diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7033 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.01,
+  "eval_steps": 500,
+  "global_step": 1000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 1e-05,
+      "grad_norm": 1.6317715205474108,
+      "learning_rate": 3e-06,
+      "loss": 10.867,
+      "step": 1
+    },
+    {
+      "epoch": 2e-05,
+      "grad_norm": 1.6173147870740345,
+      "learning_rate": 6e-06,
+      "loss": 10.8657,
+      "step": 2
+    },
+    {
+      "epoch": 3e-05,
+      "grad_norm": 1.6387509359885835,
+      "learning_rate": 9e-06,
+      "loss": 10.8658,
+      "step": 3
+    },
+    {
+      "epoch": 4e-05,
+      "grad_norm": 1.597552357815991,
+      "learning_rate": 1.2e-05,
+      "loss": 10.865,
+      "step": 4
+    },
+    {
+      "epoch": 5e-05,
+      "grad_norm": 1.6454471252189307,
+      "learning_rate": 1.5e-05,
+      "loss": 10.8617,
+      "step": 5
+    },
+    {
+      "epoch": 6e-05,
+      "grad_norm": 1.6407925722175996,
+      "learning_rate": 1.8e-05,
+      "loss": 10.8593,
+      "step": 6
+    },
+    {
+      "epoch": 7e-05,
+      "grad_norm": 1.6096088910322361,
+      "learning_rate": 2.1000000000000002e-05,
+      "loss": 10.8456,
+      "step": 7
+    },
+    {
+      "epoch": 8e-05,
+      "grad_norm": 1.4682698515009915,
+      "learning_rate": 2.4e-05,
+      "loss": 10.8184,
+      "step": 8
+    },
+    {
+      "epoch": 9e-05,
+      "grad_norm": 1.3934246427009196,
+      "learning_rate": 2.7e-05,
+      "loss": 10.8113,
+      "step": 9
+    },
+    {
+      "epoch": 0.0001,
+      "grad_norm": 1.3326087040550991,
+      "learning_rate": 3e-05,
+      "loss": 10.7969,
+      "step": 10
+    },
+    {
+      "epoch": 0.00011,
+      "grad_norm": 1.2173090118888668,
+      "learning_rate": 3.2999999999999996e-05,
+      "loss": 10.7801,
+      "step": 11
+    },
+    {
+      "epoch": 0.00012,
+      "grad_norm": 1.176457724285593,
+      "learning_rate": 3.6e-05,
+      "loss": 10.7688,
+      "step": 12
+    },
+    {
+      "epoch": 0.00013,
+      "grad_norm": 1.1304424318539597,
+      "learning_rate": 3.9e-05,
+      "loss": 10.7498,
+      "step": 13
+    },
+    {
+      "epoch": 0.00014,
+      "grad_norm": 1.1158244568462428,
+      "learning_rate": 4.2000000000000004e-05,
+      "loss": 10.739,
+      "step": 14
+    },
+    {
+      "epoch": 0.00015,
+      "grad_norm": 1.10654756415174,
+      "learning_rate": 4.4999999999999996e-05,
+      "loss": 10.7299,
+      "step": 15
+    },
+    {
+      "epoch": 0.00016,
+      "grad_norm": 1.0751770608444569,
+      "learning_rate": 4.8e-05,
+      "loss": 10.7126,
+      "step": 16
+    },
+    {
+      "epoch": 0.00017,
+      "grad_norm": 1.046143788290158,
+      "learning_rate": 5.1000000000000006e-05,
+      "loss": 10.6968,
+      "step": 17
+    },
+    {
+      "epoch": 0.00018,
+      "grad_norm": 1.0230391412556632,
+      "learning_rate": 5.4e-05,
+      "loss": 10.6806,
+      "step": 18
+    },
+    {
+      "epoch": 0.00019,
+      "grad_norm": 0.9869982733638543,
+      "learning_rate": 5.7e-05,
+      "loss": 10.6649,
+      "step": 19
+    },
+    {
+      "epoch": 0.0002,
+      "grad_norm": 0.9728818553338922,
+      "learning_rate": 6e-05,
+      "loss": 10.6526,
+      "step": 20
+    },
+    {
+      "epoch": 0.00021,
+      "grad_norm": 0.9439994749998407,
+      "learning_rate": 6.3e-05,
+      "loss": 10.639,
+      "step": 21
+    },
+    {
+      "epoch": 0.00022,
+      "grad_norm": 0.9279528910342688,
+      "learning_rate": 6.599999999999999e-05,
+      "loss": 10.6244,
+      "step": 22
+    },
+    {
+      "epoch": 0.00023,
+      "grad_norm": 0.9208663519602571,
+      "learning_rate": 6.9e-05,
+      "loss": 10.6103,
+      "step": 23
+    },
+    {
+      "epoch": 0.00024,
+      "grad_norm": 0.9178551557561957,
+      "learning_rate": 7.2e-05,
+      "loss": 10.5993,
+      "step": 24
+    },
+    {
+      "epoch": 0.00025,
+      "grad_norm": 0.9193923250060233,
+      "learning_rate": 7.500000000000001e-05,
+      "loss": 10.5847,
+      "step": 25
+    },
+    {
+      "epoch": 0.00026,
+      "grad_norm": 0.9190901609677985,
+      "learning_rate": 7.8e-05,
+      "loss": 10.5717,
+      "step": 26
+    },
+    {
+      "epoch": 0.00027,
+      "grad_norm": 0.913753327244254,
+      "learning_rate": 8.1e-05,
+      "loss": 10.5597,
+      "step": 27
+    },
+    {
+      "epoch": 0.00028,
+      "grad_norm": 0.9119625217070774,
+      "learning_rate": 8.400000000000001e-05,
+      "loss": 10.5467,
+      "step": 28
+    },
+    {
+      "epoch": 0.00029,
+      "grad_norm": 0.9131038863398008,
+      "learning_rate": 8.7e-05,
+      "loss": 10.5323,
+      "step": 29
+    },
+    {
+      "epoch": 0.0003,
+      "grad_norm": 0.9186172829723749,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 10.517,
+      "step": 30
+    },
+    {
+      "epoch": 0.00031,
+      "grad_norm": 0.9155453723962563,
+      "learning_rate": 9.3e-05,
+      "loss": 10.5024,
+      "step": 31
+    },
+    {
+      "epoch": 0.00032,
+      "grad_norm": 0.909575589137279,
+      "learning_rate": 9.6e-05,
+      "loss": 10.4882,
+      "step": 32
+    },
+    {
+      "epoch": 0.00033,
+      "grad_norm": 0.9102175928717151,
+      "learning_rate": 9.900000000000001e-05,
+      "loss": 10.4717,
+      "step": 33
+    },
+    {
+      "epoch": 0.00034,
+      "grad_norm": 0.9103991171564593,
+      "learning_rate": 0.00010200000000000001,
+      "loss": 10.4554,
+      "step": 34
+    },
+    {
+      "epoch": 0.00035,
+      "grad_norm": 0.9164468459870767,
+      "learning_rate": 0.00010500000000000002,
+      "loss": 10.4388,
+      "step": 35
+    },
+    {
+      "epoch": 0.00036,
+      "grad_norm": 0.9129850089149896,
+      "learning_rate": 0.000108,
+      "loss": 10.4213,
+      "step": 36
+    },
+    {
+      "epoch": 0.00037,
+      "grad_norm": 0.9029808327462479,
+      "learning_rate": 0.000111,
+      "loss": 10.4047,
+      "step": 37
+    },
+    {
+      "epoch": 0.00038,
+      "grad_norm": 0.9074583652458479,
+      "learning_rate": 0.000114,
+      "loss": 10.3859,
+      "step": 38
+    },
+    {
+      "epoch": 0.00039,
+      "grad_norm": 0.9101621604618185,
+      "learning_rate": 0.000117,
+      "loss": 10.3665,
+      "step": 39
+    },
+    {
+      "epoch": 0.0004,
+      "grad_norm": 0.9144345472354501,
+      "learning_rate": 0.00012,
+      "loss": 10.3443,
+      "step": 40
+    },
+    {
+      "epoch": 0.00041,
+      "grad_norm": 0.9045315909874942,
+      "learning_rate": 0.000123,
+      "loss": 10.3264,
+      "step": 41
+    },
+    {
+      "epoch": 0.00042,
+      "grad_norm": 0.9101518170592343,
+      "learning_rate": 0.000126,
+      "loss": 10.3059,
+      "step": 42
+    },
+    {
+      "epoch": 0.00043,
+      "grad_norm": 0.9147491310031046,
+      "learning_rate": 0.000129,
+      "loss": 10.2831,
+      "step": 43
+    },
+    {
+      "epoch": 0.00044,
+      "grad_norm": 0.9138166723808987,
+      "learning_rate": 0.00013199999999999998,
+      "loss": 10.2617,
+      "step": 44
+    },
+    {
+      "epoch": 0.00045,
+      "grad_norm": 0.9166634386783579,
+      "learning_rate": 0.000135,
+      "loss": 10.2388,
+      "step": 45
+    },
+    {
+      "epoch": 0.00046,
+      "grad_norm": 0.9061319871747918,
+      "learning_rate": 0.000138,
+      "loss": 10.2171,
+      "step": 46
+    },
+    {
+      "epoch": 0.00047,
+      "grad_norm": 0.9134144094551727,
+      "learning_rate": 0.000141,
+      "loss": 10.1906,
+      "step": 47
+    },
+    {
+      "epoch": 0.00048,
+      "grad_norm": 0.9111015850262806,
+      "learning_rate": 0.000144,
+      "loss": 10.1669,
+      "step": 48
+    },
+    {
+      "epoch": 0.00049,
+      "grad_norm": 0.9106086112424904,
+      "learning_rate": 0.000147,
+      "loss": 10.1439,
+      "step": 49
+    },
+    {
+      "epoch": 0.0005,
+      "grad_norm": 0.9135108001899231,
+      "learning_rate": 0.00015000000000000001,
+      "loss": 10.1178,
+      "step": 50
+    },
+    {
+      "epoch": 0.00051,
+      "grad_norm": 0.9152031108656089,
+      "learning_rate": 0.000153,
+      "loss": 10.0918,
+      "step": 51
+    },
+    {
+      "epoch": 0.00052,
+      "grad_norm": 0.9133043896843657,
+      "learning_rate": 0.000156,
+      "loss": 10.0658,
+      "step": 52
+    },
+    {
+      "epoch": 0.00053,
+      "grad_norm": 0.9039676544194273,
+      "learning_rate": 0.000159,
+      "loss": 10.0419,
+      "step": 53
+    },
+    {
+      "epoch": 0.00054,
+      "grad_norm": 0.9245050218484777,
+      "learning_rate": 0.000162,
+      "loss": 10.0099,
+      "step": 54
+    },
+    {
+      "epoch": 0.00055,
+      "grad_norm": 0.915653013423474,
+      "learning_rate": 0.000165,
+      "loss": 9.9858,
+      "step": 55
+    },
+    {
+      "epoch": 0.00056,
+      "grad_norm": 0.90743999026624,
+      "learning_rate": 0.00016800000000000002,
+      "loss": 9.9567,
+      "step": 56
+    },
+    {
+      "epoch": 0.00057,
+      "grad_norm": 0.9125740935300273,
+      "learning_rate": 0.000171,
+      "loss": 9.9292,
+      "step": 57
+    },
+    {
+      "epoch": 0.00058,
+      "grad_norm": 0.9103134473221595,
+      "learning_rate": 0.000174,
+      "loss": 9.9046,
+      "step": 58
+    },
+    {
+      "epoch": 0.00059,
+      "grad_norm": 0.91502039796166,
+      "learning_rate": 0.000177,
+      "loss": 9.8727,
+      "step": 59
+    },
+    {
+      "epoch": 0.0006,
+      "grad_norm": 0.902549468432534,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 9.8467,
+      "step": 60
+    },
+    {
+      "epoch": 0.00061,
+      "grad_norm": 0.9011753674575653,
+      "learning_rate": 0.000183,
+      "loss": 9.8184,
+      "step": 61
+    },
+    {
+      "epoch": 0.00062,
+      "grad_norm": 0.9073829944096542,
+      "learning_rate": 0.000186,
+      "loss": 9.7865,
+      "step": 62
+    },
+    {
+      "epoch": 0.00063,
+      "grad_norm": 0.900277782228101,
+      "learning_rate": 0.000189,
+      "loss": 9.7594,
+      "step": 63
+    },
+    {
+      "epoch": 0.00064,
+      "grad_norm": 0.8964836174343672,
+      "learning_rate": 0.000192,
+      "loss": 9.7292,
+      "step": 64
+    },
+    {
+      "epoch": 0.00065,
+      "grad_norm": 0.9040253232587138,
+      "learning_rate": 0.00019500000000000002,
+      "loss": 9.6969,
+      "step": 65
+    },
+    {
+      "epoch": 0.00066,
+      "grad_norm": 0.8969700410935529,
+      "learning_rate": 0.00019800000000000002,
+      "loss": 9.6728,
+      "step": 66
+    },
+    {
+      "epoch": 0.00067,
+      "grad_norm": 0.8981910064021,
+      "learning_rate": 0.000201,
+      "loss": 9.643,
+      "step": 67
+    },
+    {
+      "epoch": 0.00068,
+      "grad_norm": 0.9049113812894196,
+      "learning_rate": 0.00020400000000000003,
+      "loss": 9.61,
+      "step": 68
+    },
+    {
+      "epoch": 0.00069,
+      "grad_norm": 0.8903730719674341,
+      "learning_rate": 0.00020700000000000002,
+      "loss": 9.5824,
+      "step": 69
+    },
+    {
+      "epoch": 0.0007,
+      "grad_norm": 0.9034192157313848,
+      "learning_rate": 0.00021000000000000004,
+      "loss": 9.5456,
+      "step": 70
+    },
+    {
+      "epoch": 0.00071,
+      "grad_norm": 0.8933343163190056,
+      "learning_rate": 0.00021299999999999997,
+      "loss": 9.5189,
+      "step": 71
+    },
+    {
+      "epoch": 0.00072,
+      "grad_norm": 0.8968593008835964,
+      "learning_rate": 0.000216,
+      "loss": 9.4914,
+      "step": 72
+    },
+    {
+      "epoch": 0.00073,
+      "grad_norm": 0.8960167287531013,
+      "learning_rate": 0.00021899999999999998,
+      "loss": 9.4574,
+      "step": 73
+    },
+    {
+      "epoch": 0.00074,
+      "grad_norm": 0.8965354004907367,
+      "learning_rate": 0.000222,
+      "loss": 9.4304,
+      "step": 74
+    },
+    {
+      "epoch": 0.00075,
+      "grad_norm": 0.8923958382533757,
+      "learning_rate": 0.000225,
+      "loss": 9.3946,
+      "step": 75
+    },
+    {
+      "epoch": 0.00076,
+      "grad_norm": 0.9011054851249423,
+      "learning_rate": 0.000228,
+      "loss": 9.3667,
+      "step": 76
+    },
+    {
+      "epoch": 0.00077,
+      "grad_norm": 0.8959973023769764,
+      "learning_rate": 0.000231,
+      "loss": 9.3374,
+      "step": 77
+    },
+    {
+      "epoch": 0.00078,
+      "grad_norm": 0.8901362352309407,
+      "learning_rate": 0.000234,
+      "loss": 9.3055,
+      "step": 78
+    },
+    {
+      "epoch": 0.00079,
+      "grad_norm": 0.8931237127830396,
+      "learning_rate": 0.00023700000000000001,
+      "loss": 9.2694,
+      "step": 79
+    },
+    {
+      "epoch": 0.0008,
+      "grad_norm": 0.8937972757596185,
+      "learning_rate": 0.00024,
+      "loss": 9.2355,
+      "step": 80
+    },
+    {
+      "epoch": 0.00081,
+      "grad_norm": 0.8984186266430717,
+      "learning_rate": 0.00024300000000000002,
+      "loss": 9.2033,
+      "step": 81
+    },
+    {
+      "epoch": 0.00082,
+      "grad_norm": 0.9060711515558022,
+      "learning_rate": 0.000246,
+      "loss": 9.1789,
+      "step": 82
+    },
+    {
+      "epoch": 0.00083,
+      "grad_norm": 0.8964035250431631,
+      "learning_rate": 0.00024900000000000004,
+      "loss": 9.1434,
+      "step": 83
+    },
+    {
+      "epoch": 0.00084,
+      "grad_norm": 0.8986377804082708,
+      "learning_rate": 0.000252,
+      "loss": 9.1119,
+      "step": 84
+    },
+    {
+      "epoch": 0.00085,
+      "grad_norm": 0.8933390971667627,
+      "learning_rate": 0.000255,
+      "loss": 9.0885,
+      "step": 85
+    },
+    {
+      "epoch": 0.00086,
+      "grad_norm": 0.8953786211031305,
+      "learning_rate": 0.000258,
+      "loss": 9.0521,
+      "step": 86
+    },
+    {
+      "epoch": 0.00087,
+      "grad_norm": 0.8876773666743288,
+      "learning_rate": 0.000261,
+      "loss": 9.0302,
+      "step": 87
+    },
+    {
+      "epoch": 0.00088,
+      "grad_norm": 0.8929152197347487,
+      "learning_rate": 0.00026399999999999997,
+      "loss": 8.9927,
+      "step": 88
+    },
+    {
+      "epoch": 0.00089,
+      "grad_norm": 0.8872457913370299,
+      "learning_rate": 0.000267,
+      "loss": 8.9669,
+      "step": 89
+    },
+    {
+      "epoch": 0.0009,
+      "grad_norm": 0.8862509419810688,
+      "learning_rate": 0.00027,
+      "loss": 8.9393,
+      "step": 90
+    },
+    {
+      "epoch": 0.00091,
+      "grad_norm": 0.8807880266520192,
+      "learning_rate": 0.000273,
+      "loss": 8.9089,
+      "step": 91
+    },
+    {
+      "epoch": 0.00092,
+      "grad_norm": 0.8831207589668301,
+      "learning_rate": 0.000276,
+      "loss": 8.876,
+      "step": 92
+    },
+    {
+      "epoch": 0.00093,
+      "grad_norm": 0.883866988348631,
+      "learning_rate": 0.000279,
+      "loss": 8.8462,
+      "step": 93
+    },
+    {
+      "epoch": 0.00094,
+      "grad_norm": 0.8824853161719922,
+      "learning_rate": 0.000282,
+      "loss": 8.8199,
+      "step": 94
+    },
+    {
+      "epoch": 0.00095,
+      "grad_norm": 0.8794572899807177,
+      "learning_rate": 0.000285,
+      "loss": 8.789,
+      "step": 95
+    },
+    {
+      "epoch": 0.00096,
+      "grad_norm": 0.8879520784944948,
+      "learning_rate": 0.000288,
+      "loss": 8.7571,
+      "step": 96
+    },
+    {
+      "epoch": 0.00097,
+      "grad_norm": 0.8798042444972031,
+      "learning_rate": 0.000291,
+      "loss": 8.7321,
+      "step": 97
+    },
+    {
+      "epoch": 0.00098,
+      "grad_norm": 0.8802597894834375,
+      "learning_rate": 0.000294,
+      "loss": 8.6954,
+      "step": 98
+    },
+    {
+      "epoch": 0.00099,
+      "grad_norm": 0.8806466323910314,
+      "learning_rate": 0.000297,
+      "loss": 8.6749,
+      "step": 99
+    },
+    {
+      "epoch": 0.001,
+      "grad_norm": 0.8777097574069823,
+      "learning_rate": 0.00030000000000000003,
+      "loss": 8.6485,
+      "step": 100
+    },
+    {
+      "epoch": 0.00101,
+      "grad_norm": 0.8786755147609817,
+      "learning_rate": 0.00030300000000000005,
+      "loss": 8.614,
+      "step": 101
+    },
+    {
+      "epoch": 0.00102,
+      "grad_norm": 0.8680143868447665,
+      "learning_rate": 0.000306,
+      "loss": 8.5949,
+      "step": 102
+    },
+    {
+      "epoch": 0.00103,
+      "grad_norm": 0.8748449452730288,
+      "learning_rate": 0.000309,
+      "loss": 8.5706,
+      "step": 103
+    },
+    {
+      "epoch": 0.00104,
+      "grad_norm": 0.8670215859388973,
+      "learning_rate": 0.000312,
+      "loss": 8.5498,
+      "step": 104
+    },
+    {
+      "epoch": 0.00105,
+      "grad_norm": 0.8687292008731472,
+      "learning_rate": 0.000315,
+      "loss": 8.5231,
+      "step": 105
+    },
+    {
+      "epoch": 0.00106,
+      "grad_norm": 0.8611486197845404,
+      "learning_rate": 0.000318,
+      "loss": 8.4945,
+      "step": 106
+    },
+    {
+      "epoch": 0.00107,
+      "grad_norm": 0.8521924348129856,
+      "learning_rate": 0.000321,
+      "loss": 8.4693,
+      "step": 107
+    },
+    {
+      "epoch": 0.00108,
+      "grad_norm": 0.8581933475380797,
+      "learning_rate": 0.000324,
+      "loss": 8.4407,
+      "step": 108
+    },
+    {
+      "epoch": 0.00109,
+      "grad_norm": 0.8524146875206363,
+      "learning_rate": 0.000327,
+      "loss": 8.421,
+      "step": 109
+    },
+    {
+      "epoch": 0.0011,
+      "grad_norm": 0.8682563584613229,
+      "learning_rate": 0.00033,
+      "loss": 8.3983,
+      "step": 110
+    },
+    {
+      "epoch": 0.00111,
+      "grad_norm": 0.8871469803064874,
+      "learning_rate": 0.000333,
+      "loss": 8.3685,
+      "step": 111
+    },
+    {
+      "epoch": 0.00112,
+      "grad_norm": 0.9236879668329372,
+      "learning_rate": 0.00033600000000000004,
+      "loss": 8.3463,
+      "step": 112
+    },
+    {
+      "epoch": 0.00113,
+      "grad_norm": 0.9129864456504505,
+      "learning_rate": 0.000339,
+      "loss": 8.3063,
+      "step": 113
+    },
+    {
+      "epoch": 0.00114,
+      "grad_norm": 0.8352263501003522,
+      "learning_rate": 0.000342,
+      "loss": 8.2966,
+      "step": 114
+    },
+    {
+      "epoch": 0.00115,
+      "grad_norm": 0.8592375580090957,
+      "learning_rate": 0.00034500000000000004,
+      "loss": 8.2718,
+      "step": 115
+    },
+    {
+      "epoch": 0.00116,
+      "grad_norm": 0.8674294753896091,
+      "learning_rate": 0.000348,
+      "loss": 8.2506,
+      "step": 116
+    },
+    {
+      "epoch": 0.00117,
+      "grad_norm": 0.8276917493567776,
+      "learning_rate": 0.000351,
+      "loss": 8.2188,
+      "step": 117
+    },
+    {
+      "epoch": 0.00118,
+      "grad_norm": 0.8476744963131545,
+      "learning_rate": 0.000354,
+      "loss": 8.2045,
+      "step": 118
+    },
+    {
+      "epoch": 0.00119,
+      "grad_norm": 0.844358071644388,
+      "learning_rate": 0.000357,
+      "loss": 8.1926,
+      "step": 119
+    },
+    {
+      "epoch": 0.0012,
+      "grad_norm": 0.8166594269287538,
+      "learning_rate": 0.00035999999999999997,
+      "loss": 8.1658,
+      "step": 120
+    },
+    {
+      "epoch": 0.00121,
+      "grad_norm": 0.8239930081024902,
+      "learning_rate": 0.000363,
+      "loss": 8.1389,
+      "step": 121
+    },
+    {
+      "epoch": 0.00122,
+      "grad_norm": 0.8099951959348987,
+      "learning_rate": 0.000366,
+      "loss": 8.1225,
+      "step": 122
+    },
+    {
+      "epoch": 0.00123,
+      "grad_norm": 0.830800320388625,
+      "learning_rate": 0.000369,
+      "loss": 8.1005,
+      "step": 123
+    },
+    {
+      "epoch": 0.00124,
+      "grad_norm": 0.8139169053139192,
+      "learning_rate": 0.000372,
+      "loss": 8.0791,
+      "step": 124
+    },
+    {
+      "epoch": 0.00125,
+      "grad_norm": 0.8112246790149765,
+      "learning_rate": 0.000375,
