diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5801 @@
+{
+  "best_metric": 2.115697145462036,
+  "best_model_checkpoint": "outputs/checkpoint-615",
+  "epoch": 1.0,
+  "eval_steps": 205,
+  "global_step": 818,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0012224938875305623,
+      "grad_norm": 2.384822130203247,
+      "learning_rate": 0.001,
+      "loss": 5.1789,
+      "step": 1
+    },
+    {
+      "epoch": 0.0024449877750611247,
+      "grad_norm": 1.9184950590133667,
+      "learning_rate": 0.001,
+      "loss": 4.254,
+      "step": 2
+    },
+    {
+      "epoch": 0.003667481662591687,
+      "grad_norm": 3.020171642303467,
+      "learning_rate": 0.001,
+      "loss": 3.3665,
+      "step": 3
+    },
+    {
+      "epoch": 0.004889975550122249,
+      "grad_norm": 5.194127559661865,
+      "learning_rate": 0.001,
+      "loss": 3.067,
+      "step": 4
+    },
+    {
+      "epoch": 0.006112469437652812,
+      "grad_norm": 12.575337409973145,
+      "learning_rate": 0.001,
+      "loss": 3.0469,
+      "step": 5
+    },
+    {
+      "epoch": 0.007334963325183374,
+      "grad_norm": 3.6407251358032227,
+      "learning_rate": 0.001,
+      "loss": 2.7362,
+      "step": 6
+    },
+    {
+      "epoch": 0.008557457212713936,
+      "grad_norm": 0.9856404662132263,
+      "learning_rate": 0.001,
+      "loss": 2.6537,
+      "step": 7
+    },
+    {
+      "epoch": 0.009779951100244499,
+      "grad_norm": 1.4989922046661377,
+      "learning_rate": 0.001,
+      "loss": 2.7092,
+      "step": 8
+    },
+    {
+      "epoch": 0.011002444987775062,
+      "grad_norm": 0.7976619601249695,
+      "learning_rate": 0.001,
+      "loss": 2.49,
+      "step": 9
+    },
+    {
+      "epoch": 0.012224938875305624,
+      "grad_norm": 1.2533477544784546,
+      "learning_rate": 0.001,
+      "loss": 2.531,
+      "step": 10
+    },
+    {
+      "epoch": 0.013447432762836185,
+      "grad_norm": 3.7765350341796875,
+      "learning_rate": 0.001,
+      "loss": 2.6799,
+      "step": 11
+    },
+    {
+      "epoch": 0.014669926650366748,
+      "grad_norm": 0.6018803119659424,
+      "learning_rate": 0.001,
+      "loss": 2.5433,
+      "step": 12
+    },
+    {
+      "epoch": 0.01589242053789731,
+      "grad_norm": 0.5885470509529114,
+      "learning_rate": 0.001,
+      "loss": 2.5155,
+      "step": 13
+    },
+    {
+      "epoch": 0.017114914425427872,
+      "grad_norm": 0.8284630179405212,
+      "learning_rate": 0.001,
+      "loss": 2.4638,
+      "step": 14
+    },
+    {
+      "epoch": 0.018337408312958436,
+      "grad_norm": 0.7927239537239075,
+      "learning_rate": 0.001,
+      "loss": 2.472,
+      "step": 15
+    },
+    {
+      "epoch": 0.019559902200488997,
+      "grad_norm": 0.5843257904052734,
+      "learning_rate": 0.001,
+      "loss": 2.4512,
+      "step": 16
+    },
+    {
+      "epoch": 0.02078239608801956,
+      "grad_norm": 2.5700955390930176,
+      "learning_rate": 0.001,
+      "loss": 2.5298,
+      "step": 17
+    },
+    {
+      "epoch": 0.022004889975550123,
+      "grad_norm": 3.5902013778686523,
+      "learning_rate": 0.001,
+      "loss": 2.4943,
+      "step": 18
+    },
+    {
+      "epoch": 0.023227383863080684,
+      "grad_norm": 0.5734415054321289,
+      "learning_rate": 0.001,
+      "loss": 2.4265,
+      "step": 19
+    },
+    {
+      "epoch": 0.02444987775061125,
+      "grad_norm": 0.5567678213119507,
+      "learning_rate": 0.001,
+      "loss": 2.4346,
+      "step": 20
+    },
+    {
+      "epoch": 0.02567237163814181,
+      "grad_norm": 8.68786334991455,
+      "learning_rate": 0.001,
+      "loss": 2.4593,
+      "step": 21
+    },
+    {
+      "epoch": 0.02689486552567237,
+      "grad_norm": 0.69892418384552,
+      "learning_rate": 0.001,
+      "loss": 2.4801,
+      "step": 22
+    },
+    {
+      "epoch": 0.028117359413202935,
+      "grad_norm": 0.5525035262107849,
+      "learning_rate": 0.001,
+      "loss": 2.5337,
+      "step": 23
+    },
+    {
+      "epoch": 0.029339853300733496,
+      "grad_norm": 0.4160382151603699,
+      "learning_rate": 0.001,
+      "loss": 2.2792,
+      "step": 24
+    },
+    {
+      "epoch": 0.030562347188264057,
+      "grad_norm": 0.44313889741897583,
+      "learning_rate": 0.001,
+      "loss": 2.2945,
+      "step": 25
+    },
+    {
+      "epoch": 0.03178484107579462,
+      "grad_norm": 19.726951599121094,
+      "learning_rate": 0.001,
+      "loss": 2.4251,
+      "step": 26
+    },
+    {
+      "epoch": 0.03300733496332518,
+      "grad_norm": 1.9587205648422241,
+      "learning_rate": 0.001,
+      "loss": 2.4371,
+      "step": 27
+    },
+    {
+      "epoch": 0.034229828850855744,
+      "grad_norm": 0.6212214231491089,
+      "learning_rate": 0.001,
+      "loss": 2.3065,
+      "step": 28
+    },
+    {
+      "epoch": 0.035452322738386305,
+      "grad_norm": 4.7371978759765625,
+      "learning_rate": 0.001,
+      "loss": 2.3882,
+      "step": 29
+    },
+    {
+      "epoch": 0.03667481662591687,
+      "grad_norm": 2.593173027038574,
+      "learning_rate": 0.001,
+      "loss": 2.4037,
+      "step": 30
+    },
+    {
+      "epoch": 0.037897310513447434,
+      "grad_norm": 0.5192613005638123,
+      "learning_rate": 0.001,
+      "loss": 2.375,
+      "step": 31
+    },
+    {
+      "epoch": 0.039119804400977995,
+      "grad_norm": 0.7379730343818665,
+      "learning_rate": 0.001,
+      "loss": 2.3854,
+      "step": 32
+    },
+    {
+      "epoch": 0.040342298288508556,
+      "grad_norm": 9.484269142150879,
+      "learning_rate": 0.001,
+      "loss": 2.418,
+      "step": 33
+    },
+    {
+      "epoch": 0.04156479217603912,
+      "grad_norm": 0.47596585750579834,
+      "learning_rate": 0.001,
+      "loss": 2.2716,
+      "step": 34
+    },
+    {
+      "epoch": 0.042787286063569685,
+      "grad_norm": 0.4927333891391754,
+      "learning_rate": 0.001,
+      "loss": 2.4094,
+      "step": 35
+    },
+    {
+      "epoch": 0.044009779951100246,
+      "grad_norm": 0.5788190364837646,
+      "learning_rate": 0.001,
+      "loss": 2.2373,
+      "step": 36
+    },
+    {
+      "epoch": 0.04523227383863081,
+      "grad_norm": 0.4687283933162689,
+      "learning_rate": 0.001,
+      "loss": 2.2792,
+      "step": 37
+    },
+    {
+      "epoch": 0.04645476772616137,
+      "grad_norm": 0.5552632212638855,
+      "learning_rate": 0.001,
+      "loss": 2.3337,
+      "step": 38
+    },
+    {
+      "epoch": 0.04767726161369193,
+      "grad_norm": 0.6140040755271912,
+      "learning_rate": 0.001,
+      "loss": 2.3333,
+      "step": 39
+    },
+    {
+      "epoch": 0.0488997555012225,
+      "grad_norm": 0.4478518068790436,
+      "learning_rate": 0.001,
+      "loss": 2.2584,
+      "step": 40
+    },
+    {
+      "epoch": 0.05012224938875306,
+      "grad_norm": 0.5146190524101257,
+      "learning_rate": 0.001,
+      "loss": 2.3386,
+      "step": 41
+    },
+    {
+      "epoch": 0.05134474327628362,
+      "grad_norm": 0.3977064788341522,
+      "learning_rate": 0.001,
+      "loss": 2.2329,
+      "step": 42
+    },
+    {
+      "epoch": 0.05256723716381418,
+      "grad_norm": 0.6183723211288452,
+      "learning_rate": 0.001,
+      "loss": 2.4065,
+      "step": 43
+    },
+    {
+      "epoch": 0.05378973105134474,
+      "grad_norm": 0.4606506824493408,
+      "learning_rate": 0.001,
+      "loss": 2.3218,
+      "step": 44
+    },
+    {
+      "epoch": 0.0550122249388753,
+      "grad_norm": 0.7891462445259094,
+      "learning_rate": 0.001,
+      "loss": 2.2771,
+      "step": 45
+    },
+    {
+      "epoch": 0.05623471882640587,
+      "grad_norm": 7.153961658477783,
+      "learning_rate": 0.001,
+      "loss": 2.2543,
+      "step": 46
+    },
+    {
+      "epoch": 0.05745721271393643,
+      "grad_norm": 0.5428357720375061,
+      "learning_rate": 0.001,
+      "loss": 2.341,
+      "step": 47
+    },
+    {
+      "epoch": 0.05867970660146699,
+      "grad_norm": 0.4498843550682068,
+      "learning_rate": 0.001,
+      "loss": 2.3183,
+      "step": 48
+    },
+    {
+      "epoch": 0.05990220048899755,
+      "grad_norm": 0.41269147396087646,
+      "learning_rate": 0.001,
+      "loss": 2.2966,
+      "step": 49
+    },
+    {
+      "epoch": 0.061124694376528114,
+      "grad_norm": 0.4155029356479645,
+      "learning_rate": 0.001,
+      "loss": 2.2618,
+      "step": 50
+    },
+    {
+      "epoch": 0.06234718826405868,
+      "grad_norm": 0.41635772585868835,
+      "learning_rate": 0.001,
+      "loss": 2.3018,
+      "step": 51
+    },
+    {
+      "epoch": 0.06356968215158924,
+      "grad_norm": 0.3913130760192871,
+      "learning_rate": 0.001,
+      "loss": 2.1975,
+      "step": 52
+    },
+    {
+      "epoch": 0.0647921760391198,
+      "grad_norm": 0.4053911864757538,
+      "learning_rate": 0.001,
+      "loss": 2.2506,
+      "step": 53
+    },
+    {
+      "epoch": 0.06601466992665037,
+      "grad_norm": 0.43186694383621216,
+      "learning_rate": 0.001,
+      "loss": 2.235,
+      "step": 54
+    },
+    {
+      "epoch": 0.06723716381418093,
+      "grad_norm": 0.43836215138435364,
+      "learning_rate": 0.001,
+      "loss": 2.3089,
+      "step": 55
+    },
+    {
+      "epoch": 0.06845965770171149,
+      "grad_norm": 0.3933768570423126,
+      "learning_rate": 0.001,
+      "loss": 2.3207,
+      "step": 56
+    },
+    {
+      "epoch": 0.06968215158924206,
+      "grad_norm": 0.3999539911746979,
+      "learning_rate": 0.001,
+      "loss": 2.2643,
+      "step": 57
+    },
+    {
+      "epoch": 0.07090464547677261,
+      "grad_norm": 0.3799598813056946,
+      "learning_rate": 0.001,
+      "loss": 2.2278,
+      "step": 58
+    },
+    {
+      "epoch": 0.07212713936430318,
+      "grad_norm": 0.42073073983192444,
+      "learning_rate": 0.001,
+      "loss": 2.2424,
+      "step": 59
+    },
+    {
+      "epoch": 0.07334963325183375,
+      "grad_norm": 0.38391533493995667,
+      "learning_rate": 0.001,
+      "loss": 2.3866,
+      "step": 60
+    },
+    {
+      "epoch": 0.0745721271393643,
+      "grad_norm": 0.443452924489975,
+      "learning_rate": 0.001,
+      "loss": 2.2264,
+      "step": 61
+    },
+    {
+      "epoch": 0.07579462102689487,
+      "grad_norm": 0.4261990785598755,
+      "learning_rate": 0.001,
+      "loss": 2.3909,
+      "step": 62
+    },
+    {
+      "epoch": 0.07701711491442542,
+      "grad_norm": 0.35094141960144043,
+      "learning_rate": 0.001,
+      "loss": 2.2517,
+      "step": 63
+    },
+    {
+      "epoch": 0.07823960880195599,
+      "grad_norm": 0.44420382380485535,
+      "learning_rate": 0.001,
+      "loss": 2.2818,
+      "step": 64
+    },
+    {
+      "epoch": 0.07946210268948656,
+      "grad_norm": 0.3850550055503845,
+      "learning_rate": 0.001,
+      "loss": 2.2548,
+      "step": 65
+    },
+    {
+      "epoch": 0.08068459657701711,
+      "grad_norm": 0.39684730768203735,
+      "learning_rate": 0.001,
+      "loss": 2.3575,
+      "step": 66
+    },
+    {
+      "epoch": 0.08190709046454768,
+      "grad_norm": 0.3885946273803711,
+      "learning_rate": 0.001,
+      "loss": 2.3153,
+      "step": 67
+    },
+    {
+      "epoch": 0.08312958435207823,
+      "grad_norm": 0.39877697825431824,
+      "learning_rate": 0.001,
+      "loss": 2.3019,
+      "step": 68
+    },
+    {
+      "epoch": 0.0843520782396088,
+      "grad_norm": 0.38227578997612,
+      "learning_rate": 0.001,
+      "loss": 2.3031,
+      "step": 69
+    },
+    {
+      "epoch": 0.08557457212713937,
+      "grad_norm": 0.41848722100257874,
+      "learning_rate": 0.001,
+      "loss": 2.2451,
+      "step": 70
+    },
+    {
+      "epoch": 0.08679706601466992,
+      "grad_norm": 0.37621256709098816,
+      "learning_rate": 0.001,
+      "loss": 2.2327,
+      "step": 71
+    },
+    {
+      "epoch": 0.08801955990220049,
+      "grad_norm": 0.45229101181030273,
+      "learning_rate": 0.001,
+      "loss": 2.2395,
+      "step": 72
+    },
+    {
+      "epoch": 0.08924205378973105,
+      "grad_norm": 0.3657698631286621,
+      "learning_rate": 0.001,
+      "loss": 2.2631,
+      "step": 73
+    },
+    {
+      "epoch": 0.09046454767726161,
+      "grad_norm": 0.40659746527671814,
+      "learning_rate": 0.001,
+      "loss": 2.2561,
+      "step": 74
+    },
+    {
+      "epoch": 0.09168704156479218,
+      "grad_norm": 0.3982115089893341,
+      "learning_rate": 0.001,
+      "loss": 2.2651,
+      "step": 75
+    },
+    {
+      "epoch": 0.09290953545232274,
+      "grad_norm": 0.37506604194641113,
+      "learning_rate": 0.001,
+      "loss": 2.2049,
+      "step": 76
+    },
+    {
+      "epoch": 0.0941320293398533,
+      "grad_norm": 0.41589951515197754,
+      "learning_rate": 0.001,
+      "loss": 2.3502,
+      "step": 77
+    },
+    {
+      "epoch": 0.09535452322738386,
+      "grad_norm": 0.36884135007858276,
+      "learning_rate": 0.001,
+      "loss": 2.2662,
+      "step": 78
+    },
+    {
+      "epoch": 0.09657701711491443,
+      "grad_norm": 0.4224059283733368,
+      "learning_rate": 0.001,
+      "loss": 2.2057,
+      "step": 79
+    },
+    {
+      "epoch": 0.097799511002445,
+      "grad_norm": 0.5124974846839905,
+      "learning_rate": 0.001,
+      "loss": 2.2066,
+      "step": 80
+    },
+    {
+      "epoch": 0.09902200488997555,
+      "grad_norm": 0.4109271168708801,
+      "learning_rate": 0.001,
+      "loss": 2.2588,
+      "step": 81
+    },
+    {
+      "epoch": 0.10024449877750612,
+      "grad_norm": 0.44107335805892944,
+      "learning_rate": 0.001,
+      "loss": 2.2589,
+      "step": 82
+    },
+    {
+      "epoch": 0.10146699266503667,
+      "grad_norm": 0.4264432191848755,
+      "learning_rate": 0.001,
+      "loss": 2.123,
+      "step": 83
+    },
+    {
+      "epoch": 0.10268948655256724,
+      "grad_norm": 0.457029789686203,
+      "learning_rate": 0.001,
+      "loss": 2.3831,
+      "step": 84
+    },
+    {
+      "epoch": 0.1039119804400978,
+      "grad_norm": 0.397605836391449,
+      "learning_rate": 0.001,
+      "loss": 2.1846,
+      "step": 85
+    },
+    {
+      "epoch": 0.10513447432762836,
+      "grad_norm": 0.40626272559165955,
+      "learning_rate": 0.001,
+      "loss": 2.2036,
+      "step": 86
+    },
+    {
+      "epoch": 0.10635696821515893,
+      "grad_norm": 0.41786232590675354,
+      "learning_rate": 0.001,
+      "loss": 2.2301,
+      "step": 87
+    },
+    {
+      "epoch": 0.10757946210268948,
+      "grad_norm": 0.37623491883277893,
+      "learning_rate": 0.001,
+      "loss": 2.1944,
+      "step": 88
+    },
+    {
+      "epoch": 0.10880195599022005,
+      "grad_norm": 0.36336657404899597,
+      "learning_rate": 0.001,
+      "loss": 2.2961,
+      "step": 89
+    },
+    {
+      "epoch": 0.1100244498777506,
+      "grad_norm": 0.45478343963623047,
+      "learning_rate": 0.001,
+      "loss": 2.2176,
+      "step": 90
+    },
+    {
+      "epoch": 0.11124694376528117,
+      "grad_norm": 0.4099563956260681,
+      "learning_rate": 0.001,
+      "loss": 2.2708,
+      "step": 91
+    },
+    {
+      "epoch": 0.11246943765281174,
+      "grad_norm": 0.37966397404670715,
+      "learning_rate": 0.001,
+      "loss": 2.2438,
+      "step": 92
+    },
+    {
+      "epoch": 0.1136919315403423,
+      "grad_norm": 0.40351352095603943,
+      "learning_rate": 0.001,
+      "loss": 2.1437,
+      "step": 93
+    },
+    {
+      "epoch": 0.11491442542787286,
+      "grad_norm": 0.43552902340888977,
+      "learning_rate": 0.001,
+      "loss": 2.2331,
+      "step": 94
+    },
+    {
+      "epoch": 0.11613691931540342,
+      "grad_norm": 0.36740803718566895,
+      "learning_rate": 0.001,
+      "loss": 2.1243,
+      "step": 95
+    },
+    {
+      "epoch": 0.11735941320293398,
+      "grad_norm": 0.37405940890312195,
+      "learning_rate": 0.001,
+      "loss": 2.2194,
+      "step": 96
+    },
+    {
+      "epoch": 0.11858190709046455,
+      "grad_norm": 0.3814330995082855,
+      "learning_rate": 0.001,
+      "loss": 2.3012,
+      "step": 97
+    },
+    {
+      "epoch": 0.1198044009779951,
+      "grad_norm": 0.3979599177837372,
+      "learning_rate": 0.001,
+      "loss": 2.3228,
+      "step": 98
+    },
+    {
+      "epoch": 0.12102689486552567,
+      "grad_norm": 0.3659515380859375,
+      "learning_rate": 0.001,
+      "loss": 2.2247,
+      "step": 99
+    },
+    {
+      "epoch": 0.12224938875305623,
+      "grad_norm": 0.3673119843006134,
+      "learning_rate": 0.001,
+      "loss": 2.2416,
+      "step": 100
+    },
+    {
+      "epoch": 0.1234718826405868,
+      "grad_norm": 0.39655619859695435,
+      "learning_rate": 0.001,
+      "loss": 2.2511,
+      "step": 101
+    },
+    {
+      "epoch": 0.12469437652811736,
+      "grad_norm": 0.4136020541191101,
+      "learning_rate": 0.001,
+      "loss": 2.2613,
+      "step": 102
