+      "loss": 1.6891,
+      "step": 466500
     },
     {
-      "epoch": 26.0,
-      "eval_loss": 2.7300286293029785,
-      "eval_runtime": 0.8078,
-      "eval_samples_per_second": 618.965,
-      "eval_steps_per_second": 1.238,
-      "step": 25428
+      "epoch": 14.933486825275006,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.0007013302634944999,
+      "loss": 1.688,
+      "step": 467000
     },
     {
-      "epoch": 26.073619631901842,
-      "grad_norm": 0.357421875,
-      "learning_rate": 0.0004785276073619632,
-      "loss": 2.6382,
-      "step": 25500
+      "epoch": 14.949475569199283,
+      "grad_norm": 0.2490234375,
+      "learning_rate": 0.0007010104886160144,
+      "loss": 1.6879,
+      "step": 467500
     },
     {
-      "epoch": 26.584867075664622,
-      "grad_norm": 0.330078125,
-      "learning_rate": 0.0004683026584867076,
-      "loss": 2.632,
-      "step": 26000
+      "epoch": 14.965464313123562,
+      "grad_norm": 0.28125,
+      "learning_rate": 0.0007006907137375288,
+      "loss": 1.6838,
+      "step": 468000
+    },
+    {
+      "epoch": 14.981453057047839,
+      "grad_norm": 0.23828125,
+      "learning_rate": 0.0007003709388590432,
+      "loss": 1.688,
+      "step": 468500
+    },
+    {
+      "epoch": 14.997441800972116,
+      "grad_norm": 0.283203125,
+      "learning_rate": 0.0007000511639805577,
+      "loss": 1.678,
+      "step": 469000
     },
     {
-      "epoch": 27.0,
-      "eval_loss": 2.7241549491882324,
-      "eval_runtime": 0.8157,
-      "eval_samples_per_second": 612.964,
-      "eval_steps_per_second": 1.226,
-      "step": 26406
+      "epoch": 15.0,
+      "eval_loss": 3.3484203815460205,
+      "eval_runtime": 3.8226,
+      "eval_samples_per_second": 130.801,
+      "eval_steps_per_second": 4.186,
+      "step": 469080
     }
   ],
   "logging_steps": 500,
-  "max_steps": 48900,
+  "max_steps": 1563600,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 50,
   "save_steps": 500,
@@ -615,8 +6721,8 @@
       "attributes": {}
     }
   },
-  "total_flos": 2.553177237041664e+17,
-  "train_batch_size": 1024,
+  "total_flos": 1.32558504924672e+18,
+  "train_batch_size": 32,
   "trial_name": null,
   "trial_params": null
 }