+      "loss": 8.0547,
+      "step": 125
+    },
+    {
+      "epoch": 0.00126,
+      "grad_norm": 0.7922278873371895,
+      "learning_rate": 0.000378,
+      "loss": 8.0424,
+      "step": 126
+    },
+    {
+      "epoch": 0.00127,
+      "grad_norm": 0.7755075943975184,
+      "learning_rate": 0.000381,
+      "loss": 8.0182,
+      "step": 127
+    },
+    {
+      "epoch": 0.00128,
+      "grad_norm": 0.8028212720713388,
+      "learning_rate": 0.000384,
+      "loss": 8.0,
+      "step": 128
+    },
+    {
+      "epoch": 0.00129,
+      "grad_norm": 0.8765129391436198,
+      "learning_rate": 0.00038700000000000003,
+      "loss": 7.9904,
+      "step": 129
+    },
+    {
+      "epoch": 0.0013,
+      "grad_norm": 1.1486399983200042,
+      "learning_rate": 0.00039000000000000005,
+      "loss": 7.9724,
+      "step": 130
+    },
+    {
+      "epoch": 0.00131,
+      "grad_norm": 1.0507777578095836,
+      "learning_rate": 0.000393,
+      "loss": 7.9382,
+      "step": 131
+    },
+    {
+      "epoch": 0.00132,
+      "grad_norm": 0.7511305165281239,
+      "learning_rate": 0.00039600000000000003,
+      "loss": 7.9248,
+      "step": 132
+    },
+    {
+      "epoch": 0.00133,
+      "grad_norm": 0.8108077692812641,
+      "learning_rate": 0.00039900000000000005,
+      "loss": 7.906,
+      "step": 133
+    },
+    {
+      "epoch": 0.00134,
+      "grad_norm": 0.8301313874662418,
+      "learning_rate": 0.000402,
+      "loss": 7.8952,
+      "step": 134
+    },
+    {
+      "epoch": 0.00135,
+      "grad_norm": 0.7123568807247732,
+      "learning_rate": 0.00040500000000000003,
+      "loss": 7.8651,
+      "step": 135
+    },
+    {
+      "epoch": 0.00136,
+      "grad_norm": 0.7697556904537746,
+      "learning_rate": 0.00040800000000000005,
+      "loss": 7.8515,
+      "step": 136
+    },
+    {
+      "epoch": 0.00137,
+      "grad_norm": 0.7190977621725152,
+      "learning_rate": 0.000411,
+      "loss": 7.8299,
+      "step": 137
+    },
+    {
+      "epoch": 0.00138,
+      "grad_norm": 0.7147305335216294,
+      "learning_rate": 0.00041400000000000003,
+      "loss": 7.8066,
+      "step": 138
+    },
+    {
+      "epoch": 0.00139,
+      "grad_norm": 0.747529428120578,
+      "learning_rate": 0.00041700000000000005,
+      "loss": 7.7954,
+      "step": 139
+    },
+    {
+      "epoch": 0.0014,
+      "grad_norm": 0.6748198408281931,
+      "learning_rate": 0.00042000000000000007,
+      "loss": 7.7774,
+      "step": 140
+    },
+    {
+      "epoch": 0.00141,
+      "grad_norm": 0.662142968172009,
+      "learning_rate": 0.000423,
+      "loss": 7.7644,
+      "step": 141
+    },
+    {
+      "epoch": 0.00142,
+      "grad_norm": 0.6859796391897652,
+      "learning_rate": 0.00042599999999999995,
+      "loss": 7.7534,
+      "step": 142
+    },
+    {
+      "epoch": 0.00143,
+      "grad_norm": 0.6828514310354903,
+      "learning_rate": 0.00042899999999999997,
+      "loss": 7.7255,
+      "step": 143
+    },
+    {
+      "epoch": 0.00144,
+      "grad_norm": 0.6490687938821236,
+      "learning_rate": 0.000432,
+      "loss": 7.7078,
+      "step": 144
+    },
+    {
+      "epoch": 0.00145,
+      "grad_norm": 0.6453156151137228,
+      "learning_rate": 0.000435,
+      "loss": 7.7035,
+      "step": 145
+    },
+    {
+      "epoch": 0.00146,
+      "grad_norm": 0.7166050341593803,
+      "learning_rate": 0.00043799999999999997,
+      "loss": 7.6672,
+      "step": 146
+    },
+    {
+      "epoch": 0.00147,
+      "grad_norm": 0.7443563124698165,
+      "learning_rate": 0.000441,
+      "loss": 7.6627,
+      "step": 147
+    },
+    {
+      "epoch": 0.00148,
+      "grad_norm": 0.7124453009383569,
+      "learning_rate": 0.000444,
+      "loss": 7.648,
+      "step": 148
+    },
+    {
+      "epoch": 0.00149,
+      "grad_norm": 0.6775306354557482,
+      "learning_rate": 0.00044699999999999997,
+      "loss": 7.6419,
+      "step": 149
+    },
+    {
+      "epoch": 0.0015,
+      "grad_norm": 0.8906380813028638,
+      "learning_rate": 0.00045,
+      "loss": 7.6135,
+      "step": 150
+    },
+    {
+      "epoch": 0.00151,
+      "grad_norm": 1.2892607147030477,
+      "learning_rate": 0.000453,
+      "loss": 7.605,
+      "step": 151
+    },
+    {
+      "epoch": 0.00152,
+      "grad_norm": 0.6492280537852009,
+      "learning_rate": 0.000456,
+      "loss": 7.5875,
+      "step": 152
+    },
+    {
+      "epoch": 0.00153,
+      "grad_norm": 0.8226748559002907,
+      "learning_rate": 0.000459,
+      "loss": 7.5783,
+      "step": 153
+    },
+    {
+      "epoch": 0.00154,
+      "grad_norm": 1.073498340899344,
+      "learning_rate": 0.000462,
+      "loss": 7.5662,
+      "step": 154
+    },
+    {
+      "epoch": 0.00155,
+      "grad_norm": 0.772416311968319,
+      "learning_rate": 0.000465,
+      "loss": 7.5481,
+      "step": 155
+    },
+    {
+      "epoch": 0.00156,
+      "grad_norm": 0.7999364848474875,
+      "learning_rate": 0.000468,
+      "loss": 7.5209,
+      "step": 156
+    },
+    {
+      "epoch": 0.00157,
+      "grad_norm": 0.9435570004081977,
+      "learning_rate": 0.000471,
+      "loss": 7.5196,
+      "step": 157
+    },
+    {
+      "epoch": 0.00158,
+      "grad_norm": 0.8364296006508578,
+      "learning_rate": 0.00047400000000000003,
+      "loss": 7.4997,
+      "step": 158
+    },
+    {
+      "epoch": 0.00159,
+      "grad_norm": 0.5596750978008624,
+      "learning_rate": 0.000477,
+      "loss": 7.4825,
+      "step": 159
+    },
+    {
+      "epoch": 0.0016,
+      "grad_norm": 0.5748856794890025,
+      "learning_rate": 0.00048,
+      "loss": 7.4616,
+      "step": 160
+    },
+    {
+      "epoch": 0.00161,
+      "grad_norm": 0.5986411025981033,
+      "learning_rate": 0.00048300000000000003,
+      "loss": 7.4419,
+      "step": 161
+    },
+    {
+      "epoch": 0.00162,
+      "grad_norm": 0.5117747222458712,
+      "learning_rate": 0.00048600000000000005,
+      "loss": 7.4429,
+      "step": 162
+    },
+    {
+      "epoch": 0.00163,
+      "grad_norm": 0.5509815158764758,
+      "learning_rate": 0.0004890000000000001,
+      "loss": 7.4259,
+      "step": 163
+    },
+    {
+      "epoch": 0.00164,
+      "grad_norm": 0.46393808675226217,
+      "learning_rate": 0.000492,
+      "loss": 7.4236,
+      "step": 164
+    },
+    {
+      "epoch": 0.00165,
+      "grad_norm": 0.5390679177469344,
+      "learning_rate": 0.000495,
+      "loss": 7.4006,
+      "step": 165
+    },
+    {
+      "epoch": 0.00166,
+      "grad_norm": 0.645354949444588,
+      "learning_rate": 0.0004980000000000001,
+      "loss": 7.3773,
+      "step": 166
+    },
+    {
+      "epoch": 0.00167,
+      "grad_norm": 1.0877276724000633,
+      "learning_rate": 0.000501,
+      "loss": 7.3741,
+      "step": 167
+    },
+    {
+      "epoch": 0.00168,
+      "grad_norm": 1.0909247587015876,
+      "learning_rate": 0.000504,
+      "loss": 7.3697,
+      "step": 168
+    },
+    {
+      "epoch": 0.00169,
+      "grad_norm": 0.5084368948335112,
+      "learning_rate": 0.0005070000000000001,
+      "loss": 7.3418,
+      "step": 169
+    },
+    {
+      "epoch": 0.0017,
+      "grad_norm": 1.5053615947271437,
+      "learning_rate": 0.00051,
+      "loss": 7.3492,
+      "step": 170
+    },
+    {
+      "epoch": 0.00171,
+      "grad_norm": 0.6120595065498764,
+      "learning_rate": 0.000513,
+      "loss": 7.3094,
+      "step": 171
+    },
+    {
+      "epoch": 0.00172,
+      "grad_norm": 0.9401489103136018,
+      "learning_rate": 0.000516,
+      "loss": 7.3189,
+      "step": 172
+    },
+    {
+      "epoch": 0.00173,
+      "grad_norm": 0.5558816443971099,
+      "learning_rate": 0.0005189999999999999,
+      "loss": 7.2906,
+      "step": 173
+    },
+    {
+      "epoch": 0.00174,
+      "grad_norm": 0.7810658223347703,
+      "learning_rate": 0.000522,
+      "loss": 7.314,
+      "step": 174
+    },
+    {
+      "epoch": 0.00175,
+      "grad_norm": 0.5409187817835382,
+      "learning_rate": 0.000525,
+      "loss": 7.2727,
+      "step": 175
+    },
+    {
+      "epoch": 0.00176,
+      "grad_norm": 0.630636606756854,
+      "learning_rate": 0.0005279999999999999,
+      "loss": 7.2484,
+      "step": 176
+    },
+    {
+      "epoch": 0.00177,
+      "grad_norm": 0.5206138592215499,
+      "learning_rate": 0.000531,
+      "loss": 7.2507,
+      "step": 177
+    },
+    {
+      "epoch": 0.00178,
+      "grad_norm": 0.5577166707239251,
+      "learning_rate": 0.000534,
+      "loss": 7.2428,
+      "step": 178
+    },
+    {
+      "epoch": 0.00179,
+      "grad_norm": 0.4980094016420717,
+      "learning_rate": 0.000537,
+      "loss": 7.2333,
+      "step": 179
+    },
+    {
+      "epoch": 0.0018,
+      "grad_norm": 0.5436132024149352,
+      "learning_rate": 0.00054,
+      "loss": 7.2053,
+      "step": 180
+    },
+    {
+      "epoch": 0.00181,
+      "grad_norm": 0.4850436660416719,
+      "learning_rate": 0.000543,
+      "loss": 7.1915,
+      "step": 181
+    },
+    {
+      "epoch": 0.00182,
+      "grad_norm": 0.686701132410735,
+      "learning_rate": 0.000546,
+      "loss": 7.1909,
+      "step": 182
+    },
+    {
+      "epoch": 0.00183,
+      "grad_norm": 0.68400997444651,
+      "learning_rate": 0.000549,
+      "loss": 7.1802,
+      "step": 183
+    },
+    {
+      "epoch": 0.00184,
+      "grad_norm": 0.8396027257351396,
+      "learning_rate": 0.000552,
+      "loss": 7.171,
+      "step": 184
+    },
+    {
+      "epoch": 0.00185,
+      "grad_norm": 0.7665365037410753,
+      "learning_rate": 0.000555,
+      "loss": 7.1476,
+      "step": 185
+    },
+    {
+      "epoch": 0.00186,
+      "grad_norm": 0.5359175667027454,
+      "learning_rate": 0.000558,
+      "loss": 7.1506,
+      "step": 186
+    },
+    {
+      "epoch": 0.00187,
+      "grad_norm": 0.5513800735908173,
+      "learning_rate": 0.000561,
+      "loss": 7.1261,
+      "step": 187
+    },
+    {
+      "epoch": 0.00188,
+      "grad_norm": 0.5352093004255375,
+      "learning_rate": 0.000564,
+      "loss": 7.1044,
+      "step": 188
+    },
+    {
+      "epoch": 0.00189,
+      "grad_norm": 0.5938457818726526,
+      "learning_rate": 0.000567,
+      "loss": 7.1004,
+      "step": 189
+    },
+    {
+      "epoch": 0.0019,
+      "grad_norm": 0.5273842405001533,
+      "learning_rate": 0.00057,
+      "loss": 7.0834,
+      "step": 190
+    },
+    {
+      "epoch": 0.00191,
+      "grad_norm": 0.47487115515279366,
+      "learning_rate": 0.000573,
+      "loss": 7.0721,
+      "step": 191
+    },
+    {
+      "epoch": 0.00192,
+      "grad_norm": 0.6036837698051599,
+      "learning_rate": 0.000576,
+      "loss": 7.0655,
+      "step": 192
+    },
+    {
+      "epoch": 0.00193,
+      "grad_norm": 0.3951184212196986,
+      "learning_rate": 0.000579,
+      "loss": 7.061,
+      "step": 193
+    },
+    {
+      "epoch": 0.00194,
+      "grad_norm": 0.442083567688087,
+      "learning_rate": 0.000582,
+      "loss": 7.0548,
+      "step": 194
+    },
+    {
+      "epoch": 0.00195,
+      "grad_norm": 0.439546882682468,
+      "learning_rate": 0.000585,
+      "loss": 7.0348,
+      "step": 195
+    },
+    {
+      "epoch": 0.00196,
+      "grad_norm": 0.46247531692771043,
+      "learning_rate": 0.000588,
+      "loss": 7.0228,
+      "step": 196
+    },
+    {
+      "epoch": 0.00197,
+      "grad_norm": 0.4140335072217301,
+      "learning_rate": 0.000591,
+      "loss": 7.0171,
+      "step": 197
+    },
+    {
+      "epoch": 0.00198,
+      "grad_norm": 0.3685986320410548,
+      "learning_rate": 0.000594,
+      "loss": 7.0081,
+      "step": 198
+    },
+    {
+      "epoch": 0.00199,
+      "grad_norm": 0.4020373564129086,
+      "learning_rate": 0.0005970000000000001,
+      "loss": 6.9898,
+      "step": 199
+    },
+    {
+      "epoch": 0.002,
+      "grad_norm": 0.37126410475941546,
+      "learning_rate": 0.0006000000000000001,
+      "loss": 6.9867,
+      "step": 200
+    },
+    {
+      "epoch": 0.00201,
+      "grad_norm": 0.3773154493828028,
+      "learning_rate": 0.000603,
+      "loss": 6.9617,
+      "step": 201
+    },
+    {
+      "epoch": 0.00202,
+      "grad_norm": 0.3540017416986532,
+      "learning_rate": 0.0006060000000000001,
+      "loss": 6.9491,
+      "step": 202
+    },
+    {
+      "epoch": 0.00203,
+      "grad_norm": 0.403279648640721,
+      "learning_rate": 0.0006090000000000001,
+      "loss": 6.9534,
+      "step": 203
+    },
+    {
+      "epoch": 0.00204,
+      "grad_norm": 0.5112949618253247,
+      "learning_rate": 0.000612,
+      "loss": 6.9385,
+      "step": 204
+    },
+    {
+      "epoch": 0.00205,
+      "grad_norm": 0.7200998739972175,
+      "learning_rate": 0.000615,
+      "loss": 6.931,
+      "step": 205
+    },
+    {
+      "epoch": 0.00206,
+      "grad_norm": 1.209379808685074,
+      "learning_rate": 0.000618,
+      "loss": 6.9351,
+      "step": 206
+    },
+    {
+      "epoch": 0.00207,
+      "grad_norm": 1.0040442357645134,
+      "learning_rate": 0.000621,
+      "loss": 6.9279,
+      "step": 207
+    },
+    {
+      "epoch": 0.00208,
+      "grad_norm": 0.7878464521205251,
+      "learning_rate": 0.000624,
+      "loss": 6.9129,
+      "step": 208
+    },
+    {
+      "epoch": 0.00209,
+      "grad_norm": 1.3096135062236434,
+      "learning_rate": 0.000627,
+      "loss": 6.9067,
+      "step": 209
+    },
+    {
+      "epoch": 0.0021,
+      "grad_norm": 0.910504669978176,
+      "learning_rate": 0.00063,
+      "loss": 6.8878,
+      "step": 210
+    },
+    {
+      "epoch": 0.00211,
+      "grad_norm": 1.3015817966038044,
+      "learning_rate": 0.000633,
+      "loss": 6.8987,
+      "step": 211
+    },
+    {
+      "epoch": 0.00212,
+      "grad_norm": 0.5587575104994011,
+      "learning_rate": 0.000636,
+      "loss": 6.8696,
+      "step": 212
+    },
+    {
+      "epoch": 0.00213,
+      "grad_norm": 0.8081412049208773,
+      "learning_rate": 0.000639,
+      "loss": 6.8608,
+      "step": 213
+    },
+    {
+      "epoch": 0.00214,
+      "grad_norm": 0.6397075273457759,
+      "learning_rate": 0.000642,
+      "loss": 6.8511,
+      "step": 214
+    },
+    {
+      "epoch": 0.00215,
+      "grad_norm": 0.5875043250740225,
+      "learning_rate": 0.000645,
+      "loss": 6.8482,
+      "step": 215
+    },
+    {
+      "epoch": 0.00216,
+      "grad_norm": 0.6060774535893669,
+      "learning_rate": 0.000648,
+      "loss": 6.8413,
+      "step": 216
+    },
+    {
+      "epoch": 0.00217,
+      "grad_norm": 0.5183751970166313,
+      "learning_rate": 0.000651,
+      "loss": 6.8119,
+      "step": 217
+    },
+    {
+      "epoch": 0.00218,
+      "grad_norm": 0.5539011900924167,
+      "learning_rate": 0.000654,
+      "loss": 6.8171,
+      "step": 218
+    },
+    {
+      "epoch": 0.00219,
+      "grad_norm": 0.5004312163301685,
+      "learning_rate": 0.000657,
+      "loss": 6.7882,
+      "step": 219
+    },
+    {
+      "epoch": 0.0022,
+      "grad_norm": 0.4762494747244133,
+      "learning_rate": 0.00066,
+      "loss": 6.8062,
+      "step": 220
+    },
+    {
+      "epoch": 0.00221,
+      "grad_norm": 0.353452768224107,
+      "learning_rate": 0.0006630000000000001,
+      "loss": 6.7814,
+      "step": 221
+    },
+    {
+      "epoch": 0.00222,
+      "grad_norm": 0.3879096289808107,
+      "learning_rate": 0.000666,
+      "loss": 6.7696,
+      "step": 222
+    },
+    {
+      "epoch": 0.00223,
+      "grad_norm": 0.4253518811476648,
+      "learning_rate": 0.000669,
+      "loss": 6.7658,
+      "step": 223
+    },
+    {
+      "epoch": 0.00224,
+      "grad_norm": 0.4076338977034843,
+      "learning_rate": 0.0006720000000000001,
+      "loss": 6.7609,
+      "step": 224
+    },
+    {
+      "epoch": 0.00225,
+      "grad_norm": 0.4491881195721512,
+      "learning_rate": 0.000675,
+      "loss": 6.7489,
+      "step": 225
+    },
+    {
+      "epoch": 0.00226,
+      "grad_norm": 0.4195112098951784,
+      "learning_rate": 0.000678,
+      "loss": 6.7444,
+      "step": 226
+    },
+    {
+      "epoch": 0.00227,
+      "grad_norm": 0.371663908330708,
+      "learning_rate": 0.0006810000000000001,
+      "loss": 6.7174,
+      "step": 227
+    },
+    {
+      "epoch": 0.00228,
+      "grad_norm": 0.3462952066328263,
+      "learning_rate": 0.000684,
+      "loss": 6.7197,
+      "step": 228
+    },
+    {
+      "epoch": 0.00229,
+      "grad_norm": 0.5288966810878937,
+      "learning_rate": 0.000687,
+      "loss": 6.7178,
+      "step": 229
+    },
+    {
+      "epoch": 0.0023,
+      "grad_norm": 0.7623592789505088,
+      "learning_rate": 0.0006900000000000001,
+      "loss": 6.6993,
+      "step": 230
+    },
+    {
+      "epoch": 0.00231,
+      "grad_norm": 0.9880780315432149,
+      "learning_rate": 0.000693,
+      "loss": 6.6923,
+      "step": 231
+    },
+    {
+      "epoch": 0.00232,
+      "grad_norm": 1.028691756937799,
+      "learning_rate": 0.000696,
+      "loss": 6.7142,
+      "step": 232
+    },
+    {
+      "epoch": 0.00233,
+      "grad_norm": 0.9501761132787978,
+      "learning_rate": 0.0006990000000000001,
+      "loss": 6.6946,
+      "step": 233
+    },
+    {
+      "epoch": 0.00234,
+      "grad_norm": 0.9999298841530961,
+      "learning_rate": 0.000702,
+      "loss": 6.6948,
+      "step": 234
+    },
+    {
+      "epoch": 0.00235,
+      "grad_norm": 0.6945930996245869,
+      "learning_rate": 0.000705,
+      "loss": 6.6675,
+      "step": 235
+    },
+    {
+      "epoch": 0.00236,
+      "grad_norm": 0.6422608041910052,
+      "learning_rate": 0.000708,
+      "loss": 6.6513,
+      "step": 236
+    },
+    {
+      "epoch": 0.00237,
+      "grad_norm": 0.47985797682797676,
+      "learning_rate": 0.0007109999999999999,
+      "loss": 6.657,
+      "step": 237
+    },
+    {
+      "epoch": 0.00238,
+      "grad_norm": 0.6191094472441181,
+      "learning_rate": 0.000714,
+      "loss": 6.6482,
+      "step": 238
+    },
+    {
+      "epoch": 0.00239,
+      "grad_norm": 0.5181523072026278,
+      "learning_rate": 0.000717,
+      "loss": 6.628,
+      "step": 239
+    },
+    {
+      "epoch": 0.0024,
+      "grad_norm": 0.5894783578801835,
+      "learning_rate": 0.0007199999999999999,
+      "loss": 6.645,
+      "step": 240
+    },
+    {
+      "epoch": 0.00241,
+      "grad_norm": 0.5639698176641863,
+      "learning_rate": 0.000723,
+      "loss": 6.6279,
+      "step": 241
+    },
+    {
+      "epoch": 0.00242,
+      "grad_norm": 0.9095170652542525,
+      "learning_rate": 0.000726,
+      "loss": 6.6087,
+      "step": 242
+    },
+    {
+      "epoch": 0.00243,
+      "grad_norm": 1.3373514416355459,
+      "learning_rate": 0.000729,
+      "loss": 6.6191,
+      "step": 243
+    },
+    {
+      "epoch": 0.00244,
+      "grad_norm": 0.5837452319331187,
+      "learning_rate": 0.000732,
+      "loss": 6.5991,
+      "step": 244
+    },
+    {
+      "epoch": 0.00245,
+      "grad_norm": 1.0261084178764917,
+      "learning_rate": 0.000735,
+      "loss": 6.6035,
+      "step": 245
+    },
+    {
+      "epoch": 0.00246,
+      "grad_norm": 0.5663979442820745,
+      "learning_rate": 0.000738,
+      "loss": 6.574,
+      "step": 246
+    },
+    {
+      "epoch": 0.00247,
+      "grad_norm": 0.5896636858778472,
+      "learning_rate": 0.000741,
+      "loss": 6.5719,
+      "step": 247
+    },
+    {
+      "epoch": 0.00248,
+      "grad_norm": 0.47326080430149503,
+      "learning_rate": 0.000744,
+      "loss": 6.5757,
+      "step": 248
+    },
+    {
+      "epoch": 0.00249,
+      "grad_norm": 0.6248164732961499,
+      "learning_rate": 0.000747,
+      "loss": 6.5582,
+      "step": 249
+    },
+    {
+      "epoch": 0.0025,
+      "grad_norm": 0.48362025373458484,
+      "learning_rate": 0.00075,
+      "loss": 6.5567,
+      "step": 250