+    },
+    {
+      "epoch": 0.12591687041564792,
+      "grad_norm": 0.42136579751968384,
+      "learning_rate": 0.001,
+      "loss": 2.1932,
+      "step": 103
+    },
+    {
+      "epoch": 0.1271393643031785,
+      "grad_norm": 0.4128344655036926,
+      "learning_rate": 0.001,
+      "loss": 2.2686,
+      "step": 104
+    },
+    {
+      "epoch": 0.12836185819070906,
+      "grad_norm": 0.3856717646121979,
+      "learning_rate": 0.001,
+      "loss": 2.2565,
+      "step": 105
+    },
+    {
+      "epoch": 0.1295843520782396,
+      "grad_norm": 0.43530967831611633,
+      "learning_rate": 0.001,
+      "loss": 2.2146,
+      "step": 106
+    },
+    {
+      "epoch": 0.13080684596577016,
+      "grad_norm": 0.4071982204914093,
+      "learning_rate": 0.001,
+      "loss": 2.2356,
+      "step": 107
+    },
+    {
+      "epoch": 0.13202933985330073,
+      "grad_norm": 0.3789448142051697,
+      "learning_rate": 0.001,
+      "loss": 2.2783,
+      "step": 108
+    },
+    {
+      "epoch": 0.1332518337408313,
+      "grad_norm": 0.42017051577568054,
+      "learning_rate": 0.001,
+      "loss": 2.1768,
+      "step": 109
+    },
+    {
+      "epoch": 0.13447432762836187,
+      "grad_norm": 0.43539750576019287,
+      "learning_rate": 0.001,
+      "loss": 2.181,
+      "step": 110
+    },
+    {
+      "epoch": 0.1356968215158924,
+      "grad_norm": 0.40269771218299866,
+      "learning_rate": 0.001,
+      "loss": 2.2085,
+      "step": 111
+    },
+    {
+      "epoch": 0.13691931540342298,
+      "grad_norm": 0.42498067021369934,
+      "learning_rate": 0.001,
+      "loss": 2.2269,
+      "step": 112
+    },
+    {
+      "epoch": 0.13814180929095354,
+      "grad_norm": 0.38461583852767944,
+      "learning_rate": 0.001,
+      "loss": 2.2906,
+      "step": 113
+    },
+    {
+      "epoch": 0.1393643031784841,
+      "grad_norm": 0.4348565638065338,
+      "learning_rate": 0.001,
+      "loss": 2.295,
+      "step": 114
+    },
+    {
+      "epoch": 0.14058679706601468,
+      "grad_norm": 0.4046652019023895,
+      "learning_rate": 0.001,
+      "loss": 2.181,
+      "step": 115
+    },
+    {
+      "epoch": 0.14180929095354522,
+      "grad_norm": 0.38174059987068176,
+      "learning_rate": 0.001,
+      "loss": 2.1752,
+      "step": 116
+    },
+    {
+      "epoch": 0.1430317848410758,
+      "grad_norm": 0.5675603747367859,
+      "learning_rate": 0.001,
+      "loss": 2.3263,
+      "step": 117
+    },
+    {
+      "epoch": 0.14425427872860636,
+      "grad_norm": 0.42327845096588135,
+      "learning_rate": 0.001,
+      "loss": 2.2042,
+      "step": 118
+    },
+    {
+      "epoch": 0.14547677261613692,
+      "grad_norm": 0.5511468052864075,
+      "learning_rate": 0.001,
+      "loss": 2.2331,
+      "step": 119
+    },
+    {
+      "epoch": 0.1466992665036675,
+      "grad_norm": 0.41198548674583435,
+      "learning_rate": 0.001,
+      "loss": 2.2646,
+      "step": 120
+    },
+    {
+      "epoch": 0.14792176039119803,
+      "grad_norm": 0.5606162548065186,
+      "learning_rate": 0.001,
+      "loss": 2.303,
+      "step": 121
+    },
+    {
+      "epoch": 0.1491442542787286,
+      "grad_norm": 0.5389887690544128,
+      "learning_rate": 0.001,
+      "loss": 2.2866,
+      "step": 122
+    },
+    {
+      "epoch": 0.15036674816625917,
+      "grad_norm": 0.47240230441093445,
+      "learning_rate": 0.001,
+      "loss": 2.3047,
+      "step": 123
+    },
+    {
+      "epoch": 0.15158924205378974,
+      "grad_norm": 0.3920021653175354,
+      "learning_rate": 0.001,
+      "loss": 2.2462,
+      "step": 124
+    },
+    {
+      "epoch": 0.1528117359413203,
+      "grad_norm": 0.5402709245681763,
+      "learning_rate": 0.001,
+      "loss": 2.2036,
+      "step": 125
+    },
+    {
+      "epoch": 0.15403422982885084,
+      "grad_norm": 0.40199369192123413,
+      "learning_rate": 0.001,
+      "loss": 2.1024,
+      "step": 126
+    },
+    {
+      "epoch": 0.1552567237163814,
+      "grad_norm": 0.4106731414794922,
+      "learning_rate": 0.001,
+      "loss": 2.2969,
+      "step": 127
+    },
+    {
+      "epoch": 0.15647921760391198,
+      "grad_norm": 0.38309475779533386,
+      "learning_rate": 0.001,
+      "loss": 2.3367,
+      "step": 128
+    },
+    {
+      "epoch": 0.15770171149144255,
+      "grad_norm": 0.3637794852256775,
+      "learning_rate": 0.001,
+      "loss": 2.2432,
+      "step": 129
+    },
+    {
+      "epoch": 0.15892420537897312,
+      "grad_norm": 0.4057571589946747,
+      "learning_rate": 0.001,
+      "loss": 2.2451,
+      "step": 130
+    },
+    {
+      "epoch": 0.16014669926650366,
+      "grad_norm": 0.3911604881286621,
+      "learning_rate": 0.001,
+      "loss": 2.2577,
+      "step": 131
+    },
+    {
+      "epoch": 0.16136919315403422,
+      "grad_norm": 0.4547809660434723,
+      "learning_rate": 0.001,
+      "loss": 2.2029,
+      "step": 132
+    },
+    {
+      "epoch": 0.1625916870415648,
+      "grad_norm": 0.3767680823802948,
+      "learning_rate": 0.001,
+      "loss": 2.3067,
+      "step": 133
+    },
+    {
+      "epoch": 0.16381418092909536,
+      "grad_norm": 0.36867696046829224,
+      "learning_rate": 0.001,
+      "loss": 2.1743,
+      "step": 134
+    },
+    {
+      "epoch": 0.16503667481662593,
+      "grad_norm": 0.3789012134075165,
+      "learning_rate": 0.001,
+      "loss": 2.1533,
+      "step": 135
+    },
+    {
+      "epoch": 0.16625916870415647,
+      "grad_norm": 0.3715672194957733,
+      "learning_rate": 0.001,
+      "loss": 2.1754,
+      "step": 136
+    },
+    {
+      "epoch": 0.16748166259168704,
+      "grad_norm": 0.4021635055541992,
+      "learning_rate": 0.001,
+      "loss": 2.16,
+      "step": 137
+    },
+    {
+      "epoch": 0.1687041564792176,
+      "grad_norm": 0.401578426361084,
+      "learning_rate": 0.001,
+      "loss": 2.245,
+      "step": 138
+    },
+    {
+      "epoch": 0.16992665036674817,
+      "grad_norm": 0.3970203697681427,
+      "learning_rate": 0.001,
+      "loss": 2.2107,
+      "step": 139
+    },
+    {
+      "epoch": 0.17114914425427874,
+      "grad_norm": 0.4019182026386261,
+      "learning_rate": 0.001,
+      "loss": 2.2779,
+      "step": 140
+    },
+    {
+      "epoch": 0.17237163814180928,
+      "grad_norm": 0.42052340507507324,
+      "learning_rate": 0.001,
+      "loss": 2.2395,
+      "step": 141
+    },
+    {
+      "epoch": 0.17359413202933985,
+      "grad_norm": 0.450577050447464,
+      "learning_rate": 0.001,
+      "loss": 2.3603,
+      "step": 142
+    },
+    {
+      "epoch": 0.17481662591687042,
+      "grad_norm": 0.37792298197746277,
+      "learning_rate": 0.001,
+      "loss": 2.1671,
+      "step": 143
+    },
+    {
+      "epoch": 0.17603911980440098,
+      "grad_norm": 0.3733505308628082,
+      "learning_rate": 0.001,
+      "loss": 2.1905,
+      "step": 144
+    },
+    {
+      "epoch": 0.17726161369193155,
+      "grad_norm": 0.376338928937912,
+      "learning_rate": 0.001,
+      "loss": 2.1718,
+      "step": 145
+    },
+    {
+      "epoch": 0.1784841075794621,
+      "grad_norm": 0.3833712637424469,
+      "learning_rate": 0.001,
+      "loss": 2.2652,
+      "step": 146
+    },
+    {
+      "epoch": 0.17970660146699266,
+      "grad_norm": 0.37632736563682556,
+      "learning_rate": 0.001,
+      "loss": 2.1476,
+      "step": 147
+    },
+    {
+      "epoch": 0.18092909535452323,
+      "grad_norm": 0.40533486008644104,
+      "learning_rate": 0.001,
+      "loss": 2.1678,
+      "step": 148
+    },
+    {
+      "epoch": 0.1821515892420538,
+      "grad_norm": 0.4485403597354889,
+      "learning_rate": 0.001,
+      "loss": 2.2362,
+      "step": 149
+    },
+    {
+      "epoch": 0.18337408312958436,
+      "grad_norm": 0.41201671957969666,
+      "learning_rate": 0.001,
+      "loss": 2.2703,
+      "step": 150
+    },
+    {
+      "epoch": 0.1845965770171149,
+      "grad_norm": 0.38323187828063965,
+      "learning_rate": 0.001,
+      "loss": 2.2436,
+      "step": 151
+    },
+    {
+      "epoch": 0.18581907090464547,
+      "grad_norm": 0.4398377537727356,
+      "learning_rate": 0.001,
+      "loss": 2.3432,
+      "step": 152
+    },
+    {
+      "epoch": 0.18704156479217604,
+      "grad_norm": 0.45339325070381165,
+      "learning_rate": 0.001,
+      "loss": 2.1949,
+      "step": 153
+    },
+    {
+      "epoch": 0.1882640586797066,
+      "grad_norm": 0.44392064213752747,
+      "learning_rate": 0.001,
+      "loss": 2.0682,
+      "step": 154
+    },
+    {
+      "epoch": 0.18948655256723718,
+      "grad_norm": 0.4051275849342346,
+      "learning_rate": 0.001,
+      "loss": 2.2742,
+      "step": 155
+    },
+    {
+      "epoch": 0.19070904645476772,
+      "grad_norm": 0.47490188479423523,
+      "learning_rate": 0.001,
+      "loss": 2.2003,
+      "step": 156
+    },
+    {
+      "epoch": 0.19193154034229828,
+      "grad_norm": 0.4116264879703522,
+      "learning_rate": 0.001,
+      "loss": 2.264,
+      "step": 157
+    },
+    {
+      "epoch": 0.19315403422982885,
+      "grad_norm": 0.3978973627090454,
+      "learning_rate": 0.001,
+      "loss": 2.1993,
+      "step": 158
+    },
+    {
+      "epoch": 0.19437652811735942,
+      "grad_norm": 0.4235959053039551,
+      "learning_rate": 0.001,
+      "loss": 2.2614,
+      "step": 159
+    },
+    {
+      "epoch": 0.19559902200489,
+      "grad_norm": 0.4421234428882599,
+      "learning_rate": 0.001,
+      "loss": 2.1779,
+      "step": 160
+    },
+    {
+      "epoch": 0.19682151589242053,
+      "grad_norm": 0.37337085604667664,
+      "learning_rate": 0.001,
+      "loss": 2.1856,
+      "step": 161
+    },
+    {
+      "epoch": 0.1980440097799511,
+      "grad_norm": 0.44785186648368835,
+      "learning_rate": 0.001,
+      "loss": 2.251,
+      "step": 162
+    },
+    {
+      "epoch": 0.19926650366748166,
+      "grad_norm": 0.4330941438674927,
+      "learning_rate": 0.001,
+      "loss": 2.1669,
+      "step": 163
+    },
+    {
+      "epoch": 0.20048899755501223,
+      "grad_norm": 0.38546377420425415,
+      "learning_rate": 0.001,
+      "loss": 2.1478,
+      "step": 164
+    },
+    {
+      "epoch": 0.2017114914425428,
+      "grad_norm": 0.3848446309566498,
+      "learning_rate": 0.001,
+      "loss": 2.1803,
+      "step": 165
+    },
+    {
+      "epoch": 0.20293398533007334,
+      "grad_norm": 0.41835349798202515,
+      "learning_rate": 0.001,
+      "loss": 2.2478,
+      "step": 166
+    },
+    {
+      "epoch": 0.2041564792176039,
+      "grad_norm": 0.3915431797504425,
+      "learning_rate": 0.001,
+      "loss": 2.168,
+      "step": 167
+    },
+    {
+      "epoch": 0.20537897310513448,
+      "grad_norm": 0.40515565872192383,
+      "learning_rate": 0.001,
+      "loss": 2.2641,
+      "step": 168
+    },
+    {
+      "epoch": 0.20660146699266504,
+      "grad_norm": 0.38732442259788513,
+      "learning_rate": 0.001,
+      "loss": 2.2896,
+      "step": 169
+    },
+    {
+      "epoch": 0.2078239608801956,
+      "grad_norm": 0.3959033489227295,
+      "learning_rate": 0.001,
+      "loss": 2.1561,
+      "step": 170
+    },
+    {
+      "epoch": 0.20904645476772615,
+      "grad_norm": 0.43555209040641785,
+      "learning_rate": 0.001,
+      "loss": 2.2183,
+      "step": 171
+    },
+    {
+      "epoch": 0.21026894865525672,
+      "grad_norm": 0.43636706471443176,
+      "learning_rate": 0.001,
+      "loss": 2.1785,
+      "step": 172
+    },
+    {
+      "epoch": 0.2114914425427873,
+      "grad_norm": 0.4059796631336212,
+      "learning_rate": 0.001,
+      "loss": 2.1839,
+      "step": 173
+    },
+    {
+      "epoch": 0.21271393643031786,
+      "grad_norm": 0.4584046006202698,
+      "learning_rate": 0.001,
+      "loss": 2.2967,
+      "step": 174
+    },
+    {
+      "epoch": 0.2139364303178484,
+      "grad_norm": 0.41237303614616394,
+      "learning_rate": 0.001,
+      "loss": 2.319,
+      "step": 175
+    },
+    {
+      "epoch": 0.21515892420537897,
+      "grad_norm": 0.44713032245635986,
+      "learning_rate": 0.001,
+      "loss": 2.2066,
+      "step": 176
+    },
+    {
+      "epoch": 0.21638141809290953,
+      "grad_norm": 0.4673137962818146,
+      "learning_rate": 0.001,
+      "loss": 2.1519,
+      "step": 177
+    },
+    {
+      "epoch": 0.2176039119804401,
+      "grad_norm": 0.47857725620269775,
+      "learning_rate": 0.001,
+      "loss": 2.2214,
+      "step": 178
+    },
+    {
+      "epoch": 0.21882640586797067,
+      "grad_norm": 0.40556231141090393,
+      "learning_rate": 0.001,
+      "loss": 2.1443,
+      "step": 179
+    },
+    {
+      "epoch": 0.2200488997555012,
+      "grad_norm": 0.40253594517707825,
+      "learning_rate": 0.001,
+      "loss": 2.209,
+      "step": 180
+    },
+    {
+      "epoch": 0.22127139364303178,
+      "grad_norm": 0.44094711542129517,
+      "learning_rate": 0.001,
+      "loss": 2.142,
+      "step": 181
+    },
+    {
+      "epoch": 0.22249388753056235,
+      "grad_norm": 0.40753334760665894,
+      "learning_rate": 0.001,
+      "loss": 2.1561,
+      "step": 182
+    },
+    {
+      "epoch": 0.2237163814180929,
+      "grad_norm": 0.49447524547576904,
+      "learning_rate": 0.001,
+      "loss": 2.2386,
+      "step": 183
+    },
+    {
+      "epoch": 0.22493887530562348,
+      "grad_norm": 0.42737525701522827,
+      "learning_rate": 0.001,
+      "loss": 2.2404,
+      "step": 184
+    },
+    {
+      "epoch": 0.22616136919315402,
+      "grad_norm": 0.4718843698501587,
+      "learning_rate": 0.001,
+      "loss": 2.179,
+      "step": 185
+    },
+    {
+      "epoch": 0.2273838630806846,
+      "grad_norm": 0.3781047761440277,
+      "learning_rate": 0.001,
+      "loss": 2.2067,
+      "step": 186
+    },
+    {
+      "epoch": 0.22860635696821516,
+      "grad_norm": 0.5081697702407837,
+      "learning_rate": 0.001,
+      "loss": 2.1621,
+      "step": 187
+    },
+    {
+      "epoch": 0.22982885085574573,
+      "grad_norm": 0.4034336507320404,
+      "learning_rate": 0.001,
+      "loss": 2.1191,
+      "step": 188
+    },
+    {
+      "epoch": 0.2310513447432763,
+      "grad_norm": 0.45380139350891113,
+      "learning_rate": 0.001,
+      "loss": 2.167,
+      "step": 189
+    },
+    {
+      "epoch": 0.23227383863080683,
+      "grad_norm": 0.47250330448150635,
+      "learning_rate": 0.001,
+      "loss": 2.2126,
+      "step": 190
+    },
+    {
+      "epoch": 0.2334963325183374,
+      "grad_norm": 0.43483713269233704,
+      "learning_rate": 0.001,
+      "loss": 2.2488,
+      "step": 191
+    },
+    {
+      "epoch": 0.23471882640586797,
+      "grad_norm": 0.5420922636985779,
+      "learning_rate": 0.001,
+      "loss": 2.2842,
+      "step": 192
+    },
+    {
+      "epoch": 0.23594132029339854,
+      "grad_norm": 0.46761560440063477,
+      "learning_rate": 0.001,
+      "loss": 2.2363,
+      "step": 193
+    },
+    {
+      "epoch": 0.2371638141809291,
+      "grad_norm": 0.43423908948898315,
+      "learning_rate": 0.001,
+      "loss": 2.1634,
+      "step": 194
+    },
+    {
+      "epoch": 0.23838630806845965,
+      "grad_norm": 0.4111950397491455,
+      "learning_rate": 0.001,
+      "loss": 2.1722,
+      "step": 195
+    },
+    {
+      "epoch": 0.2396088019559902,
+      "grad_norm": 0.3922325670719147,
+      "learning_rate": 0.001,
+      "loss": 2.1681,
+      "step": 196
+    },
+    {
+      "epoch": 0.24083129584352078,
+      "grad_norm": 0.45113420486450195,
+      "learning_rate": 0.001,
+      "loss": 2.1986,
+      "step": 197
+    },
+    {
+      "epoch": 0.24205378973105135,
+      "grad_norm": 0.42804262042045593,
+      "learning_rate": 0.001,
+      "loss": 2.2062,
+      "step": 198
+    },
+    {
+      "epoch": 0.24327628361858192,
+      "grad_norm": 0.4039282500743866,
+      "learning_rate": 0.001,
+      "loss": 2.2656,
+      "step": 199
+    },
+    {
+      "epoch": 0.24449877750611246,
+      "grad_norm": 0.4062311351299286,
+      "learning_rate": 0.001,
+      "loss": 2.2483,
+      "step": 200
+    },
+    {
+      "epoch": 0.24572127139364303,
+      "grad_norm": 0.4116600751876831,
+      "learning_rate": 0.001,
+      "loss": 2.1896,
+      "step": 201
+    },
+    {
+      "epoch": 0.2469437652811736,
+      "grad_norm": 0.44408801198005676,
+      "learning_rate": 0.001,
+      "loss": 2.1628,
+      "step": 202
+    },
+    {
+      "epoch": 0.24816625916870416,
+      "grad_norm": 0.4627872109413147,
+      "learning_rate": 0.001,
+      "loss": 2.1843,
+      "step": 203
+    },
+    {
+      "epoch": 0.24938875305623473,
+      "grad_norm": 0.4298408031463623,
+      "learning_rate": 0.001,
+      "loss": 2.1324,
+      "step": 204
+    },
+    {
+      "epoch": 0.2506112469437653,
+      "grad_norm": 0.4020983576774597,
+      "learning_rate": 0.001,
+      "loss": 2.2135,
+      "step": 205
+    },
+    {
+      "epoch": 0.2506112469437653,
+      "eval_loss": 2.182542324066162,
+      "eval_runtime": 69.0489,
+      "eval_samples_per_second": 151.559,
+      "eval_steps_per_second": 18.958,
+      "step": 205
+    },
+    {
+      "epoch": 0.25183374083129584,
+      "grad_norm": 0.6257454752922058,
+      "learning_rate": 0.001,
+      "loss": 2.1383,
+      "step": 206
+    },
+    {
+      "epoch": 0.2530562347188264,
+      "grad_norm": 0.4335342347621918,
+      "learning_rate": 0.001,
+      "loss": 2.2015,
+      "step": 207
+    },
+    {
+      "epoch": 0.254278728606357,
+      "grad_norm": 0.4160003364086151,
+      "learning_rate": 0.001,
+      "loss": 2.2308,
+      "step": 208
+    },
+    {
+      "epoch": 0.2555012224938875,
+      "grad_norm": 0.40494173765182495,
+      "learning_rate": 0.001,
+      "loss": 2.3659,
+      "step": 209
+    },
+    {
+      "epoch": 0.2567237163814181,
+      "grad_norm": 0.4622797966003418,
+      "learning_rate": 0.001,
+      "loss": 2.1818,
+      "step": 210
+    },
+    {
+      "epoch": 0.25794621026894865,
+      "grad_norm": 0.4066208004951477,
+      "learning_rate": 0.001,
+      "loss": 2.2376,
+      "step": 211
+    },
+    {
+      "epoch": 0.2591687041564792,
+      "grad_norm": 0.5869137048721313,
+      "learning_rate": 0.001,
+      "loss": 2.1886,
+      "step": 212
+    },
+    {
+      "epoch": 0.2603911980440098,
+      "grad_norm": 0.38838991522789,
+      "learning_rate": 0.001,
+      "loss": 2.2406,
+      "step": 213
+    },
+    {
+      "epoch": 0.2616136919315403,
+      "grad_norm": 0.4655460715293884,
+      "learning_rate": 0.001,
+      "loss": 2.2619,
+      "step": 214
+    },
+    {
+      "epoch": 0.2628361858190709,
+      "grad_norm": 0.46997955441474915,
+      "learning_rate": 0.001,
+      "loss": 2.1855,
+      "step": 215
+    },
+    {
+      "epoch": 0.26405867970660146,
+      "grad_norm": 0.4898509979248047,
+      "learning_rate": 0.001,
+      "loss": 2.2856,
+      "step": 216
+    },
+    {
+      "epoch": 0.265281173594132,
+      "grad_norm": 0.4614812731742859,
+      "learning_rate": 0.001,
+      "loss": 2.2034,
+      "step": 217
+    },
+    {
+      "epoch": 0.2665036674816626,
+      "grad_norm": 0.41966143250465393,
+      "learning_rate": 0.001,
+      "loss": 2.1964,
+      "step": 218
+    },
+    {
+      "epoch": 0.26772616136919314,
+      "grad_norm": 0.3985699713230133,
+      "learning_rate": 0.001,
+      "loss": 2.15,
+      "step": 219
+    },
+    {
+      "epoch": 0.26894865525672373,
+      "grad_norm": 0.42294928431510925,
+      "learning_rate": 0.001,
+      "loss": 2.2558,
+      "step": 220
+    },
+    {
+      "epoch": 0.2701711491442543,
+      "grad_norm": 0.4656440019607544,
+      "learning_rate": 0.001,
+      "loss": 2.2737,
+      "step": 221
+    },
+    {
+      "epoch": 0.2713936430317848,
+      "grad_norm": 0.4101804196834564,
+      "learning_rate": 0.001,
+      "loss": 2.2278,
+      "step": 222
+    },
+    {
+      "epoch": 0.2726161369193154,
+      "grad_norm": 0.5069494843482971,
+      "learning_rate": 0.001,
+      "loss": 2.1685,
+      "step": 223
+    },
+    {
+      "epoch": 0.27383863080684595,
+      "grad_norm": 0.42884835600852966,
+      "learning_rate": 0.001,
+      "loss": 2.1298,
+      "step": 224
+    },
+    {
+      "epoch": 0.27506112469437655,
+      "grad_norm": 0.508474588394165,
+      "learning_rate": 0.001,
+      "loss": 2.0927,
+      "step": 225
+    },
+    {
+      "epoch": 0.2762836185819071,
+      "grad_norm": 0.42199471592903137,
+      "learning_rate": 0.001,
+      "loss": 2.1884,
+      "step": 226
+    },
+    {
+      "epoch": 0.2775061124694376,
+      "grad_norm": 0.45699968934059143,
+      "learning_rate": 0.001,
+      "loss": 2.3467,
+      "step": 227
+    },
+    {
+      "epoch": 0.2787286063569682,
+      "grad_norm": 0.45894646644592285,
+      "learning_rate": 0.001,
+      "loss": 2.1106,
+      "step": 228
+    },
+    {
+      "epoch": 0.27995110024449876,
+      "grad_norm": 0.42262643575668335,
+      "learning_rate": 0.001,
+      "loss": 2.1146,
+      "step": 229
+    },
+    {
+      "epoch": 0.28117359413202936,
+      "grad_norm": 0.4845474064350128,
+      "learning_rate": 0.001,
+      "loss": 2.2117,
+      "step": 230
+    },
+    {
+      "epoch": 0.2823960880195599,
+      "grad_norm": 0.42860740423202515,
+      "learning_rate": 0.001,
+      "loss": 2.1881,
+      "step": 231
+    },
+    {
+      "epoch": 0.28361858190709044,
+      "grad_norm": 0.5572071671485901,
+      "learning_rate": 0.001,
+      "loss": 2.2525,
+      "step": 232
+    },
+    {
+      "epoch": 0.28484107579462103,
+      "grad_norm": 0.4548279941082001,
+      "learning_rate": 0.001,
+      "loss": 2.1974,
+      "step": 233
+    },
+    {
+      "epoch": 0.2860635696821516,
+      "grad_norm": 0.4927968382835388,
+      "learning_rate": 0.001,
+      "loss": 2.2972,
+      "step": 234
+    },
+    {
+      "epoch": 0.28728606356968217,
+      "grad_norm": 0.42159003019332886,
+      "learning_rate": 0.001,
+      "loss": 2.1951,
+      "step": 235
+    },
+    {
+      "epoch": 0.2885085574572127,
+      "grad_norm": 0.480201780796051,
+      "learning_rate": 0.001,
+      "loss": 2.1974,
+      "step": 236
+    },
+    {
+      "epoch": 0.28973105134474325,
+      "grad_norm": 0.4612812399864197,
+      "learning_rate": 0.001,
+      "loss": 2.2359,
+      "step": 237
+    },
+    {
+      "epoch": 0.29095354523227385,
+      "grad_norm": 0.411996454000473,
+      "learning_rate": 0.001,
+      "loss": 2.1805,
+      "step": 238
+    },
+    {
+      "epoch": 0.2921760391198044,
+      "grad_norm": 0.4566625952720642,
+      "learning_rate": 0.001,
+      "loss": 2.1987,
+      "step": 239
+    },
+    {
+      "epoch": 0.293398533007335,
+      "grad_norm": 0.41748395562171936,
+      "learning_rate": 0.001,
+      "loss": 2.2479,
+      "step": 240
+    },
+    {
+      "epoch": 0.2946210268948655,
+      "grad_norm": 0.4833139479160309,
+      "learning_rate": 0.001,
+      "loss": 2.2413,
+      "step": 241
+    },
+    {
+      "epoch": 0.29584352078239606,
+      "grad_norm": 0.45390957593917847,
+      "learning_rate": 0.001,
+      "loss": 2.2535,
+      "step": 242
+    },
+    {
+      "epoch": 0.29706601466992666,
+      "grad_norm": 0.49068349599838257,
+      "learning_rate": 0.001,
+      "loss": 2.2128,
+      "step": 243
+    },
+    {
+      "epoch": 0.2982885085574572,
+      "grad_norm": 0.48676741123199463,
+      "learning_rate": 0.001,
+      "loss": 2.17,
+      "step": 244
+    },
+    {
+      "epoch": 0.2995110024449878,
+      "grad_norm": 0.42673414945602417,
+      "learning_rate": 0.001,
+      "loss": 2.0798,
+      "step": 245
+    },
+    {
+      "epoch": 0.30073349633251834,
+      "grad_norm": 0.45610859990119934,
+      "learning_rate": 0.001,
+      "loss": 2.2348,
+      "step": 246
+    },
+    {
+      "epoch": 0.3019559902200489,
+      "grad_norm": 0.48875653743743896,
+      "learning_rate": 0.001,
+      "loss": 2.1261,
+      "step": 247
+    },
+    {
+      "epoch": 0.30317848410757947,
+      "grad_norm": 0.48908236622810364,
+      "learning_rate": 0.001,
+      "loss": 2.1355,
+      "step": 248
+    },
+    {
+      "epoch": 0.30440097799511,
+      "grad_norm": 0.4941488206386566,
+      "learning_rate": 0.001,
+      "loss": 2.2139,
+      "step": 249
+    },
+    {
+      "epoch": 0.3056234718826406,
+      "grad_norm": 0.4283684194087982,
+      "learning_rate": 0.001,
+      "loss": 2.1801,
+      "step": 250
+    },
+    {
+      "epoch": 0.30684596577017115,
+      "grad_norm": 0.4848301112651825,
+      "learning_rate": 0.001,
+      "loss": 2.1626,
+      "step": 251
+    },
+    {
+      "epoch": 0.3080684596577017,
+      "grad_norm": 0.38889309763908386,
+      "learning_rate": 0.001,
+      "loss": 2.1595,
+      "step": 252
+    },
+    {
+      "epoch": 0.3092909535452323,
+      "grad_norm": 0.46130338311195374,
+      "learning_rate": 0.001,
+      "loss": 2.1576,
+      "step": 253
+    },
+    {
+      "epoch": 0.3105134474327628,
+      "grad_norm": 0.45563745498657227,
+      "learning_rate": 0.001,
+      "loss": 2.2491,
+      "step": 254
+    },
+    {
+      "epoch": 0.3117359413202934,
+      "grad_norm": 0.47916093468666077,
+      "learning_rate": 0.001,
+      "loss": 2.2009,
+      "step": 255
+    },
+    {
+      "epoch": 0.31295843520782396,
+      "grad_norm": 0.49458763003349304,
+      "learning_rate": 0.001,
+      "loss": 2.2011,
+      "step": 256
+    },
+    {
+      "epoch": 0.3141809290953545,
+      "grad_norm": 0.4567379653453827,
+      "learning_rate": 0.001,
+      "loss": 2.218,
+      "step": 257
+    },
+    {
+      "epoch": 0.3154034229828851,
+      "grad_norm": 0.5005376935005188,
+      "learning_rate": 0.001,
+      "loss": 2.2661,
+      "step": 258
+    },
+    {
+      "epoch": 0.31662591687041564,
+      "grad_norm": 0.4395696222782135,
+      "learning_rate": 0.001,
+      "loss": 2.2497,
+      "step": 259
+    },
+    {
+      "epoch": 0.31784841075794623,
+      "grad_norm": 0.48259636759757996,
+      "learning_rate": 0.001,
+      "loss": 2.2019,
+      "step": 260
+    },
+    {
+      "epoch": 0.31907090464547677,
+      "grad_norm": 0.3922736942768097,
+      "learning_rate": 0.001,
+      "loss": 2.2352,
+      "step": 261
+    },
+    {
+      "epoch": 0.3202933985330073,
+      "grad_norm": 0.528638482093811,
+      "learning_rate": 0.001,
+      "loss": 2.1914,
+      "step": 262
+    },
+    {
+      "epoch": 0.3215158924205379,
+      "grad_norm": 0.50613933801651,
+      "learning_rate": 0.001,
+      "loss": 2.2448,
+      "step": 263
+    },
+    {
+      "epoch": 0.32273838630806845,
+      "grad_norm": 0.4539433717727661,
+      "learning_rate": 0.001,
+      "loss": 2.1952,
+      "step": 264
+    },
+    {
+      "epoch": 0.32396088019559904,
+      "grad_norm": 0.5500141382217407,
+      "learning_rate": 0.001,
+      "loss": 2.2235,
+      "step": 265
+    },
+    {
+      "epoch": 0.3251833740831296,
+      "grad_norm": 0.4346414804458618,
+      "learning_rate": 0.001,
+      "loss": 2.1393,
+      "step": 266
+    },
+    {
+      "epoch": 0.3264058679706601,
+      "grad_norm": 0.47232580184936523,
+      "learning_rate": 0.001,
+      "loss": 2.2068,
+      "step": 267
+    },
+    {
+      "epoch": 0.3276283618581907,
+      "grad_norm": 0.43433475494384766,
+      "learning_rate": 0.001,
+      "loss": 2.2397,
+      "step": 268
+    },
+    {
+      "epoch": 0.32885085574572126,
+      "grad_norm": 0.44475042819976807,
+      "learning_rate": 0.001,
+      "loss": 2.1364,
+      "step": 269
+    },
+    {
+      "epoch": 0.33007334963325186,
+      "grad_norm": 0.5216724276542664,
+      "learning_rate": 0.001,
+      "loss": 2.2284,
+      "step": 270
+    },
+    {
+      "epoch": 0.3312958435207824,
+      "grad_norm": 0.4575565755367279,
+      "learning_rate": 0.001,
+      "loss": 2.2487,
+      "step": 271
+    },
+    {
+      "epoch": 0.33251833740831294,
+      "grad_norm": 0.46748054027557373,
+      "learning_rate": 0.001,
+      "loss": 2.2365,
+      "step": 272
+    },
+    {
+      "epoch": 0.33374083129584353,
+      "grad_norm": 0.4664430618286133,
+      "learning_rate": 0.001,
+      "loss": 2.2382,
+      "step": 273
+    },
+    {
+      "epoch": 0.33496332518337407,
+      "grad_norm": 0.4394036531448364,
+      "learning_rate": 0.001,
+      "loss": 2.1527,
+      "step": 274
+    },
+    {
+      "epoch": 0.33618581907090467,
+      "grad_norm": 0.40508896112442017,
+      "learning_rate": 0.001,
+      "loss": 2.241,
+      "step": 275
+    },
+    {
+      "epoch": 0.3374083129584352,
+      "grad_norm": 0.4276115298271179,
+      "learning_rate": 0.001,
+      "loss": 2.1951,
+      "step": 276
+    },
+    {
+      "epoch": 0.33863080684596575,
+      "grad_norm": 0.4394541382789612,
+      "learning_rate": 0.001,
+      "loss": 2.191,
+      "step": 277
+    },
+    {
+      "epoch": 0.33985330073349634,
+      "grad_norm": 0.47757041454315186,
+      "learning_rate": 0.001,
+      "loss": 2.1546,
+      "step": 278
+    },
+    {
+      "epoch": 0.3410757946210269,
+      "grad_norm": 0.43888697028160095,
+      "learning_rate": 0.001,
+      "loss": 2.1439,
+      "step": 279
+    },
+    {
+      "epoch": 0.3422982885085575,
+      "grad_norm": 0.43808552622795105,
+      "learning_rate": 0.001,
+      "loss": 2.2013,
+      "step": 280
+    },
+    {
+      "epoch": 0.343520782396088,
+      "grad_norm": 0.441026508808136,
+      "learning_rate": 0.001,
+      "loss": 2.2773,
+      "step": 281
+    },
+    {
+      "epoch": 0.34474327628361856,
+      "grad_norm": 0.7664148807525635,
+      "learning_rate": 0.001,
+      "loss": 2.1903,
+      "step": 282
+    },
+    {
+      "epoch": 0.34596577017114916,
+      "grad_norm": 0.4416791796684265,
+      "learning_rate": 0.001,
+      "loss": 2.2962,
+      "step": 283
+    },
+    {
+      "epoch": 0.3471882640586797,
+      "grad_norm": 0.6133376955986023,
+      "learning_rate": 0.001,
+      "loss": 2.1729,
+      "step": 284
+    },
+    {
+      "epoch": 0.3484107579462103,
+      "grad_norm": 0.446942001581192,
+      "learning_rate": 0.001,
+      "loss": 2.201,
+      "step": 285
+    },
+    {
+      "epoch": 0.34963325183374083,
+      "grad_norm": 0.4628114700317383,
+      "learning_rate": 0.001,
+      "loss": 2.1246,
+      "step": 286
+    },
+    {
+      "epoch": 0.3508557457212714,
+      "grad_norm": 0.5027017593383789,
+      "learning_rate": 0.001,
+      "loss": 2.1607,
+      "step": 287
+    },
+    {
+      "epoch": 0.35207823960880197,
+      "grad_norm": 0.4833141267299652,
+      "learning_rate": 0.001,
+      "loss": 2.2449,
+      "step": 288
+    },
+    {
+      "epoch": 0.3533007334963325,
+      "grad_norm": 0.44004133343696594,
+      "learning_rate": 0.001,
+      "loss": 2.2503,
+      "step": 289
+    },
+    {
+      "epoch": 0.3545232273838631,
+      "grad_norm": 0.44651323556900024,
+      "learning_rate": 0.001,
+      "loss": 2.1717,
+      "step": 290
+    },
+    {
+      "epoch": 0.35574572127139364,
+      "grad_norm": 0.42337921261787415,
+      "learning_rate": 0.001,
+      "loss": 2.1461,
+      "step": 291
+    },
+    {
+      "epoch": 0.3569682151589242,
+      "grad_norm": 0.42791426181793213,
+      "learning_rate": 0.001,
+      "loss": 2.1238,
+      "step": 292
+    },
+    {
+      "epoch": 0.3581907090464548,
+      "grad_norm": 0.46949905157089233,
+      "learning_rate": 0.001,
+      "loss": 2.1909,
+      "step": 293
+    },
+    {
+      "epoch": 0.3594132029339853,
+      "grad_norm": 0.49196457862854004,
+      "learning_rate": 0.001,
+      "loss": 2.1772,
+      "step": 294
+    },
+    {
+      "epoch": 0.3606356968215159,
+      "grad_norm": 0.47412702441215515,
+      "learning_rate": 0.001,
+      "loss": 2.225,
+      "step": 295
+    },
+    {
+      "epoch": 0.36185819070904646,
+      "grad_norm": 0.46963196992874146,
+      "learning_rate": 0.001,
+      "loss": 2.2759,
+      "step": 296
+    },
+    {
+      "epoch": 0.363080684596577,
+      "grad_norm": 0.49510377645492554,
+      "learning_rate": 0.001,
+      "loss": 2.2625,
+      "step": 297
+    },
+    {
+      "epoch": 0.3643031784841076,
+      "grad_norm": 0.4306149482727051,
+      "learning_rate": 0.001,
+      "loss": 2.1667,
+      "step": 298
+    },
+    {
+      "epoch": 0.36552567237163813,
+      "grad_norm": 0.4583509862422943,
+      "learning_rate": 0.001,
+      "loss": 2.2367,
+      "step": 299
+    },
+    {
+      "epoch": 0.36674816625916873,
+      "grad_norm": 0.49827685952186584,
+      "learning_rate": 0.001,
+      "loss": 2.2043,
+      "step": 300
+    },
+    {
+      "epoch": 0.36797066014669927,
+      "grad_norm": 0.5875521898269653,
+      "learning_rate": 0.001,
+      "loss": 2.2675,
+      "step": 301
+    },
+    {
+      "epoch": 0.3691931540342298,
+      "grad_norm": 0.4727851152420044,
+      "learning_rate": 0.001,
+      "loss": 2.1613,
+      "step": 302
+    },
+    {
+      "epoch": 0.3704156479217604,
+      "grad_norm": 0.4806399941444397,
+      "learning_rate": 0.001,
+      "loss": 2.1929,
+      "step": 303
+    },
+    {
+      "epoch": 0.37163814180929094,
+      "grad_norm": 0.711615800857544,
+      "learning_rate": 0.001,
+      "loss": 2.2085,
+      "step": 304
+    },
+    {
+      "epoch": 0.37286063569682154,
+      "grad_norm": 0.5827208757400513,
+      "learning_rate": 0.001,
+      "loss": 2.2931,
+      "step": 305
+    },
+    {
+      "epoch": 0.3740831295843521,
+      "grad_norm": 0.4817729592323303,
+      "learning_rate": 0.001,
+      "loss": 2.1518,
+      "step": 306
+    },
+    {
+      "epoch": 0.3753056234718826,
+      "grad_norm": 0.4607214033603668,
+      "learning_rate": 0.001,
+      "loss": 2.2376,