+    },
+    {
+      "epoch": 0.00251,
+      "grad_norm": 0.4858269118610639,
+      "learning_rate": 0.000753,
+      "loss": 6.5227,
+      "step": 251
+    },
+    {
+      "epoch": 0.00252,
+      "grad_norm": 0.41862369731289734,
+      "learning_rate": 0.000756,
+      "loss": 6.5302,
+      "step": 252
+    },
+    {
+      "epoch": 0.00253,
+      "grad_norm": 0.47200180330590646,
+      "learning_rate": 0.000759,
+      "loss": 6.522,
+      "step": 253
+    },
+    {
+      "epoch": 0.00254,
+      "grad_norm": 0.3651546358356223,
+      "learning_rate": 0.000762,
+      "loss": 6.5136,
+      "step": 254
+    },
+    {
+      "epoch": 0.00255,
+      "grad_norm": 0.4285192093499382,
+      "learning_rate": 0.0007650000000000001,
+      "loss": 6.5149,
+      "step": 255
+    },
+    {
+      "epoch": 0.00256,
+      "grad_norm": 0.42767490819993825,
+      "learning_rate": 0.000768,
+      "loss": 6.4917,
+      "step": 256
+    },
+    {
+      "epoch": 0.00257,
+      "grad_norm": 0.48143237273687123,
+      "learning_rate": 0.000771,
+      "loss": 6.4974,
+      "step": 257
+    },
+    {
+      "epoch": 0.00258,
+      "grad_norm": 0.5467090762573645,
+      "learning_rate": 0.0007740000000000001,
+      "loss": 6.4744,
+      "step": 258
+    },
+    {
+      "epoch": 0.00259,
+      "grad_norm": 0.6750367098607939,
+      "learning_rate": 0.000777,
+      "loss": 6.4781,
+      "step": 259
+    },
+    {
+      "epoch": 0.0026,
+      "grad_norm": 0.8320211245129605,
+      "learning_rate": 0.0007800000000000001,
+      "loss": 6.4681,
+      "step": 260
+    },
+    {
+      "epoch": 0.00261,
+      "grad_norm": 0.905841412497731,
+      "learning_rate": 0.0007830000000000001,
+      "loss": 6.4809,
+      "step": 261
+    },
+    {
+      "epoch": 0.00262,
+      "grad_norm": 1.0687360869781928,
+      "learning_rate": 0.000786,
+      "loss": 6.4647,
+      "step": 262
+    },
+    {
+      "epoch": 0.00263,
+      "grad_norm": 1.2051408681263374,
+      "learning_rate": 0.0007890000000000001,
+      "loss": 6.4874,
+      "step": 263
+    },
+    {
+      "epoch": 0.00264,
+      "grad_norm": 0.8690142485653533,
+      "learning_rate": 0.0007920000000000001,
+      "loss": 6.455,
+      "step": 264
+    },
+    {
+      "epoch": 0.00265,
+      "grad_norm": 1.2774066489819682,
+      "learning_rate": 0.000795,
+      "loss": 6.4641,
+      "step": 265
+    },
+    {
+      "epoch": 0.00266,
+      "grad_norm": 0.7288978979341997,
+      "learning_rate": 0.0007980000000000001,
+      "loss": 6.4454,
+      "step": 266
+    },
+    {
+      "epoch": 0.00267,
+      "grad_norm": 0.6506287971604123,
+      "learning_rate": 0.0008010000000000001,
+      "loss": 6.4371,
+      "step": 267
+    },
+    {
+      "epoch": 0.00268,
+      "grad_norm": 0.6866545943797145,
+      "learning_rate": 0.000804,
+      "loss": 6.4338,
+      "step": 268
+    },
+    {
+      "epoch": 0.00269,
+      "grad_norm": 1.0440114151727509,
+      "learning_rate": 0.0008070000000000001,
+      "loss": 6.4227,
+      "step": 269
+    },
+    {
+      "epoch": 0.0027,
+      "grad_norm": 1.170351969791303,
+      "learning_rate": 0.0008100000000000001,
+      "loss": 6.4362,
+      "step": 270
+    },
+    {
+      "epoch": 0.00271,
+      "grad_norm": 0.6832222674646221,
+      "learning_rate": 0.000813,
+      "loss": 6.4253,
+      "step": 271
+    },
+    {
+      "epoch": 0.00272,
+      "grad_norm": 0.617936452008115,
+      "learning_rate": 0.0008160000000000001,
+      "loss": 6.4063,
+      "step": 272
+    },
+    {
+      "epoch": 0.00273,
+      "grad_norm": 0.6750066600068259,
+      "learning_rate": 0.0008190000000000001,
+      "loss": 6.4008,
+      "step": 273
+    },
+    {
+      "epoch": 0.00274,
+      "grad_norm": 0.6315560378177079,
+      "learning_rate": 0.000822,
+      "loss": 6.379,
+      "step": 274
+    },
+    {
+      "epoch": 0.00275,
+      "grad_norm": 0.6774266517760174,
+      "learning_rate": 0.0008250000000000001,
+      "loss": 6.3852,
+      "step": 275
+    },
+    {
+      "epoch": 0.00276,
+      "grad_norm": 0.7272895822836024,
+      "learning_rate": 0.0008280000000000001,
+      "loss": 6.3794,
+      "step": 276
+    },
+    {
+      "epoch": 0.00277,
+      "grad_norm": 0.6752471037637485,
+      "learning_rate": 0.0008310000000000001,
+      "loss": 6.3735,
+      "step": 277
+    },
+    {
+      "epoch": 0.00278,
+      "grad_norm": 0.5678457826039285,
+      "learning_rate": 0.0008340000000000001,
+      "loss": 6.354,
+      "step": 278
+    },
+    {
+      "epoch": 0.00279,
+      "grad_norm": 0.4611700189072147,
+      "learning_rate": 0.0008370000000000001,
+      "loss": 6.3529,
+      "step": 279
+    },
+    {
+      "epoch": 0.0028,
+      "grad_norm": 0.343285643042232,
+      "learning_rate": 0.0008400000000000001,
+      "loss": 6.3329,
+      "step": 280
+    },
+    {
+      "epoch": 0.00281,
+      "grad_norm": 0.4519631747446028,
+      "learning_rate": 0.0008430000000000001,
+      "loss": 6.3253,
+      "step": 281
+    },
+    {
+      "epoch": 0.00282,
+      "grad_norm": 0.3255189118052276,
+      "learning_rate": 0.000846,
+      "loss": 6.3232,
+      "step": 282
+    },
+    {
+      "epoch": 0.00283,
+      "grad_norm": 0.4297016476682907,
+      "learning_rate": 0.0008489999999999999,
+      "loss": 6.3149,
+      "step": 283
+    },
+    {
+      "epoch": 0.00284,
+      "grad_norm": 0.37515936714697207,
+      "learning_rate": 0.0008519999999999999,
+      "loss": 6.3058,
+      "step": 284
+    },
+    {
+      "epoch": 0.00285,
+      "grad_norm": 0.3458870104505622,
+      "learning_rate": 0.000855,
+      "loss": 6.3075,
+      "step": 285
+    },
+    {
+      "epoch": 0.00286,
+      "grad_norm": 0.39943359732325345,
+      "learning_rate": 0.0008579999999999999,
+      "loss": 6.2857,
+      "step": 286
+    },
+    {
+      "epoch": 0.00287,
+      "grad_norm": 0.3631751365570726,
+      "learning_rate": 0.000861,
+      "loss": 6.2875,
+      "step": 287
+    },
+    {
+      "epoch": 0.00288,
+      "grad_norm": 0.552280032213235,
+      "learning_rate": 0.000864,
+      "loss": 6.2707,
+      "step": 288
+    },
+    {
+      "epoch": 0.00289,
+      "grad_norm": 0.9396362484724781,
+      "learning_rate": 0.0008669999999999999,
+      "loss": 6.2923,
+      "step": 289
+    },
+    {
+      "epoch": 0.0029,
+      "grad_norm": 1.3959756273937387,
+      "learning_rate": 0.00087,
+      "loss": 6.2945,
+      "step": 290
+    },
+    {
+      "epoch": 0.00291,
+      "grad_norm": 0.6791923155853262,
+      "learning_rate": 0.000873,
+      "loss": 6.2663,
+      "step": 291
+    },
+    {
+      "epoch": 0.00292,
+      "grad_norm": 0.9879732835250642,
+      "learning_rate": 0.0008759999999999999,
+      "loss": 6.2794,
+      "step": 292
+    },
+    {
+      "epoch": 0.00293,
+      "grad_norm": 1.0140384817226566,
+      "learning_rate": 0.000879,
+      "loss": 6.2652,
+      "step": 293
+    },
+    {
+      "epoch": 0.00294,
+      "grad_norm": 0.9889225551341856,
+      "learning_rate": 0.000882,
+      "loss": 6.2822,
+      "step": 294
+    },
+    {
+      "epoch": 0.00295,
+      "grad_norm": 1.0429477871581094,
+      "learning_rate": 0.0008849999999999999,
+      "loss": 6.2468,
+      "step": 295
+    },
+    {
+      "epoch": 0.00296,
+      "grad_norm": 1.1774929945794055,
+      "learning_rate": 0.000888,
+      "loss": 6.2705,
+      "step": 296
+    },
+    {
+      "epoch": 0.00297,
+      "grad_norm": 0.7364156162079134,
+      "learning_rate": 0.000891,
+      "loss": 6.2278,
+      "step": 297
+    },
+    {
+      "epoch": 0.00298,
+      "grad_norm": 0.8424638476384282,
+      "learning_rate": 0.0008939999999999999,
+      "loss": 6.2455,
+      "step": 298
+    },
+    {
+      "epoch": 0.00299,
+      "grad_norm": 0.8668489286879963,
+      "learning_rate": 0.000897,
+      "loss": 6.2225,
+      "step": 299
+    },
+    {
+      "epoch": 0.003,
+      "grad_norm": 0.9039057952602142,
+      "learning_rate": 0.0009,
+      "loss": 6.2236,
+      "step": 300
+    },
+    {
+      "epoch": 0.00301,
+      "grad_norm": 1.0344858724084711,
+      "learning_rate": 0.0009029999999999999,
+      "loss": 6.2222,
+      "step": 301
+    },
+    {
+      "epoch": 0.00302,
+      "grad_norm": 0.801399966246171,
+      "learning_rate": 0.000906,
+      "loss": 6.2007,
+      "step": 302
+    },
+    {
+      "epoch": 0.00303,
+      "grad_norm": 0.7276255151675343,
+      "learning_rate": 0.000909,
+      "loss": 6.2106,
+      "step": 303
+    },
+    {
+      "epoch": 0.00304,
+      "grad_norm": 0.8306162070729353,
+      "learning_rate": 0.000912,
+      "loss": 6.205,
+      "step": 304
+    },
+    {
+      "epoch": 0.00305,
+      "grad_norm": 0.7650178489304597,
+      "learning_rate": 0.000915,
+      "loss": 6.2045,
+      "step": 305
+    },
+    {
+      "epoch": 0.00306,
+      "grad_norm": 0.7024963687074245,
+      "learning_rate": 0.000918,
+      "loss": 6.1878,
+      "step": 306
+    },
+    {
+      "epoch": 0.00307,
+      "grad_norm": 0.5687961336654864,
+      "learning_rate": 0.000921,
+      "loss": 6.1778,
+      "step": 307
+    },
+    {
+      "epoch": 0.00308,
+      "grad_norm": 0.4515866134049927,
+      "learning_rate": 0.000924,
+      "loss": 6.1586,
+      "step": 308
+    },
+    {
+      "epoch": 0.00309,
+      "grad_norm": 0.5454081565882548,
+      "learning_rate": 0.000927,
+      "loss": 6.1763,
+      "step": 309
+    },
+    {
+      "epoch": 0.0031,
+      "grad_norm": 0.4033959865123679,
+      "learning_rate": 0.00093,
+      "loss": 6.1549,
+      "step": 310
+    },
+    {
+      "epoch": 0.00311,
+      "grad_norm": 0.5004494164305024,
+      "learning_rate": 0.000933,
+      "loss": 6.1478,
+      "step": 311
+    },
+    {
+      "epoch": 0.00312,
+      "grad_norm": 0.470361361901935,
+      "learning_rate": 0.000936,
+      "loss": 6.1275,
+      "step": 312
+    },
+    {
+      "epoch": 0.00313,
+      "grad_norm": 0.4973667290148138,
+      "learning_rate": 0.0009390000000000001,
+      "loss": 6.1399,
+      "step": 313
+    },
+    {
+      "epoch": 0.00314,
+      "grad_norm": 0.47398584855820086,
+      "learning_rate": 0.000942,
+      "loss": 6.1453,
+      "step": 314
+    },
+    {
+      "epoch": 0.00315,
+      "grad_norm": 0.39081338977861474,
+      "learning_rate": 0.000945,
+      "loss": 6.1206,
+      "step": 315
+    },
+    {
+      "epoch": 0.00316,
+      "grad_norm": 0.308452368547838,
+      "learning_rate": 0.0009480000000000001,
+      "loss": 6.0925,
+      "step": 316
+    },
+    {
+      "epoch": 0.00317,
+      "grad_norm": 0.33735261718955184,
+      "learning_rate": 0.000951,
+      "loss": 6.1112,
+      "step": 317
+    },
+    {
+      "epoch": 0.00318,
+      "grad_norm": 0.3843792856632324,
+      "learning_rate": 0.000954,
+      "loss": 6.1055,
+      "step": 318
+    },
+    {
+      "epoch": 0.00319,
+      "grad_norm": 0.45015697169720664,
+      "learning_rate": 0.0009570000000000001,
+      "loss": 6.0951,
+      "step": 319
+    },
+    {
+      "epoch": 0.0032,
+      "grad_norm": 0.6347836654615971,
+      "learning_rate": 0.00096,
+      "loss": 6.097,
+      "step": 320
+    },
+    {
+      "epoch": 0.00321,
+      "grad_norm": 0.9264517947757075,
+      "learning_rate": 0.000963,
+      "loss": 6.0955,
+      "step": 321
+    },
+    {
+      "epoch": 0.00322,
+      "grad_norm": 1.3311662164937155,
+      "learning_rate": 0.0009660000000000001,
+      "loss": 6.0984,
+      "step": 322
+    },
+    {
+      "epoch": 0.00323,
+      "grad_norm": 0.7192768486088142,
+      "learning_rate": 0.000969,
+      "loss": 6.081,
+      "step": 323
+    },
+    {
+      "epoch": 0.00324,
+      "grad_norm": 0.8815686637890671,
+      "learning_rate": 0.0009720000000000001,
+      "loss": 6.0921,
+      "step": 324
+    },
+    {
+      "epoch": 0.00325,
+      "grad_norm": 0.9399377971403509,
+      "learning_rate": 0.0009750000000000001,
+      "loss": 6.0559,
+      "step": 325
+    },
+    {
+      "epoch": 0.00326,
+      "grad_norm": 1.1161084705724094,
+      "learning_rate": 0.0009780000000000001,
+      "loss": 6.0866,
+      "step": 326
+    },
+    {
+      "epoch": 0.00327,
+      "grad_norm": 0.9798577225908143,
+      "learning_rate": 0.000981,
+      "loss": 6.08,
+      "step": 327
+    },
+    {
+      "epoch": 0.00328,
+      "grad_norm": 1.4594200076446016,
+      "learning_rate": 0.000984,
+      "loss": 6.0949,
+      "step": 328
+    },
+    {
+      "epoch": 0.00329,
+      "grad_norm": 0.7530414187502621,
+      "learning_rate": 0.000987,
+      "loss": 6.0592,
+      "step": 329
+    },
+    {
+      "epoch": 0.0033,
+      "grad_norm": 0.9983029089238832,
+      "learning_rate": 0.00099,
+      "loss": 6.0707,
+      "step": 330
+    },
+    {
+      "epoch": 0.00331,
+      "grad_norm": 0.835334238631822,
+      "learning_rate": 0.0009930000000000002,
+      "loss": 6.0609,
+      "step": 331
+    },
+    {
+      "epoch": 0.00332,
+      "grad_norm": 1.2633352311024129,
+      "learning_rate": 0.0009960000000000001,
+      "loss": 6.0632,
+      "step": 332
+    },
+    {
+      "epoch": 0.00333,
+      "grad_norm": 0.9415989578137678,
+      "learning_rate": 0.000999,
+      "loss": 6.0388,
+      "step": 333
+    },
+    {
+      "epoch": 0.00334,
+      "grad_norm": 0.752057040577036,
+      "learning_rate": 0.001002,
+      "loss": 6.0434,
+      "step": 334
+    },
+    {
+      "epoch": 0.00335,
+      "grad_norm": 0.6368712711427076,
+      "learning_rate": 0.001005,
+      "loss": 6.039,
+      "step": 335
+    },
+    {
+      "epoch": 0.00336,
+      "grad_norm": 0.5277321252477392,
+      "learning_rate": 0.001008,
+      "loss": 6.0255,
+      "step": 336
+    },
+    {
+      "epoch": 0.00337,
+      "grad_norm": 0.545640031716998,
+      "learning_rate": 0.0010110000000000002,
+      "loss": 6.0051,
+      "step": 337
+    },
+    {
+      "epoch": 0.00338,
+      "grad_norm": 0.5605817392374447,
+      "learning_rate": 0.0010140000000000001,
+      "loss": 6.0183,
+      "step": 338
+    },
+    {
+      "epoch": 0.00339,
+      "grad_norm": 0.4963166785854256,
+      "learning_rate": 0.0010170000000000001,
+      "loss": 6.0214,
+      "step": 339
+    },
+    {
+      "epoch": 0.0034,
+      "grad_norm": 0.496225629593103,
+      "learning_rate": 0.00102,
+      "loss": 5.9915,
+      "step": 340
+    },
+    {
+      "epoch": 0.00341,
+      "grad_norm": 0.45013704339915594,
+      "learning_rate": 0.001023,
+      "loss": 5.9719,
+      "step": 341
+    },
+    {
+      "epoch": 0.00342,
+      "grad_norm": 0.37899658405778774,
+      "learning_rate": 0.001026,
+      "loss": 5.9765,
+      "step": 342
+    },
+    {
+      "epoch": 0.00343,
+      "grad_norm": 0.4216072327604619,
+      "learning_rate": 0.0010290000000000002,
+      "loss": 5.9773,
+      "step": 343
+    },
+    {
+      "epoch": 0.00344,
+      "grad_norm": 0.4964437378733662,
+      "learning_rate": 0.001032,
+      "loss": 5.9678,
+      "step": 344
+    },
+    {
+      "epoch": 0.00345,
+      "grad_norm": 0.6031142308234815,
+      "learning_rate": 0.001035,
+      "loss": 5.9605,
+      "step": 345
+    },
+    {
+      "epoch": 0.00346,
+      "grad_norm": 0.6349778009023861,
+      "learning_rate": 0.0010379999999999999,
+      "loss": 5.9675,
+      "step": 346
+    },
+    {
+      "epoch": 0.00347,
+      "grad_norm": 0.6213366174245898,
+      "learning_rate": 0.001041,
+      "loss": 5.9393,
+      "step": 347
+    },
+    {
+      "epoch": 0.00348,
+      "grad_norm": 0.684857817555668,
+      "learning_rate": 0.001044,
+      "loss": 5.9499,
+      "step": 348
+    },
+    {
+      "epoch": 0.00349,
+      "grad_norm": 0.8421661368559449,
+      "learning_rate": 0.001047,
+      "loss": 5.9398,
+      "step": 349
+    },
+    {
+      "epoch": 0.0035,
+      "grad_norm": 0.8062802732253019,
+      "learning_rate": 0.00105,
+      "loss": 5.9568,
+      "step": 350
+    },
+    {
+      "epoch": 0.00351,
+      "grad_norm": 0.7669273491138234,
+      "learning_rate": 0.001053,
+      "loss": 5.9549,
+      "step": 351
+    },
+    {
+      "epoch": 0.00352,
+      "grad_norm": 1.0673384228730578,
+      "learning_rate": 0.0010559999999999999,
+      "loss": 5.9491,
+      "step": 352
+    },
+    {
+      "epoch": 0.00353,
+      "grad_norm": 1.313057737158452,
+      "learning_rate": 0.001059,
+      "loss": 5.9596,
+      "step": 353
+    },
+    {
+      "epoch": 0.00354,
+      "grad_norm": 0.8748293544965601,
+      "learning_rate": 0.001062,
+      "loss": 5.9228,
+      "step": 354
+    },
+    {
+      "epoch": 0.00355,
+      "grad_norm": 0.7617258072087358,
+      "learning_rate": 0.001065,
+      "loss": 5.9348,
+      "step": 355
+    },
+    {
+      "epoch": 0.00356,
+      "grad_norm": 0.9561923051718775,
+      "learning_rate": 0.001068,
+      "loss": 5.9306,
+      "step": 356
+    },
+    {
+      "epoch": 0.00357,
+      "grad_norm": 0.8631216371911231,
+      "learning_rate": 0.001071,
+      "loss": 5.9199,
+      "step": 357
+    },
+    {
+      "epoch": 0.00358,
+      "grad_norm": 0.9525953351992685,
+      "learning_rate": 0.001074,
+      "loss": 5.9184,
+      "step": 358
+    },
+    {
+      "epoch": 0.00359,
+      "grad_norm": 0.6858307664974747,
+      "learning_rate": 0.001077,
+      "loss": 5.8973,
+      "step": 359
+    },
+    {
+      "epoch": 0.0036,
+      "grad_norm": 0.7167037627907079,
+      "learning_rate": 0.00108,
+      "loss": 5.9065,
+      "step": 360
+    },
+    {
+      "epoch": 0.00361,
+      "grad_norm": 0.7969047134484166,
+      "learning_rate": 0.001083,
+      "loss": 5.8986,
+      "step": 361
+    },
+    {
+      "epoch": 0.00362,
+      "grad_norm": 0.8186204145394074,
+      "learning_rate": 0.001086,
+      "loss": 5.888,
+      "step": 362
+    },
+    {
+      "epoch": 0.00363,
+      "grad_norm": 0.7436182013517663,
+      "learning_rate": 0.001089,
+      "loss": 5.8766,
+      "step": 363
+    },
+    {
+      "epoch": 0.00364,
+      "grad_norm": 0.6658023261534547,
+      "learning_rate": 0.001092,
+      "loss": 5.8872,
+      "step": 364
+    },
+    {
+      "epoch": 0.00365,
+      "grad_norm": 0.7206668214896482,
+      "learning_rate": 0.001095,
+      "loss": 5.8872,
+      "step": 365
+    },
+    {
+      "epoch": 0.00366,
+      "grad_norm": 0.6345413922647961,
+      "learning_rate": 0.001098,
+      "loss": 5.8617,
+      "step": 366
+    },
+    {
+      "epoch": 0.00367,
+      "grad_norm": 0.6094924231218852,
+      "learning_rate": 0.001101,
+      "loss": 5.8618,
+      "step": 367
+    },
+    {
+      "epoch": 0.00368,
+      "grad_norm": 0.7732260654201254,
+      "learning_rate": 0.001104,
+      "loss": 5.8653,
+      "step": 368
+    },
+    {
+      "epoch": 0.00369,
+      "grad_norm": 0.9451132874875877,
+      "learning_rate": 0.001107,
+      "loss": 5.8779,
+      "step": 369
+    },
+    {
+      "epoch": 0.0037,
+      "grad_norm": 1.047387946938651,
+      "learning_rate": 0.00111,
+      "loss": 5.8603,
+      "step": 370
+    },
+    {
+      "epoch": 0.00371,
+      "grad_norm": 0.8709497379832931,
+      "learning_rate": 0.001113,
+      "loss": 5.8262,
+      "step": 371
+    },
+    {
+      "epoch": 0.00372,
+      "grad_norm": 0.6993621521276565,
+      "learning_rate": 0.001116,
+      "loss": 5.8434,
+      "step": 372
+    },
+    {
+      "epoch": 0.00373,
+      "grad_norm": 0.6889019297226876,
+      "learning_rate": 0.001119,
+      "loss": 5.8344,
+      "step": 373
+    },
+    {
+      "epoch": 0.00374,
+      "grad_norm": 0.8187808414499578,
+      "learning_rate": 0.001122,
+      "loss": 5.8424,
+      "step": 374
+    },
+    {
+      "epoch": 0.00375,
+      "grad_norm": 0.7395556739972736,