+      "step": 307
+    },
+    {
+      "epoch": 0.3765281173594132,
+      "grad_norm": 0.5439569354057312,
+      "learning_rate": 0.001,
+      "loss": 2.214,
+      "step": 308
+    },
+    {
+      "epoch": 0.37775061124694376,
+      "grad_norm": 0.553310751914978,
+      "learning_rate": 0.001,
+      "loss": 2.1845,
+      "step": 309
+    },
+    {
+      "epoch": 0.37897310513447435,
+      "grad_norm": 0.5218061208724976,
+      "learning_rate": 0.001,
+      "loss": 2.1877,
+      "step": 310
+    },
+    {
+      "epoch": 0.3801955990220049,
+      "grad_norm": 0.44088688492774963,
+      "learning_rate": 0.001,
+      "loss": 2.1382,
+      "step": 311
+    },
+    {
+      "epoch": 0.38141809290953543,
+      "grad_norm": 0.46488451957702637,
+      "learning_rate": 0.001,
+      "loss": 2.2819,
+      "step": 312
+    },
+    {
+      "epoch": 0.38264058679706603,
+      "grad_norm": 0.46994853019714355,
+      "learning_rate": 0.001,
+      "loss": 2.1534,
+      "step": 313
+    },
+    {
+      "epoch": 0.38386308068459657,
+      "grad_norm": 0.47553324699401855,
+      "learning_rate": 0.001,
+      "loss": 2.2419,
+      "step": 314
+    },
+    {
+      "epoch": 0.38508557457212717,
+      "grad_norm": 0.4419028162956238,
+      "learning_rate": 0.001,
+      "loss": 2.0614,
+      "step": 315
+    },
+    {
+      "epoch": 0.3863080684596577,
+      "grad_norm": 0.482204794883728,
+      "learning_rate": 0.001,
+      "loss": 2.1259,
+      "step": 316
+    },
+    {
+      "epoch": 0.38753056234718825,
+      "grad_norm": 0.46489477157592773,
+      "learning_rate": 0.001,
+      "loss": 2.1542,
+      "step": 317
+    },
+    {
+      "epoch": 0.38875305623471884,
+      "grad_norm": 0.613818883895874,
+      "learning_rate": 0.001,
+      "loss": 2.151,
+      "step": 318
+    },
+    {
+      "epoch": 0.3899755501222494,
+      "grad_norm": 0.48042118549346924,
+      "learning_rate": 0.001,
+      "loss": 2.2729,
+      "step": 319
+    },
+    {
+      "epoch": 0.39119804400978,
+      "grad_norm": 0.5772485733032227,
+      "learning_rate": 0.001,
+      "loss": 2.1729,
+      "step": 320
+    },
+    {
+      "epoch": 0.3924205378973105,
+      "grad_norm": 0.9449200630187988,
+      "learning_rate": 0.001,
+      "loss": 2.2178,
+      "step": 321
+    },
+    {
+      "epoch": 0.39364303178484106,
+      "grad_norm": 0.4307827651500702,
+      "learning_rate": 0.001,
+      "loss": 2.1567,
+      "step": 322
+    },
+    {
+      "epoch": 0.39486552567237165,
+      "grad_norm": 0.5241131782531738,
+      "learning_rate": 0.001,
+      "loss": 2.1624,
+      "step": 323
+    },
+    {
+      "epoch": 0.3960880195599022,
+      "grad_norm": 0.5168140530586243,
+      "learning_rate": 0.001,
+      "loss": 2.2424,
+      "step": 324
+    },
+    {
+      "epoch": 0.3973105134474328,
+      "grad_norm": 0.47847840189933777,
+      "learning_rate": 0.001,
+      "loss": 2.1865,
+      "step": 325
+    },
+    {
+      "epoch": 0.39853300733496333,
+      "grad_norm": 0.559695303440094,
+      "learning_rate": 0.001,
+      "loss": 2.1683,
+      "step": 326
+    },
+    {
+      "epoch": 0.39975550122249387,
+      "grad_norm": 0.48280563950538635,
+      "learning_rate": 0.001,
+      "loss": 2.1433,
+      "step": 327
+    },
+    {
+      "epoch": 0.40097799511002447,
+      "grad_norm": 0.5894402861595154,
+      "learning_rate": 0.001,
+      "loss": 2.1939,
+      "step": 328
+    },
+    {
+      "epoch": 0.402200488997555,
+      "grad_norm": 0.4633193016052246,
+      "learning_rate": 0.001,
+      "loss": 2.2026,
+      "step": 329
+    },
+    {
+      "epoch": 0.4034229828850856,
+      "grad_norm": 0.502029538154602,
+      "learning_rate": 0.001,
+      "loss": 2.1984,
+      "step": 330
+    },
+    {
+      "epoch": 0.40464547677261614,
+      "grad_norm": 0.6570725440979004,
+      "learning_rate": 0.001,
+      "loss": 2.2083,
+      "step": 331
+    },
+    {
+      "epoch": 0.4058679706601467,
+      "grad_norm": 0.459811270236969,
+      "learning_rate": 0.001,
+      "loss": 2.2403,
+      "step": 332
+    },
+    {
+      "epoch": 0.4070904645476773,
+      "grad_norm": 0.6034518480300903,
+      "learning_rate": 0.001,
+      "loss": 2.1935,
+      "step": 333
+    },
+    {
+      "epoch": 0.4083129584352078,
+      "grad_norm": 0.47477465867996216,
+      "learning_rate": 0.001,
+      "loss": 2.2074,
+      "step": 334
+    },
+    {
+      "epoch": 0.4095354523227384,
+      "grad_norm": 0.5079094767570496,
+      "learning_rate": 0.001,
+      "loss": 2.2006,
+      "step": 335
+    },
+    {
+      "epoch": 0.41075794621026895,
+      "grad_norm": 0.5319966077804565,
+      "learning_rate": 0.001,
+      "loss": 2.1931,
+      "step": 336
+    },
+    {
+      "epoch": 0.4119804400977995,
+      "grad_norm": 0.4573460519313812,
+      "learning_rate": 0.001,
+      "loss": 2.2233,
+      "step": 337
+    },
+    {
+      "epoch": 0.4132029339853301,
+      "grad_norm": 0.47283318638801575,
+      "learning_rate": 0.001,
+      "loss": 2.1454,
+      "step": 338
+    },
+    {
+      "epoch": 0.41442542787286063,
+      "grad_norm": 0.43014955520629883,
+      "learning_rate": 0.001,
+      "loss": 2.1942,
+      "step": 339
+    },
+    {
+      "epoch": 0.4156479217603912,
+      "grad_norm": 0.4784499406814575,
+      "learning_rate": 0.001,
+      "loss": 2.1523,
+      "step": 340
+    },
+    {
+      "epoch": 0.41687041564792177,
+      "grad_norm": 0.44560521841049194,
+      "learning_rate": 0.001,
+      "loss": 2.1879,
+      "step": 341
+    },
+    {
+      "epoch": 0.4180929095354523,
+      "grad_norm": 0.42554497718811035,
+      "learning_rate": 0.001,
+      "loss": 2.1782,
+      "step": 342
+    },
+    {
+      "epoch": 0.4193154034229829,
+      "grad_norm": 0.6102154850959778,
+      "learning_rate": 0.001,
+      "loss": 2.2327,
+      "step": 343
+    },
+    {
+      "epoch": 0.42053789731051344,
+      "grad_norm": 0.45047393441200256,
+      "learning_rate": 0.001,
+      "loss": 2.2005,
+      "step": 344
+    },
+    {
+      "epoch": 0.421760391198044,
+      "grad_norm": 0.4865048825740814,
+      "learning_rate": 0.001,
+      "loss": 2.3121,
+      "step": 345
+    },
+    {
+      "epoch": 0.4229828850855746,
+      "grad_norm": 0.4858083426952362,
+      "learning_rate": 0.001,
+      "loss": 2.0612,
+      "step": 346
+    },
+    {
+      "epoch": 0.4242053789731051,
+      "grad_norm": 0.4916556477546692,
+      "learning_rate": 0.001,
+      "loss": 2.2065,
+      "step": 347
+    },
+    {
+      "epoch": 0.4254278728606357,
+      "grad_norm": 0.6283455491065979,
+      "learning_rate": 0.001,
+      "loss": 2.2506,
+      "step": 348
+    },
+    {
+      "epoch": 0.42665036674816625,
+      "grad_norm": 0.6359683275222778,
+      "learning_rate": 0.001,
+      "loss": 2.2236,
+      "step": 349
+    },
+    {
+      "epoch": 0.4278728606356968,
+      "grad_norm": 0.5602818727493286,
+      "learning_rate": 0.001,
+      "loss": 2.2154,
+      "step": 350
+    },
+    {
+      "epoch": 0.4290953545232274,
+      "grad_norm": 0.5634006261825562,
+      "learning_rate": 0.001,
+      "loss": 2.0715,
+      "step": 351
+    },
+    {
+      "epoch": 0.43031784841075793,
+      "grad_norm": 0.4602808654308319,
+      "learning_rate": 0.001,
+      "loss": 2.1809,
+      "step": 352
+    },
+    {
+      "epoch": 0.4315403422982885,
+      "grad_norm": 0.4815332591533661,
+      "learning_rate": 0.001,
+      "loss": 2.1637,
+      "step": 353
+    },
+    {
+      "epoch": 0.43276283618581907,
+      "grad_norm": 0.49894168972969055,
+      "learning_rate": 0.001,
+      "loss": 2.2466,
+      "step": 354
+    },
+    {
+      "epoch": 0.4339853300733496,
+      "grad_norm": 0.4808076322078705,
+      "learning_rate": 0.001,
+      "loss": 2.2192,
+      "step": 355
+    },
+    {
+      "epoch": 0.4352078239608802,
+      "grad_norm": 0.45529651641845703,
+      "learning_rate": 0.001,
+      "loss": 2.2279,
+      "step": 356
+    },
+    {
+      "epoch": 0.43643031784841074,
+      "grad_norm": 0.4804287552833557,
+      "learning_rate": 0.001,
+      "loss": 2.1756,
+      "step": 357
+    },
+    {
+      "epoch": 0.43765281173594134,
+      "grad_norm": 0.5576850175857544,
+      "learning_rate": 0.001,
+      "loss": 2.1637,
+      "step": 358
+    },
+    {
+      "epoch": 0.4388753056234719,
+      "grad_norm": 0.5175532698631287,
+      "learning_rate": 0.001,
+      "loss": 2.0891,
+      "step": 359
+    },
+    {
+      "epoch": 0.4400977995110024,
+      "grad_norm": 0.6010396480560303,
+      "learning_rate": 0.001,
+      "loss": 2.2515,
+      "step": 360
+    },
+    {
+      "epoch": 0.441320293398533,
+      "grad_norm": 0.6106312274932861,
+      "learning_rate": 0.001,
+      "loss": 2.1245,
+      "step": 361
+    },
+    {
+      "epoch": 0.44254278728606355,
+      "grad_norm": 0.5079056620597839,
+      "learning_rate": 0.001,
+      "loss": 2.1149,
+      "step": 362
+    },
+    {
+      "epoch": 0.44376528117359415,
+      "grad_norm": 0.5122853517532349,
+      "learning_rate": 0.001,
+      "loss": 2.175,
+      "step": 363
+    },
+    {
+      "epoch": 0.4449877750611247,
+      "grad_norm": 0.5180224776268005,
+      "learning_rate": 0.001,
+      "loss": 2.2279,
+      "step": 364
+    },
+    {
+      "epoch": 0.44621026894865523,
+      "grad_norm": 0.4952344298362732,
+      "learning_rate": 0.001,
+      "loss": 2.1543,
+      "step": 365
+    },
+    {
+      "epoch": 0.4474327628361858,
+      "grad_norm": 0.5394139885902405,
+      "learning_rate": 0.001,
+      "loss": 2.0964,
+      "step": 366
+    },
+    {
+      "epoch": 0.44865525672371637,
+      "grad_norm": 0.4426160454750061,
+      "learning_rate": 0.001,
+      "loss": 2.1959,
+      "step": 367
+    },
+    {
+      "epoch": 0.44987775061124696,
+      "grad_norm": 0.446934312582016,
+      "learning_rate": 0.001,
+      "loss": 2.2799,
+      "step": 368
+    },
+    {
+      "epoch": 0.4511002444987775,
+      "grad_norm": 0.6091988682746887,
+      "learning_rate": 0.001,
+      "loss": 2.2826,
+      "step": 369
+    },
+    {
+      "epoch": 0.45232273838630804,
+      "grad_norm": 0.6177111268043518,
+      "learning_rate": 0.001,
+      "loss": 2.1416,
+      "step": 370
+    },
+    {
+      "epoch": 0.45354523227383864,
+      "grad_norm": 0.6030283570289612,
+      "learning_rate": 0.001,
+      "loss": 2.1534,
+      "step": 371
+    },
+    {
+      "epoch": 0.4547677261613692,
+      "grad_norm": 0.7201472520828247,
+      "learning_rate": 0.001,
+      "loss": 2.1527,
+      "step": 372
+    },
+    {
+      "epoch": 0.4559902200488998,
+      "grad_norm": 0.5519202947616577,
+      "learning_rate": 0.001,
+      "loss": 2.1276,
+      "step": 373
+    },
+    {
+      "epoch": 0.4572127139364303,
+      "grad_norm": 0.665658712387085,
+      "learning_rate": 0.001,
+      "loss": 2.1326,
+      "step": 374
+    },
+    {
+      "epoch": 0.45843520782396086,
+      "grad_norm": 0.5527313947677612,
+      "learning_rate": 0.001,
+      "loss": 2.1636,
+      "step": 375
+    },
+    {
+      "epoch": 0.45965770171149145,
+      "grad_norm": 0.5481876134872437,
+      "learning_rate": 0.001,
+      "loss": 2.0782,
+      "step": 376
+    },
+    {
+      "epoch": 0.460880195599022,
+      "grad_norm": 0.5887396335601807,
+      "learning_rate": 0.001,
+      "loss": 2.1269,
+      "step": 377
+    },
+    {
+      "epoch": 0.4621026894865526,
+      "grad_norm": 0.47340717911720276,
+      "learning_rate": 0.001,
+      "loss": 2.1033,
+      "step": 378
+    },
+    {
+      "epoch": 0.4633251833740831,
+      "grad_norm": 0.5240494012832642,
+      "learning_rate": 0.001,
+      "loss": 2.1931,
+      "step": 379
+    },
+    {
+      "epoch": 0.46454767726161367,
+      "grad_norm": 0.7044278979301453,
+      "learning_rate": 0.001,
+      "loss": 2.2165,
+      "step": 380
+    },
+    {
+      "epoch": 0.46577017114914426,
+      "grad_norm": 0.4907272458076477,
+      "learning_rate": 0.001,
+      "loss": 2.1903,
+      "step": 381
+    },
+    {
+      "epoch": 0.4669926650366748,
+      "grad_norm": 0.588350236415863,
+      "learning_rate": 0.001,
+      "loss": 2.209,
+      "step": 382
+    },
+    {
+      "epoch": 0.4682151589242054,
+      "grad_norm": 0.43188443779945374,
+      "learning_rate": 0.001,
+      "loss": 2.2638,
+      "step": 383
+    },
+    {
+      "epoch": 0.46943765281173594,
+      "grad_norm": 0.5474991798400879,
+      "learning_rate": 0.001,
+      "loss": 2.1538,
+      "step": 384
+    },
+    {
+      "epoch": 0.4706601466992665,
+      "grad_norm": 0.4939529597759247,
+      "learning_rate": 0.001,
+      "loss": 2.1797,
+      "step": 385
+    },
+    {
+      "epoch": 0.4718826405867971,
+      "grad_norm": 0.46284395456314087,
+      "learning_rate": 0.001,
+      "loss": 2.1734,
+      "step": 386
+    },
+    {
+      "epoch": 0.4731051344743276,
+      "grad_norm": 0.4820462465286255,
+      "learning_rate": 0.001,
+      "loss": 2.2029,
+      "step": 387
+    },
+    {
+      "epoch": 0.4743276283618582,
+      "grad_norm": 0.5413796901702881,
+      "learning_rate": 0.001,
+      "loss": 2.2625,
+      "step": 388
+    },
+    {
+      "epoch": 0.47555012224938875,
+      "grad_norm": 0.5737091898918152,
+      "learning_rate": 0.001,
+      "loss": 2.2101,
+      "step": 389
+    },
+    {
+      "epoch": 0.4767726161369193,
+      "grad_norm": 0.6295710802078247,
+      "learning_rate": 0.001,
+      "loss": 2.2646,
+      "step": 390
+    },
+    {
+      "epoch": 0.4779951100244499,
+      "grad_norm": 0.5470510125160217,
+      "learning_rate": 0.001,
+      "loss": 2.2067,
+      "step": 391
+    },
+    {
+      "epoch": 0.4792176039119804,
+      "grad_norm": 0.7382400631904602,
+      "learning_rate": 0.001,
+      "loss": 2.1721,
+      "step": 392
+    },
+    {
+      "epoch": 0.480440097799511,
+      "grad_norm": 0.5777731537818909,
+      "learning_rate": 0.001,
+      "loss": 2.1658,
+      "step": 393
+    },
+    {
+      "epoch": 0.48166259168704156,
+      "grad_norm": 0.6507386565208435,
+      "learning_rate": 0.001,
+      "loss": 2.2107,
+      "step": 394
+    },
+    {
+      "epoch": 0.4828850855745721,
+      "grad_norm": 0.5742748379707336,
+      "learning_rate": 0.001,
+      "loss": 2.2308,
+      "step": 395
+    },
+    {
+      "epoch": 0.4841075794621027,
+      "grad_norm": 0.4340347349643707,
+      "learning_rate": 0.001,
+      "loss": 2.1719,
+      "step": 396
+    },
+    {
+      "epoch": 0.48533007334963324,
+      "grad_norm": 0.5274593234062195,
+      "learning_rate": 0.001,
+      "loss": 2.1833,
+      "step": 397
+    },
+    {
+      "epoch": 0.48655256723716384,
+      "grad_norm": 0.4945325553417206,
+      "learning_rate": 0.001,
+      "loss": 2.2275,
+      "step": 398
+    },
+    {
+      "epoch": 0.4877750611246944,
+      "grad_norm": 0.5010286569595337,
+      "learning_rate": 0.001,
+      "loss": 2.0742,
+      "step": 399
+    },
+    {
+      "epoch": 0.4889975550122249,
+      "grad_norm": 0.5433893799781799,
+      "learning_rate": 0.001,
+      "loss": 2.1805,
+      "step": 400
+    },
+    {
+      "epoch": 0.4902200488997555,
+      "grad_norm": 0.5283117294311523,
+      "learning_rate": 0.001,
+      "loss": 2.1663,
+      "step": 401
+    },
+    {
+      "epoch": 0.49144254278728605,
+      "grad_norm": 0.5704575181007385,
+      "learning_rate": 0.001,
+      "loss": 2.2328,
+      "step": 402
+    },
+    {
+      "epoch": 0.49266503667481665,
+      "grad_norm": 0.45715004205703735,
+      "learning_rate": 0.001,
+      "loss": 2.184,
+      "step": 403
+    },
+    {
+      "epoch": 0.4938875305623472,
+      "grad_norm": 0.5323057174682617,
+      "learning_rate": 0.001,
+      "loss": 2.1889,
+      "step": 404
+    },
+    {
+      "epoch": 0.49511002444987773,
+      "grad_norm": 0.5370688438415527,
+      "learning_rate": 0.001,
+      "loss": 2.1373,
+      "step": 405
+    },
+    {
+      "epoch": 0.4963325183374083,
+      "grad_norm": 0.45251837372779846,
+      "learning_rate": 0.001,
+      "loss": 2.1459,
+      "step": 406
+    },
+    {
+      "epoch": 0.49755501222493886,
+      "grad_norm": 0.4832480549812317,
+      "learning_rate": 0.001,
+      "loss": 2.1073,
+      "step": 407
+    },
+    {
+      "epoch": 0.49877750611246946,
+      "grad_norm": 0.5315654873847961,
+      "learning_rate": 0.001,
+      "loss": 2.16,
+      "step": 408
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 0.5699260830879211,
+      "learning_rate": 0.001,
+      "loss": 2.2989,
+      "step": 409
+    },
+    {
+      "epoch": 0.5012224938875306,
+      "grad_norm": 0.49974632263183594,
+      "learning_rate": 0.001,
+      "loss": 2.1772,
+      "step": 410
+    },
+    {
+      "epoch": 0.5012224938875306,
+      "eval_loss": 2.1469788551330566,
+      "eval_runtime": 69.1494,
+      "eval_samples_per_second": 151.339,
+      "eval_steps_per_second": 18.93,
+      "step": 410
+    },
+    {
+      "epoch": 0.5024449877750611,
+      "grad_norm": 0.5740631818771362,
+      "learning_rate": 0.001,
+      "loss": 2.1783,
+      "step": 411
+    },
+    {
+      "epoch": 0.5036674816625917,
+      "grad_norm": 0.5050974488258362,
+      "learning_rate": 0.001,
+      "loss": 2.1696,
+      "step": 412
+    },
+    {
+      "epoch": 0.5048899755501223,
+      "grad_norm": 0.6442350149154663,
+      "learning_rate": 0.001,
+      "loss": 2.1755,
+      "step": 413
+    },
+    {
+      "epoch": 0.5061124694376528,
+      "grad_norm": 0.5595468282699585,
+      "learning_rate": 0.001,
+      "loss": 2.2138,
+      "step": 414
+    },
+    {
+      "epoch": 0.5073349633251834,
+      "grad_norm": 0.5576492547988892,
+      "learning_rate": 0.001,
+      "loss": 2.1927,
+      "step": 415
+    },
+    {
+      "epoch": 0.508557457212714,
+      "grad_norm": 1.1910855770111084,
+      "learning_rate": 0.001,
+      "loss": 2.1443,
+      "step": 416
+    },
+    {
+      "epoch": 0.5097799511002445,
+      "grad_norm": 0.5421645045280457,
+      "learning_rate": 0.001,
+      "loss": 2.1744,
+      "step": 417
+    },
+    {
+      "epoch": 0.511002444987775,
+      "grad_norm": 0.6474606394767761,
+      "learning_rate": 0.001,
+      "loss": 2.2957,
+      "step": 418
+    },
+    {
+      "epoch": 0.5122249388753056,
+      "grad_norm": 0.5093860626220703,
+      "learning_rate": 0.001,
+      "loss": 2.2137,
+      "step": 419
+    },
+    {
+      "epoch": 0.5134474327628362,
+      "grad_norm": 0.6543591022491455,
+      "learning_rate": 0.001,
+      "loss": 2.2671,
+      "step": 420
+    },
+    {
+      "epoch": 0.5146699266503667,
+      "grad_norm": 0.4889623522758484,
+      "learning_rate": 0.001,
+      "loss": 2.1874,
+      "step": 421
+    },
+    {
+      "epoch": 0.5158924205378973,
+      "grad_norm": 0.5349574089050293,
+      "learning_rate": 0.001,
+      "loss": 2.1826,
+      "step": 422
+    },
+    {
+      "epoch": 0.5171149144254279,
+      "grad_norm": 0.5584796071052551,
+      "learning_rate": 0.001,
+      "loss": 2.2241,
+      "step": 423
+    },
+    {
+      "epoch": 0.5183374083129584,
+      "grad_norm": 0.49419817328453064,
+      "learning_rate": 0.001,
+      "loss": 2.2039,
+      "step": 424
+    },
+    {
+      "epoch": 0.519559902200489,
+      "grad_norm": 0.6226582527160645,
+      "learning_rate": 0.001,
+      "loss": 2.1601,
+      "step": 425
+    },
+    {
+      "epoch": 0.5207823960880196,
+      "grad_norm": 0.4907011091709137,
+      "learning_rate": 0.001,
+      "loss": 2.1829,
+      "step": 426
+    },
+    {
+      "epoch": 0.5220048899755502,
+      "grad_norm": 0.6074104309082031,
+      "learning_rate": 0.001,
+      "loss": 2.1575,
+      "step": 427
+    },
+    {
+      "epoch": 0.5232273838630807,
+      "grad_norm": 0.5580583810806274,
+      "learning_rate": 0.001,
+      "loss": 2.1722,
+      "step": 428
+    },
+    {
+      "epoch": 0.5244498777506112,
+      "grad_norm": 0.5270822644233704,
+      "learning_rate": 0.001,
+      "loss": 2.2356,
+      "step": 429
+    },
+    {
+      "epoch": 0.5256723716381418,
+      "grad_norm": 0.5240160226821899,
+      "learning_rate": 0.001,
+      "loss": 2.187,
+      "step": 430
+    },
+    {
+      "epoch": 0.5268948655256723,
+      "grad_norm": 0.6706748604774475,
+      "learning_rate": 0.001,
+      "loss": 2.2156,
+      "step": 431
+    },
+    {
+      "epoch": 0.5281173594132029,
+      "grad_norm": 0.5267993807792664,
+      "learning_rate": 0.001,
+      "loss": 2.205,
+      "step": 432
+    },
+    {
+      "epoch": 0.5293398533007335,
+      "grad_norm": 0.6394602656364441,
+      "learning_rate": 0.001,
+      "loss": 2.203,
+      "step": 433
+    },
+    {
+      "epoch": 0.530562347188264,
+      "grad_norm": 0.544459879398346,
+      "learning_rate": 0.001,
+      "loss": 2.3439,
+      "step": 434
+    },
+    {
+      "epoch": 0.5317848410757946,
+      "grad_norm": 0.5763698220252991,
+      "learning_rate": 0.001,
+      "loss": 2.1619,
+      "step": 435
+    },
+    {
+      "epoch": 0.5330073349633252,
+      "grad_norm": 0.539488673210144,
+      "learning_rate": 0.001,
+      "loss": 2.2853,
+      "step": 436
+    },
+    {
+      "epoch": 0.5342298288508558,
+      "grad_norm": 0.5510901212692261,
+      "learning_rate": 0.001,
+      "loss": 2.2645,
+      "step": 437
+    },
+    {
+      "epoch": 0.5354523227383863,
+      "grad_norm": 0.5045850276947021,
+      "learning_rate": 0.001,
+      "loss": 2.2893,
+      "step": 438
+    },
+    {
+      "epoch": 0.5366748166259169,
+      "grad_norm": 0.48548564314842224,
+      "learning_rate": 0.001,
+      "loss": 2.3173,
+      "step": 439
+    },
+    {
+      "epoch": 0.5378973105134475,
+      "grad_norm": 0.5125932693481445,
+      "learning_rate": 0.001,
+      "loss": 2.2716,
+      "step": 440
+    },
+    {
+      "epoch": 0.539119804400978,
+      "grad_norm": 0.5173934102058411,
+      "learning_rate": 0.001,
+      "loss": 2.1719,
+      "step": 441
+    },
+    {
+      "epoch": 0.5403422982885085,
+      "grad_norm": 0.546996533870697,
+      "learning_rate": 0.001,
+      "loss": 2.3323,
+      "step": 442
+    },
+    {
+      "epoch": 0.5415647921760391,
+      "grad_norm": 0.48216158151626587,
+      "learning_rate": 0.001,
+      "loss": 2.1881,
+      "step": 443
+    },
+    {
+      "epoch": 0.5427872860635696,
+      "grad_norm": 0.44798579812049866,
+      "learning_rate": 0.001,
+      "loss": 2.1785,
+      "step": 444
+    },
+    {
+      "epoch": 0.5440097799511002,
+      "grad_norm": 0.4922938346862793,
+      "learning_rate": 0.001,
+      "loss": 2.1886,
+      "step": 445
+    },
+    {
+      "epoch": 0.5452322738386308,
+      "grad_norm": 0.564640998840332,
+      "learning_rate": 0.001,
+      "loss": 2.1519,
+      "step": 446
+    },
+    {
+      "epoch": 0.5464547677261614,
+      "grad_norm": 0.5763362050056458,
+      "learning_rate": 0.001,
+      "loss": 2.2067,
+      "step": 447
+    },
+    {
+      "epoch": 0.5476772616136919,
+      "grad_norm": 0.5354210138320923,
+      "learning_rate": 0.001,
+      "loss": 2.2425,
+      "step": 448
+    },
+    {
+      "epoch": 0.5488997555012225,
+      "grad_norm": 0.6192667484283447,
+      "learning_rate": 0.001,
+      "loss": 2.2426,
+      "step": 449
+    },
+    {
+      "epoch": 0.5501222493887531,
+      "grad_norm": 0.4782339632511139,
+      "learning_rate": 0.001,
+      "loss": 2.1758,
+      "step": 450
+    },
+    {
+      "epoch": 0.5513447432762836,
+      "grad_norm": 0.5111038684844971,
+      "learning_rate": 0.001,
+      "loss": 2.1927,
+      "step": 451
+    },
+    {
+      "epoch": 0.5525672371638142,
+      "grad_norm": 0.5513277649879456,
+      "learning_rate": 0.001,
+      "loss": 2.2836,
+      "step": 452
+    },
+    {
+      "epoch": 0.5537897310513448,
+      "grad_norm": 0.6026435494422913,
+      "learning_rate": 0.001,
+      "loss": 2.2628,
+      "step": 453
+    },
+    {
+      "epoch": 0.5550122249388753,
+      "grad_norm": 0.6273881793022156,
+      "learning_rate": 0.001,
+      "loss": 2.1867,
+      "step": 454
+    },
+    {
+      "epoch": 0.5562347188264058,
+      "grad_norm": 0.5088536143302917,
+      "learning_rate": 0.001,
+      "loss": 2.1296,
+      "step": 455
+    },
+    {
+      "epoch": 0.5574572127139364,
+      "grad_norm": 0.5127514600753784,
+      "learning_rate": 0.001,
+      "loss": 2.2092,
+      "step": 456
+    },
+    {
+      "epoch": 0.558679706601467,
+      "grad_norm": 0.6852807402610779,
+      "learning_rate": 0.001,
+      "loss": 2.2453,
+      "step": 457
+    },
+    {
+      "epoch": 0.5599022004889975,
+      "grad_norm": 0.5020545125007629,
+      "learning_rate": 0.001,
+      "loss": 2.2399,
+      "step": 458
+    },
+    {
+      "epoch": 0.5611246943765281,
+      "grad_norm": 0.5110552906990051,
+      "learning_rate": 0.001,
+      "loss": 2.1945,
+      "step": 459
+    },
+    {
+      "epoch": 0.5623471882640587,
+      "grad_norm": 0.6412564516067505,
+      "learning_rate": 0.001,
+      "loss": 2.1798,
+      "step": 460
+    },
+    {
+      "epoch": 0.5635696821515892,
+      "grad_norm": 0.4610752463340759,
+      "learning_rate": 0.001,
+      "loss": 2.1207,
+      "step": 461
+    },
+    {
+      "epoch": 0.5647921760391198,
+      "grad_norm": 0.6350216865539551,
+      "learning_rate": 0.001,
+      "loss": 2.3013,
+      "step": 462
+    },
+    {
+      "epoch": 0.5660146699266504,
+      "grad_norm": 0.5525195002555847,
+      "learning_rate": 0.001,
+      "loss": 2.1299,
+      "step": 463
+    },
+    {
+      "epoch": 0.5672371638141809,
+      "grad_norm": 0.5612464547157288,
+      "learning_rate": 0.001,
+      "loss": 2.2332,
+      "step": 464
+    },
+    {
+      "epoch": 0.5684596577017115,
+      "grad_norm": 0.5840879678726196,
+      "learning_rate": 0.001,
+      "loss": 2.259,
+      "step": 465
+    },
+    {
+      "epoch": 0.5696821515892421,
+      "grad_norm": 0.558628261089325,
+      "learning_rate": 0.001,
+      "loss": 2.1456,
+      "step": 466
+    },
+    {
+      "epoch": 0.5709046454767727,
+      "grad_norm": 0.5693490505218506,
+      "learning_rate": 0.001,
+      "loss": 2.195,
+      "step": 467
+    },
+    {
+      "epoch": 0.5721271393643031,
+      "grad_norm": 0.48814135789871216,
+      "learning_rate": 0.001,
+      "loss": 2.1886,
+      "step": 468
+    },
+    {
+      "epoch": 0.5733496332518337,
+      "grad_norm": 0.528095006942749,
+      "learning_rate": 0.001,
+      "loss": 2.2137,
+      "step": 469
+    },
+    {
+      "epoch": 0.5745721271393643,
+      "grad_norm": 0.5991528034210205,
+      "learning_rate": 0.001,
+      "loss": 2.1803,
+      "step": 470
+    },
+    {
+      "epoch": 0.5757946210268948,
+      "grad_norm": 0.5528990626335144,
+      "learning_rate": 0.001,
+      "loss": 2.1444,
+      "step": 471
+    },
+    {
+      "epoch": 0.5770171149144254,
+      "grad_norm": 0.5375529527664185,
+      "learning_rate": 0.001,
+      "loss": 2.1947,
+      "step": 472
+    },
+    {
+      "epoch": 0.578239608801956,
+      "grad_norm": 0.47309327125549316,
+      "learning_rate": 0.001,
+      "loss": 2.1808,
+      "step": 473
+    },
+    {
+      "epoch": 0.5794621026894865,
+      "grad_norm": 0.49250733852386475,
+      "learning_rate": 0.001,
+      "loss": 2.2268,
+      "step": 474
+    },
+    {
+      "epoch": 0.5806845965770171,
+      "grad_norm": 0.5195334553718567,
+      "learning_rate": 0.001,
+      "loss": 2.1929,
+      "step": 475
+    },
+    {
+      "epoch": 0.5819070904645477,
+      "grad_norm": 0.48456084728240967,
+      "learning_rate": 0.001,
+      "loss": 2.1217,
+      "step": 476
+    },
+    {
+      "epoch": 0.5831295843520783,
+      "grad_norm": 0.534269392490387,
+      "learning_rate": 0.001,
+      "loss": 2.1721,
+      "step": 477
+    },
+    {
+      "epoch": 0.5843520782396088,
+      "grad_norm": 0.5373458862304688,
+      "learning_rate": 0.001,
+      "loss": 2.3162,
+      "step": 478
+    },
+    {
+      "epoch": 0.5855745721271394,
+      "grad_norm": 0.5691243410110474,
+      "learning_rate": 0.001,
+      "loss": 2.215,
+      "step": 479
+    },
+    {
+      "epoch": 0.58679706601467,
+      "grad_norm": 0.5426737666130066,
+      "learning_rate": 0.001,
+      "loss": 2.1483,
+      "step": 480
+    },
+    {
+      "epoch": 0.5880195599022005,
+      "grad_norm": 0.5185466408729553,
+      "learning_rate": 0.001,
+      "loss": 2.1977,
+      "step": 481
+    },
+    {
+      "epoch": 0.589242053789731,
+      "grad_norm": 0.49309492111206055,
+      "learning_rate": 0.001,
+      "loss": 2.1648,
+      "step": 482
+    },
+    {
+      "epoch": 0.5904645476772616,
+      "grad_norm": 0.4933290183544159,
+      "learning_rate": 0.001,
+      "loss": 2.2618,
+      "step": 483
+    },
+    {
+      "epoch": 0.5916870415647921,
+      "grad_norm": 0.44854751229286194,
+      "learning_rate": 0.001,
+      "loss": 2.1669,
+      "step": 484
+    },
+    {
+      "epoch": 0.5929095354523227,
+      "grad_norm": 0.580627977848053,
+      "learning_rate": 0.001,
+      "loss": 2.212,
+      "step": 485
+    },
+    {
+      "epoch": 0.5941320293398533,
+      "grad_norm": 0.6000897884368896,
+      "learning_rate": 0.001,
+      "loss": 2.1923,
+      "step": 486
+    },
+    {
+      "epoch": 0.5953545232273839,
+      "grad_norm": 0.5180191397666931,
+      "learning_rate": 0.001,
+      "loss": 2.1981,
+      "step": 487
+    },
+    {
+      "epoch": 0.5965770171149144,
+      "grad_norm": 0.6499658823013306,
+      "learning_rate": 0.001,
+      "loss": 2.1392,
+      "step": 488
+    },
+    {
+      "epoch": 0.597799511002445,
+      "grad_norm": 0.5067393183708191,
+      "learning_rate": 0.001,
+      "loss": 2.2073,
+      "step": 489
+    },
+    {
+      "epoch": 0.5990220048899756,
+      "grad_norm": 0.5518601536750793,
+      "learning_rate": 0.001,
+      "loss": 2.1892,
+      "step": 490
+    },
+    {
+      "epoch": 0.6002444987775061,
+      "grad_norm": 0.5741767287254333,
+      "learning_rate": 0.001,
+      "loss": 2.2325,
+      "step": 491
+    },
+    {
+      "epoch": 0.6014669926650367,
+      "grad_norm": 0.47276920080184937,
+      "learning_rate": 0.001,
+      "loss": 2.1709,
+      "step": 492
+    },
+    {
+      "epoch": 0.6026894865525673,
+      "grad_norm": 0.6871250867843628,
+      "learning_rate": 0.001,
+      "loss": 2.2172,
+      "step": 493
+    },
+    {
+      "epoch": 0.6039119804400978,
+      "grad_norm": 0.6001262664794922,
+      "learning_rate": 0.001,
+      "loss": 2.1843,
+      "step": 494
+    },
+    {
+      "epoch": 0.6051344743276283,
+      "grad_norm": 0.6246949434280396,
+      "learning_rate": 0.001,
+      "loss": 2.2578,
+      "step": 495
+    },
+    {
+      "epoch": 0.6063569682151589,
+      "grad_norm": 0.599838137626648,
+      "learning_rate": 0.001,
+      "loss": 2.2204,
+      "step": 496
+    },
+    {
+      "epoch": 0.6075794621026895,
+      "grad_norm": 0.6918209195137024,
+      "learning_rate": 0.001,
+      "loss": 2.2305,
+      "step": 497
+    },
+    {
+      "epoch": 0.60880195599022,
+      "grad_norm": 0.5067518949508667,
+      "learning_rate": 0.001,
+      "loss": 2.2096,
+      "step": 498
+    },
+    {
+      "epoch": 0.6100244498777506,
+      "grad_norm": 0.6585003733634949,
+      "learning_rate": 0.001,
+      "loss": 2.1348,
+      "step": 499
+    },
+    {
+      "epoch": 0.6112469437652812,
+      "grad_norm": 0.5794422626495361,
+      "learning_rate": 0.001,
+      "loss": 2.1785,
+      "step": 500
+    },
+    {
+      "epoch": 0.6124694376528117,
+      "grad_norm": 0.5732021927833557,
+      "learning_rate": 0.001,
+      "loss": 2.1643,
+      "step": 501
+    },
+    {
+      "epoch": 0.6136919315403423,
+      "grad_norm": 0.5366373062133789,
+      "learning_rate": 0.001,
+      "loss": 2.1448,
+      "step": 502
+    },
+    {
+      "epoch": 0.6149144254278729,
+      "grad_norm": 0.5503267049789429,
+      "learning_rate": 0.001,
+      "loss": 2.1848,
+      "step": 503
+    },
+    {
+      "epoch": 0.6161369193154034,
+      "grad_norm": 0.5450462102890015,
+      "learning_rate": 0.001,
+      "loss": 2.1748,
+      "step": 504
+    },
+    {
+      "epoch": 0.617359413202934,
+      "grad_norm": 0.5018944144248962,
+      "learning_rate": 0.001,
+      "loss": 2.2055,
+      "step": 505
+    },
+    {
+      "epoch": 0.6185819070904646,
+      "grad_norm": 0.6032975316047668,
+      "learning_rate": 0.001,
+      "loss": 2.1492,
+      "step": 506
+    },
+    {
+      "epoch": 0.6198044009779952,
+      "grad_norm": 0.5328817367553711,
+      "learning_rate": 0.001,
+      "loss": 2.2036,
+      "step": 507
+    },
+    {
+      "epoch": 0.6210268948655256,
+      "grad_norm": 0.5828500986099243,
+      "learning_rate": 0.001,
+      "loss": 2.1992,
+      "step": 508
+    },
+    {
+      "epoch": 0.6222493887530562,
+      "grad_norm": 0.5160496234893799,
+      "learning_rate": 0.001,
+      "loss": 2.1988,
+      "step": 509
+    },
+    {
+      "epoch": 0.6234718826405868,
+      "grad_norm": 0.6087934374809265,
+      "learning_rate": 0.001,
+      "loss": 2.1834,
+      "step": 510
+    },
+    {
+      "epoch": 0.6246943765281173,
+      "grad_norm": 0.5729974508285522,
+      "learning_rate": 0.001,
+      "loss": 2.1361,
+      "step": 511
+    },
+    {
+      "epoch": 0.6259168704156479,
+      "grad_norm": 0.5637314319610596,
+      "learning_rate": 0.001,
+      "loss": 2.2719,
+      "step": 512
+    },
+    {
+      "epoch": 0.6271393643031785,
+      "grad_norm": 0.7113996744155884,