+      "learning_rate": 0.0011250000000000001,
+      "loss": 5.8404,
+      "step": 375
+    },
+    {
+      "epoch": 0.00376,
+      "grad_norm": 0.5027446756427529,
+      "learning_rate": 0.001128,
+      "loss": 5.8296,
+      "step": 376
+    },
+    {
+      "epoch": 0.00377,
+      "grad_norm": 0.45297103786338255,
+      "learning_rate": 0.001131,
+      "loss": 5.8239,
+      "step": 377
+    },
+    {
+      "epoch": 0.00378,
+      "grad_norm": 0.5150298222384522,
+      "learning_rate": 0.001134,
+      "loss": 5.82,
+      "step": 378
+    },
+    {
+      "epoch": 0.00379,
+      "grad_norm": 0.4216428009711753,
+      "learning_rate": 0.001137,
+      "loss": 5.8036,
+      "step": 379
+    },
+    {
+      "epoch": 0.0038,
+      "grad_norm": 0.43574801532624385,
+      "learning_rate": 0.00114,
+      "loss": 5.8234,
+      "step": 380
+    },
+    {
+      "epoch": 0.00381,
+      "grad_norm": 0.4737821231317218,
+      "learning_rate": 0.0011430000000000001,
+      "loss": 5.7908,
+      "step": 381
+    },
+    {
+      "epoch": 0.00382,
+      "grad_norm": 0.5236457098681065,
+      "learning_rate": 0.001146,
+      "loss": 5.7778,
+      "step": 382
+    },
+    {
+      "epoch": 0.00383,
+      "grad_norm": 0.5006340354259897,
+      "learning_rate": 0.001149,
+      "loss": 5.787,
+      "step": 383
+    },
+    {
+      "epoch": 0.00384,
+      "grad_norm": 0.5762330042414852,
+      "learning_rate": 0.001152,
+      "loss": 5.7915,
+      "step": 384
+    },
+    {
+      "epoch": 0.00385,
+      "grad_norm": 0.6623935085819848,
+      "learning_rate": 0.001155,
+      "loss": 5.7808,
+      "step": 385
+    },
+    {
+      "epoch": 0.00386,
+      "grad_norm": 0.6780931079980513,
+      "learning_rate": 0.001158,
+      "loss": 5.7864,
+      "step": 386
+    },
+    {
+      "epoch": 0.00387,
+      "grad_norm": 0.6283557380392781,
+      "learning_rate": 0.0011610000000000001,
+      "loss": 5.76,
+      "step": 387
+    },
+    {
+      "epoch": 0.00388,
+      "grad_norm": 0.8661947460887196,
+      "learning_rate": 0.001164,
+      "loss": 5.7642,
+      "step": 388
+    },
+    {
+      "epoch": 0.00389,
+      "grad_norm": 1.1877567285678448,
+      "learning_rate": 0.001167,
+      "loss": 5.7898,
+      "step": 389
+    },
+    {
+      "epoch": 0.0039,
+      "grad_norm": 0.5285139722189788,
+      "learning_rate": 0.00117,
+      "loss": 5.7491,
+      "step": 390
+    },
+    {
+      "epoch": 0.00391,
+      "grad_norm": 0.6430118692881639,
+      "learning_rate": 0.001173,
+      "loss": 5.7622,
+      "step": 391
+    },
+    {
+      "epoch": 0.00392,
+      "grad_norm": 0.6748100403552353,
+      "learning_rate": 0.001176,
+      "loss": 5.7713,
+      "step": 392
+    },
+    {
+      "epoch": 0.00393,
+      "grad_norm": 0.7344199345621275,
+      "learning_rate": 0.0011790000000000001,
+      "loss": 5.7448,
+      "step": 393
+    },
+    {
+      "epoch": 0.00394,
+      "grad_norm": 0.8611653440542995,
+      "learning_rate": 0.001182,
+      "loss": 5.7377,
+      "step": 394
+    },
+    {
+      "epoch": 0.00395,
+      "grad_norm": 1.0200403565527223,
+      "learning_rate": 0.001185,
+      "loss": 5.7297,
+      "step": 395
+    },
+    {
+      "epoch": 0.00396,
+      "grad_norm": 1.3219082304761296,
+      "learning_rate": 0.001188,
+      "loss": 5.7571,
+      "step": 396
+    },
+    {
+      "epoch": 0.00397,
+      "grad_norm": 1.0648841185793536,
+      "learning_rate": 0.001191,
+      "loss": 5.7452,
+      "step": 397
+    },
+    {
+      "epoch": 0.00398,
+      "grad_norm": 0.8818852137031271,
+      "learning_rate": 0.0011940000000000002,
+      "loss": 5.7476,
+      "step": 398
+    },
+    {
+      "epoch": 0.00399,
+      "grad_norm": 0.7229082271352473,
+      "learning_rate": 0.0011970000000000001,
+      "loss": 5.7455,
+      "step": 399
+    },
+    {
+      "epoch": 0.004,
+      "grad_norm": 0.7464785543636239,
+      "learning_rate": 0.0012000000000000001,
+      "loss": 5.7335,
+      "step": 400
+    },
+    {
+      "epoch": 0.00401,
+      "grad_norm": 0.788008045277313,
+      "learning_rate": 0.001203,
+      "loss": 5.7147,
+      "step": 401
+    },
+    {
+      "epoch": 0.00402,
+      "grad_norm": 0.8844811042429299,
+      "learning_rate": 0.001206,
+      "loss": 5.7255,
+      "step": 402
+    },
+    {
+      "epoch": 0.00403,
+      "grad_norm": 0.8786909754931423,
+      "learning_rate": 0.001209,
+      "loss": 5.7201,
+      "step": 403
+    },
+    {
+      "epoch": 0.00404,
+      "grad_norm": 0.9763167280044874,
+      "learning_rate": 0.0012120000000000002,
+      "loss": 5.7188,
+      "step": 404
+    },
+    {
+      "epoch": 0.00405,
+      "grad_norm": 0.9708490732732808,
+      "learning_rate": 0.0012150000000000002,
+      "loss": 5.7216,
+      "step": 405
+    },
+    {
+      "epoch": 0.00406,
+      "grad_norm": 0.7953769652219763,
+      "learning_rate": 0.0012180000000000001,
+      "loss": 5.6987,
+      "step": 406
+    },
+    {
+      "epoch": 0.00407,
+      "grad_norm": 0.7231086032909518,
+      "learning_rate": 0.0012209999999999999,
+      "loss": 5.6945,
+      "step": 407
+    },
+    {
+      "epoch": 0.00408,
+      "grad_norm": 0.6539281657127057,
+      "learning_rate": 0.001224,
+      "loss": 5.7108,
+      "step": 408
+    },
+    {
+      "epoch": 0.00409,
+      "grad_norm": 0.48725126694443294,
+      "learning_rate": 0.001227,
+      "loss": 5.6827,
+      "step": 409
+    },
+    {
+      "epoch": 0.0041,
+      "grad_norm": 0.5935989194477762,
+      "learning_rate": 0.00123,
+      "loss": 5.6771,
+      "step": 410
+    },
+    {
+      "epoch": 0.00411,
+      "grad_norm": 0.6524672694061662,
+      "learning_rate": 0.001233,
+      "loss": 5.6945,
+      "step": 411
+    },
+    {
+      "epoch": 0.00412,
+      "grad_norm": 0.5686252583506046,
+      "learning_rate": 0.001236,
+      "loss": 5.6796,
+      "step": 412
+    },
+    {
+      "epoch": 0.00413,
+      "grad_norm": 0.5206794027062916,
+      "learning_rate": 0.0012389999999999999,
+      "loss": 5.676,
+      "step": 413
+    },
+    {
+      "epoch": 0.00414,
+      "grad_norm": 0.5100681259570212,
+      "learning_rate": 0.001242,
+      "loss": 5.6561,
+      "step": 414
+    },
+    {
+      "epoch": 0.00415,
+      "grad_norm": 0.5323494958966052,
+      "learning_rate": 0.001245,
+      "loss": 5.6572,
+      "step": 415
+    },
+    {
+      "epoch": 0.00416,
+      "grad_norm": 0.593833682795931,
+      "learning_rate": 0.001248,
+      "loss": 5.6419,
+      "step": 416
+    },
+    {
+      "epoch": 0.00417,
+      "grad_norm": 0.5881502982050868,
+      "learning_rate": 0.001251,
+      "loss": 5.6711,
+      "step": 417
+    },
+    {
+      "epoch": 0.00418,
+      "grad_norm": 0.6571826832460801,
+      "learning_rate": 0.001254,
+      "loss": 5.6451,
+      "step": 418
+    },
+    {
+      "epoch": 0.00419,
+      "grad_norm": 0.8242389183582979,
+      "learning_rate": 0.0012569999999999999,
+      "loss": 5.6393,
+      "step": 419
+    },
+    {
+      "epoch": 0.0042,
+      "grad_norm": 0.7322278119135938,
+      "learning_rate": 0.00126,
+      "loss": 5.6531,
+      "step": 420
+    },
+    {
+      "epoch": 0.00421,
+      "grad_norm": 0.6080486142741918,
+      "learning_rate": 0.001263,
+      "loss": 5.6415,
+      "step": 421
+    },
+    {
+      "epoch": 0.00422,
+      "grad_norm": 0.6271805214461816,
+      "learning_rate": 0.001266,
+      "loss": 5.6327,
+      "step": 422
+    },
+    {
+      "epoch": 0.00423,
+      "grad_norm": 0.5934268893913589,
+      "learning_rate": 0.001269,
+      "loss": 5.6286,
+      "step": 423
+    },
+    {
+      "epoch": 0.00424,
+      "grad_norm": 0.6795296945522938,
+      "learning_rate": 0.001272,
+      "loss": 5.6299,
+      "step": 424
+    },
+    {
+      "epoch": 0.00425,
+      "grad_norm": 0.7845276448161947,
+      "learning_rate": 0.001275,
+      "loss": 5.6344,
+      "step": 425
+    },
+    {
+      "epoch": 0.00426,
+      "grad_norm": 0.9176562117983922,
+      "learning_rate": 0.001278,
+      "loss": 5.6402,
+      "step": 426
+    },
+    {
+      "epoch": 0.00427,
+      "grad_norm": 1.0306205830362438,
+      "learning_rate": 0.001281,
+      "loss": 5.6293,
+      "step": 427
+    },
+    {
+      "epoch": 0.00428,
+      "grad_norm": 1.0252381178312269,
+      "learning_rate": 0.001284,
+      "loss": 5.6086,
+      "step": 428
+    },
+    {
+      "epoch": 0.00429,
+      "grad_norm": 1.3332162612637855,
+      "learning_rate": 0.001287,
+      "loss": 5.6393,
+      "step": 429
+    },
+    {
+      "epoch": 0.0043,
+      "grad_norm": 0.8174291079939794,
+      "learning_rate": 0.00129,
+      "loss": 5.6202,
+      "step": 430
+    },
+    {
+      "epoch": 0.00431,
+      "grad_norm": 0.821028463418781,
+      "learning_rate": 0.001293,
+      "loss": 5.6102,
+      "step": 431
+    },
+    {
+      "epoch": 0.00432,
+      "grad_norm": 0.7475137161763143,
+      "learning_rate": 0.001296,
+      "loss": 5.6163,
+      "step": 432
+    },
+    {
+      "epoch": 0.00433,
+      "grad_norm": 0.7571870052992741,
+      "learning_rate": 0.001299,
+      "loss": 5.604,
+      "step": 433
+    },
+    {
+      "epoch": 0.00434,
+      "grad_norm": 1.0543560255015263,
+      "learning_rate": 0.001302,
+      "loss": 5.6184,
+      "step": 434
+    },
+    {
+      "epoch": 0.00435,
+      "grad_norm": 0.8758758304266553,
+      "learning_rate": 0.001305,
+      "loss": 5.6066,
+      "step": 435
+    },
+    {
+      "epoch": 0.00436,
+      "grad_norm": 0.9137062421440546,
+      "learning_rate": 0.001308,
+      "loss": 5.5859,
+      "step": 436
+    },
+    {
+      "epoch": 0.00437,
+      "grad_norm": 1.301736875083812,
+      "learning_rate": 0.001311,
+      "loss": 5.6173,
+      "step": 437
+    },
+    {
+      "epoch": 0.00438,
+      "grad_norm": 0.9356953917037294,
+      "learning_rate": 0.001314,
+      "loss": 5.5919,
+      "step": 438
+    },
+    {
+      "epoch": 0.00439,
+      "grad_norm": 0.8522821992819578,
+      "learning_rate": 0.001317,
+      "loss": 5.6158,
+      "step": 439
+    },
+    {
+      "epoch": 0.0044,
+      "grad_norm": 0.7182761753028103,
+      "learning_rate": 0.00132,
+      "loss": 5.5821,
+      "step": 440
+    },
+    {
+      "epoch": 0.00441,
+      "grad_norm": 0.6293266892726601,
+      "learning_rate": 0.001323,
+      "loss": 5.577,
+      "step": 441
+    },
+    {
+      "epoch": 0.00442,
+      "grad_norm": 0.8189921873128356,
+      "learning_rate": 0.0013260000000000001,
+      "loss": 5.5762,
+      "step": 442
+    },
+    {
+      "epoch": 0.00443,
+      "grad_norm": 1.0212422821054057,
+      "learning_rate": 0.001329,
+      "loss": 5.5904,
+      "step": 443
+    },
+    {
+      "epoch": 0.00444,
+      "grad_norm": 0.8951785018901781,
+      "learning_rate": 0.001332,
+      "loss": 5.5851,
+      "step": 444
+    },
+    {
+      "epoch": 0.00445,
+      "grad_norm": 0.7868769140150608,
+      "learning_rate": 0.001335,
+      "loss": 5.5661,
+      "step": 445
+    },
+    {
+      "epoch": 0.00446,
+      "grad_norm": 0.790102612629763,
+      "learning_rate": 0.001338,
+      "loss": 5.5718,
+      "step": 446
+    },
+    {
+      "epoch": 0.00447,
+      "grad_norm": 0.8396194874372788,
+      "learning_rate": 0.001341,
+      "loss": 5.5716,
+      "step": 447
+    },
+    {
+      "epoch": 0.00448,
+      "grad_norm": 0.9120841535821665,
+      "learning_rate": 0.0013440000000000001,
+      "loss": 5.5589,
+      "step": 448
+    },
+    {
+      "epoch": 0.00449,
+      "grad_norm": 0.8573073152890212,
+      "learning_rate": 0.001347,
+      "loss": 5.5603,
+      "step": 449
+    },
+    {
+      "epoch": 0.0045,
+      "grad_norm": 0.7420732830276576,
+      "learning_rate": 0.00135,
+      "loss": 5.5551,
+      "step": 450
+    },
+    {
+      "epoch": 0.00451,
+      "grad_norm": 0.7054051762730813,
+      "learning_rate": 0.001353,
+      "loss": 5.5451,
+      "step": 451
+    },
+    {
+      "epoch": 0.00452,
+      "grad_norm": 0.5383194985068459,
+      "learning_rate": 0.001356,
+      "loss": 5.5229,
+      "step": 452
+    },
+    {
+      "epoch": 0.00453,
+      "grad_norm": 0.5845224529530345,
+      "learning_rate": 0.001359,
+      "loss": 5.5144,
+      "step": 453
+    },
+    {
+      "epoch": 0.00454,
+      "grad_norm": 0.5197277965757966,
+      "learning_rate": 0.0013620000000000001,
+      "loss": 5.5269,
+      "step": 454
+    },
+    {
+      "epoch": 0.00455,
+      "grad_norm": 0.44901215490386587,
+      "learning_rate": 0.0013650000000000001,
+      "loss": 5.5227,
+      "step": 455
+    },
+    {
+      "epoch": 0.00456,
+      "grad_norm": 0.40823042735319937,
+      "learning_rate": 0.001368,
+      "loss": 5.5155,
+      "step": 456
+    },
+    {
+      "epoch": 0.00457,
+      "grad_norm": 0.4077054091161063,
+      "learning_rate": 0.001371,
+      "loss": 5.5045,
+      "step": 457
+    },
+    {
+      "epoch": 0.00458,
+      "grad_norm": 0.4051721280659754,
+      "learning_rate": 0.001374,
+      "loss": 5.5115,
+      "step": 458
+    },
+    {
+      "epoch": 0.00459,
+      "grad_norm": 0.3819222938644956,
+      "learning_rate": 0.0013770000000000002,
+      "loss": 5.4986,
+      "step": 459
+    },
+    {
+      "epoch": 0.0046,
+      "grad_norm": 0.4096791477469412,
+      "learning_rate": 0.0013800000000000002,
+      "loss": 5.4851,
+      "step": 460
+    },
+    {
+      "epoch": 0.00461,
+      "grad_norm": 0.44167018915391465,
+      "learning_rate": 0.0013830000000000001,
+      "loss": 5.4805,
+      "step": 461
+    },
+    {
+      "epoch": 0.00462,
+      "grad_norm": 0.49346826910496583,
+      "learning_rate": 0.001386,
+      "loss": 5.4928,
+      "step": 462
+    },
+    {
+      "epoch": 0.00463,
+      "grad_norm": 0.463489633996114,
+      "learning_rate": 0.001389,
+      "loss": 5.4652,
+      "step": 463
+    },
+    {
+      "epoch": 0.00464,
+      "grad_norm": 0.507205937400668,
+      "learning_rate": 0.001392,
+      "loss": 5.4859,
+      "step": 464
+    },
+    {
+      "epoch": 0.00465,
+      "grad_norm": 0.5318639728571777,
+      "learning_rate": 0.0013950000000000002,
+      "loss": 5.4572,
+      "step": 465
+    },
+    {
+      "epoch": 0.00466,
+      "grad_norm": 0.6489252257104292,
+      "learning_rate": 0.0013980000000000002,
+      "loss": 5.4792,
+      "step": 466
+    },
+    {
+      "epoch": 0.00467,
+      "grad_norm": 0.7957789059873086,
+      "learning_rate": 0.0014010000000000001,
+      "loss": 5.4727,
+      "step": 467
+    },
+    {
+      "epoch": 0.00468,
+      "grad_norm": 0.7621042481808248,
+      "learning_rate": 0.001404,
+      "loss": 5.462,
+      "step": 468
+    },
+    {
+      "epoch": 0.00469,
+      "grad_norm": 0.573822393217689,
+      "learning_rate": 0.001407,
+      "loss": 5.457,
+      "step": 469
+    },
+    {
+      "epoch": 0.0047,
+      "grad_norm": 0.7543746323219513,
+      "learning_rate": 0.00141,
+      "loss": 5.4709,
+      "step": 470
+    },
+    {
+      "epoch": 0.00471,
+      "grad_norm": 0.8641651762100534,
+      "learning_rate": 0.001413,
+      "loss": 5.4567,
+      "step": 471
+    },
+    {
+      "epoch": 0.00472,
+      "grad_norm": 0.9026674726231502,
+      "learning_rate": 0.001416,
+      "loss": 5.4699,
+      "step": 472
+    },
+    {
+      "epoch": 0.00473,
+      "grad_norm": 0.8212885101559565,
+      "learning_rate": 0.001419,
+      "loss": 5.4778,
+      "step": 473
+    },
+    {
+      "epoch": 0.00474,
+      "grad_norm": 0.8914030740906659,
+      "learning_rate": 0.0014219999999999999,
+      "loss": 5.461,
+      "step": 474
+    },
+    {
+      "epoch": 0.00475,
+      "grad_norm": 1.0570094425693455,
+      "learning_rate": 0.001425,
+      "loss": 5.4652,
+      "step": 475
+    },
+    {
+      "epoch": 0.00476,
+      "grad_norm": 0.9736444976311589,
+      "learning_rate": 0.001428,
+      "loss": 5.4875,
+      "step": 476
+    },
+    {
+      "epoch": 0.00477,
+      "grad_norm": 1.1550380737092787,
+      "learning_rate": 0.001431,
+      "loss": 5.4568,
+      "step": 477
+    },
+    {
+      "epoch": 0.00478,
+      "grad_norm": 0.848321570803796,
+      "learning_rate": 0.001434,
+      "loss": 5.4695,
+      "step": 478
+    },
+    {
+      "epoch": 0.00479,
+      "grad_norm": 0.9517827225501269,
+      "learning_rate": 0.001437,
+      "loss": 5.4501,
+      "step": 479
+    },
+    {
+      "epoch": 0.0048,
+      "grad_norm": 1.0883787540754652,
+      "learning_rate": 0.0014399999999999999,
+      "loss": 5.4562,
+      "step": 480
+    },
+    {
+      "epoch": 0.00481,
+      "grad_norm": 0.9422991164230814,
+      "learning_rate": 0.001443,
+      "loss": 5.4516,
+      "step": 481
+    },
+    {
+      "epoch": 0.00482,
+      "grad_norm": 0.9159499791385636,
+      "learning_rate": 0.001446,
+      "loss": 5.4273,
+      "step": 482
+    },
+    {
+      "epoch": 0.00483,
+      "grad_norm": 0.9688645055474768,
+      "learning_rate": 0.001449,
+      "loss": 5.44,
+      "step": 483
+    },
+    {
+      "epoch": 0.00484,
+      "grad_norm": 1.1114303023214132,
+      "learning_rate": 0.001452,
+      "loss": 5.4474,
+      "step": 484
+    },
+    {
+      "epoch": 0.00485,
+      "grad_norm": 0.9051569573634253,
+      "learning_rate": 0.001455,
+      "loss": 5.4468,
+      "step": 485
+    },
+    {
+      "epoch": 0.00486,
+      "grad_norm": 0.9247454458721566,
+      "learning_rate": 0.001458,
+      "loss": 5.4217,
+      "step": 486
+    },
+    {
+      "epoch": 0.00487,
+      "grad_norm": 0.852928162562673,
+      "learning_rate": 0.001461,
+      "loss": 5.4339,
+      "step": 487
+    },
+    {
+      "epoch": 0.00488,
+      "grad_norm": 0.8142513932978471,
+      "learning_rate": 0.001464,
+      "loss": 5.41,
+      "step": 488
+    },
+    {
+      "epoch": 0.00489,
+      "grad_norm": 0.9329231917883894,
+      "learning_rate": 0.001467,
+      "loss": 5.403,
+      "step": 489
+    },
+    {
+      "epoch": 0.0049,
+      "grad_norm": 0.9855383918059676,
+      "learning_rate": 0.00147,
+      "loss": 5.4217,
+      "step": 490
+    },
+    {
+      "epoch": 0.00491,
+      "grad_norm": 0.7649189111405409,
+      "learning_rate": 0.001473,
+      "loss": 5.4131,
+      "step": 491
+    },
+    {
+      "epoch": 0.00492,
+      "grad_norm": 0.7909135941762935,
+      "learning_rate": 0.001476,
+      "loss": 5.4088,
+      "step": 492
+    },
+    {
+      "epoch": 0.00493,
+      "grad_norm": 0.9895574066894314,
+      "learning_rate": 0.001479,
+      "loss": 5.412,
+      "step": 493
+    },
+    {
+      "epoch": 0.00494,
+      "grad_norm": 1.132991334341666,
+      "learning_rate": 0.001482,
+      "loss": 5.429,
+      "step": 494
+    },
+    {
+      "epoch": 0.00495,
+      "grad_norm": 0.761507737933228,
+      "learning_rate": 0.001485,
+      "loss": 5.4134,
+      "step": 495
+    },
+    {
+      "epoch": 0.00496,
+      "grad_norm": 0.9726159326361932,
+      "learning_rate": 0.001488,
+      "loss": 5.4067,
+      "step": 496
+    },
+    {
+      "epoch": 0.00497,
+      "grad_norm": 1.1482864163713484,
+      "learning_rate": 0.001491,
+      "loss": 5.3997,
+      "step": 497
+    },
+    {
+      "epoch": 0.00498,
+      "grad_norm": 0.8327332046897746,
+      "learning_rate": 0.001494,
+      "loss": 5.4147,
+      "step": 498
+    },
+    {
+      "epoch": 0.00499,
+      "grad_norm": 0.725916611519047,
+      "learning_rate": 0.001497,
+      "loss": 5.3792,
+      "step": 499
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 0.6719782846245154,
+      "learning_rate": 0.0015,
+      "loss": 5.3842,
+      "step": 500
+    },
+    {
+      "epoch": 0.00501,