+      "learning_rate": 0.001,
+      "loss": 2.2268,
+      "step": 513
+    },
+    {
+      "epoch": 0.628361858190709,
+      "grad_norm": 0.5220823884010315,
+      "learning_rate": 0.001,
+      "loss": 2.2301,
+      "step": 514
+    },
+    {
+      "epoch": 0.6295843520782396,
+      "grad_norm": 0.5641618371009827,
+      "learning_rate": 0.001,
+      "loss": 2.1943,
+      "step": 515
+    },
+    {
+      "epoch": 0.6308068459657702,
+      "grad_norm": 0.587079644203186,
+      "learning_rate": 0.001,
+      "loss": 2.1861,
+      "step": 516
+    },
+    {
+      "epoch": 0.6320293398533008,
+      "grad_norm": 0.7265164256095886,
+      "learning_rate": 0.001,
+      "loss": 2.1136,
+      "step": 517
+    },
+    {
+      "epoch": 0.6332518337408313,
+      "grad_norm": 0.5387188196182251,
+      "learning_rate": 0.001,
+      "loss": 2.1507,
+      "step": 518
+    },
+    {
+      "epoch": 0.6344743276283619,
+      "grad_norm": 0.7700957655906677,
+      "learning_rate": 0.001,
+      "loss": 2.2727,
+      "step": 519
+    },
+    {
+      "epoch": 0.6356968215158925,
+      "grad_norm": 0.526703953742981,
+      "learning_rate": 0.001,
+      "loss": 2.1255,
+      "step": 520
+    },
+    {
+      "epoch": 0.636919315403423,
+      "grad_norm": 0.5161398649215698,
+      "learning_rate": 0.001,
+      "loss": 2.1244,
+      "step": 521
+    },
+    {
+      "epoch": 0.6381418092909535,
+      "grad_norm": 0.5797749757766724,
+      "learning_rate": 0.001,
+      "loss": 2.2473,
+      "step": 522
+    },
+    {
+      "epoch": 0.6393643031784841,
+      "grad_norm": 0.5566680431365967,
+      "learning_rate": 0.001,
+      "loss": 2.1835,
+      "step": 523
+    },
+    {
+      "epoch": 0.6405867970660146,
+      "grad_norm": 0.5569880604743958,
+      "learning_rate": 0.001,
+      "loss": 2.155,
+      "step": 524
+    },
+    {
+      "epoch": 0.6418092909535452,
+      "grad_norm": 0.5413252711296082,
+      "learning_rate": 0.001,
+      "loss": 2.2778,
+      "step": 525
+    },
+    {
+      "epoch": 0.6430317848410758,
+      "grad_norm": 0.4530647397041321,
+      "learning_rate": 0.001,
+      "loss": 2.1888,
+      "step": 526
+    },
+    {
+      "epoch": 0.6442542787286064,
+      "grad_norm": 0.5370563864707947,
+      "learning_rate": 0.001,
+      "loss": 2.242,
+      "step": 527
+    },
+    {
+      "epoch": 0.6454767726161369,
+      "grad_norm": 0.5274892449378967,
+      "learning_rate": 0.001,
+      "loss": 2.2493,
+      "step": 528
+    },
+    {
+      "epoch": 0.6466992665036675,
+      "grad_norm": 0.5889971852302551,
+      "learning_rate": 0.001,
+      "loss": 2.1055,
+      "step": 529
+    },
+    {
+      "epoch": 0.6479217603911981,
+      "grad_norm": 0.5334299206733704,
+      "learning_rate": 0.001,
+      "loss": 2.2133,
+      "step": 530
+    },
+    {
+      "epoch": 0.6491442542787286,
+      "grad_norm": 0.5480680465698242,
+      "learning_rate": 0.001,
+      "loss": 2.1973,
+      "step": 531
+    },
+    {
+      "epoch": 0.6503667481662592,
+      "grad_norm": 0.5218977332115173,
+      "learning_rate": 0.001,
+      "loss": 2.1938,
+      "step": 532
+    },
+    {
+      "epoch": 0.6515892420537898,
+      "grad_norm": 0.5179240107536316,
+      "learning_rate": 0.001,
+      "loss": 2.1262,
+      "step": 533
+    },
+    {
+      "epoch": 0.6528117359413202,
+      "grad_norm": 0.6605428457260132,
+      "learning_rate": 0.001,
+      "loss": 2.2826,
+      "step": 534
+    },
+    {
+      "epoch": 0.6540342298288508,
+      "grad_norm": 0.5191709995269775,
+      "learning_rate": 0.001,
+      "loss": 2.1649,
+      "step": 535
+    },
+    {
+      "epoch": 0.6552567237163814,
+      "grad_norm": 0.5000888109207153,
+      "learning_rate": 0.001,
+      "loss": 2.1345,
+      "step": 536
+    },
+    {
+      "epoch": 0.656479217603912,
+      "grad_norm": 0.5836233496665955,
+      "learning_rate": 0.001,
+      "loss": 2.1304,
+      "step": 537
+    },
+    {
+      "epoch": 0.6577017114914425,
+      "grad_norm": 0.672378420829773,
+      "learning_rate": 0.001,
+      "loss": 2.1928,
+      "step": 538
+    },
+    {
+      "epoch": 0.6589242053789731,
+      "grad_norm": 0.5583218932151794,
+      "learning_rate": 0.001,
+      "loss": 2.1173,
+      "step": 539
+    },
+    {
+      "epoch": 0.6601466992665037,
+      "grad_norm": 0.6046769618988037,
+      "learning_rate": 0.001,
+      "loss": 2.1138,
+      "step": 540
+    },
+    {
+      "epoch": 0.6613691931540342,
+      "grad_norm": 0.6253052353858948,
+      "learning_rate": 0.001,
+      "loss": 2.1529,
+      "step": 541
+    },
+    {
+      "epoch": 0.6625916870415648,
+      "grad_norm": 0.7345244884490967,
+      "learning_rate": 0.001,
+      "loss": 2.1945,
+      "step": 542
+    },
+    {
+      "epoch": 0.6638141809290954,
+      "grad_norm": 0.5053750276565552,
+      "learning_rate": 0.001,
+      "loss": 2.1233,
+      "step": 543
+    },
+    {
+      "epoch": 0.6650366748166259,
+      "grad_norm": 0.6003459095954895,
+      "learning_rate": 0.001,
+      "loss": 2.1655,
+      "step": 544
+    },
+    {
+      "epoch": 0.6662591687041565,
+      "grad_norm": 0.9038470387458801,
+      "learning_rate": 0.001,
+      "loss": 2.198,
+      "step": 545
+    },
+    {
+      "epoch": 0.6674816625916871,
+      "grad_norm": 0.6527109146118164,
+      "learning_rate": 0.001,
+      "loss": 2.2085,
+      "step": 546
+    },
+    {
+      "epoch": 0.6687041564792175,
+      "grad_norm": 0.6161382794380188,
+      "learning_rate": 0.001,
+      "loss": 2.0986,
+      "step": 547
+    },
+    {
+      "epoch": 0.6699266503667481,
+      "grad_norm": 0.5980858206748962,
+      "learning_rate": 0.001,
+      "loss": 2.2287,
+      "step": 548
+    },
+    {
+      "epoch": 0.6711491442542787,
+      "grad_norm": 0.5564101934432983,
+      "learning_rate": 0.001,
+      "loss": 2.2339,
+      "step": 549
+    },
+    {
+      "epoch": 0.6723716381418093,
+      "grad_norm": 0.5963722467422485,
+      "learning_rate": 0.001,
+      "loss": 2.1952,
+      "step": 550
+    },
+    {
+      "epoch": 0.6735941320293398,
+      "grad_norm": 0.5184721350669861,
+      "learning_rate": 0.001,
+      "loss": 2.1839,
+      "step": 551
+    },
+    {
+      "epoch": 0.6748166259168704,
+      "grad_norm": 0.5719816088676453,
+      "learning_rate": 0.001,
+      "loss": 2.1343,
+      "step": 552
+    },
+    {
+      "epoch": 0.676039119804401,
+      "grad_norm": 0.49916115403175354,
+      "learning_rate": 0.001,
+      "loss": 2.2026,
+      "step": 553
+    },
+    {
+      "epoch": 0.6772616136919315,
+      "grad_norm": 0.7603865265846252,
+      "learning_rate": 0.001,
+      "loss": 2.1293,
+      "step": 554
+    },
+    {
+      "epoch": 0.6784841075794621,
+      "grad_norm": 0.5817961096763611,
+      "learning_rate": 0.001,
+      "loss": 2.3228,
+      "step": 555
+    },
+    {
+      "epoch": 0.6797066014669927,
+      "grad_norm": 0.6436595916748047,
+      "learning_rate": 0.001,
+      "loss": 2.1515,
+      "step": 556
+    },
+    {
+      "epoch": 0.6809290953545232,
+      "grad_norm": 0.600204586982727,
+      "learning_rate": 0.001,
+      "loss": 2.2553,
+      "step": 557
+    },
+    {
+      "epoch": 0.6821515892420538,
+      "grad_norm": 0.5750772953033447,
+      "learning_rate": 0.001,
+      "loss": 2.2018,
+      "step": 558
+    },
+    {
+      "epoch": 0.6833740831295844,
+      "grad_norm": 0.6340410709381104,
+      "learning_rate": 0.001,
+      "loss": 2.2321,
+      "step": 559
+    },
+    {
+      "epoch": 0.684596577017115,
+      "grad_norm": 0.5944398641586304,
+      "learning_rate": 0.001,
+      "loss": 2.1864,
+      "step": 560
+    },
+    {
+      "epoch": 0.6858190709046454,
+      "grad_norm": 0.5228376984596252,
+      "learning_rate": 0.001,
+      "loss": 2.2398,
+      "step": 561
+    },
+    {
+      "epoch": 0.687041564792176,
+      "grad_norm": 0.7118151187896729,
+      "learning_rate": 0.001,
+      "loss": 2.3087,
+      "step": 562
+    },
+    {
+      "epoch": 0.6882640586797066,
+      "grad_norm": 0.6326483488082886,
+      "learning_rate": 0.001,
+      "loss": 2.1648,
+      "step": 563
+    },
+    {
+      "epoch": 0.6894865525672371,
+      "grad_norm": 0.6230838894844055,
+      "learning_rate": 0.001,
+      "loss": 2.1223,
+      "step": 564
+    },
+    {
+      "epoch": 0.6907090464547677,
+      "grad_norm": 0.5977362990379333,
+      "learning_rate": 0.001,
+      "loss": 2.2783,
+      "step": 565
+    },
+    {
+      "epoch": 0.6919315403422983,
+      "grad_norm": 0.5887629389762878,
+      "learning_rate": 0.001,
+      "loss": 2.03,
+      "step": 566
+    },
+    {
+      "epoch": 0.6931540342298288,
+      "grad_norm": 0.5934235453605652,
+      "learning_rate": 0.001,
+      "loss": 2.228,
+      "step": 567
+    },
+    {
+      "epoch": 0.6943765281173594,
+      "grad_norm": 0.5208036303520203,
+      "learning_rate": 0.001,
+      "loss": 2.3098,
+      "step": 568
+    },
+    {
+      "epoch": 0.69559902200489,
+      "grad_norm": 0.6961516737937927,
+      "learning_rate": 0.001,
+      "loss": 2.1223,
+      "step": 569
+    },
+    {
+      "epoch": 0.6968215158924206,
+      "grad_norm": 0.6044623255729675,
+      "learning_rate": 0.001,
+      "loss": 2.2311,
+      "step": 570
+    },
+    {
+      "epoch": 0.6980440097799511,
+      "grad_norm": 0.5864453911781311,
+      "learning_rate": 0.001,
+      "loss": 2.2621,
+      "step": 571
+    },
+    {
+      "epoch": 0.6992665036674817,
+      "grad_norm": 0.6569441556930542,
+      "learning_rate": 0.001,
+      "loss": 2.3103,
+      "step": 572
+    },
+    {
+      "epoch": 0.7004889975550123,
+      "grad_norm": 0.6718661189079285,
+      "learning_rate": 0.001,
+      "loss": 2.1022,
+      "step": 573
+    },
+    {
+      "epoch": 0.7017114914425427,
+      "grad_norm": 0.6719921231269836,
+      "learning_rate": 0.001,
+      "loss": 2.2599,
+      "step": 574
+    },
+    {
+      "epoch": 0.7029339853300733,
+      "grad_norm": 0.5817815661430359,
+      "learning_rate": 0.001,
+      "loss": 2.1637,
+      "step": 575
+    },
+    {
+      "epoch": 0.7041564792176039,
+      "grad_norm": 0.648612380027771,
+      "learning_rate": 0.001,
+      "loss": 2.2681,
+      "step": 576
+    },
+    {
+      "epoch": 0.7053789731051344,
+      "grad_norm": 0.6213855147361755,
+      "learning_rate": 0.001,
+      "loss": 2.1772,
+      "step": 577
+    },
+    {
+      "epoch": 0.706601466992665,
+      "grad_norm": 0.6786682605743408,
+      "learning_rate": 0.001,
+      "loss": 2.1969,
+      "step": 578
+    },
+    {
+      "epoch": 0.7078239608801956,
+      "grad_norm": 0.6275045275688171,
+      "learning_rate": 0.001,
+      "loss": 2.259,
+      "step": 579
+    },
+    {
+      "epoch": 0.7090464547677262,
+      "grad_norm": 0.6425896286964417,
+      "learning_rate": 0.001,
+      "loss": 2.263,
+      "step": 580
+    },
+    {
+      "epoch": 0.7102689486552567,
+      "grad_norm": 0.6005223989486694,
+      "learning_rate": 0.001,
+      "loss": 2.2789,
+      "step": 581
+    },
+    {
+      "epoch": 0.7114914425427873,
+      "grad_norm": 0.8881091475486755,
+      "learning_rate": 0.001,
+      "loss": 2.0891,
+      "step": 582
+    },
+    {
+      "epoch": 0.7127139364303179,
+      "grad_norm": 0.511212170124054,
+      "learning_rate": 0.001,
+      "loss": 2.1824,
+      "step": 583
+    },
+    {
+      "epoch": 0.7139364303178484,
+      "grad_norm": 0.5520249009132385,
+      "learning_rate": 0.001,
+      "loss": 2.2404,
+      "step": 584
+    },
+    {
+      "epoch": 0.715158924205379,
+      "grad_norm": 0.5928168296813965,
+      "learning_rate": 0.001,
+      "loss": 2.2272,
+      "step": 585
+    },
+    {
+      "epoch": 0.7163814180929096,
+      "grad_norm": 0.5859750509262085,
+      "learning_rate": 0.001,
+      "loss": 2.1354,
+      "step": 586
+    },
+    {
+      "epoch": 0.71760391198044,
+      "grad_norm": 0.5234774947166443,
+      "learning_rate": 0.001,
+      "loss": 2.2706,
+      "step": 587
+    },
+    {
+      "epoch": 0.7188264058679706,
+      "grad_norm": 0.6386865973472595,
+      "learning_rate": 0.001,
+      "loss": 2.2589,
+      "step": 588
+    },
+    {
+      "epoch": 0.7200488997555012,
+      "grad_norm": 0.6570860147476196,
+      "learning_rate": 0.001,
+      "loss": 2.1486,
+      "step": 589
+    },
+    {
+      "epoch": 0.7212713936430318,
+      "grad_norm": 0.6303104162216187,
+      "learning_rate": 0.001,
+      "loss": 2.1699,
+      "step": 590
+    },
+    {
+      "epoch": 0.7224938875305623,
+      "grad_norm": 0.6067413091659546,
+      "learning_rate": 0.001,
+      "loss": 2.1678,
+      "step": 591
+    },
+    {
+      "epoch": 0.7237163814180929,
+      "grad_norm": 0.7316265106201172,
+      "learning_rate": 0.001,
+      "loss": 2.1259,
+      "step": 592
+    },
+    {
+      "epoch": 0.7249388753056235,
+      "grad_norm": 0.5220648050308228,
+      "learning_rate": 0.001,
+      "loss": 2.2057,
+      "step": 593
+    },
+    {
+      "epoch": 0.726161369193154,
+      "grad_norm": 0.5792797803878784,
+      "learning_rate": 0.001,
+      "loss": 2.1987,
+      "step": 594
+    },
+    {
+      "epoch": 0.7273838630806846,
+      "grad_norm": 0.885420024394989,
+      "learning_rate": 0.001,
+      "loss": 2.2314,
+      "step": 595
+    },
+    {
+      "epoch": 0.7286063569682152,
+      "grad_norm": 0.5722411274909973,
+      "learning_rate": 0.001,
+      "loss": 2.094,
+      "step": 596
+    },
+    {
+      "epoch": 0.7298288508557457,
+      "grad_norm": 0.5585153698921204,
+      "learning_rate": 0.001,
+      "loss": 2.2757,
+      "step": 597
+    },
+    {
+      "epoch": 0.7310513447432763,
+      "grad_norm": 0.5866743922233582,
+      "learning_rate": 0.001,
+      "loss": 2.2134,
+      "step": 598
+    },
+    {
+      "epoch": 0.7322738386308069,
+      "grad_norm": 0.5329750180244446,
+      "learning_rate": 0.001,
+      "loss": 2.2929,
+      "step": 599
+    },
+    {
+      "epoch": 0.7334963325183375,
+      "grad_norm": 0.5793975591659546,
+      "learning_rate": 0.001,
+      "loss": 2.1822,
+      "step": 600
+    },
+    {
+      "epoch": 0.7347188264058679,
+      "grad_norm": 0.621846616268158,
+      "learning_rate": 0.001,
+      "loss": 2.2078,
+      "step": 601
+    },
+    {
+      "epoch": 0.7359413202933985,
+      "grad_norm": 0.6965752840042114,
+      "learning_rate": 0.001,
+      "loss": 2.1822,
+      "step": 602
+    },
+    {
+      "epoch": 0.7371638141809291,
+      "grad_norm": 0.626397430896759,
+      "learning_rate": 0.001,
+      "loss": 2.21,
+      "step": 603
+    },
+    {
+      "epoch": 0.7383863080684596,
+      "grad_norm": 0.8464508056640625,
+      "learning_rate": 0.001,
+      "loss": 2.18,
+      "step": 604
+    },
+    {
+      "epoch": 0.7396088019559902,
+      "grad_norm": 0.5484370589256287,
+      "learning_rate": 0.001,
+      "loss": 2.12,
+      "step": 605
+    },
+    {
+      "epoch": 0.7408312958435208,
+      "grad_norm": 0.7459133267402649,
+      "learning_rate": 0.001,
+      "loss": 2.1339,
+      "step": 606
+    },
+    {
+      "epoch": 0.7420537897310513,
+      "grad_norm": 0.7169046998023987,
+      "learning_rate": 0.001,
+      "loss": 2.296,
+      "step": 607
+    },
+    {
+      "epoch": 0.7432762836185819,
+      "grad_norm": 0.5717725157737732,
+      "learning_rate": 0.001,
+      "loss": 2.163,
+      "step": 608
+    },
+    {
+      "epoch": 0.7444987775061125,
+      "grad_norm": 0.6337913274765015,
+      "learning_rate": 0.001,
+      "loss": 2.1702,
+      "step": 609
+    },
+    {
+      "epoch": 0.7457212713936431,
+      "grad_norm": 0.6826694011688232,
+      "learning_rate": 0.001,
+      "loss": 2.2204,
+      "step": 610
+    },
+    {
+      "epoch": 0.7469437652811736,
+      "grad_norm": 0.6060184836387634,
+      "learning_rate": 0.001,
+      "loss": 2.1545,
+      "step": 611
+    },
+    {
+      "epoch": 0.7481662591687042,
+      "grad_norm": 0.6665242910385132,
+      "learning_rate": 0.001,
+      "loss": 2.2457,
+      "step": 612
+    },
+    {
+      "epoch": 0.7493887530562348,
+      "grad_norm": 0.5639733076095581,
+      "learning_rate": 0.001,
+      "loss": 2.1641,
+      "step": 613
+    },
+    {
+      "epoch": 0.7506112469437652,
+      "grad_norm": 0.6882290840148926,
+      "learning_rate": 0.001,
+      "loss": 2.1655,
+      "step": 614
+    },
+    {
+      "epoch": 0.7518337408312958,
+      "grad_norm": 0.5488794445991516,
+      "learning_rate": 0.001,
+      "loss": 2.2228,
+      "step": 615
+    },
+    {
+      "epoch": 0.7518337408312958,
+      "eval_loss": 2.115697145462036,
+      "eval_runtime": 70.7202,
+      "eval_samples_per_second": 147.978,
+      "eval_steps_per_second": 18.51,
+      "step": 615