+      "grad_norm": 0.7463076928465905,
+      "learning_rate": 0.001503,
+      "loss": 5.3796,
+      "step": 501
+    },
+    {
+      "epoch": 0.00502,
+      "grad_norm": 0.8408104186601356,
+      "learning_rate": 0.001506,
+      "loss": 5.389,
+      "step": 502
+    },
+    {
+      "epoch": 0.00503,
+      "grad_norm": 0.860909611402231,
+      "learning_rate": 0.0015090000000000001,
+      "loss": 5.3926,
+      "step": 503
+    },
+    {
+      "epoch": 0.00504,
+      "grad_norm": 0.7026490274993983,
+      "learning_rate": 0.001512,
+      "loss": 5.3646,
+      "step": 504
+    },
+    {
+      "epoch": 0.00505,
+      "grad_norm": 0.7184807991697565,
+      "learning_rate": 0.001515,
+      "loss": 5.3547,
+      "step": 505
+    },
+    {
+      "epoch": 0.00506,
+      "grad_norm": 0.7839475253802514,
+      "learning_rate": 0.001518,
+      "loss": 5.3611,
+      "step": 506
+    },
+    {
+      "epoch": 0.00507,
+      "grad_norm": 0.7039499110993044,
+      "learning_rate": 0.001521,
+      "loss": 5.3522,
+      "step": 507
+    },
+    {
+      "epoch": 0.00508,
+      "grad_norm": 0.5587706859739108,
+      "learning_rate": 0.001524,
+      "loss": 5.3512,
+      "step": 508
+    },
+    {
+      "epoch": 0.00509,
+      "grad_norm": 0.4952941325228141,
+      "learning_rate": 0.0015270000000000001,
+      "loss": 5.326,
+      "step": 509
+    },
+    {
+      "epoch": 0.0051,
+      "grad_norm": 0.5131490664852795,
+      "learning_rate": 0.0015300000000000001,
+      "loss": 5.3428,
+      "step": 510
+    },
+    {
+      "epoch": 0.00511,
+      "grad_norm": 0.5539213487194597,
+      "learning_rate": 0.001533,
+      "loss": 5.3196,
+      "step": 511
+    },
+    {
+      "epoch": 0.00512,
+      "grad_norm": 0.5937876471208409,
+      "learning_rate": 0.001536,
+      "loss": 5.3158,
+      "step": 512
+    },
+    {
+      "epoch": 0.00513,
+      "grad_norm": 0.5441672327389838,
+      "learning_rate": 0.001539,
+      "loss": 5.327,
+      "step": 513
+    },
+    {
+      "epoch": 0.00514,
+      "grad_norm": 0.5371789671410057,
+      "learning_rate": 0.001542,
+      "loss": 5.3038,
+      "step": 514
+    },
+    {
+      "epoch": 0.00515,
+      "grad_norm": 0.5194765862771661,
+      "learning_rate": 0.0015450000000000001,
+      "loss": 5.3109,
+      "step": 515
+    },
+    {
+      "epoch": 0.00516,
+      "grad_norm": 0.5575198815834714,
+      "learning_rate": 0.0015480000000000001,
+      "loss": 5.3083,
+      "step": 516
+    },
+    {
+      "epoch": 0.00517,
+      "grad_norm": 0.5237583962443445,
+      "learning_rate": 0.001551,
+      "loss": 5.2962,
+      "step": 517
+    },
+    {
+      "epoch": 0.00518,
+      "grad_norm": 0.5237181167506534,
+      "learning_rate": 0.001554,
+      "loss": 5.2786,
+      "step": 518
+    },
+    {
+      "epoch": 0.00519,
+      "grad_norm": 0.6652146969315359,
+      "learning_rate": 0.001557,
+      "loss": 5.2847,
+      "step": 519
+    },
+    {
+      "epoch": 0.0052,
+      "grad_norm": 0.8817258231902962,
+      "learning_rate": 0.0015600000000000002,
+      "loss": 5.2966,
+      "step": 520
+    },
+    {
+      "epoch": 0.00521,
+      "grad_norm": 0.9047592596397305,
+      "learning_rate": 0.0015630000000000002,
+      "loss": 5.2732,
+      "step": 521
+    },
+    {
+      "epoch": 0.00522,
+      "grad_norm": 0.6139926424217688,
+      "learning_rate": 0.0015660000000000001,
+      "loss": 5.2701,
+      "step": 522
+    },
+    {
+      "epoch": 0.00523,
+      "grad_norm": 0.6292464658556638,
+      "learning_rate": 0.001569,
+      "loss": 5.2826,
+      "step": 523
+    },
+    {
+      "epoch": 0.00524,
+      "grad_norm": 0.752030715547053,
+      "learning_rate": 0.001572,
+      "loss": 5.2856,
+      "step": 524
+    },
+    {
+      "epoch": 0.00525,
+      "grad_norm": 0.929770536798091,
+      "learning_rate": 0.001575,
+      "loss": 5.287,
+      "step": 525
+    },
+    {
+      "epoch": 0.00526,
+      "grad_norm": 0.887370520628206,
+      "learning_rate": 0.0015780000000000002,
+      "loss": 5.2635,
+      "step": 526
+    },
+    {
+      "epoch": 0.00527,
+      "grad_norm": 0.7819104471156305,
+      "learning_rate": 0.0015810000000000002,
+      "loss": 5.2728,
+      "step": 527
+    },
+    {
+      "epoch": 0.00528,
+      "grad_norm": 0.9038037239389326,
+      "learning_rate": 0.0015840000000000001,
+      "loss": 5.2534,
+      "step": 528
+    },
+    {
+      "epoch": 0.00529,
+      "grad_norm": 0.8898068580069259,
+      "learning_rate": 0.001587,
+      "loss": 5.265,
+      "step": 529
+    },
+    {
+      "epoch": 0.0053,
+      "grad_norm": 1.0010848916209774,
+      "learning_rate": 0.00159,
+      "loss": 5.2764,
+      "step": 530
+    },
+    {
+      "epoch": 0.00531,
+      "grad_norm": 1.0010984765594055,
+      "learning_rate": 0.001593,
+      "loss": 5.2677,
+      "step": 531
+    },
+    {
+      "epoch": 0.00532,
+      "grad_norm": 1.026224264336229,
+      "learning_rate": 0.0015960000000000002,
+      "loss": 5.2779,
+      "step": 532
+    },
+    {
+      "epoch": 0.00533,
+      "grad_norm": 0.9759289805508353,
+      "learning_rate": 0.0015990000000000002,
+      "loss": 5.2678,
+      "step": 533
+    },
+    {
+      "epoch": 0.00534,
+      "grad_norm": 1.1376558518204782,
+      "learning_rate": 0.0016020000000000001,
+      "loss": 5.2612,
+      "step": 534
+    },
+    {
+      "epoch": 0.00535,
+      "grad_norm": 1.0517455067238486,
+      "learning_rate": 0.001605,
+      "loss": 5.2672,
+      "step": 535
+    },
+    {
+      "epoch": 0.00536,
+      "grad_norm": 0.9398243649272562,
+      "learning_rate": 0.001608,
+      "loss": 5.2627,
+      "step": 536
+    },
+    {
+      "epoch": 0.00537,
+      "grad_norm": 0.9512995727424398,
+      "learning_rate": 0.0016110000000000002,
+      "loss": 5.2558,
+      "step": 537
+    },
+    {
+      "epoch": 0.00538,
+      "grad_norm": 1.2746657535312511,
+      "learning_rate": 0.0016140000000000002,
+      "loss": 5.2451,
+      "step": 538
+    },
+    {
+      "epoch": 0.00539,
+      "grad_norm": 1.067518036012326,
+      "learning_rate": 0.0016170000000000002,
+      "loss": 5.2415,
+      "step": 539
+    },
+    {
+      "epoch": 0.0054,
+      "grad_norm": 1.1859628541053806,
+      "learning_rate": 0.0016200000000000001,
+      "loss": 5.2618,
+      "step": 540
+    },
+    {
+      "epoch": 0.00541,
+      "grad_norm": 0.8882936824028492,
+      "learning_rate": 0.001623,
+      "loss": 5.2308,
+      "step": 541
+    },
+    {
+      "epoch": 0.00542,
+      "grad_norm": 0.8517075205384302,
+      "learning_rate": 0.001626,
+      "loss": 5.2545,
+      "step": 542
+    },
+    {
+      "epoch": 0.00543,
+      "grad_norm": 0.8283552605034004,
+      "learning_rate": 0.0016290000000000002,
+      "loss": 5.2098,
+      "step": 543
+    },
+    {
+      "epoch": 0.00544,
+      "grad_norm": 0.9087829134911761,
+      "learning_rate": 0.0016320000000000002,
+      "loss": 5.2265,
+      "step": 544
+    },
+    {
+      "epoch": 0.00545,
+      "grad_norm": 0.8034144620978907,
+      "learning_rate": 0.0016350000000000002,
+      "loss": 5.2348,
+      "step": 545
+    },
+    {
+      "epoch": 0.00546,
+      "grad_norm": 0.7091235133563132,
+      "learning_rate": 0.0016380000000000001,
+      "loss": 5.2004,
+      "step": 546
+    },
+    {
+      "epoch": 0.00547,
+      "grad_norm": 0.6683331586466694,
+      "learning_rate": 0.001641,
+      "loss": 5.1915,
+      "step": 547
+    },
+    {
+      "epoch": 0.00548,
+      "grad_norm": 0.5441552662279447,
+      "learning_rate": 0.001644,
+      "loss": 5.1843,
+      "step": 548
+    },
+    {
+      "epoch": 0.00549,
+      "grad_norm": 0.5462993858197037,
+      "learning_rate": 0.0016470000000000002,
+      "loss": 5.1806,
+      "step": 549
+    },
+    {
+      "epoch": 0.0055,
+      "grad_norm": 0.5691406737163984,
+      "learning_rate": 0.0016500000000000002,
+      "loss": 5.1914,
+      "step": 550
+    },
+    {
+      "epoch": 0.00551,
+      "grad_norm": 0.57998143312047,
+      "learning_rate": 0.0016530000000000002,
+      "loss": 5.1789,
+      "step": 551
+    },
+    {
+      "epoch": 0.00552,
+      "grad_norm": 0.5286482487653069,
+      "learning_rate": 0.0016560000000000001,
+      "loss": 5.158,
+      "step": 552
+    },
+    {
+      "epoch": 0.00553,
+      "grad_norm": 0.48759014943874474,
+      "learning_rate": 0.001659,
+      "loss": 5.1465,
+      "step": 553
+    },
+    {
+      "epoch": 0.00554,
+      "grad_norm": 0.4792524526847805,
+      "learning_rate": 0.0016620000000000003,
+      "loss": 5.1537,
+      "step": 554
+    },
+    {
+      "epoch": 0.00555,
+      "grad_norm": 0.5054837577144806,
+      "learning_rate": 0.0016650000000000002,
+      "loss": 5.1496,
+      "step": 555
+    },
+    {
+      "epoch": 0.00556,
+      "grad_norm": 0.5148667639200912,
+      "learning_rate": 0.0016680000000000002,
+      "loss": 5.1288,
+      "step": 556
+    },
+    {
+      "epoch": 0.00557,
+      "grad_norm": 0.5122706189056161,
+      "learning_rate": 0.0016710000000000002,
+      "loss": 5.1153,
+      "step": 557
+    },
+    {
+      "epoch": 0.00558,
+      "grad_norm": 0.6046224327964763,
+      "learning_rate": 0.0016740000000000001,
+      "loss": 5.1358,
+      "step": 558
+    },
+    {
+      "epoch": 0.00559,
+      "grad_norm": 0.6851636176151574,
+      "learning_rate": 0.001677,
+      "loss": 5.1268,
+      "step": 559
+    },
+    {
+      "epoch": 0.0056,
+      "grad_norm": 0.6922547745331437,
+      "learning_rate": 0.0016800000000000003,
+      "loss": 5.096,
+      "step": 560
+    },
+    {
+      "epoch": 0.00561,
+      "grad_norm": 0.7411075733344746,
+      "learning_rate": 0.0016830000000000003,
+      "loss": 5.1176,
+      "step": 561
+    },
+    {
+      "epoch": 0.00562,
+      "grad_norm": 0.8981627852593407,
+      "learning_rate": 0.0016860000000000002,
+      "loss": 5.1206,
+      "step": 562
+    },
+    {
+      "epoch": 0.00563,
+      "grad_norm": 1.1260200882948381,
+      "learning_rate": 0.001689,
+      "loss": 5.1239,
+      "step": 563
+    },
+    {
+      "epoch": 0.00564,
+      "grad_norm": 1.1027210513289374,
+      "learning_rate": 0.001692,
+      "loss": 5.113,
+      "step": 564
+    },
+    {
+      "epoch": 0.00565,
+      "grad_norm": 0.890213024695838,
+      "learning_rate": 0.001695,
+      "loss": 5.1071,
+      "step": 565
+    },
+    {
+      "epoch": 0.00566,
+      "grad_norm": 0.8627979775394023,
+      "learning_rate": 0.0016979999999999999,
+      "loss": 5.0976,
+      "step": 566
+    },
+    {
+      "epoch": 0.00567,
+      "grad_norm": 0.9823945407746334,
+      "learning_rate": 0.0017009999999999998,
+      "loss": 5.101,
+      "step": 567
+    },
+    {
+      "epoch": 0.00568,
+      "grad_norm": 0.9715849000742567,
+      "learning_rate": 0.0017039999999999998,
+      "loss": 5.1139,
+      "step": 568
+    },
+    {
+      "epoch": 0.00569,
+      "grad_norm": 0.9989913016974431,
+      "learning_rate": 0.001707,
+      "loss": 5.0861,
+      "step": 569
+    },
+    {
+      "epoch": 0.0057,
+      "grad_norm": 1.0927877338999235,
+      "learning_rate": 0.00171,
+      "loss": 5.1262,
+      "step": 570
+    },
+    {
+      "epoch": 0.00571,
+      "grad_norm": 1.0328737042176641,
+      "learning_rate": 0.001713,
+      "loss": 5.1192,
+      "step": 571
+    },
+    {
+      "epoch": 0.00572,
+      "grad_norm": 1.3722388350456287,
+      "learning_rate": 0.0017159999999999999,
+      "loss": 5.1049,
+      "step": 572
+    },
+    {
+      "epoch": 0.00573,
+      "grad_norm": 0.9525979203379623,
+      "learning_rate": 0.0017189999999999998,
+      "loss": 5.1081,
+      "step": 573
+    },
+    {
+      "epoch": 0.00574,
+      "grad_norm": 1.0626128534442882,
+      "learning_rate": 0.001722,
+      "loss": 5.1048,
+      "step": 574
+    },
+    {
+      "epoch": 0.00575,
+      "grad_norm": 0.9331527599734185,
+      "learning_rate": 0.001725,
+      "loss": 5.074,
+      "step": 575
+    },
+    {
+      "epoch": 0.00576,
+      "grad_norm": 0.9277735357052385,
+      "learning_rate": 0.001728,
+      "loss": 5.0843,
+      "step": 576
+    },
+    {
+      "epoch": 0.00577,
+      "grad_norm": 0.9070967517565243,
+      "learning_rate": 0.001731,
+      "loss": 5.0908,
+      "step": 577
+    },
+    {
+      "epoch": 0.00578,
+      "grad_norm": 0.8451551366430134,
+      "learning_rate": 0.0017339999999999999,
+      "loss": 5.0704,
+      "step": 578
+    },
+    {
+      "epoch": 0.00579,
+      "grad_norm": 0.7590562073625285,
+      "learning_rate": 0.0017369999999999998,
+      "loss": 5.058,
+      "step": 579
+    },
+    {
+      "epoch": 0.0058,
+      "grad_norm": 0.6385326977156373,
+      "learning_rate": 0.00174,
+      "loss": 5.0662,
+      "step": 580
+    },
+    {
+      "epoch": 0.00581,
+      "grad_norm": 0.5982129206576257,
+      "learning_rate": 0.001743,
+      "loss": 5.0624,
+      "step": 581
+    },
+    {
+      "epoch": 0.00582,
+      "grad_norm": 0.7358321954275717,
+      "learning_rate": 0.001746,
+      "loss": 5.0371,
+      "step": 582
+    },
+    {
+      "epoch": 0.00583,
+      "grad_norm": 0.790704997209863,
+      "learning_rate": 0.001749,
+      "loss": 5.0641,
+      "step": 583
+    },
+    {
+      "epoch": 0.00584,
+      "grad_norm": 0.7351931257233056,
+      "learning_rate": 0.0017519999999999999,
+      "loss": 5.0549,
+      "step": 584
+    },
+    {
+      "epoch": 0.00585,
+      "grad_norm": 0.6167540927458872,
+      "learning_rate": 0.0017549999999999998,
+      "loss": 5.0362,
+      "step": 585
+    },
+    {
+      "epoch": 0.00586,
+      "grad_norm": 0.6143004439139178,
+      "learning_rate": 0.001758,
+      "loss": 5.0316,
+      "step": 586
+    },
+    {
+      "epoch": 0.00587,
+      "grad_norm": 0.619830997492515,
+      "learning_rate": 0.001761,
+      "loss": 5.018,
+      "step": 587
+    },
+    {
+      "epoch": 0.00588,
+      "grad_norm": 0.676754014852622,
+      "learning_rate": 0.001764,
+      "loss": 5.0113,
+      "step": 588
+    },
+    {
+      "epoch": 0.00589,
+      "grad_norm": 0.8152367183756798,
+      "learning_rate": 0.001767,
+      "loss": 5.0063,
+      "step": 589
+    },
+    {
+      "epoch": 0.0059,
+      "grad_norm": 0.8067730505459064,
+      "learning_rate": 0.0017699999999999999,
+      "loss": 5.0009,
+      "step": 590
+    },
+    {
+      "epoch": 0.00591,
+      "grad_norm": 0.7857115694134444,
+      "learning_rate": 0.001773,
+      "loss": 5.0014,
+      "step": 591
+    },
+    {
+      "epoch": 0.00592,
+      "grad_norm": 1.0305248086573016,
+      "learning_rate": 0.001776,
+      "loss": 5.0384,
+      "step": 592
+    },
+    {
+      "epoch": 0.00593,
+      "grad_norm": 1.1034322728795254,
+      "learning_rate": 0.001779,
+      "loss": 5.0387,
+      "step": 593
+    },
+    {
+      "epoch": 0.00594,
+      "grad_norm": 0.796782800390064,
+      "learning_rate": 0.001782,
+      "loss": 5.0138,
+      "step": 594
+    },
+    {
+      "epoch": 0.00595,
+      "grad_norm": 0.947951394101208,
+      "learning_rate": 0.001785,
+      "loss": 5.0288,
+      "step": 595
+    },
+    {
+      "epoch": 0.00596,
+      "grad_norm": 0.948001541672118,
+      "learning_rate": 0.0017879999999999999,
+      "loss": 5.0236,
+      "step": 596
+    },
+    {
+      "epoch": 0.00597,
+      "grad_norm": 0.8614311642183788,
+      "learning_rate": 0.001791,
+      "loss": 4.983,
+      "step": 597
+    },
+    {
+      "epoch": 0.00598,
+      "grad_norm": 0.7822502963641336,
+      "learning_rate": 0.001794,
+      "loss": 5.0032,
+      "step": 598
+    },
+    {
+      "epoch": 0.00599,
+      "grad_norm": 0.662502314672726,
+      "learning_rate": 0.001797,
+      "loss": 4.9824,
+      "step": 599
+    },
+    {
+      "epoch": 0.006,
+      "grad_norm": 0.612691797581944,
+      "learning_rate": 0.0018,
+      "loss": 4.9865,
+      "step": 600
+    },
+    {
+      "epoch": 0.00601,
+      "grad_norm": 0.6872791409552744,
+      "learning_rate": 0.001803,
+      "loss": 4.9768,
+      "step": 601
+    },
+    {
+      "epoch": 0.00602,
+      "grad_norm": 0.6805787880992017,
+      "learning_rate": 0.0018059999999999999,
+      "loss": 4.9611,
+      "step": 602
+    },
+    {
+      "epoch": 0.00603,
+      "grad_norm": 0.6726518476757681,
+      "learning_rate": 0.001809,
+      "loss": 4.9863,
+      "step": 603
+    },
+    {
+      "epoch": 0.00604,
+      "grad_norm": 0.6569683723636671,
+      "learning_rate": 0.001812,
+      "loss": 4.9594,
+      "step": 604
+    },
+    {
+      "epoch": 0.00605,
+      "grad_norm": 0.7635898698708745,
+      "learning_rate": 0.001815,
+      "loss": 4.9733,
+      "step": 605
+    },
+    {
+      "epoch": 0.00606,
+      "grad_norm": 0.7455554600622951,
+      "learning_rate": 0.001818,
+      "loss": 4.9743,
+      "step": 606
+    },
+    {
+      "epoch": 0.00607,
+      "grad_norm": 0.7493825131119356,
+      "learning_rate": 0.001821,
+      "loss": 4.968,
+      "step": 607
+    },
+    {
+      "epoch": 0.00608,
+      "grad_norm": 0.7957828614510537,
+      "learning_rate": 0.001824,
+      "loss": 4.9536,
+      "step": 608
+    },
+    {
+      "epoch": 0.00609,
+      "grad_norm": 0.8304899179729943,
+      "learning_rate": 0.001827,
+      "loss": 4.9564,
+      "step": 609
+    },
+    {
+      "epoch": 0.0061,
+      "grad_norm": 0.8109281284885322,
+      "learning_rate": 0.00183,
+      "loss": 4.9141,
+      "step": 610
+    },
+    {
+      "epoch": 0.00611,
+      "grad_norm": 0.7476158062529945,
+      "learning_rate": 0.001833,
+      "loss": 4.936,
+      "step": 611
+    },
+    {
+      "epoch": 0.00612,
+      "grad_norm": 0.7704555466663776,
+      "learning_rate": 0.001836,
+      "loss": 4.9262,
+      "step": 612
+    },
+    {
+      "epoch": 0.00613,
+      "grad_norm": 0.7179188189545482,
+      "learning_rate": 0.001839,
+      "loss": 4.9285,
+      "step": 613
+    },
+    {
+      "epoch": 0.00614,
+      "grad_norm": 0.6487628040476678,
+      "learning_rate": 0.001842,
+      "loss": 4.9483,
+      "step": 614
+    },
+    {
+      "epoch": 0.00615,
+      "grad_norm": 0.6346341426047015,
+      "learning_rate": 0.001845,
+      "loss": 4.9336,
+      "step": 615
+    },
+    {
+      "epoch": 0.00616,
+      "grad_norm": 0.6919482296512011,
+      "learning_rate": 0.001848,
+      "loss": 4.92,
+      "step": 616
+    },
+    {
+      "epoch": 0.00617,
+      "grad_norm": 0.8470324326487232,
+      "learning_rate": 0.001851,
+      "loss": 4.937,
+      "step": 617
+    },
+    {
+      "epoch": 0.00618,
+      "grad_norm": 0.9780451420741652,
+      "learning_rate": 0.001854,
+      "loss": 4.918,
+      "step": 618
+    },
+    {
+      "epoch": 0.00619,
+      "grad_norm": 1.0393940937639112,
+      "learning_rate": 0.001857,
+      "loss": 4.9393,
+      "step": 619
+    },
+    {
+      "epoch": 0.0062,
+      "grad_norm": 1.0864888594503201,
+      "learning_rate": 0.00186,
+      "loss": 4.9315,
+      "step": 620
+    },
+    {
+      "epoch": 0.00621,
+      "grad_norm": 1.1243573748169302,
+      "learning_rate": 0.001863,
+      "loss": 4.931,
+      "step": 621
+    },
+    {
+      "epoch": 0.00622,
+      "grad_norm": 1.2701399731469665,
+      "learning_rate": 0.001866,
+      "loss": 4.9502,
+      "step": 622
+    },
+    {
+      "epoch": 0.00623,
+      "grad_norm": 0.8273605340089976,
+      "learning_rate": 0.001869,
+      "loss": 4.9139,
+      "step": 623
+    },
+    {
+      "epoch": 0.00624,
+      "grad_norm": 0.9643311427939416,
+      "learning_rate": 0.001872,
+      "loss": 4.9253,
+      "step": 624
+    },
+    {
+      "epoch": 0.00625,
+      "grad_norm": 1.2310098040523034,
+      "learning_rate": 0.001875,
+      "loss": 4.9539,
+      "step": 625
+    },
+    {
+      "epoch": 0.00626,
+      "grad_norm": 0.8268848993203812,
+      "learning_rate": 0.0018780000000000001,
+      "loss": 4.9199,
+      "step": 626
+    },
+    {
+      "epoch": 0.00627,
+      "grad_norm": 0.9575639973355906,
+      "learning_rate": 0.001881,
+      "loss": 4.909,
+      "step": 627
+    },
+    {
+      "epoch": 0.00628,
+      "grad_norm": 0.9391450767019313,
+      "learning_rate": 0.001884,
+      "loss": 4.9487,
+      "step": 628
+    },
+    {
+      "epoch": 0.00629,
+      "grad_norm": 0.8387975784827085,
+      "learning_rate": 0.001887,
+      "loss": 4.8976,
+      "step": 629
+    },
+    {
+      "epoch": 0.0063,
+      "grad_norm": 0.7745062985579545,
+      "learning_rate": 0.00189,
+      "loss": 4.9062,
+      "step": 630
+    },
+    {
+      "epoch": 0.00631,
+      "grad_norm": 0.8213366097020988,
+      "learning_rate": 0.0018930000000000002,
+      "loss": 4.9006,
+      "step": 631
+    },
+    {
+      "epoch": 0.00632,
+      "grad_norm": 0.8670608316656828,
+      "learning_rate": 0.0018960000000000001,
+      "loss": 4.9226,
+      "step": 632
+    },
+    {
+      "epoch": 0.00633,
+      "grad_norm": 0.9183940724690223,
+      "learning_rate": 0.001899,
+      "loss": 4.9055,
+      "step": 633
+    },
+    {
+      "epoch": 0.00634,
+      "grad_norm": 0.8953336214105924,
+      "learning_rate": 0.001902,
+      "loss": 4.8931,
+      "step": 634
+    },
+    {
+      "epoch": 0.00635,
+      "grad_norm": 0.8869843611928877,
+      "learning_rate": 0.001905,
+      "loss": 4.8851,
+      "step": 635
+    },
+    {
+      "epoch": 0.00636,
+      "grad_norm": 1.014617544478078,
+      "learning_rate": 0.001908,
+      "loss": 4.911,
+      "step": 636
+    },
+    {
+      "epoch": 0.00637,
+      "grad_norm": 0.8681767489690893,
+      "learning_rate": 0.0019110000000000002,
+      "loss": 4.909,
+      "step": 637
+    },
+    {
+      "epoch": 0.00638,
+      "grad_norm": 0.9792692499883113,
+      "learning_rate": 0.0019140000000000001,
+      "loss": 4.9073,
+      "step": 638
+    },
+    {
+      "epoch": 0.00639,
+      "grad_norm": 0.8702384519183142,
+      "learning_rate": 0.001917,
+      "loss": 4.8724,
+      "step": 639
+    },
+    {
+      "epoch": 0.0064,
+      "grad_norm": 0.7656979903996737,
+      "learning_rate": 0.00192,
+      "loss": 4.8809,
+      "step": 640
+    },
+    {
+      "epoch": 0.00641,
+      "grad_norm": 0.7895438458212233,
+      "learning_rate": 0.001923,
+      "loss": 4.8762,
+      "step": 641
+    },
+    {
+      "epoch": 0.00642,
+      "grad_norm": 0.8281699495124627,
+      "learning_rate": 0.001926,
+      "loss": 4.8749,
+      "step": 642
+    },
+    {
+      "epoch": 0.00643,
+      "grad_norm": 0.7845360432833325,
+      "learning_rate": 0.0019290000000000002,
+      "loss": 4.8856,
+      "step": 643
+    },
+    {
+      "epoch": 0.00644,
+      "grad_norm": 0.6857666802048429,
+      "learning_rate": 0.0019320000000000001,
+      "loss": 4.8583,
+      "step": 644
+    },
+    {
+      "epoch": 0.00645,
+      "grad_norm": 0.5056941566313082,
+      "learning_rate": 0.001935,
+      "loss": 4.8522,
+      "step": 645
+    },
+    {
+      "epoch": 0.00646,
+      "grad_norm": 0.47964630717910517,
+      "learning_rate": 0.001938,
+      "loss": 4.8628,
+      "step": 646
+    },
+    {
+      "epoch": 0.00647,
+      "grad_norm": 0.519046406898008,
+      "learning_rate": 0.001941,
+      "loss": 4.8552,
+      "step": 647
+    },
+    {
+      "epoch": 0.00648,
+      "grad_norm": 0.4935185080219269,
+      "learning_rate": 0.0019440000000000002,
+      "loss": 4.8488,
+      "step": 648
+    },
+    {
+      "epoch": 0.00649,
+      "grad_norm": 0.5205599594018984,
+      "learning_rate": 0.0019470000000000002,
+      "loss": 4.8333,
+      "step": 649
+    },
+    {
+      "epoch": 0.0065,
+      "grad_norm": 0.45295352715192905,
+      "learning_rate": 0.0019500000000000001,
+      "loss": 4.8617,
+      "step": 650
+    },
+    {
+      "epoch": 0.00651,
+      "grad_norm": 0.4378817965282719,
+      "learning_rate": 0.001953,
+      "loss": 4.84,
+      "step": 651
+    },
+    {
+      "epoch": 0.00652,
+      "grad_norm": 0.48395025642257355,
+      "learning_rate": 0.0019560000000000003,
+      "loss": 4.8266,
+      "step": 652
+    },
+    {
+      "epoch": 0.00653,
+      "grad_norm": 0.5194845024092328,
+      "learning_rate": 0.0019590000000000002,
+      "loss": 4.8294,
+      "step": 653
+    },
+    {
+      "epoch": 0.00654,
+      "grad_norm": 0.5707068593609262,
+      "learning_rate": 0.001962,
+      "loss": 4.8026,
+      "step": 654
+    },
+    {
+      "epoch": 0.00655,
+      "grad_norm": 0.6725172209521785,
+      "learning_rate": 0.001965,
+      "loss": 4.834,
+      "step": 655
+    },
+    {
+      "epoch": 0.00656,
+      "grad_norm": 0.7729568432311581,
+      "learning_rate": 0.001968,
+      "loss": 4.8398,
+      "step": 656
+    },
+    {
+      "epoch": 0.00657,
+      "grad_norm": 0.9182317536069756,
+      "learning_rate": 0.001971,
+      "loss": 4.8175,
+      "step": 657
+    },
+    {
+      "epoch": 0.00658,
+      "grad_norm": 0.9241834469814327,
+      "learning_rate": 0.001974,
+      "loss": 4.8482,
+      "step": 658
+    },
+    {
+      "epoch": 0.00659,
+      "grad_norm": 0.7554289117841486,
+      "learning_rate": 0.001977,
+      "loss": 4.8199,
+      "step": 659
+    },
+    {
+      "epoch": 0.0066,
+      "grad_norm": 0.786222644002383,
+      "learning_rate": 0.00198,
+      "loss": 4.8336,
+      "step": 660
+    },
+    {
+      "epoch": 0.00661,
+      "grad_norm": 0.7486956414167568,
+      "learning_rate": 0.001983,
+      "loss": 4.8035,
+      "step": 661
+    },
+    {
+      "epoch": 0.00662,
+      "grad_norm": 0.8224936329580866,
+      "learning_rate": 0.0019860000000000004,
+      "loss": 4.8038,
+      "step": 662
+    },
+    {
+      "epoch": 0.00663,
+      "grad_norm": 1.0333891900276588,
+      "learning_rate": 0.0019890000000000003,
+      "loss": 4.8216,
+      "step": 663
+    },
+    {
+      "epoch": 0.00664,
+      "grad_norm": 1.0454184895583787,
+      "learning_rate": 0.0019920000000000003,
+      "loss": 4.826,
+      "step": 664
+    },
+    {
+      "epoch": 0.00665,
+      "grad_norm": 0.9684168839706867,
+      "learning_rate": 0.0019950000000000002,
+      "loss": 4.7952,
+      "step": 665
+    },
+    {
+      "epoch": 0.00666,
+      "grad_norm": 1.0353944208678731,
+      "learning_rate": 0.001998,
+      "loss": 4.8213,
+      "step": 666
+    },
+    {
+      "epoch": 0.00667,
+      "grad_norm": 0.9360230286599108,
+      "learning_rate": 0.002001,
+      "loss": 4.8166,
+      "step": 667
+    },
+    {
+      "epoch": 0.00668,
+      "grad_norm": 0.7256915823313501,
+      "learning_rate": 0.002004,
+      "loss": 4.83,
+      "step": 668
+    },
+    {
+      "epoch": 0.00669,
+      "grad_norm": 0.8394069732991961,
+      "learning_rate": 0.002007,
+      "loss": 4.8494,
+      "step": 669
+    },
+    {
+      "epoch": 0.0067,
+      "grad_norm": 1.077577977883947,
+      "learning_rate": 0.00201,
+      "loss": 4.8532,
+      "step": 670
+    },
+    {
+      "epoch": 0.00671,
+      "grad_norm": 0.944781077919628,
+      "learning_rate": 0.002013,
+      "loss": 4.8236,
+      "step": 671
+    },
+    {
+      "epoch": 0.00672,
+      "grad_norm": 0.7024086289087288,
+      "learning_rate": 0.002016,
+      "loss": 4.7966,
+      "step": 672
+    },
+    {
+      "epoch": 0.00673,
+      "grad_norm": 0.6931174948166909,
+      "learning_rate": 0.002019,
+      "loss": 4.809,
+      "step": 673
+    },
+    {
+      "epoch": 0.00674,
+      "grad_norm": 0.6108100930625565,
+      "learning_rate": 0.0020220000000000004,
+      "loss": 4.8005,
+      "step": 674
+    },
+    {
+      "epoch": 0.00675,
+      "grad_norm": 0.5498762136107717,
+      "learning_rate": 0.0020250000000000003,
+      "loss": 4.8171,
+      "step": 675
+    },
+    {
+      "epoch": 0.00676,
+      "grad_norm": 0.6244881777878202,
+      "learning_rate": 0.0020280000000000003,
+      "loss": 4.7951,
+      "step": 676
+    },
+    {
+      "epoch": 0.00677,
+      "grad_norm": 0.5815667573949113,
+      "learning_rate": 0.0020310000000000003,
+      "loss": 4.7913,
+      "step": 677
+    },
+    {
+      "epoch": 0.00678,
+      "grad_norm": 0.521491074875552,
+      "learning_rate": 0.0020340000000000002,
+      "loss": 4.7959,
+      "step": 678
+    },
+    {
+      "epoch": 0.00679,
+      "grad_norm": 0.5801192243287052,
+      "learning_rate": 0.002037,
+      "loss": 4.7864,
+      "step": 679
+    },
+    {
+      "epoch": 0.0068,
+      "grad_norm": 0.6398788453868361,
+      "learning_rate": 0.00204,
+      "loss": 4.7875,
+      "step": 680
+    },
+    {
+      "epoch": 0.00681,
+      "grad_norm": 0.6456567912241706,
+      "learning_rate": 0.002043,
+      "loss": 4.7807,
+      "step": 681
+    },
+    {
+      "epoch": 0.00682,
+      "grad_norm": 0.6889535888719373,
+      "learning_rate": 0.002046,
+      "loss": 4.7776,
+      "step": 682
+    },
+    {
+      "epoch": 0.00683,
+      "grad_norm": 0.68132819276425,
+      "learning_rate": 0.002049,
+      "loss": 4.7612,
+      "step": 683
+    },
+    {
+      "epoch": 0.00684,
+      "grad_norm": 0.6724100911918068,
+      "learning_rate": 0.002052,
+      "loss": 4.7814,
+      "step": 684
+    },
+    {
+      "epoch": 0.00685,
+      "grad_norm": 0.815928121483196,
+      "learning_rate": 0.0020550000000000004,
+      "loss": 4.7913,
+      "step": 685
+    },
+    {
+      "epoch": 0.00686,
+      "grad_norm": 0.7537150884203632,
+      "learning_rate": 0.0020580000000000004,
+      "loss": 4.7584,
+      "step": 686
+    },
+    {
+      "epoch": 0.00687,
+      "grad_norm": 0.7326770278926477,
+      "learning_rate": 0.0020610000000000003,
+      "loss": 4.7655,
+      "step": 687
+    },
+    {
+      "epoch": 0.00688,
+      "grad_norm": 0.7145131916015852,
+      "learning_rate": 0.002064,
+      "loss": 4.7631,
+      "step": 688
+    },
+    {
+      "epoch": 0.00689,
+      "grad_norm": 0.694828893634318,
+      "learning_rate": 0.002067,
+      "loss": 4.7685,
+      "step": 689
+    },
+    {
+      "epoch": 0.0069,
+      "grad_norm": 0.7402628986076095,
+      "learning_rate": 0.00207,
+      "loss": 4.7629,
+      "step": 690
+    },
+    {
+      "epoch": 0.00691,
+      "grad_norm": 0.7809038473743037,
+      "learning_rate": 0.0020729999999999998,
+      "loss": 4.734,
+      "step": 691
+    },
+    {
+      "epoch": 0.00692,
+      "grad_norm": 0.9321737974219415,
+      "learning_rate": 0.0020759999999999997,
+      "loss": 4.7623,
+      "step": 692
+    },
+    {
+      "epoch": 0.00693,
+      "grad_norm": 1.0940875671633443,
+      "learning_rate": 0.0020789999999999997,
+      "loss": 4.788,
+      "step": 693
+    },
+    {
+      "epoch": 0.00694,
+      "grad_norm": 0.9030072391970948,
+      "learning_rate": 0.002082,
+      "loss": 4.7582,
+      "step": 694
+    },
+    {
+      "epoch": 0.00695,
+      "grad_norm": 0.8857004379777869,
+      "learning_rate": 0.002085,
+      "loss": 4.7649,
+      "step": 695
+    },
+    {
+      "epoch": 0.00696,
+      "grad_norm": 0.8991843577664433,
+      "learning_rate": 0.002088,
+      "loss": 4.7712,
+      "step": 696
+    },
+    {
+      "epoch": 0.00697,
+      "grad_norm": 1.0094726318481275,
+      "learning_rate": 0.002091,
+      "loss": 4.7869,
+      "step": 697
+    },
+    {
+      "epoch": 0.00698,
+      "grad_norm": 1.09376996979269,
+      "learning_rate": 0.002094,
+      "loss": 4.7773,
+      "step": 698
+    },
+    {
+      "epoch": 0.00699,
+      "grad_norm": 0.8766228601427865,
+      "learning_rate": 0.002097,
+      "loss": 4.7548,
+      "step": 699
+    },
+    {
+      "epoch": 0.007,
+      "grad_norm": 0.8580408760742435,
+      "learning_rate": 0.0021,
+      "loss": 4.7811,
+      "step": 700
+    },
+    {
+      "epoch": 0.00701,
+      "grad_norm": 0.7398805390269545,
+      "learning_rate": 0.002103,
+      "loss": 4.7496,
+      "step": 701
+    },
+    {
+      "epoch": 0.00702,
+      "grad_norm": 0.8329034135969645,
+      "learning_rate": 0.002106,
+      "loss": 4.762,
+      "step": 702
+    },
+    {
+      "epoch": 0.00703,
+      "grad_norm": 0.8505784581176613,
+      "learning_rate": 0.0021089999999999998,
+      "loss": 4.7427,
+      "step": 703
+    },
+    {
+      "epoch": 0.00704,
+      "grad_norm": 0.8003313240374758,
+      "learning_rate": 0.0021119999999999997,
+      "loss": 4.7516,
+      "step": 704
+    },
+    {
+      "epoch": 0.00705,
+      "grad_norm": 0.7109629500256387,
+      "learning_rate": 0.002115,
+      "loss": 4.7635,
+      "step": 705
+    },
+    {
+      "epoch": 0.00706,
+      "grad_norm": 0.6289559048526784,
+      "learning_rate": 0.002118,
+      "loss": 4.7473,
+      "step": 706
+    },
+    {
+      "epoch": 0.00707,
+      "grad_norm": 0.5482382237711074,
+      "learning_rate": 0.002121,
+      "loss": 4.7193,
+      "step": 707
+    },
+    {
+      "epoch": 0.00708,
+      "grad_norm": 0.5598390325768088,
+      "learning_rate": 0.002124,
+      "loss": 4.7015,
+      "step": 708
+    },
+    {
+      "epoch": 0.00709,
+      "grad_norm": 0.5529551760214286,
+      "learning_rate": 0.002127,
+      "loss": 4.7331,
+      "step": 709
+    },
+    {
+      "epoch": 0.0071,
+      "grad_norm": 0.6671084870548735,
+      "learning_rate": 0.00213,
+      "loss": 4.7271,
+      "step": 710
+    },
+    {
+      "epoch": 0.00711,
+      "grad_norm": 1.1336515667563218,
+      "learning_rate": 0.002133,
+      "loss": 4.7225,
+      "step": 711
+    },
+    {
+      "epoch": 0.00712,
+      "grad_norm": 0.8878745395415352,
+      "learning_rate": 0.002136,
+      "loss": 4.761,
+      "step": 712
+    },
+    {
+      "epoch": 0.00713,
+      "grad_norm": 0.6767692274074788,
+      "learning_rate": 0.002139,
+      "loss": 4.7547,
+      "step": 713
+    },
+    {
+      "epoch": 0.00714,
+      "grad_norm": 0.7698487468076475,
+      "learning_rate": 0.002142,
+      "loss": 4.7544,
+      "step": 714
+    },
+    {
+      "epoch": 0.00715,
+      "grad_norm": 0.6189384208707256,
+      "learning_rate": 0.0021449999999999998,
+      "loss": 4.7077,
+      "step": 715
+    },
+    {
+      "epoch": 0.00716,
+      "grad_norm": 0.5968344999809513,
+      "learning_rate": 0.002148,
+      "loss": 4.7064,
+      "step": 716
+    },
+    {
+      "epoch": 0.00717,
+      "grad_norm": 0.5117629580972962,
+      "learning_rate": 0.002151,
+      "loss": 4.7111,
+      "step": 717
+    },
+    {
+      "epoch": 0.00718,
+      "grad_norm": 0.5586203984201494,
+      "learning_rate": 0.002154,
+      "loss": 4.7079,
+      "step": 718
+    },
+    {
+      "epoch": 0.00719,
+      "grad_norm": 0.5075479633130432,
+      "learning_rate": 0.002157,
+      "loss": 4.7121,
+      "step": 719
+    },
+    {
+      "epoch": 0.0072,
+      "grad_norm": 0.4840284295584723,
+      "learning_rate": 0.00216,
+      "loss": 4.7045,
+      "step": 720
+    },
+    {
+      "epoch": 0.00721,
+      "grad_norm": 0.4971167150526394,
+      "learning_rate": 0.002163,
+      "loss": 4.7056,
+      "step": 721
+    },
+    {
+      "epoch": 0.00722,
+      "grad_norm": 0.5347204422021664,
+      "learning_rate": 0.002166,
+      "loss": 4.6807,
+      "step": 722
+    },
+    {
+      "epoch": 0.00723,
+      "grad_norm": 0.5424206330484346,
+      "learning_rate": 0.002169,
+      "loss": 4.6942,
+      "step": 723
+    },
+    {
+      "epoch": 0.00724,
+      "grad_norm": 0.5219918511718452,
+      "learning_rate": 0.002172,
+      "loss": 4.6912,
+      "step": 724
+    },
+    {
+      "epoch": 0.00725,
+      "grad_norm": 0.4975444213381179,
+      "learning_rate": 0.002175,
+      "loss": 4.6834,
+      "step": 725
+    },
+    {
+      "epoch": 0.00726,
+      "grad_norm": 0.5840922524665152,
+      "learning_rate": 0.002178,
+      "loss": 4.7008,
+      "step": 726
+    },
+    {
+      "epoch": 0.00727,
+      "grad_norm": 0.7633698368434719,
+      "learning_rate": 0.0021809999999999998,
+      "loss": 4.6777,
+      "step": 727
+    },
+    {
+      "epoch": 0.00728,
+      "grad_norm": 0.9052599955462425,
+      "learning_rate": 0.002184,
+      "loss": 4.7084,
+      "step": 728
+    },
+    {
+      "epoch": 0.00729,
+      "grad_norm": 0.9052862109699649,
+      "learning_rate": 0.002187,
+      "loss": 4.7065,
+      "step": 729
+    },
+    {
+      "epoch": 0.0073,
+      "grad_norm": 0.9788098741089467,
+      "learning_rate": 0.00219,
+      "loss": 4.7163,
+      "step": 730
+    },
+    {
+      "epoch": 0.00731,
+      "grad_norm": 0.9218858714566311,
+      "learning_rate": 0.002193,
+      "loss": 4.7193,
+      "step": 731
+    },
+    {
+      "epoch": 0.00732,
+      "grad_norm": 1.1412145043722297,
+      "learning_rate": 0.002196,
+      "loss": 4.7513,
+      "step": 732
+    },
+    {
+      "epoch": 0.00733,
+      "grad_norm": 1.0851884555194036,
+      "learning_rate": 0.002199,
+      "loss": 4.7135,
+      "step": 733
+    },
+    {
+      "epoch": 0.00734,
+      "grad_norm": 0.9295060612046938,
+      "learning_rate": 0.002202,
+      "loss": 4.745,
+      "step": 734
+    },
+    {
+      "epoch": 0.00735,
+      "grad_norm": 1.1744848346988581,
+      "learning_rate": 0.002205,
+      "loss": 4.766,
+      "step": 735
+    },
+    {
+      "epoch": 0.00736,
+      "grad_norm": 0.8278931321658332,
+      "learning_rate": 0.002208,
+      "loss": 4.7413,
+      "step": 736
+    },
+    {
+      "epoch": 0.00737,
+      "grad_norm": 0.8170748474782394,
+      "learning_rate": 0.002211,
+      "loss": 4.701,
+      "step": 737
+    },
+    {
+      "epoch": 0.00738,
+      "grad_norm": 0.772640277783885,
+      "learning_rate": 0.002214,
+      "loss": 4.7083,
+      "step": 738
+    },
+    {
+      "epoch": 0.00739,
+      "grad_norm": 0.6328057525928463,
+      "learning_rate": 0.0022170000000000002,
+      "loss": 4.6886,
+      "step": 739
+    },
+    {
+      "epoch": 0.0074,
+      "grad_norm": 0.6456777453177256,
+      "learning_rate": 0.00222,
+      "loss": 4.7001,
+      "step": 740
+    },
+    {
+      "epoch": 0.00741,
+      "grad_norm": 0.5912580097221507,
+      "learning_rate": 0.002223,
+      "loss": 4.6652,
+      "step": 741
+    },
+    {
+      "epoch": 0.00742,
+      "grad_norm": 0.4622238730689768,
+      "learning_rate": 0.002226,
+      "loss": 4.6972,
+      "step": 742
+    },
+    {
+      "epoch": 0.00743,
+      "grad_norm": 0.4904966133460914,
+      "learning_rate": 0.002229,
+      "loss": 4.6585,
+      "step": 743
+    },
+    {
+      "epoch": 0.00744,
+      "grad_norm": 0.5611948954376664,
+      "learning_rate": 0.002232,
+      "loss": 4.6643,
+      "step": 744
+    },
+    {
+      "epoch": 0.00745,
+      "grad_norm": 0.6222946629538655,
+      "learning_rate": 0.002235,
+      "loss": 4.6777,
+      "step": 745
+    },
+    {
+      "epoch": 0.00746,
+      "grad_norm": 0.6678221674623648,
+      "learning_rate": 0.002238,
+      "loss": 4.6735,
+      "step": 746
+    },
+    {
+      "epoch": 0.00747,
+      "grad_norm": 0.7945455347777014,
+      "learning_rate": 0.002241,
+      "loss": 4.6637,
+      "step": 747
+    },
+    {
+      "epoch": 0.00748,
+      "grad_norm": 0.9027593299924651,
+      "learning_rate": 0.002244,
+      "loss": 4.683,
+      "step": 748
+    },
+    {
+      "epoch": 0.00749,
+      "grad_norm": 0.8224517051186598,
+      "learning_rate": 0.002247,
+      "loss": 4.6612,
+      "step": 749
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 0.6737194715747743,
+      "learning_rate": 0.0022500000000000003,
+      "loss": 4.6955,