+    },
+    {
+      "epoch": 0.7530562347188264,
+      "grad_norm": 0.5853757262229919,
+      "learning_rate": 0.001,
+      "loss": 2.1722,
+      "step": 616
+    },
+    {
+      "epoch": 0.7542787286063569,
+      "grad_norm": 0.6095851063728333,
+      "learning_rate": 0.001,
+      "loss": 2.1482,
+      "step": 617
+    },
+    {
+      "epoch": 0.7555012224938875,
+      "grad_norm": 0.7331103682518005,
+      "learning_rate": 0.001,
+      "loss": 2.2136,
+      "step": 618
+    },
+    {
+      "epoch": 0.7567237163814181,
+      "grad_norm": 0.5856543779373169,
+      "learning_rate": 0.001,
+      "loss": 2.2382,
+      "step": 619
+    },
+    {
+      "epoch": 0.7579462102689487,
+      "grad_norm": 0.8980780839920044,
+      "learning_rate": 0.001,
+      "loss": 2.1312,
+      "step": 620
+    },
+    {
+      "epoch": 0.7591687041564792,
+      "grad_norm": 0.5309576988220215,
+      "learning_rate": 0.001,
+      "loss": 2.1447,
+      "step": 621
+    },
+    {
+      "epoch": 0.7603911980440098,
+      "grad_norm": 0.5389253497123718,
+      "learning_rate": 0.001,
+      "loss": 2.1772,
+      "step": 622
+    },
+    {
+      "epoch": 0.7616136919315404,
+      "grad_norm": 0.6142221093177795,
+      "learning_rate": 0.001,
+      "loss": 2.1545,
+      "step": 623
+    },
+    {
+      "epoch": 0.7628361858190709,
+      "grad_norm": 0.6199280023574829,
+      "learning_rate": 0.001,
+      "loss": 2.1463,
+      "step": 624
+    },
+    {
+      "epoch": 0.7640586797066015,
+      "grad_norm": 0.6900845170021057,
+      "learning_rate": 0.001,
+      "loss": 2.1927,
+      "step": 625
+    },
+    {
+      "epoch": 0.7652811735941321,
+      "grad_norm": 0.4952549338340759,
+      "learning_rate": 0.001,
+      "loss": 2.2629,
+      "step": 626
+    },
+    {
+      "epoch": 0.7665036674816625,
+      "grad_norm": 0.7719544172286987,
+      "learning_rate": 0.001,
+      "loss": 2.2246,
+      "step": 627
+    },
+    {
+      "epoch": 0.7677261613691931,
+      "grad_norm": 0.6267576813697815,
+      "learning_rate": 0.001,
+      "loss": 2.1653,
+      "step": 628
+    },
+    {
+      "epoch": 0.7689486552567237,
+      "grad_norm": 0.5654335618019104,
+      "learning_rate": 0.001,
+      "loss": 2.2629,
+      "step": 629
+    },
+    {
+      "epoch": 0.7701711491442543,
+      "grad_norm": 0.5660375952720642,
+      "learning_rate": 0.001,
+      "loss": 2.1891,
+      "step": 630
+    },
+    {
+      "epoch": 0.7713936430317848,
+      "grad_norm": 0.5151835083961487,
+      "learning_rate": 0.001,
+      "loss": 2.1307,
+      "step": 631
+    },
+    {
+      "epoch": 0.7726161369193154,
+      "grad_norm": 0.615497350692749,
+      "learning_rate": 0.001,
+      "loss": 2.2797,
+      "step": 632
+    },
+    {
+      "epoch": 0.773838630806846,
+      "grad_norm": 0.627841055393219,
+      "learning_rate": 0.001,
+      "loss": 2.2369,
+      "step": 633
+    },
+    {
+      "epoch": 0.7750611246943765,
+      "grad_norm": 0.5649968981742859,
+      "learning_rate": 0.001,
+      "loss": 2.1355,
+      "step": 634
+    },
+    {
+      "epoch": 0.7762836185819071,
+      "grad_norm": 0.6311205625534058,
+      "learning_rate": 0.001,
+      "loss": 2.2487,
+      "step": 635
+    },
+    {
+      "epoch": 0.7775061124694377,
+      "grad_norm": 0.6173393130302429,
+      "learning_rate": 0.001,
+      "loss": 2.2116,
+      "step": 636
+    },
+    {
+      "epoch": 0.7787286063569682,
+      "grad_norm": 0.706934928894043,
+      "learning_rate": 0.001,
+      "loss": 2.1385,
+      "step": 637
+    },
+    {
+      "epoch": 0.7799511002444988,
+      "grad_norm": 0.6243404746055603,
+      "learning_rate": 0.001,
+      "loss": 2.2883,
+      "step": 638
+    },
+    {
+      "epoch": 0.7811735941320294,
+      "grad_norm": 0.6092289686203003,
+      "learning_rate": 0.001,
+      "loss": 2.248,
+      "step": 639
+    },
+    {
+      "epoch": 0.78239608801956,
+      "grad_norm": 0.6141385436058044,
+      "learning_rate": 0.001,
+      "loss": 2.3563,
+      "step": 640
+    },
+    {
+      "epoch": 0.7836185819070904,
+      "grad_norm": 0.5421451926231384,
+      "learning_rate": 0.001,
+      "loss": 2.2518,
+      "step": 641
+    },
+    {
+      "epoch": 0.784841075794621,
+      "grad_norm": 0.5878483653068542,
+      "learning_rate": 0.001,
+      "loss": 2.1793,
+      "step": 642
+    },
+    {
+      "epoch": 0.7860635696821516,
+      "grad_norm": 0.5330148339271545,
+      "learning_rate": 0.001,
+      "loss": 2.2476,
+      "step": 643
+    },
+    {
+      "epoch": 0.7872860635696821,
+      "grad_norm": 0.5393016934394836,
+      "learning_rate": 0.001,
+      "loss": 2.2135,
+      "step": 644
+    },
+    {
+      "epoch": 0.7885085574572127,
+      "grad_norm": 0.5908885598182678,
+      "learning_rate": 0.001,
+      "loss": 2.0988,
+      "step": 645
+    },
+    {
+      "epoch": 0.7897310513447433,
+      "grad_norm": 0.5079434514045715,
+      "learning_rate": 0.001,
+      "loss": 2.1475,
+      "step": 646
+    },
+    {
+      "epoch": 0.7909535452322738,
+      "grad_norm": 0.4999438226222992,
+      "learning_rate": 0.001,
+      "loss": 2.1897,
+      "step": 647
+    },
+    {
+      "epoch": 0.7921760391198044,
+      "grad_norm": 0.5183396339416504,
+      "learning_rate": 0.001,
+      "loss": 2.2245,
+      "step": 648
+    },
+    {
+      "epoch": 0.793398533007335,
+      "grad_norm": 0.5283822417259216,
+      "learning_rate": 0.001,
+      "loss": 2.238,
+      "step": 649
+    },
+    {
+      "epoch": 0.7946210268948656,
+      "grad_norm": 0.6713541150093079,
+      "learning_rate": 0.001,
+      "loss": 2.1341,
+      "step": 650
+    },
+    {
+      "epoch": 0.7958435207823961,
+      "grad_norm": 0.6195135712623596,
+      "learning_rate": 0.001,
+      "loss": 2.2756,
+      "step": 651
+    },
+    {
+      "epoch": 0.7970660146699267,
+      "grad_norm": 1.1329959630966187,
+      "learning_rate": 0.001,
+      "loss": 2.1616,
+      "step": 652
+    },
+    {
+      "epoch": 0.7982885085574573,
+      "grad_norm": 0.6215202212333679,
+      "learning_rate": 0.001,
+      "loss": 2.1917,
+      "step": 653
+    },
+    {
+      "epoch": 0.7995110024449877,
+      "grad_norm": 0.9555675983428955,
+      "learning_rate": 0.001,
+      "loss": 2.1186,
+      "step": 654
+    },
+    {
+      "epoch": 0.8007334963325183,
+      "grad_norm": 0.6120454668998718,
+      "learning_rate": 0.001,
+      "loss": 2.1704,
+      "step": 655
+    },
+    {
+      "epoch": 0.8019559902200489,
+      "grad_norm": 0.7060715556144714,
+      "learning_rate": 0.001,
+      "loss": 2.2127,
+      "step": 656
+    },
+    {
+      "epoch": 0.8031784841075794,
+      "grad_norm": 0.6769619584083557,
+      "learning_rate": 0.001,
+      "loss": 2.2006,
+      "step": 657
+    },
+    {
+      "epoch": 0.80440097799511,
+      "grad_norm": 0.5458390712738037,
+      "learning_rate": 0.001,
+      "loss": 2.2003,
+      "step": 658
+    },
+    {
+      "epoch": 0.8056234718826406,
+      "grad_norm": 0.6172414422035217,
+      "learning_rate": 0.001,
+      "loss": 2.221,
+      "step": 659
+    },
+    {
+      "epoch": 0.8068459657701712,
+      "grad_norm": 0.8861773014068604,
+      "learning_rate": 0.001,
+      "loss": 2.1507,
+      "step": 660
+    },
+    {
+      "epoch": 0.8080684596577017,
+      "grad_norm": 0.5643443465232849,
+      "learning_rate": 0.001,
+      "loss": 2.1747,
+      "step": 661
+    },
+    {
+      "epoch": 0.8092909535452323,
+      "grad_norm": 0.6740103363990784,
+      "learning_rate": 0.001,
+      "loss": 2.211,
+      "step": 662
+    },
+    {
+      "epoch": 0.8105134474327629,
+      "grad_norm": 0.6159763336181641,
+      "learning_rate": 0.001,
+      "loss": 2.311,
+      "step": 663
+    },
+    {
+      "epoch": 0.8117359413202934,
+      "grad_norm": 0.7396857142448425,
+      "learning_rate": 0.001,
+      "loss": 2.1574,
+      "step": 664
+    },
+    {
+      "epoch": 0.812958435207824,
+      "grad_norm": 0.7003642320632935,
+      "learning_rate": 0.001,
+      "loss": 2.1753,
+      "step": 665
+    },
+    {
+      "epoch": 0.8141809290953546,
+      "grad_norm": 0.735837996006012,
+      "learning_rate": 0.001,
+      "loss": 2.1408,
+      "step": 666
+    },
+    {
+      "epoch": 0.815403422982885,
+      "grad_norm": 0.6381567716598511,
+      "learning_rate": 0.001,
+      "loss": 2.1237,
+      "step": 667
+    },
+    {
+      "epoch": 0.8166259168704156,
+      "grad_norm": 0.6023222804069519,
+      "learning_rate": 0.001,
+      "loss": 2.2082,
+      "step": 668
+    },
+    {
+      "epoch": 0.8178484107579462,
+      "grad_norm": 0.7556263208389282,
+      "learning_rate": 0.001,
+      "loss": 2.1482,
+      "step": 669
+    },
+    {
+      "epoch": 0.8190709046454768,
+      "grad_norm": 0.685670018196106,
+      "learning_rate": 0.001,
+      "loss": 2.1409,
+      "step": 670
+    },
+    {
+      "epoch": 0.8202933985330073,
+      "grad_norm": 0.6988388895988464,
+      "learning_rate": 0.001,
+      "loss": 2.215,
+      "step": 671
+    },
+    {
+      "epoch": 0.8215158924205379,
+      "grad_norm": 0.7381739616394043,
+      "learning_rate": 0.001,
+      "loss": 2.1533,
+      "step": 672
+    },
+    {
+      "epoch": 0.8227383863080685,
+      "grad_norm": 0.6375918388366699,
+      "learning_rate": 0.001,
+      "loss": 2.1447,
+      "step": 673
+    },
+    {
+      "epoch": 0.823960880195599,
+      "grad_norm": 0.8125602006912231,
+      "learning_rate": 0.001,
+      "loss": 2.2065,
+      "step": 674
+    },
+    {
+      "epoch": 0.8251833740831296,
+      "grad_norm": 0.6340590119361877,
+      "learning_rate": 0.001,
+      "loss": 2.2653,
+      "step": 675
+    },
+    {
+      "epoch": 0.8264058679706602,
+      "grad_norm": 0.6410722732543945,
+      "learning_rate": 0.001,
+      "loss": 2.2131,
+      "step": 676
+    },
+    {
+      "epoch": 0.8276283618581907,
+      "grad_norm": 0.7203342318534851,
+      "learning_rate": 0.001,
+      "loss": 2.1969,
+      "step": 677
+    },
+    {
+      "epoch": 0.8288508557457213,
+      "grad_norm": 0.5189094543457031,
+      "learning_rate": 0.001,
+      "loss": 2.0506,
+      "step": 678
+    },
+    {
+      "epoch": 0.8300733496332519,
+      "grad_norm": 0.5186660885810852,
+      "learning_rate": 0.001,
+      "loss": 2.0897,
+      "step": 679
+    },
+    {
+      "epoch": 0.8312958435207825,
+      "grad_norm": 0.5656071901321411,
+      "learning_rate": 0.001,
+      "loss": 2.2073,
+      "step": 680
+    },
+    {
+      "epoch": 0.8325183374083129,
+      "grad_norm": 0.621357262134552,
+      "learning_rate": 0.001,
+      "loss": 2.3196,
+      "step": 681
+    },
+    {
+      "epoch": 0.8337408312958435,
+      "grad_norm": 0.6449140906333923,
+      "learning_rate": 0.001,
+      "loss": 2.1453,
+      "step": 682
+    },
+    {
+      "epoch": 0.8349633251833741,
+      "grad_norm": 0.5833041071891785,
+      "learning_rate": 0.001,
+      "loss": 2.1055,
+      "step": 683
+    },
+    {
+      "epoch": 0.8361858190709046,
+      "grad_norm": 0.5668090581893921,
+      "learning_rate": 0.001,
+      "loss": 2.2212,
+      "step": 684
+    },
+    {
+      "epoch": 0.8374083129584352,
+      "grad_norm": 0.611060619354248,
+      "learning_rate": 0.001,
+      "loss": 2.2337,
+      "step": 685
+    },
+    {
+      "epoch": 0.8386308068459658,
+      "grad_norm": 0.607295036315918,
+      "learning_rate": 0.001,
+      "loss": 2.1919,
+      "step": 686
+    },
+    {
+      "epoch": 0.8398533007334963,
+      "grad_norm": 0.6494394540786743,
+      "learning_rate": 0.001,
+      "loss": 2.1755,
+      "step": 687
+    },
+    {
+      "epoch": 0.8410757946210269,
+      "grad_norm": 0.650504469871521,
+      "learning_rate": 0.001,
+      "loss": 2.2382,
+      "step": 688
+    },
+    {
+      "epoch": 0.8422982885085575,
+      "grad_norm": 0.5527434349060059,
+      "learning_rate": 0.001,
+      "loss": 2.1191,
+      "step": 689
+    },
+    {
+      "epoch": 0.843520782396088,
+      "grad_norm": 0.6322658061981201,
+      "learning_rate": 0.001,
+      "loss": 2.1291,
+      "step": 690
+    },
+    {
+      "epoch": 0.8447432762836186,
+      "grad_norm": 0.6053141355514526,
+      "learning_rate": 0.001,
+      "loss": 2.169,
+      "step": 691
+    },
+    {
+      "epoch": 0.8459657701711492,
+      "grad_norm": 0.670330822467804,
+      "learning_rate": 0.001,
+      "loss": 2.1265,
+      "step": 692
+    },
+    {
+      "epoch": 0.8471882640586798,
+      "grad_norm": 0.810947835445404,
+      "learning_rate": 0.001,
+      "loss": 2.2218,
+      "step": 693
+    },
+    {
+      "epoch": 0.8484107579462102,
+      "grad_norm": 0.6541447639465332,
+      "learning_rate": 0.001,
+      "loss": 2.1557,
+      "step": 694
+    },
+    {
+      "epoch": 0.8496332518337408,
+      "grad_norm": 0.5765539407730103,
+      "learning_rate": 0.001,
+      "loss": 2.1568,
+      "step": 695
+    },
+    {
+      "epoch": 0.8508557457212714,
+      "grad_norm": 0.9142853617668152,
+      "learning_rate": 0.001,
+      "loss": 2.1001,
+      "step": 696
+    },
+    {
+      "epoch": 0.8520782396088019,
+      "grad_norm": 0.8350043296813965,
+      "learning_rate": 0.001,
+      "loss": 2.0955,
+      "step": 697
+    },
+    {
+      "epoch": 0.8533007334963325,
+      "grad_norm": 0.7408676743507385,
+      "learning_rate": 0.001,
+      "loss": 2.2104,
+      "step": 698
+    },
+    {
+      "epoch": 0.8545232273838631,
+      "grad_norm": 0.6313987970352173,
+      "learning_rate": 0.001,
+      "loss": 2.228,
+      "step": 699
+    },
+    {
+      "epoch": 0.8557457212713936,
+      "grad_norm": 0.503736138343811,
+      "learning_rate": 0.001,
+      "loss": 2.2701,
+      "step": 700
+    },
+    {
+      "epoch": 0.8569682151589242,
+      "grad_norm": 0.6567993760108948,
+      "learning_rate": 0.001,
+      "loss": 2.2163,
+      "step": 701
+    },
+    {
+      "epoch": 0.8581907090464548,
+      "grad_norm": 0.6446906328201294,
+      "learning_rate": 0.001,
+      "loss": 2.1723,
+      "step": 702
+    },
+    {
+      "epoch": 0.8594132029339854,
+      "grad_norm": 0.5486587882041931,
+      "learning_rate": 0.001,
+      "loss": 2.1876,
+      "step": 703
+    },
+    {
+      "epoch": 0.8606356968215159,
+      "grad_norm": 0.5563488602638245,
+      "learning_rate": 0.001,
+      "loss": 2.1077,
+      "step": 704
+    },
+    {
+      "epoch": 0.8618581907090465,
+      "grad_norm": 0.8015148043632507,
+      "learning_rate": 0.001,
+      "loss": 2.1492,
+      "step": 705
+    },
+    {
+      "epoch": 0.863080684596577,
+      "grad_norm": 0.5513556003570557,
+      "learning_rate": 0.001,
+      "loss": 2.207,
+      "step": 706
+    },
+    {
+      "epoch": 0.8643031784841075,
+      "grad_norm": 0.604444682598114,
+      "learning_rate": 0.001,
+      "loss": 2.2217,
+      "step": 707
+    },
+    {
+      "epoch": 0.8655256723716381,
+      "grad_norm": 0.5867708325386047,
+      "learning_rate": 0.001,
+      "loss": 2.236,
+      "step": 708
+    },
+    {
+      "epoch": 0.8667481662591687,
+      "grad_norm": 0.5340182781219482,
+      "learning_rate": 0.001,
+      "loss": 2.1848,
+      "step": 709
+    },
+    {
+      "epoch": 0.8679706601466992,
+      "grad_norm": 0.5035935044288635,
+      "learning_rate": 0.001,
+      "loss": 2.1546,
+      "step": 710
+    },
+    {
+      "epoch": 0.8691931540342298,
+      "grad_norm": 0.5513314604759216,
+      "learning_rate": 0.001,
+      "loss": 2.2459,
+      "step": 711
+    },
+    {
+      "epoch": 0.8704156479217604,
+      "grad_norm": 0.6878560781478882,
+      "learning_rate": 0.001,
+      "loss": 2.2302,
+      "step": 712
+    },
+    {
+      "epoch": 0.871638141809291,
+      "grad_norm": 0.5138952136039734,
+      "learning_rate": 0.001,
+      "loss": 2.186,
+      "step": 713
+    },
+    {
+      "epoch": 0.8728606356968215,
+      "grad_norm": 0.6376103162765503,
+      "learning_rate": 0.001,
+      "loss": 2.1625,
+      "step": 714
+    },
+    {
+      "epoch": 0.8740831295843521,
+      "grad_norm": 0.5404793620109558,
+      "learning_rate": 0.001,
+      "loss": 2.2442,
+      "step": 715
+    },
+    {
+      "epoch": 0.8753056234718827,
+      "grad_norm": 0.6397867798805237,
+      "learning_rate": 0.001,
+      "loss": 2.197,
+      "step": 716
+    },
+    {
+      "epoch": 0.8765281173594132,
+      "grad_norm": 0.5790246725082397,
+      "learning_rate": 0.001,
+      "loss": 2.1914,
+      "step": 717
+    },
+    {
+      "epoch": 0.8777506112469438,
+      "grad_norm": 0.5128389000892639,
+      "learning_rate": 0.001,
+      "loss": 2.2063,
+      "step": 718
+    },
+    {
+      "epoch": 0.8789731051344744,
+      "grad_norm": 0.5515012145042419,
+      "learning_rate": 0.001,
+      "loss": 2.1923,
+      "step": 719
+    },
+    {
+      "epoch": 0.8801955990220048,
+      "grad_norm": 0.5543470978736877,
+      "learning_rate": 0.001,
+      "loss": 2.2026,
+      "step": 720
+    },
+    {
+      "epoch": 0.8814180929095354,
+      "grad_norm": 0.7619017362594604,
+      "learning_rate": 0.001,
+      "loss": 2.2105,
+      "step": 721
+    },
+    {
+      "epoch": 0.882640586797066,
+      "grad_norm": 0.6313949227333069,
+      "learning_rate": 0.001,
+      "loss": 2.1459,
+      "step": 722
+    },
+    {
+      "epoch": 0.8838630806845966,
+      "grad_norm": 0.550726056098938,
+      "learning_rate": 0.001,
+      "loss": 2.3319,
+      "step": 723
+    },
+    {
+      "epoch": 0.8850855745721271,
+      "grad_norm": 0.7315700650215149,
+      "learning_rate": 0.001,
+      "loss": 2.2009,
+      "step": 724
+    },
+    {
+      "epoch": 0.8863080684596577,
+      "grad_norm": 0.705125629901886,
+      "learning_rate": 0.001,
+      "loss": 2.1458,
+      "step": 725
+    },
+    {
+      "epoch": 0.8875305623471883,
+      "grad_norm": 0.545971691608429,
+      "learning_rate": 0.001,
+      "loss": 2.161,
+      "step": 726
+    },
+    {
+      "epoch": 0.8887530562347188,
+      "grad_norm": 0.6012575030326843,
+      "learning_rate": 0.001,
+      "loss": 2.1907,
+      "step": 727
+    },
+    {
+      "epoch": 0.8899755501222494,
+      "grad_norm": 0.5882204174995422,
+      "learning_rate": 0.001,
+      "loss": 2.2835,
+      "step": 728
+    },
+    {
+      "epoch": 0.89119804400978,
+      "grad_norm": 0.5331866145133972,
+      "learning_rate": 0.001,
+      "loss": 2.2505,
+      "step": 729
+    },
+    {
+      "epoch": 0.8924205378973105,
+      "grad_norm": 0.5553421974182129,
+      "learning_rate": 0.001,
+      "loss": 2.1752,
+      "step": 730
+    },
+    {
+      "epoch": 0.8936430317848411,
+      "grad_norm": 0.576580822467804,
+      "learning_rate": 0.001,
+      "loss": 2.0457,
+      "step": 731
+    },
+    {
+      "epoch": 0.8948655256723717,
+      "grad_norm": 0.6258639693260193,
+      "learning_rate": 0.001,
+      "loss": 2.1565,
+      "step": 732
+    },
+    {
+      "epoch": 0.8960880195599022,
+      "grad_norm": 0.5421616435050964,
+      "learning_rate": 0.001,
+      "loss": 2.1701,
+      "step": 733
+    },
+    {
+      "epoch": 0.8973105134474327,
+      "grad_norm": 0.6916106939315796,
+      "learning_rate": 0.001,
+      "loss": 2.1443,
+      "step": 734
+    },
+    {
+      "epoch": 0.8985330073349633,
+      "grad_norm": 0.5956131219863892,
+      "learning_rate": 0.001,
+      "loss": 2.1897,
+      "step": 735
+    },
+    {
+      "epoch": 0.8997555012224939,
+      "grad_norm": 0.5249125361442566,
+      "learning_rate": 0.001,
+      "loss": 2.2231,
+      "step": 736
+    },
+    {
+      "epoch": 0.9009779951100244,
+      "grad_norm": 0.5568667650222778,
+      "learning_rate": 0.001,
+      "loss": 2.1758,
+      "step": 737
+    },
+    {
+      "epoch": 0.902200488997555,
+      "grad_norm": 0.5609161853790283,
+      "learning_rate": 0.001,
+      "loss": 2.1774,
+      "step": 738
+    },
+    {
+      "epoch": 0.9034229828850856,
+      "grad_norm": 0.5504893064498901,
+      "learning_rate": 0.001,
+      "loss": 2.1974,
+      "step": 739
+    },
+    {
+      "epoch": 0.9046454767726161,
+      "grad_norm": 0.526772141456604,
+      "learning_rate": 0.001,
+      "loss": 2.1867,
+      "step": 740
+    },
+    {
+      "epoch": 0.9058679706601467,
+      "grad_norm": 0.6314900517463684,
+      "learning_rate": 0.001,
+      "loss": 2.1958,
+      "step": 741
+    },
+    {
+      "epoch": 0.9070904645476773,
+      "grad_norm": 0.4864460229873657,
+      "learning_rate": 0.001,
+      "loss": 2.2753,
+      "step": 742
+    },
+    {
+      "epoch": 0.9083129584352079,
+      "grad_norm": 0.5816671848297119,
+      "learning_rate": 0.001,
+      "loss": 2.2385,
+      "step": 743
+    },
+    {
+      "epoch": 0.9095354523227384,
+      "grad_norm": 0.6227999329566956,
+      "learning_rate": 0.001,
+      "loss": 2.3065,
+      "step": 744
+    },
+    {
+      "epoch": 0.910757946210269,
+      "grad_norm": 0.5527008771896362,
+      "learning_rate": 0.001,
+      "loss": 2.184,
+      "step": 745
+    },
+    {
+      "epoch": 0.9119804400977995,
+      "grad_norm": 0.5377302169799805,
+      "learning_rate": 0.001,
+      "loss": 2.2083,
+      "step": 746
+    },
+    {
+      "epoch": 0.91320293398533,
+      "grad_norm": 0.7717643976211548,
+      "learning_rate": 0.001,
+      "loss": 2.2884,
+      "step": 747
+    },
+    {
+      "epoch": 0.9144254278728606,
+      "grad_norm": 0.5529236793518066,
+      "learning_rate": 0.001,
+      "loss": 2.1217,
+      "step": 748
+    },
+    {
+      "epoch": 0.9156479217603912,
+      "grad_norm": 0.557788074016571,
+      "learning_rate": 0.001,
+      "loss": 2.2013,
+      "step": 749
+    },
+    {
+      "epoch": 0.9168704156479217,
+      "grad_norm": 0.5905176401138306,
+      "learning_rate": 0.001,
+      "loss": 2.1523,
+      "step": 750
+    },
+    {
+      "epoch": 0.9180929095354523,
+      "grad_norm": 0.7020503878593445,
+      "learning_rate": 0.001,
+      "loss": 2.2644,
+      "step": 751
+    },
+    {
+      "epoch": 0.9193154034229829,
+      "grad_norm": 0.6087464094161987,
+      "learning_rate": 0.001,
+      "loss": 2.164,
+      "step": 752
+    },
+    {
+      "epoch": 0.9205378973105135,
+      "grad_norm": 0.7976818680763245,
+      "learning_rate": 0.001,
+      "loss": 2.2657,
+      "step": 753
+    },
+    {
+      "epoch": 0.921760391198044,
+      "grad_norm": 0.5396186709403992,
+      "learning_rate": 0.001,
+      "loss": 2.205,
+      "step": 754
+    },
+    {
+      "epoch": 0.9229828850855746,
+      "grad_norm": 0.7185906171798706,
+      "learning_rate": 0.001,
+      "loss": 2.1775,
+      "step": 755
+    },
+    {
+      "epoch": 0.9242053789731052,
+      "grad_norm": 0.5456767082214355,
+      "learning_rate": 0.001,
+      "loss": 2.1751,
+      "step": 756
+    },
+    {
+      "epoch": 0.9254278728606357,
+      "grad_norm": 0.6924658417701721,
+      "learning_rate": 0.001,
+      "loss": 2.1324,
+      "step": 757
+    },
+    {
+      "epoch": 0.9266503667481663,
+      "grad_norm": 0.5632027983665466,
+      "learning_rate": 0.001,
+      "loss": 2.1198,
+      "step": 758
+    },
+    {
+      "epoch": 0.9278728606356969,
+      "grad_norm": 0.6390364170074463,
+      "learning_rate": 0.001,
+      "loss": 2.261,
+      "step": 759
+    },
+    {
+      "epoch": 0.9290953545232273,
+      "grad_norm": 0.7371255159378052,
+      "learning_rate": 0.001,
+      "loss": 2.2915,
+      "step": 760
+    },
+    {
+      "epoch": 0.9303178484107579,
+      "grad_norm": 0.665230393409729,
+      "learning_rate": 0.001,
+      "loss": 2.1735,
+      "step": 761
+    },
+    {
+      "epoch": 0.9315403422982885,
+      "grad_norm": 0.5170220732688904,
+      "learning_rate": 0.001,
+      "loss": 2.251,
+      "step": 762
+    },
+    {
+      "epoch": 0.9327628361858191,
+      "grad_norm": 0.6142858266830444,
+      "learning_rate": 0.001,
+      "loss": 2.2155,
+      "step": 763
+    },
+    {
+      "epoch": 0.9339853300733496,
+      "grad_norm": 0.5615746974945068,
+      "learning_rate": 0.001,
+      "loss": 2.1899,
+      "step": 764
+    },
+    {
+      "epoch": 0.9352078239608802,
+      "grad_norm": 0.6055252552032471,
+      "learning_rate": 0.001,
+      "loss": 2.2205,
+      "step": 765
+    },
+    {
+      "epoch": 0.9364303178484108,
+      "grad_norm": 0.5220616459846497,
+      "learning_rate": 0.001,
+      "loss": 2.2125,
+      "step": 766
+    },
+    {
+      "epoch": 0.9376528117359413,
+      "grad_norm": 0.665969729423523,
+      "learning_rate": 0.001,
+      "loss": 2.2579,
+      "step": 767
+    },
+    {
+      "epoch": 0.9388753056234719,
+      "grad_norm": 0.8221463561058044,
+      "learning_rate": 0.001,
+      "loss": 2.2587,
+      "step": 768
+    },
+    {
+      "epoch": 0.9400977995110025,
+      "grad_norm": 0.5516085624694824,
+      "learning_rate": 0.001,
+      "loss": 2.1706,
+      "step": 769
+    },
+    {
+      "epoch": 0.941320293398533,
+      "grad_norm": 0.6319237351417542,
+      "learning_rate": 0.001,
+      "loss": 2.1506,
+      "step": 770
+    },
+    {
+      "epoch": 0.9425427872860636,
+      "grad_norm": 0.6836983561515808,
+      "learning_rate": 0.001,
+      "loss": 2.172,
+      "step": 771
+    },
+    {
+      "epoch": 0.9437652811735942,
+      "grad_norm": 0.5751057863235474,
+      "learning_rate": 0.001,
+      "loss": 2.1617,
+      "step": 772
+    },
+    {
+      "epoch": 0.9449877750611247,
+      "grad_norm": 0.7755267024040222,
+      "learning_rate": 0.001,
+      "loss": 2.1039,
+      "step": 773
+    },
+    {
+      "epoch": 0.9462102689486552,
+      "grad_norm": 0.5571411848068237,
+      "learning_rate": 0.001,
+      "loss": 2.1565,
+      "step": 774
+    },
+    {
+      "epoch": 0.9474327628361858,
+      "grad_norm": 0.5482550263404846,
+      "learning_rate": 0.001,
+      "loss": 2.2251,
+      "step": 775
+    },
+    {
+      "epoch": 0.9486552567237164,
+      "grad_norm": 0.5580353140830994,
+      "learning_rate": 0.001,
+      "loss": 2.2191,
+      "step": 776
+    },
+    {
+      "epoch": 0.9498777506112469,
+      "grad_norm": 0.5179866552352905,
+      "learning_rate": 0.001,
+      "loss": 2.1786,
+      "step": 777
+    },
+    {
+      "epoch": 0.9511002444987775,
+      "grad_norm": 0.6428946256637573,
+      "learning_rate": 0.001,
+      "loss": 2.2259,
+      "step": 778
+    },
+    {
+      "epoch": 0.9523227383863081,
+      "grad_norm": 0.5078992247581482,
+      "learning_rate": 0.001,
+      "loss": 2.1951,
+      "step": 779
+    },
+    {
+      "epoch": 0.9535452322738386,
+      "grad_norm": 0.6398012638092041,
+      "learning_rate": 0.001,
+      "loss": 2.1634,
+      "step": 780
+    },
+    {
+      "epoch": 0.9547677261613692,
+      "grad_norm": 0.6948238611221313,
+      "learning_rate": 0.001,
+      "loss": 2.2215,
+      "step": 781
+    },
+    {
+      "epoch": 0.9559902200488998,
+      "grad_norm": 0.6511239409446716,
+      "learning_rate": 0.001,
+      "loss": 2.305,
+      "step": 782
+    },
+    {
+      "epoch": 0.9572127139364304,
+      "grad_norm": 0.5249418020248413,
+      "learning_rate": 0.001,
+      "loss": 2.1631,
+      "step": 783
+    },
+    {
+      "epoch": 0.9584352078239609,
+      "grad_norm": 0.7171691656112671,
+      "learning_rate": 0.001,
+      "loss": 2.2198,
+      "step": 784
+    },
+    {
+      "epoch": 0.9596577017114915,
+      "grad_norm": 0.637071967124939,
+      "learning_rate": 0.001,
+      "loss": 2.2273,
+      "step": 785
+    },
+    {
+      "epoch": 0.960880195599022,
+      "grad_norm": 0.6645258069038391,
+      "learning_rate": 0.001,
+      "loss": 2.2887,
+      "step": 786
+    },
+    {
+      "epoch": 0.9621026894865525,
+      "grad_norm": 0.7602251172065735,
+      "learning_rate": 0.001,
+      "loss": 2.2839,
+      "step": 787
+    },
+    {
+      "epoch": 0.9633251833740831,
+      "grad_norm": 0.5988993644714355,
+      "learning_rate": 0.001,
+      "loss": 2.291,
+      "step": 788
+    },
+    {
+      "epoch": 0.9645476772616137,
+      "grad_norm": 0.6392702460289001,
+      "learning_rate": 0.001,
+      "loss": 2.1837,
+      "step": 789
+    },
+    {
+      "epoch": 0.9657701711491442,
+      "grad_norm": 0.5164241790771484,
+      "learning_rate": 0.001,
+      "loss": 2.1945,
+      "step": 790
+    },
+    {
+      "epoch": 0.9669926650366748,
+      "grad_norm": 0.6533142924308777,
+      "learning_rate": 0.001,
+      "loss": 2.258,
+      "step": 791
+    },
+    {
+      "epoch": 0.9682151589242054,
+      "grad_norm": 0.5421776175498962,
+      "learning_rate": 0.001,
+      "loss": 2.2432,
+      "step": 792
+    },
+    {
+      "epoch": 0.969437652811736,
+      "grad_norm": 0.9301859140396118,
+      "learning_rate": 0.001,
+      "loss": 2.175,
+      "step": 793
+    },
+    {
+      "epoch": 0.9706601466992665,
+      "grad_norm": 0.5094099640846252,
+      "learning_rate": 0.001,
+      "loss": 2.0958,
+      "step": 794
+    },
+    {
+      "epoch": 0.9718826405867971,
+      "grad_norm": 0.6982297897338867,
+      "learning_rate": 0.001,
+      "loss": 2.2227,
+      "step": 795
+    },
+    {
+      "epoch": 0.9731051344743277,
+      "grad_norm": 0.6126400232315063,
+      "learning_rate": 0.001,
+      "loss": 2.0977,
+      "step": 796
+    },
+    {
+      "epoch": 0.9743276283618582,
+      "grad_norm": 0.572688102722168,
+      "learning_rate": 0.001,
+      "loss": 2.1711,
+      "step": 797
+    },
+    {
+      "epoch": 0.9755501222493888,
+      "grad_norm": 0.5334901213645935,
+      "learning_rate": 0.001,
+      "loss": 2.177,
+      "step": 798
+    },
+    {
+      "epoch": 0.9767726161369193,
+      "grad_norm": 0.6569558382034302,
+      "learning_rate": 0.001,
+      "loss": 2.0747,
+      "step": 799
+    },
+    {
+      "epoch": 0.9779951100244498,
+      "grad_norm": 0.5903536677360535,
+      "learning_rate": 0.001,
+      "loss": 2.2039,
+      "step": 800
+    },
+    {
+      "epoch": 0.9792176039119804,
+      "grad_norm": 0.695564329624176,
+      "learning_rate": 0.001,
+      "loss": 2.1656,
+      "step": 801
+    },
+    {
+      "epoch": 0.980440097799511,
+      "grad_norm": 0.6315183043479919,
+      "learning_rate": 0.001,
+      "loss": 2.2292,
+      "step": 802
+    },
+    {
+      "epoch": 0.9816625916870416,
+      "grad_norm": 0.7556746006011963,
+      "learning_rate": 0.001,
+      "loss": 2.2316,
+      "step": 803
+    },
+    {
+      "epoch": 0.9828850855745721,
+      "grad_norm": 0.8313256502151489,
+      "learning_rate": 0.001,
+      "loss": 2.1614,
+      "step": 804
+    },
+    {
+      "epoch": 0.9841075794621027,
+      "grad_norm": 0.6595815420150757,
+      "learning_rate": 0.001,
+      "loss": 2.139,
+      "step": 805
+    },
+    {
+      "epoch": 0.9853300733496333,
+      "grad_norm": 0.6900365352630615,
+      "learning_rate": 0.001,
+      "loss": 2.1342,
+      "step": 806
+    },
+    {
+      "epoch": 0.9865525672371638,
+      "grad_norm": 0.9232804775238037,
+      "learning_rate": 0.001,
+      "loss": 2.2395,
+      "step": 807
+    },
+    {
+      "epoch": 0.9877750611246944,
+      "grad_norm": 0.6544688940048218,
+      "learning_rate": 0.001,
+      "loss": 2.1565,
+      "step": 808
+    },
+    {
+      "epoch": 0.988997555012225,
+      "grad_norm": 0.6720030903816223,
+      "learning_rate": 0.001,
+      "loss": 2.2108,
+      "step": 809
+    },
+    {
+      "epoch": 0.9902200488997555,
+      "grad_norm": 0.653714120388031,
+      "learning_rate": 0.001,
+      "loss": 2.121,
+      "step": 810
+    },
+    {
+      "epoch": 0.991442542787286,
+      "grad_norm": 0.5552472472190857,
+      "learning_rate": 0.001,
+      "loss": 2.1173,
+      "step": 811
+    },
+    {
+      "epoch": 0.9926650366748166,
+      "grad_norm": 0.5896117091178894,
+      "learning_rate": 0.001,
+      "loss": 2.2343,
+      "step": 812
+    },
+    {
+      "epoch": 0.9938875305623472,
+      "grad_norm": 0.5029270052909851,
+      "learning_rate": 0.001,
+      "loss": 2.1824,
+      "step": 813
+    },
+    {
+      "epoch": 0.9951100244498777,
+      "grad_norm": 0.6343913674354553,
+      "learning_rate": 0.001,
+      "loss": 2.1458,
+      "step": 814
+    },
+    {
+      "epoch": 0.9963325183374083,
+      "grad_norm": 0.5997858047485352,
+      "learning_rate": 0.001,
+      "loss": 2.1901,
+      "step": 815
+    },
+    {
+      "epoch": 0.9975550122249389,
+      "grad_norm": 0.5781143307685852,
+      "learning_rate": 0.001,
+      "loss": 2.1817,
+      "step": 816
+    },
+    {
+      "epoch": 0.9987775061124694,
+      "grad_norm": 0.6393863558769226,
+      "learning_rate": 0.001,
+      "loss": 2.1763,
+      "step": 817
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.7026780247688293,
+      "learning_rate": 0.001,
+      "loss": 2.2279,
+      "step": 818
+    },
+    {
+      "epoch": 1.0,
+      "step": 818,
+      "total_flos": 3.557919619724083e+17,
+      "train_loss": 2.2239257339160425,
+      "train_runtime": 2338.1817,
+      "train_samples_per_second": 44.756,
+      "train_steps_per_second": 0.35
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 818,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 205,
+  "stateful_callbacks": {
+    "EarlyStoppingCallback": {
+      "args": {
+        "early_stopping_patience": 5,
+        "early_stopping_threshold": 0.0
+      },
+      "attributes": {
+        "early_stopping_patience_counter": 0
+      }
+    },
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.557919619724083e+17,
+  "train_batch_size": 128,
+  "trial_name": null,
+  "trial_params": null
+}