+      "step": 750
+    },
+    {
+      "epoch": 0.00751,
+      "grad_norm": 0.7232822830562855,
+      "learning_rate": 0.0022530000000000002,
+      "loss": 4.6893,
+      "step": 751
+    },
+    {
+      "epoch": 0.00752,
+      "grad_norm": 0.6449225627821611,
+      "learning_rate": 0.002256,
+      "loss": 4.637,
+      "step": 752
+    },
+    {
+      "epoch": 0.00753,
+      "grad_norm": 0.6558342053271724,
+      "learning_rate": 0.002259,
+      "loss": 4.6782,
+      "step": 753
+    },
+    {
+      "epoch": 0.00754,
+      "grad_norm": 0.7056102529212358,
+      "learning_rate": 0.002262,
+      "loss": 4.6906,
+      "step": 754
+    },
+    {
+      "epoch": 0.00755,
+      "grad_norm": 0.9043185481447295,
+      "learning_rate": 0.002265,
+      "loss": 4.6606,
+      "step": 755
+    },
+    {
+      "epoch": 0.00756,
+      "grad_norm": 0.8770375990036015,
+      "learning_rate": 0.002268,
+      "loss": 4.6551,
+      "step": 756
+    },
+    {
+      "epoch": 0.00757,
+      "grad_norm": 0.784655485815756,
+      "learning_rate": 0.002271,
+      "loss": 4.6655,
+      "step": 757
+    },
+    {
+      "epoch": 0.00758,
+      "grad_norm": 0.6912079950422929,
+      "learning_rate": 0.002274,
+      "loss": 4.655,
+      "step": 758
+    },
+    {
+      "epoch": 0.00759,
+      "grad_norm": 0.6281371231708264,
+      "learning_rate": 0.002277,
+      "loss": 4.6629,
+      "step": 759
+    },
+    {
+      "epoch": 0.0076,
+      "grad_norm": 0.5992557531443571,
+      "learning_rate": 0.00228,
+      "loss": 4.6463,
+      "step": 760
+    },
+    {
+      "epoch": 0.00761,
+      "grad_norm": 0.6034804931159479,
+      "learning_rate": 0.002283,
+      "loss": 4.6146,
+      "step": 761
+    },
+    {
+      "epoch": 0.00762,
+      "grad_norm": 0.6060668832918386,
+      "learning_rate": 0.0022860000000000003,
+      "loss": 4.5953,
+      "step": 762
+    },
+    {
+      "epoch": 0.00763,
+      "grad_norm": 0.5816892837581542,
+      "learning_rate": 0.0022890000000000002,
+      "loss": 4.6459,
+      "step": 763
+    },
+    {
+      "epoch": 0.00764,
+      "grad_norm": 0.599786845424844,
+      "learning_rate": 0.002292,
+      "loss": 4.6554,
+      "step": 764
+    },
+    {
+      "epoch": 0.00765,
+      "grad_norm": 0.6640243443171511,
+      "learning_rate": 0.002295,
+      "loss": 4.6451,
+      "step": 765
+    },
+    {
+      "epoch": 0.00766,
+      "grad_norm": 0.6663176934335011,
+      "learning_rate": 0.002298,
+      "loss": 4.6438,
+      "step": 766
+    },
+    {
+      "epoch": 0.00767,
+      "grad_norm": 0.6384162747773746,
+      "learning_rate": 0.002301,
+      "loss": 4.6392,
+      "step": 767
+    },
+    {
+      "epoch": 0.00768,
+      "grad_norm": 0.8537658256313299,
+      "learning_rate": 0.002304,
+      "loss": 4.6457,
+      "step": 768
+    },
+    {
+      "epoch": 0.00769,
+      "grad_norm": 1.0098120726516953,
+      "learning_rate": 0.002307,
+      "loss": 4.6694,
+      "step": 769
+    },
+    {
+      "epoch": 0.0077,
+      "grad_norm": 0.7423003820485509,
+      "learning_rate": 0.00231,
+      "loss": 4.6223,
+      "step": 770
+    },
+    {
+      "epoch": 0.00771,
+      "grad_norm": 0.6691351211965346,
+      "learning_rate": 0.002313,
+      "loss": 4.6569,
+      "step": 771
+    },
+    {
+      "epoch": 0.00772,
+      "grad_norm": 0.8185609860643415,
+      "learning_rate": 0.002316,
+      "loss": 4.6531,
+      "step": 772
+    },
+    {
+      "epoch": 0.00773,
+      "grad_norm": 1.0039613061235502,
+      "learning_rate": 0.0023190000000000003,
+      "loss": 4.6664,
+      "step": 773
+    },
+    {
+      "epoch": 0.00774,
+      "grad_norm": 0.8500607234716588,
+      "learning_rate": 0.0023220000000000003,
+      "loss": 4.6845,
+      "step": 774
+    },
+    {
+      "epoch": 0.00775,
+      "grad_norm": 0.7057192127032131,
+      "learning_rate": 0.0023250000000000002,
+      "loss": 4.6688,
+      "step": 775
+    },
+    {
+      "epoch": 0.00776,
+      "grad_norm": 0.8527617729361273,
+      "learning_rate": 0.002328,
+      "loss": 4.6732,
+      "step": 776
+    },
+    {
+      "epoch": 0.00777,
+      "grad_norm": 0.7987783133918631,
+      "learning_rate": 0.002331,
+      "loss": 4.6469,
+      "step": 777
+    },
+    {
+      "epoch": 0.00778,
+      "grad_norm": 0.8221364860163118,
+      "learning_rate": 0.002334,
+      "loss": 4.6645,
+      "step": 778
+    },
+    {
+      "epoch": 0.00779,
+      "grad_norm": 0.8901185821524535,
+      "learning_rate": 0.002337,
+      "loss": 4.6243,
+      "step": 779
+    },
+    {
+      "epoch": 0.0078,
+      "grad_norm": 0.889956887452623,
+      "learning_rate": 0.00234,
+      "loss": 4.646,
+      "step": 780
+    },
+    {
+      "epoch": 0.00781,
+      "grad_norm": 0.763195026878423,
+      "learning_rate": 0.002343,
+      "loss": 4.6465,
+      "step": 781
+    },
+    {
+      "epoch": 0.00782,
+      "grad_norm": 0.6941360203492353,
+      "learning_rate": 0.002346,
+      "loss": 4.6378,
+      "step": 782
+    },
+    {
+      "epoch": 0.00783,
+      "grad_norm": 0.6504570243922513,
+      "learning_rate": 0.002349,
+      "loss": 4.6195,
+      "step": 783
+    },
+    {
+      "epoch": 0.00784,
+      "grad_norm": 0.5612286515783087,
+      "learning_rate": 0.002352,
+      "loss": 4.6293,
+      "step": 784
+    },
+    {
+      "epoch": 0.00785,
+      "grad_norm": 0.6491841258536649,
+      "learning_rate": 0.0023550000000000003,
+      "loss": 4.6258,
+      "step": 785
+    },
+    {
+      "epoch": 0.00786,
+      "grad_norm": 0.7618604140557245,
+      "learning_rate": 0.0023580000000000003,
+      "loss": 4.6353,
+      "step": 786
+    },
+    {
+      "epoch": 0.00787,
+      "grad_norm": 0.9289251225559676,
+      "learning_rate": 0.0023610000000000003,
+      "loss": 4.6478,
+      "step": 787
+    },
+    {
+      "epoch": 0.00788,
+      "grad_norm": 1.0160410825626025,
+      "learning_rate": 0.002364,
+      "loss": 4.6569,
+      "step": 788
+    },
+    {
+      "epoch": 0.00789,
+      "grad_norm": 0.9904632547103471,
+      "learning_rate": 0.002367,
+      "loss": 4.6841,
+      "step": 789
+    },
+    {
+      "epoch": 0.0079,
+      "grad_norm": 0.978549262142993,
+      "learning_rate": 0.00237,
+      "loss": 4.6525,
+      "step": 790
+    },
+    {
+      "epoch": 0.00791,
+      "grad_norm": 0.7626000247604355,
+      "learning_rate": 0.002373,
+      "loss": 4.6008,
+      "step": 791
+    },
+    {
+      "epoch": 0.00792,
+      "grad_norm": 0.7032433100729657,
+      "learning_rate": 0.002376,
+      "loss": 4.6361,
+      "step": 792
+    },
+    {
+      "epoch": 0.00793,
+      "grad_norm": 0.6164393933780811,
+      "learning_rate": 0.002379,
+      "loss": 4.6263,
+      "step": 793
+    },
+    {
+      "epoch": 0.00794,
+      "grad_norm": 0.6849838631141278,
+      "learning_rate": 0.002382,
+      "loss": 4.6402,
+      "step": 794
+    },
+    {
+      "epoch": 0.00795,
+      "grad_norm": 0.7541453624973187,
+      "learning_rate": 0.002385,
+      "loss": 4.6181,
+      "step": 795
+    },
+    {
+      "epoch": 0.00796,
+      "grad_norm": 0.7900314117496404,
+      "learning_rate": 0.0023880000000000004,
+      "loss": 4.6154,
+      "step": 796
+    },
+    {
+      "epoch": 0.00797,
+      "grad_norm": 0.8032302751261208,
+      "learning_rate": 0.0023910000000000003,
+      "loss": 4.6598,
+      "step": 797
+    },
+    {
+      "epoch": 0.00798,
+      "grad_norm": 0.7037506066431396,
+      "learning_rate": 0.0023940000000000003,
+      "loss": 4.6206,
+      "step": 798
+    },
+    {
+      "epoch": 0.00799,
+      "grad_norm": 0.6873143294249761,
+      "learning_rate": 0.0023970000000000003,
+      "loss": 4.6565,
+      "step": 799
+    },
+    {
+      "epoch": 0.008,
+      "grad_norm": 0.5611968184575689,
+      "learning_rate": 0.0024000000000000002,
+      "loss": 4.6002,
+      "step": 800
+    },
+    {
+      "epoch": 0.00801,
+      "grad_norm": 0.6431524925156966,
+      "learning_rate": 0.002403,
+      "loss": 4.6326,
+      "step": 801
+    },
+    {
+      "epoch": 0.00802,
+      "grad_norm": 0.664361851854149,
+      "learning_rate": 0.002406,
+      "loss": 4.624,
+      "step": 802
+    },
+    {
+      "epoch": 0.00803,
+      "grad_norm": 0.7385532765716779,
+      "learning_rate": 0.002409,
+      "loss": 4.6034,
+      "step": 803
+    },
+    {
+      "epoch": 0.00804,
+      "grad_norm": 0.6763586866376322,
+      "learning_rate": 0.002412,
+      "loss": 4.609,
+      "step": 804
+    },
+    {
+      "epoch": 0.00805,
+      "grad_norm": 0.4984516329675046,
+      "learning_rate": 0.002415,
+      "loss": 4.5958,
+      "step": 805
+    },
+    {
+      "epoch": 0.00806,
+      "grad_norm": 0.6422151194456196,
+      "learning_rate": 0.002418,
+      "loss": 4.6202,
+      "step": 806
+    },
+    {
+      "epoch": 0.00807,
+      "grad_norm": 0.6520347813599764,
+      "learning_rate": 0.0024210000000000004,
+      "loss": 4.6111,
+      "step": 807
+    },
+    {
+      "epoch": 0.00808,
+      "grad_norm": 0.6466117231395049,
+      "learning_rate": 0.0024240000000000004,
+      "loss": 4.5938,
+      "step": 808
+    },
+    {
+      "epoch": 0.00809,
+      "grad_norm": 0.7173763198404414,
+      "learning_rate": 0.0024270000000000003,
+      "loss": 4.5875,
+      "step": 809
+    },
+    {
+      "epoch": 0.0081,
+      "grad_norm": 0.7630708316428362,
+      "learning_rate": 0.0024300000000000003,
+      "loss": 4.6257,
+      "step": 810
+    },
+    {
+      "epoch": 0.00811,
+      "grad_norm": 0.6597258634486322,
+      "learning_rate": 0.0024330000000000003,
+      "loss": 4.5916,
+      "step": 811
+    },
+    {
+      "epoch": 0.00812,
+      "grad_norm": 0.5596135689444884,
+      "learning_rate": 0.0024360000000000002,
+      "loss": 4.5639,
+      "step": 812
+    },
+    {
+      "epoch": 0.00813,
+      "grad_norm": 0.5455291997717118,
+      "learning_rate": 0.0024389999999999998,
+      "loss": 4.6039,
+      "step": 813
+    },
+    {
+      "epoch": 0.00814,
+      "grad_norm": 0.643394229982758,
+      "learning_rate": 0.0024419999999999997,
+      "loss": 4.6125,
+      "step": 814
+    },
+    {
+      "epoch": 0.00815,
+      "grad_norm": 0.7154153020141445,
+      "learning_rate": 0.0024449999999999997,
+      "loss": 4.6204,
+      "step": 815
+    },
+    {
+      "epoch": 0.00816,
+      "grad_norm": 0.7061775094598466,
+      "learning_rate": 0.002448,
+      "loss": 4.6005,
+      "step": 816
+    },
+    {
+      "epoch": 0.00817,
+      "grad_norm": 0.7497949470997279,
+      "learning_rate": 0.002451,
+      "loss": 4.6054,
+      "step": 817
+    },
+    {
+      "epoch": 0.00818,
+      "grad_norm": 0.8303821582765404,
+      "learning_rate": 0.002454,
+      "loss": 4.6047,
+      "step": 818
+    },
+    {
+      "epoch": 0.00819,
+      "grad_norm": 0.7870772660210968,
+      "learning_rate": 0.002457,
+      "loss": 4.6129,
+      "step": 819
+    },
+    {
+      "epoch": 0.0082,
+      "grad_norm": 0.8045799279983024,
+      "learning_rate": 0.00246,
+      "loss": 4.5959,
+      "step": 820
+    },
+    {
+      "epoch": 0.00821,
+      "grad_norm": 0.6847814476141592,
+      "learning_rate": 0.002463,
+      "loss": 4.5753,
+      "step": 821
+    },
+    {
+      "epoch": 0.00822,
+      "grad_norm": 0.6767342952639601,
+      "learning_rate": 0.002466,
+      "loss": 4.5743,
+      "step": 822
+    },
+    {
+      "epoch": 0.00823,
+      "grad_norm": 0.7814972414925468,
+      "learning_rate": 0.002469,
+      "loss": 4.5892,
+      "step": 823
+    },
+    {
+      "epoch": 0.00824,
+      "grad_norm": 0.7970591329920511,
+      "learning_rate": 0.002472,
+      "loss": 4.6124,
+      "step": 824
+    },
+    {
+      "epoch": 0.00825,
+      "grad_norm": 0.7237059608391628,
+      "learning_rate": 0.0024749999999999998,
+      "loss": 4.6158,
+      "step": 825
+    },
+    {
+      "epoch": 0.00826,
+      "grad_norm": 0.8825063151832095,
+      "learning_rate": 0.0024779999999999997,
+      "loss": 4.5838,
+      "step": 826
+    },
+    {
+      "epoch": 0.00827,
+      "grad_norm": 0.895197450200386,
+      "learning_rate": 0.002481,
+      "loss": 4.6224,
+      "step": 827
+    },
+    {
+      "epoch": 0.00828,
+      "grad_norm": 0.7760924275661016,
+      "learning_rate": 0.002484,
+      "loss": 4.6087,
+      "step": 828
+    },
+    {
+      "epoch": 0.00829,
+      "grad_norm": 0.8835354563964355,
+      "learning_rate": 0.002487,
+      "loss": 4.6157,
+      "step": 829
+    },
+    {
+      "epoch": 0.0083,
+      "grad_norm": 0.8984013262238549,
+      "learning_rate": 0.00249,
+      "loss": 4.6232,
+      "step": 830
+    },
+    {
+      "epoch": 0.00831,
+      "grad_norm": 0.8653834946214679,
+      "learning_rate": 0.002493,
+      "loss": 4.6038,
+      "step": 831
+    },
+    {
+      "epoch": 0.00832,
+      "grad_norm": 0.8641563564291244,
+      "learning_rate": 0.002496,
+      "loss": 4.6089,
+      "step": 832
+    },
+    {
+      "epoch": 0.00833,
+      "grad_norm": 0.8849940351412556,
+      "learning_rate": 0.002499,
+      "loss": 4.6069,
+      "step": 833
+    },
+    {
+      "epoch": 0.00834,
+      "grad_norm": 0.9333438912141526,
+      "learning_rate": 0.002502,
+      "loss": 4.6165,
+      "step": 834
+    },
+    {
+      "epoch": 0.00835,
+      "grad_norm": 0.8853225829488065,
+      "learning_rate": 0.002505,
+      "loss": 4.6108,
+      "step": 835
+    },
+    {
+      "epoch": 0.00836,
+      "grad_norm": 0.9042165265638229,
+      "learning_rate": 0.002508,
+      "loss": 4.6433,
+      "step": 836
+    },
+    {
+      "epoch": 0.00837,
+      "grad_norm": 0.8034248166459536,
+      "learning_rate": 0.0025109999999999998,
+      "loss": 4.5745,
+      "step": 837
+    },
+    {
+      "epoch": 0.00838,
+      "grad_norm": 0.8086998635281932,
+      "learning_rate": 0.0025139999999999997,
+      "loss": 4.6114,
+      "step": 838
+    },
+    {
+      "epoch": 0.00839,
+      "grad_norm": 0.587375456181757,
+      "learning_rate": 0.002517,
+      "loss": 4.5954,
+      "step": 839
+    },
+    {
+      "epoch": 0.0084,
+      "grad_norm": 0.6561997865962053,
+      "learning_rate": 0.00252,
+      "loss": 4.5818,
+      "step": 840
+    },
+    {
+      "epoch": 0.00841,
+      "grad_norm": 0.6860624256233325,
+      "learning_rate": 0.002523,
+      "loss": 4.5813,
+      "step": 841
+    },
+    {
+      "epoch": 0.00842,
+      "grad_norm": 0.6384378914075728,
+      "learning_rate": 0.002526,
+      "loss": 4.5783,
+      "step": 842
+    },
+    {
+      "epoch": 0.00843,
+      "grad_norm": 0.5397302637540236,
+      "learning_rate": 0.002529,
+      "loss": 4.5728,
+      "step": 843
+    },
+    {
+      "epoch": 0.00844,
+      "grad_norm": 0.5753282450552862,
+      "learning_rate": 0.002532,
+      "loss": 4.5912,
+      "step": 844
+    },
+    {
+      "epoch": 0.00845,
+      "grad_norm": 0.605188801053789,
+      "learning_rate": 0.002535,
+      "loss": 4.5588,
+      "step": 845
+    },
+    {
+      "epoch": 0.00846,
+      "grad_norm": 0.523231421872222,
+      "learning_rate": 0.002538,
+      "loss": 4.5771,
+      "step": 846
+    },
+    {
+      "epoch": 0.00847,
+      "grad_norm": 0.4442194357732046,
+      "learning_rate": 0.002541,
+      "loss": 4.5539,
+      "step": 847
+    },
+    {
+      "epoch": 0.00848,
+      "grad_norm": 0.4340595323225716,
+      "learning_rate": 0.002544,
+      "loss": 4.5562,
+      "step": 848
+    },
+    {
+      "epoch": 0.00849,
+      "grad_norm": 0.40987686572717924,
+      "learning_rate": 0.002547,
+      "loss": 4.5471,
+      "step": 849
+    },
+    {
+      "epoch": 0.0085,
+      "grad_norm": 0.38823479672015115,
+      "learning_rate": 0.00255,
+      "loss": 4.5547,
+      "step": 850
+    },
+    {
+      "epoch": 0.00851,
+      "grad_norm": 0.42997075207568086,
+      "learning_rate": 0.002553,
+      "loss": 4.5733,
+      "step": 851
+    },
+    {
+      "epoch": 0.00852,
+      "grad_norm": 0.5192269912554525,
+      "learning_rate": 0.002556,
+      "loss": 4.5624,
+      "step": 852
+    },
+    {
+      "epoch": 0.00853,
+      "grad_norm": 0.6626364019198889,
+      "learning_rate": 0.002559,
+      "loss": 4.5465,
+      "step": 853
+    },
+    {
+      "epoch": 0.00854,
+      "grad_norm": 0.9475750496728129,
+      "learning_rate": 0.002562,
+      "loss": 4.5544,
+      "step": 854
+    },
+    {
+      "epoch": 0.00855,
+      "grad_norm": 0.9196861875673602,
+      "learning_rate": 0.002565,
+      "loss": 4.5554,
+      "step": 855
+    },
+    {
+      "epoch": 0.00856,
+      "grad_norm": 0.6972970172901616,
+      "learning_rate": 0.002568,
+      "loss": 4.5747,
+      "step": 856
+    },
+    {
+      "epoch": 0.00857,
+      "grad_norm": 0.7425487517589463,
+      "learning_rate": 0.002571,
+      "loss": 4.5717,
+      "step": 857
+    },
+    {
+      "epoch": 0.00858,
+      "grad_norm": 0.678911102843075,
+      "learning_rate": 0.002574,
+      "loss": 4.5576,
+      "step": 858
+    },
+    {
+      "epoch": 0.00859,
+      "grad_norm": 0.6680695283315139,
+      "learning_rate": 0.002577,
+      "loss": 4.5494,
+      "step": 859
+    },
+    {
+      "epoch": 0.0086,
+      "grad_norm": 0.7323389776740598,
+      "learning_rate": 0.00258,
+      "loss": 4.5797,
+      "step": 860
+    },
+    {
+      "epoch": 0.00861,
+      "grad_norm": 0.6564756555111457,
+      "learning_rate": 0.0025830000000000002,
+      "loss": 4.5928,
+      "step": 861
+    },
+    {
+      "epoch": 0.00862,
+      "grad_norm": 0.6458920384469554,
+      "learning_rate": 0.002586,
+      "loss": 4.5402,
+      "step": 862
+    },
+    {
+      "epoch": 0.00863,
+      "grad_norm": 0.56307833723133,
+      "learning_rate": 0.002589,
+      "loss": 4.5659,
+      "step": 863
+    },
+    {
+      "epoch": 0.00864,
+      "grad_norm": 0.6301527121328465,
+      "learning_rate": 0.002592,
+      "loss": 4.5169,
+      "step": 864
+    },
+    {
+      "epoch": 0.00865,
+      "grad_norm": 0.5741592962502369,
+      "learning_rate": 0.002595,
+      "loss": 4.5342,
+      "step": 865
+    },
+    {
+      "epoch": 0.00866,
+      "grad_norm": 0.6185631950272453,
+      "learning_rate": 0.002598,
+      "loss": 4.5624,
+      "step": 866
+    },
+    {
+      "epoch": 0.00867,
+      "grad_norm": 0.7226342011827854,
+      "learning_rate": 0.002601,
+      "loss": 4.5538,
+      "step": 867
+    },
+    {
+      "epoch": 0.00868,
+      "grad_norm": 0.6871063796143311,
+      "learning_rate": 0.002604,
+      "loss": 4.5354,
+      "step": 868
+    },
+    {
+      "epoch": 0.00869,
+      "grad_norm": 0.6146544871717051,
+      "learning_rate": 0.002607,
+      "loss": 4.5411,
+      "step": 869
+    },
+    {
+      "epoch": 0.0087,
+      "grad_norm": 0.5850681754527672,
+      "learning_rate": 0.00261,
+      "loss": 4.5279,
+      "step": 870
+    },
+    {
+      "epoch": 0.00871,
+      "grad_norm": 0.6619050810997609,
+      "learning_rate": 0.002613,
+      "loss": 4.5496,
+      "step": 871
+    },
+    {
+      "epoch": 0.00872,
+      "grad_norm": 0.6309510333250544,
+      "learning_rate": 0.002616,
+      "loss": 4.5312,
+      "step": 872
+    },
+    {
+      "epoch": 0.00873,
+      "grad_norm": 0.620458951387438,
+      "learning_rate": 0.0026190000000000002,
+      "loss": 4.5409,
+      "step": 873
+    },
+    {
+      "epoch": 0.00874,
+      "grad_norm": 0.7675582961358233,
+      "learning_rate": 0.002622,
+      "loss": 4.5679,
+      "step": 874
+    },
+    {
+      "epoch": 0.00875,
+      "grad_norm": 0.8413199287183839,
+      "learning_rate": 0.002625,
+      "loss": 4.5217,
+      "step": 875
+    },
+    {
+      "epoch": 0.00876,
+      "grad_norm": 0.6439305106538762,
+      "learning_rate": 0.002628,
+      "loss": 4.5435,
+      "step": 876
+    },
+    {
+      "epoch": 0.00877,
+      "grad_norm": 0.6155973146282673,
+      "learning_rate": 0.002631,
+      "loss": 4.5607,
+      "step": 877
+    },
+    {
+      "epoch": 0.00878,
+      "grad_norm": 0.6572664983016872,
+      "learning_rate": 0.002634,
+      "loss": 4.4959,
+      "step": 878
+    },
+    {
+      "epoch": 0.00879,
+      "grad_norm": 0.678308084092591,
+      "learning_rate": 0.002637,
+      "loss": 4.526,
+      "step": 879
+    },
+    {
+      "epoch": 0.0088,
+      "grad_norm": 0.682644138366955,
+      "learning_rate": 0.00264,
+      "loss": 4.5397,
+      "step": 880
+    },
+    {
+      "epoch": 0.00881,
+      "grad_norm": 0.6916960221367301,
+      "learning_rate": 0.002643,
+      "loss": 4.5424,
+      "step": 881
+    },
+    {
+      "epoch": 0.00882,
+      "grad_norm": 0.8625989079873627,
+      "learning_rate": 0.002646,
+      "loss": 4.5523,
+      "step": 882
+    },
+    {
+      "epoch": 0.00883,
+      "grad_norm": 0.9115056000231379,
+      "learning_rate": 0.002649,
+      "loss": 4.5538,
+      "step": 883
+    },
+    {
+      "epoch": 0.00884,
+      "grad_norm": 1.1386535317654105,
+      "learning_rate": 0.0026520000000000003,
+      "loss": 4.5528,
+      "step": 884
+    },
+    {
+      "epoch": 0.00885,
+      "grad_norm": 0.9401738564991132,
+      "learning_rate": 0.0026550000000000002,
+      "loss": 4.5478,
+      "step": 885
+    },
+    {
+      "epoch": 0.00886,
+      "grad_norm": 1.0261918984073035,
+      "learning_rate": 0.002658,
+      "loss": 4.5785,
+      "step": 886
+    },
+    {
+      "epoch": 0.00887,
+      "grad_norm": 1.2427377192417122,
+      "learning_rate": 0.002661,
+      "loss": 4.6,
+      "step": 887
+    },
+    {
+      "epoch": 0.00888,
+      "grad_norm": 1.061546131863703,
+      "learning_rate": 0.002664,
+      "loss": 4.5885,
+      "step": 888
+    },
+    {
+      "epoch": 0.00889,
+      "grad_norm": 0.907593068920597,
+      "learning_rate": 0.002667,
+      "loss": 4.535,
+      "step": 889
+    },
+    {
+      "epoch": 0.0089,
+      "grad_norm": 0.8129313603021157,
+      "learning_rate": 0.00267,
+      "loss": 4.5855,
+      "step": 890
+    },
+    {
+      "epoch": 0.00891,
+      "grad_norm": 0.8985699151758613,
+      "learning_rate": 0.002673,
+      "loss": 4.5762,
+      "step": 891
+    },
+    {
+      "epoch": 0.00892,
+      "grad_norm": 0.9328845387585828,
+      "learning_rate": 0.002676,
+      "loss": 4.5962,
+      "step": 892
+    },
+    {
+      "epoch": 0.00893,
+      "grad_norm": 0.9658499469408992,
+      "learning_rate": 0.002679,
+      "loss": 4.5743,
+      "step": 893
+    },
+    {
+      "epoch": 0.00894,
+      "grad_norm": 0.9262060176852961,
+      "learning_rate": 0.002682,
+      "loss": 4.5624,
+      "step": 894
+    },
+    {
+      "epoch": 0.00895,
+      "grad_norm": 1.0340643070320608,
+      "learning_rate": 0.0026850000000000003,
+      "loss": 4.5798,
+      "step": 895
+    },
+    {
+      "epoch": 0.00896,
+      "grad_norm": 0.961568519948311,
+      "learning_rate": 0.0026880000000000003,
+      "loss": 4.5724,
+      "step": 896
+    },
+    {
+      "epoch": 0.00897,
+      "grad_norm": 0.9522979094666822,
+      "learning_rate": 0.0026910000000000002,
+      "loss": 4.5901,
+      "step": 897
+    },
+    {
+      "epoch": 0.00898,
+      "grad_norm": 0.90558868323236,
+      "learning_rate": 0.002694,
+      "loss": 4.5802,
+      "step": 898
+    },
+    {
+      "epoch": 0.00899,
+      "grad_norm": 0.7752291080346148,
+      "learning_rate": 0.002697,
+      "loss": 4.5848,
+      "step": 899
+    },
+    {
+      "epoch": 0.009,
+      "grad_norm": 0.7198757797557334,
+      "learning_rate": 0.0027,
+      "loss": 4.5998,
+      "step": 900
+    },
+    {
+      "epoch": 0.00901,
+      "grad_norm": 0.7328950523365488,
+      "learning_rate": 0.002703,
+      "loss": 4.5552,
+      "step": 901
+    },
+    {
+      "epoch": 0.00902,
+      "grad_norm": 0.5974121750232518,
+      "learning_rate": 0.002706,
+      "loss": 4.558,
+      "step": 902
+    },
+    {
+      "epoch": 0.00903,
+      "grad_norm": 0.5753314236304967,
+      "learning_rate": 0.002709,
+      "loss": 4.5534,
+      "step": 903
+    },
+    {
+      "epoch": 0.00904,
+      "grad_norm": 0.6128643778394687,
+      "learning_rate": 0.002712,
+      "loss": 4.5329,
+      "step": 904
+    },
+    {
+      "epoch": 0.00905,
+      "grad_norm": 0.5691916638541509,
+      "learning_rate": 0.002715,
+      "loss": 4.5446,
+      "step": 905
+    },
+    {
+      "epoch": 0.00906,
+      "grad_norm": 0.5122473095411203,
+      "learning_rate": 0.002718,
+      "loss": 4.5538,
+      "step": 906
+    },
+    {
+      "epoch": 0.00907,
+      "grad_norm": 0.4479956748150508,
+      "learning_rate": 0.0027210000000000003,
+      "loss": 4.5603,
+      "step": 907
+    },
+    {
+      "epoch": 0.00908,
+      "grad_norm": 0.4387217017050332,
+      "learning_rate": 0.0027240000000000003,
+      "loss": 4.5304,
+      "step": 908
+    },
+    {
+      "epoch": 0.00909,
+      "grad_norm": 0.4358600325368994,
+      "learning_rate": 0.0027270000000000003,
+      "loss": 4.5454,
+      "step": 909
+    },
+    {
+      "epoch": 0.0091,
+      "grad_norm": 0.3849373702796595,
+      "learning_rate": 0.0027300000000000002,
+      "loss": 4.5477,
+      "step": 910
+    },
+    {
+      "epoch": 0.00911,
+      "grad_norm": 0.4574591880095763,
+      "learning_rate": 0.002733,
+      "loss": 4.5038,
+      "step": 911
+    },
+    {
+      "epoch": 0.00912,
+      "grad_norm": 0.6121326743360266,
+      "learning_rate": 0.002736,
+      "loss": 4.5131,
+      "step": 912
+    },
+    {
+      "epoch": 0.00913,
+      "grad_norm": 0.6845161679319078,
+      "learning_rate": 0.002739,
+      "loss": 4.513,
+      "step": 913
+    },
+    {
+      "epoch": 0.00914,
+      "grad_norm": 0.7333722528690678,
+      "learning_rate": 0.002742,
+      "loss": 4.5432,
+      "step": 914
+    },
+    {
+      "epoch": 0.00915,
+      "grad_norm": 0.6714442153045657,
+      "learning_rate": 0.002745,
+      "loss": 4.5048,
+      "step": 915
+    },
+    {
+      "epoch": 0.00916,
+      "grad_norm": 0.5861682792098877,
+      "learning_rate": 0.002748,
+      "loss": 4.4768,
+      "step": 916
+    },
+    {
+      "epoch": 0.00917,
+      "grad_norm": 0.7568015257717708,
+      "learning_rate": 0.002751,
+      "loss": 4.4985,
+      "step": 917
+    },
+    {
+      "epoch": 0.00918,
+      "grad_norm": 0.67845085346177,
+      "learning_rate": 0.0027540000000000004,
+      "loss": 4.5185,
+      "step": 918
+    },
+    {
+      "epoch": 0.00919,
+      "grad_norm": 0.6376249584521229,
+      "learning_rate": 0.0027570000000000003,
+      "loss": 4.4976,
+      "step": 919
+    },
+    {
+      "epoch": 0.0092,
+      "grad_norm": 0.5953295490629925,
+      "learning_rate": 0.0027600000000000003,
+      "loss": 4.5104,
+      "step": 920
+    },
+    {
+      "epoch": 0.00921,
+      "grad_norm": 0.5470113372090761,
+      "learning_rate": 0.0027630000000000003,
+      "loss": 4.4892,
+      "step": 921
+    },
+    {
+      "epoch": 0.00922,
+      "grad_norm": 0.5586775976924615,
+      "learning_rate": 0.0027660000000000002,
+      "loss": 4.5007,
+      "step": 922
+    },
+    {
+      "epoch": 0.00923,
+      "grad_norm": 0.5602976125001545,
+      "learning_rate": 0.002769,
+      "loss": 4.4767,
+      "step": 923
+    },
+    {
+      "epoch": 0.00924,
+      "grad_norm": 0.5512257756150963,
+      "learning_rate": 0.002772,
+      "loss": 4.4992,
+      "step": 924
+    },
+    {
+      "epoch": 0.00925,
+      "grad_norm": 0.6104014154605217,
+      "learning_rate": 0.002775,
+      "loss": 4.4804,
+      "step": 925
+    },
+    {
+      "epoch": 0.00926,
+      "grad_norm": 0.627882987139063,
+      "learning_rate": 0.002778,
+      "loss": 4.4994,
+      "step": 926
+    },
+    {
+      "epoch": 0.00927,
+      "grad_norm": 0.5637389395797978,
+      "learning_rate": 0.002781,
+      "loss": 4.4582,
+      "step": 927
+    },
+    {
+      "epoch": 0.00928,
+      "grad_norm": 0.5793012970445044,
+      "learning_rate": 0.002784,
+      "loss": 4.505,
+      "step": 928
+    },
+    {
+      "epoch": 0.00929,
+      "grad_norm": 0.5538600481661593,
+      "learning_rate": 0.0027870000000000004,
+      "loss": 4.4867,
+      "step": 929
+    },
+    {
+      "epoch": 0.0093,
+      "grad_norm": 0.4915384406563099,
+      "learning_rate": 0.0027900000000000004,
+      "loss": 4.4652,
+      "step": 930
+    },
+    {
+      "epoch": 0.00931,
+      "grad_norm": 0.5291025545236148,
+      "learning_rate": 0.0027930000000000003,
+      "loss": 4.4796,
+      "step": 931
+    },
+    {
+      "epoch": 0.00932,
+      "grad_norm": 0.5189072536981689,
+      "learning_rate": 0.0027960000000000003,
+      "loss": 4.4881,
+      "step": 932
+    },
+    {
+      "epoch": 0.00933,
+      "grad_norm": 0.4391692113712953,
+      "learning_rate": 0.0027990000000000003,
+      "loss": 4.4495,
+      "step": 933
+    },
+    {
+      "epoch": 0.00934,
+      "grad_norm": 0.5067956527040228,
+      "learning_rate": 0.0028020000000000002,
+      "loss": 4.4811,
+      "step": 934
+    },
+    {
+      "epoch": 0.00935,
+      "grad_norm": 0.5752082887405049,
+      "learning_rate": 0.002805,
+      "loss": 4.4893,
+      "step": 935
+    },
+    {
+      "epoch": 0.00936,
+      "grad_norm": 0.8597610157431208,
+      "learning_rate": 0.002808,
+      "loss": 4.5012,
+      "step": 936
+    },
+    {
+      "epoch": 0.00937,
+      "grad_norm": 1.0923006968336397,
+      "learning_rate": 0.002811,
+      "loss": 4.5462,
+      "step": 937
+    },
+    {
+      "epoch": 0.00938,
+      "grad_norm": 0.8211742236965874,
+      "learning_rate": 0.002814,
+      "loss": 4.5052,
+      "step": 938
+    },
+    {
+      "epoch": 0.00939,
+      "grad_norm": 0.699189062953733,
+      "learning_rate": 0.002817,
+      "loss": 4.4989,
+      "step": 939
+    },
+    {
+      "epoch": 0.0094,
+      "grad_norm": 0.6251265662758575,
+      "learning_rate": 0.00282,
+      "loss": 4.495,
+      "step": 940
+    },
+    {
+      "epoch": 0.00941,
+      "grad_norm": 0.8614570213641495,
+      "learning_rate": 0.002823,
+      "loss": 4.4823,
+      "step": 941
+    },
+    {
+      "epoch": 0.00942,
+      "grad_norm": 0.9710835908546839,
+      "learning_rate": 0.002826,
+      "loss": 4.5282,
+      "step": 942
+    },
+    {
+      "epoch": 0.00943,
+      "grad_norm": 0.8255156288963836,
+      "learning_rate": 0.002829,
+      "loss": 4.5175,
+      "step": 943
+    },
+    {
+      "epoch": 0.00944,
+      "grad_norm": 0.8529303551550831,
+      "learning_rate": 0.002832,
+      "loss": 4.5233,
+      "step": 944
+    },
+    {
+      "epoch": 0.00945,
+      "grad_norm": 0.8805636588256177,
+      "learning_rate": 0.002835,
+      "loss": 4.5307,
+      "step": 945
+    },
+    {
+      "epoch": 0.00946,
+      "grad_norm": 0.8815394682869104,
+      "learning_rate": 0.002838,
+      "loss": 4.545,
+      "step": 946
+    },
+    {
+      "epoch": 0.00947,
+      "grad_norm": 0.9792537736888025,
+      "learning_rate": 0.0028409999999999998,
+      "loss": 4.5259,
+      "step": 947
+    },
+    {
+      "epoch": 0.00948,
+      "grad_norm": 1.16480627354709,
+      "learning_rate": 0.0028439999999999997,
+      "loss": 4.5258,
+      "step": 948
+    },
+    {
+      "epoch": 0.00949,
+      "grad_norm": 0.9311558570406092,
+      "learning_rate": 0.002847,
+      "loss": 4.5678,
+      "step": 949
+    },
+    {
+      "epoch": 0.0095,
+      "grad_norm": 0.9151500308490546,
+      "learning_rate": 0.00285,
+      "loss": 4.5566,
+      "step": 950
+    },
+    {
+      "epoch": 0.00951,
+      "grad_norm": 0.8788038193040383,
+      "learning_rate": 0.002853,
+      "loss": 4.539,
+      "step": 951
+    },
+    {
+      "epoch": 0.00952,
+      "grad_norm": 0.6883343465241535,
+      "learning_rate": 0.002856,
+      "loss": 4.5492,
+      "step": 952
+    },
+    {
+      "epoch": 0.00953,
+      "grad_norm": 0.7428316161957196,
+      "learning_rate": 0.002859,
+      "loss": 4.4912,
+      "step": 953
+    },
+    {
+      "epoch": 0.00954,
+      "grad_norm": 0.6589075049765231,
+      "learning_rate": 0.002862,
+      "loss": 4.5478,
+      "step": 954
+    },
+    {
+      "epoch": 0.00955,
+      "grad_norm": 0.6374906159512967,
+      "learning_rate": 0.002865,
+      "loss": 4.5305,
+      "step": 955
+    },
+    {
+      "epoch": 0.00956,
+      "grad_norm": 0.7037593758453927,
+      "learning_rate": 0.002868,
+      "loss": 4.5123,
+      "step": 956
+    },
+    {
+      "epoch": 0.00957,
+      "grad_norm": 0.7575837270625414,
+      "learning_rate": 0.002871,
+      "loss": 4.5364,
+      "step": 957
+    },
+    {
+      "epoch": 0.00958,
+      "grad_norm": 0.8136537306721928,
+      "learning_rate": 0.002874,
+      "loss": 4.5221,
+      "step": 958
+    },
+    {
+      "epoch": 0.00959,
+      "grad_norm": 0.7529303499273822,
+      "learning_rate": 0.002877,
+      "loss": 4.5134,
+      "step": 959
+    },
+    {
+      "epoch": 0.0096,
+      "grad_norm": 0.7723176518474248,
+      "learning_rate": 0.0028799999999999997,
+      "loss": 4.5128,
+      "step": 960
+    },
+    {
+      "epoch": 0.00961,
+      "grad_norm": 0.6826701565045171,
+      "learning_rate": 0.002883,
+      "loss": 4.4802,
+      "step": 961
+    },
+    {
+      "epoch": 0.00962,
+      "grad_norm": 0.6022941634910601,
+      "learning_rate": 0.002886,
+      "loss": 4.5086,
+      "step": 962
+    },
+    {
+      "epoch": 0.00963,
+      "grad_norm": 0.5423609760386227,
+      "learning_rate": 0.002889,
+      "loss": 4.5025,
+      "step": 963
+    },
+    {
+      "epoch": 0.00964,
+      "grad_norm": 0.5430502092899187,
+      "learning_rate": 0.002892,
+      "loss": 4.493,
+      "step": 964
+    },
+    {
+      "epoch": 0.00965,
+      "grad_norm": 0.5046734063237629,
+      "learning_rate": 0.002895,
+      "loss": 4.4643,
+      "step": 965
+    },
+    {
+      "epoch": 0.00966,
+      "grad_norm": 0.4625107724617308,
+      "learning_rate": 0.002898,
+      "loss": 4.4758,
+      "step": 966
+    },
+    {
+      "epoch": 0.00967,
+      "grad_norm": 0.4403845150464526,
+      "learning_rate": 0.002901,
+      "loss": 4.5117,
+      "step": 967
+    },
+    {
+      "epoch": 0.00968,
+      "grad_norm": 0.40739936008519134,
+      "learning_rate": 0.002904,
+      "loss": 4.4717,
+      "step": 968
+    },
+    {
+      "epoch": 0.00969,
+      "grad_norm": 0.41620271754109195,
+      "learning_rate": 0.002907,
+      "loss": 4.5059,
+      "step": 969
+    },
+    {
+      "epoch": 0.0097,
+      "grad_norm": 0.3884080596123629,
+      "learning_rate": 0.00291,
+      "loss": 4.4561,
+      "step": 970
+    },
+    {
+      "epoch": 0.00971,
+      "grad_norm": 0.40963591049792675,
+      "learning_rate": 0.002913,
+      "loss": 4.4624,
+      "step": 971
+    },
+    {
+      "epoch": 0.00972,
+      "grad_norm": 0.427860611994679,
+      "learning_rate": 0.002916,
+      "loss": 4.4693,
+      "step": 972
+    },
+    {
+      "epoch": 0.00973,
+      "grad_norm": 0.6266830722962007,
+      "learning_rate": 0.002919,
+      "loss": 4.4689,
+      "step": 973
+    },
+    {
+      "epoch": 0.00974,
+      "grad_norm": 0.7795241673951117,
+      "learning_rate": 0.002922,
+      "loss": 4.4734,
+      "step": 974
+    },
+    {
+      "epoch": 0.00975,
+      "grad_norm": 0.7392088002205494,
+      "learning_rate": 0.002925,
+      "loss": 4.4915,
+      "step": 975
+    },
+    {
+      "epoch": 0.00976,
+      "grad_norm": 0.6090919745791438,
+      "learning_rate": 0.002928,
+      "loss": 4.4688,
+      "step": 976
+    },
+    {
+      "epoch": 0.00977,
+      "grad_norm": 0.6721667952819499,
+      "learning_rate": 0.002931,
+      "loss": 4.5114,
+      "step": 977
+    },
+    {
+      "epoch": 0.00978,
+      "grad_norm": 0.7157395676490952,
+      "learning_rate": 0.002934,
+      "loss": 4.4828,
+      "step": 978
+    },
+    {
+      "epoch": 0.00979,
+      "grad_norm": 0.6673604586868628,
+      "learning_rate": 0.002937,
+      "loss": 4.4724,
+      "step": 979
+    },
+    {
+      "epoch": 0.0098,
+      "grad_norm": 0.6241827666303977,
+      "learning_rate": 0.00294,
+      "loss": 4.4641,
+      "step": 980
+    },
+    {
+      "epoch": 0.00981,
+      "grad_norm": 0.6477735028588306,
+      "learning_rate": 0.002943,
+      "loss": 4.479,
+      "step": 981
+    },
+    {
+      "epoch": 0.00982,
+      "grad_norm": 0.5903229373466107,
+      "learning_rate": 0.002946,
+      "loss": 4.4717,
+      "step": 982
+    },
+    {
+      "epoch": 0.00983,
+      "grad_norm": 0.5367081021491611,
+      "learning_rate": 0.0029490000000000002,
+      "loss": 4.4684,
+      "step": 983
+    },
+    {
+      "epoch": 0.00984,
+      "grad_norm": 0.55855952891459,
+      "learning_rate": 0.002952,
+      "loss": 4.455,
+      "step": 984
+    },
+    {
+      "epoch": 0.00985,
+      "grad_norm": 0.6566800838357816,
+      "learning_rate": 0.002955,
+      "loss": 4.4789,
+      "step": 985
+    },
+    {
+      "epoch": 0.00986,
+      "grad_norm": 0.7602203626787503,
+      "learning_rate": 0.002958,
+      "loss": 4.4461,
+      "step": 986
+    },
+    {
+      "epoch": 0.00987,
+      "grad_norm": 0.7515002206199092,
+      "learning_rate": 0.002961,
+      "loss": 4.4656,
+      "step": 987
+    },
+    {
+      "epoch": 0.00988,
+      "grad_norm": 0.7510028045825011,
+      "learning_rate": 0.002964,
+      "loss": 4.497,
+      "step": 988
+    },
+    {
+      "epoch": 0.00989,
+      "grad_norm": 0.6686524266681442,
+      "learning_rate": 0.002967,
+      "loss": 4.438,
+      "step": 989
+    },
+    {
+      "epoch": 0.0099,
+      "grad_norm": 0.7110181433203292,
+      "learning_rate": 0.00297,
+      "loss": 4.4759,
+      "step": 990
+    },
+    {
+      "epoch": 0.00991,
+      "grad_norm": 0.9487700940206629,
+      "learning_rate": 0.002973,
+      "loss": 4.4909,
+      "step": 991
+    },
+    {
+      "epoch": 0.00992,
+      "grad_norm": 1.0128664486589338,
+      "learning_rate": 0.002976,
+      "loss": 4.4936,
+      "step": 992
+    },
+    {
+      "epoch": 0.00993,
+      "grad_norm": 0.776315556008045,
+      "learning_rate": 0.002979,
+      "loss": 4.4927,
+      "step": 993
+    },
+    {
+      "epoch": 0.00994,
+      "grad_norm": 0.6954881784010936,
+      "learning_rate": 0.002982,
+      "loss": 4.4894,
+      "step": 994
+    },
+    {
+      "epoch": 0.00995,
+      "grad_norm": 0.8343170542469196,
+      "learning_rate": 0.0029850000000000002,
+      "loss": 4.4966,
+      "step": 995
+    },
+    {
+      "epoch": 0.00996,
+      "grad_norm": 0.8064256733311401,
+      "learning_rate": 0.002988,
+      "loss": 4.4974,
+      "step": 996
+    },
+    {
+      "epoch": 0.00997,
+      "grad_norm": 0.7600837252115415,
+      "learning_rate": 0.002991,
+      "loss": 4.4859,
+      "step": 997
+    },
+    {
+      "epoch": 0.00998,
+      "grad_norm": 0.7413851219310601,
+      "learning_rate": 0.002994,
+      "loss": 4.4878,
+      "step": 998
+    },
+    {
+      "epoch": 0.00999,
+      "grad_norm": 0.7789142980889687,
+      "learning_rate": 0.002997,
+      "loss": 4.4774,
+      "step": 999
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 0.5973093879700552,
+      "learning_rate": 0.003,
+      "loss": 4.4903,
+      "step": 1000
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 100000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.9643642855424e+16,
+  "train_batch_size": 1024,
+  "trial_name": null,
+  "trial_params": null
+}